An artificial intelligence-based coastal building environment adaptability evaluation system

Abstract

The present application relates to the technical field of building state evaluation, in particular to a coastal building environment adaptability evaluation system based on artificial intelligence, the present application extracts features from multi-dimensional pretreated data and historical disease text data according to data types through a feature extraction module to obtain multi-dimensional basic features; multi-factor collaborative processing is performed on the multi-dimensional basic features to obtain coupled features; recursive feature elimination is used to eliminate features from the multi-dimensional basic features and the coupled features to obtain a core feature matrix; and the core feature matrix is analyzed and processed through a preset environment adaptability evaluation model to obtain a building environment adaptability evaluation report, thereby improving the accuracy of coastal building environment adaptability evaluation, and solving the problem that traditional evaluation methods mostly use single-factor analysis and are difficult to quantify the cumulative damage of multi-field coupling.

Description

Technical Field

[0001] This invention relates to the field of building condition assessment technology, and more specifically to an artificial intelligence-based coastal building environmental adaptability assessment system. Background Technology

[0002] Coastal buildings are situated in a complex environment formed by the ocean, atmosphere, and geology, characterized by strong corrosivity, high dynamism, and sudden disasters. Their environmental adaptability assessment faces multiple bottlenecks, primarily including the following issues: First, the environmental mechanisms are complex and highly coupled. The effects of coastal environments such as salt spray corrosion, typhoon loads, and storm surges overlap, creating nonlinear damage effects. For example, salt spray-induced steel corrosion reduces structural bearing capacity, making buildings more susceptible to instability under typhoon conditions. Traditional assessment methods often employ single-factor analysis, making it difficult to quantify the cumulative damage from multi-field coupling. Simultaneously, long-term changes such as sea-level rise and coastline shifts further increase the spatiotemporal complexity of the assessment.

[0003] Second, there are inherent deficiencies in data acquisition and processing for the environmental adaptability assessment of coastal buildings. The assessment data coverage is low, and monitoring of key components of coastal buildings (such as splash zones and foundation piles) is difficult, with reliance on manual inspection leading to numerous data gaps. Furthermore, the correlation is low, with environmental monitoring (salinity, wind speed), structural monitoring (stress, displacement), and material monitoring (corrosion rate) data stored in a scattered manner, lacking a spatiotemporal alignment mechanism. Third, utilization is low, as unstructured data such as historical damage data and maintenance records have not been effectively mined. In addition, the scarcity of extreme disaster events (such as super typhoons) results in insufficient generalization ability of the assessment model.

[0004] Third, traditional assessment methods lack adaptability. Existing empirical formulas, such as corrosion rate calculation formulas, are difficult to adapt to regional differences; numerical simulation methods are time-consuming and cannot meet the needs of real-time assessment; expert scoring methods are highly subjective and consistency is difficult to guarantee. At the same time, assessment results are mostly static conclusions, lacking dynamic early warning and decision support capabilities.

[0005] Therefore, the present invention provides an artificial intelligence-based coastal building environment adaptability assessment system to solve the above problems. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, this invention provides an artificial intelligence-based coastal building environmental adaptability assessment system to solve the problem that traditional assessment methods often use single-factor analysis and are difficult to quantify the cumulative damage of multi-field coupling.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An artificial intelligence-based coastal building environment adaptability assessment system includes: The data acquisition module acquires multi-dimensional monitoring data and historical disease text data; The data processing module uses edge nodes to preprocess multi-dimensional monitoring data to obtain multi-dimensional preprocessed data; based on a preset three-dimensional building model, it performs spatial mapping and temporal alignment on the multi-dimensional preprocessed data to obtain spatiotemporally aligned multi-dimensional preprocessed data. The feature extraction module extracts features from multidimensional preprocessed data and historical disease text data according to data type to obtain multidimensional basic features; it performs multi-factor collaborative processing on the multidimensional basic features to obtain coupled features; and it uses a recursive feature elimination method to remove features from the multidimensional basic features and coupled features to obtain the core feature matrix. The core feature matrix is ​​analyzed and processed using a pre-set environmental adaptability assessment model to obtain a building environmental adaptability assessment report that includes building status assessment results, environmental adaptability assessment results, and maintenance decisions.

[0008] Preferably, the step of using edge nodes to preprocess multi-dimensional monitoring data to obtain multi-dimensional preprocessed data includes: using the DBSCAN clustering algorithm to process outliers in spatially analyzed sensor data, and using the Isolation Forest algorithm to process outliers in time-series environmental data to remove outliers; when the missing time of a data item is less than a first preset time, using the K-nearest neighbor algorithm to analyze and select similar monitoring data to supplement the data; when the missing time of a data item is greater than or equal to the first preset time, using a preset Transformer prediction model to analyze historical monitoring data and use the predicted data as supplementary data.

[0009] Preferably, the step of extracting features from multidimensional preprocessed data and historical disease text data according to data type to obtain multidimensional basic features includes: extracting multi-frequency features from time-series data to obtain multi-frequency time-series features; using the CLAHE algorithm to enhance image data, extracting image features using a pre-trained ResNet50 model, and compressing image features through PCA dimensionality reduction to obtain key visual features; using the BERT model to perform NER recognition on historical disease text data to extract key entity features, mapping key entity features to key numerical features, and concatenating key numerical features with sensor data in multidimensional preprocessed data to obtain disease-related features.

[0010] Preferably, the multi-factor collaborative processing of multi-dimensional basic features to obtain coupled features includes: calculating the mutual information value between multi-frequency time-series features, key visual features, and disease-related features using mutual information analysis; selecting basic features corresponding to highly correlated factor pairs with mutual information values ​​greater than preset information values ​​to obtain factor-related features; performing coupling processing on the factor-related features according to preset coupling feature construction rules to obtain initial coupled features; and using the XGBoost model to rank the feature importance of the initial coupled features and filtering them based on preset conditions to obtain coupled features.

[0011] Preferably, the formula for calculating the mutual information value includes: Mutual information values ​​of discrete basic features The calculation formula is: , in, For a discrete basic feature in a multidimensional basic feature set, This is another discrete basic feature among multidimensional basic features. for The corresponding value, for The corresponding value, for and Joint probability quality at time, for Marginal probability quality at time, for Marginal probability quality at that time; Mutual information value of continuous basic features The calculation formula is: , in, For example, a continuous basic feature in a multidimensional set of basic features. It is another continuous basic feature among multidimensional basic features. for The corresponding value, for The corresponding value, for and Joint probability density at time, for Marginal probability density at time, for Marginal probability density at time.

[0012] Preferably, the process of coupling the factor correlation features according to the preset coupling feature construction rules to obtain the initial coupling features includes: performing product-type coupling on salinity features and humidity features based on building damage mechanisms to obtain product-type features; performing power-type coupling on wind speed features and building height features based on fluid mechanics and structural mechanics laws to obtain power-type features; performing product processing on the average salinity value of the first preset time period and the peak humidity value of the second preset time period to obtain time-series features; and integrating the product-type features, power-type features, and time-series features to obtain the initial coupling features.

[0013] Preferably, the step of analyzing and processing the core feature matrix using a preset environmental adaptability assessment model to obtain a building environmental adaptability assessment report that includes building condition assessment results, environmental adaptability assessment results, and maintenance decisions includes: analyzing the core feature matrix using a steel reinforcement corrosion depth prediction model to obtain corrosion depth prediction results; analyzing the core feature matrix using a typhoon load-based structural vertex displacement prediction model to obtain structural vertex displacement time-series prediction results; using a multi-factor assessment model to assess the durability, disaster resistance, and ecological properties of the core feature matrix to obtain multi-factor assessment results; and integrating the corrosion depth prediction results, structural vertex displacement time-series prediction results, and multi-factor assessment results to obtain a building environmental adaptability assessment report.

[0014] Preferably, the step of using a steel reinforcement corrosion depth prediction model to analyze the core feature matrix and obtain corrosion depth prediction results includes: obtaining core temporal features in the core feature matrix, using LSTM to capture the long-term dependencies of the core temporal features to obtain initial prediction results; and filtering the initial prediction results based on physical constraints to obtain corrosion depth prediction results.

[0015] Preferably, the method of using a typhoon-loaded structural vertex displacement prediction model to analyze the core feature matrix and obtain the temporal prediction result of the structural vertex displacement includes: obtaining cloud map features, wind speed features, and displacement features from the core feature matrix; using a CNN layer to extract the spatial structure features of the cloud map features; using an LSTM layer to capture the temporal variation features of the wind speed features; and performing a fusion correlation analysis on the spatial structure features, temporal variation features, and displacement features to obtain the temporal prediction result of the structural vertex displacement.

[0016] Preferably, the method of using a multi-factor evaluation model to assess the durability, disaster resistance, and ecological properties of the core feature matrix to obtain multi-factor evaluation results includes: acquiring the core features of each indicator; mapping each indicator to a first interval using the efficacy coefficient method to obtain a quantitative score for each indicator; assigning weights to each indicator using a combined weighting method; analyzing the correlation between indicators based on the core features, the quantitative scores and weights of each indicator using the Pearson correlation coefficient to construct an indicator correlation graph; and analyzing the indicator correlation graph using a graph neural network model to capture the synergistic effects between indicators and output multi-factor evaluation results containing each indicator.

[0017] The beneficial effects of this invention are as follows: 1. This invention extracts features from multidimensional preprocessed data and historical disease text data according to data type using a feature extraction module to obtain multidimensional basic features; performs multi-factor collaborative processing on the multidimensional basic features to obtain coupled features; employs a recursive feature elimination method to remove features from the multidimensional basic features and coupled features to obtain a core feature matrix; and analyzes and processes the core feature matrix through a preset environmental adaptability assessment model to obtain a building environmental adaptability assessment report. This improves the accuracy of building environmental adaptability assessment in coastal areas and solves the problem that traditional assessment methods often use single-factor analysis, making it difficult to quantify the cumulative damage of multi-field coupling.

[0018] 2. Due to inherent defects in data acquisition and processing for the environmental adaptability assessment of coastal buildings, the assessment data coverage is low, and the monitoring of key parts of coastal buildings (such as splash zones and foundation piles) is difficult. Reliance on manual inspection leads to many data gaps. Through the data processing module of this application, this invention can process outliers according to data type, fill in missing data, and convert the original data into usable samples, laying the foundation for subsequent data analysis.

[0019] 3. The multi-factor evaluation model of this invention solves the problems of traditional weighted summation method and single classification model in handling mutual influence between indicators and weight bias by using complex correlation of multiple indicators, fusion of subjective and objective weights and graph neural network. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of an artificial intelligence-based coastal building environmental adaptability assessment system according to the present invention. Detailed Implementation

[0021] The following will refer to the attached reference. Figure 1 The various embodiments of the present invention will be described in detail below. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0022] An AI-based coastal building environmental adaptability assessment system, as shown in the attached document. Figure 1 As shown, it includes: a data acquisition module, which acquires multi-dimensional monitoring data and historical disease text data; the multi-dimensional monitoring data includes stress / strain data collected by grating fiber sensors deployed in the foundation, vibration data collected by piezoelectric sensors, corrosion data represented by chloride ion concentration collected by corrosion probes, and images of surface defects such as cracks and spalling obtained by weekly inspections by drones equipped with hyperspectral cameras, which are then combined with LiDAR to generate three-dimensional point cloud data; and weather data such as typhoon, salinity, and tides obtained by connecting to a meteorological platform.

[0023] The data processing module uses edge nodes to preprocess multi-dimensional monitoring data to obtain multi-dimensional preprocessed data; based on a preset three-dimensional building model, it performs spatial mapping and temporal alignment on the multi-dimensional preprocessed data to obtain spatiotemporally aligned multi-dimensional preprocessed data.

[0024] Edge nodes are used to preprocess multi-dimensional monitoring data to obtain multi-dimensional preprocessed data; edge computing devices are deployed at the monitoring site to complete data cleaning, outlier removal, and missing value supplementation to obtain multi-dimensional preprocessed data; and then the data is transmitted to the cloud to reduce the bandwidth consumption of data transmission.

[0025] Based on a pre-set 3D building model, spatial mapping and temporal alignment are performed on the multidimensional preprocessed data to obtain spatiotemporally aligned multidimensional preprocessed data.

[0026] Specifically, based on a pre-set 3D building model, each sensor and drone's shooting area is assigned a component ID to ensure a one-to-one correspondence between the data and the physical location in the 3D building model. Furthermore, data such as structural stress and salt spray concentration are synchronized using meteorological data timestamps as a benchmark.

[0027] The feature extraction module extracts features from multidimensional preprocessed data and historical disease text data according to data type to obtain multidimensional basic features; it then performs multi-factor collaborative processing on the multidimensional basic features to obtain coupled features; finally, it uses a recursive feature elimination method to remove features from the multidimensional basic features and coupled features to obtain the core feature matrix.

[0028] Among them, the multidimensional basic features include multi-frequency temporal features, key visual features, and disease-related features.

[0029] A recursive feature elimination method is used to remove multidimensional basic features and coupled features to obtain a core feature matrix. Specifically, redundant features are gradually removed using the structural damage level as a label, and the 10 features with the lowest importance are removed. Then, the 5 features in the remaining 30 dimensions that have a correlation with the target label that is less than a preset correlation threshold are removed, resulting in a core feature matrix formed by 25 features.

[0030] Through the feature extraction module, this invention can provide key feature data for subsequent steps, which helps to solve the problems of high dimensionality, weak correlation and strong coupling of multidimensional preprocessed data, and can reduce the amount of computation in subsequent steps.

[0031] The comprehensive assessment module uses a pre-set environmental adaptability assessment model to analyze and process the core feature matrix, resulting in a building environmental adaptability assessment report that includes building status assessment results, environmental adaptability assessment results, and maintenance decisions.

[0032] This invention extracts features from multidimensional preprocessed data and historical disease text data according to data type using a feature extraction module to obtain multidimensional basic features; it then performs multi-factor collaborative processing on the multidimensional basic features to obtain coupled features; a recursive feature elimination method is used to remove features from the multidimensional basic features and coupled features to obtain a core feature matrix; and finally, a pre-set environmental adaptability assessment model is used to analyze and process the core feature matrix to obtain a building environmental adaptability assessment report. This improves the accuracy of building environmental adaptability assessment in coastal areas and solves the problem that traditional assessment methods often use single-factor analysis, making it difficult to quantify the cumulative damage of multi-field coupling.

[0033] In one embodiment of the present invention, the preprocessing of multi-dimensional monitoring data using edge nodes to obtain multi-dimensional preprocessed data includes: using the DBSCAN clustering algorithm to process outliers in spatially analyzed sensor data, and using the isolated forest algorithm to process outliers in time-series environmental data to remove outliers; when the missing time of a data item is less than a first preset time, using the K-nearest neighbor algorithm to analyze and select monitoring data of the same type to supplement the data; when the missing time of a data item is greater than or equal to the first preset time, using a preset Transformer prediction model to analyze historical monitoring data to use the predicted data as supplementary data.

[0034] Specifically, the DBSCAN clustering algorithm is used to process outliers in the sensor data of spatial analysis. For example, for strain sensors on different floors of a building, points that deviate from the cluster are automatically marked as outliers based on the preset sensor locations and differences in monitoring values. Combined with physical constraints such as the ultimate stress of concrete, it is determined whether the problem is a faulty equipment or a real structural anomaly.

[0035] The Isolation Forest algorithm is used to handle outliers in time-series environmental data, such as salinity and wind speed data. The Isolation Forest algorithm quickly isolates outliers by randomly splitting the data.

[0036] For short-term missing data, such as data missing within 6 hours, based on the spatiotemporal correlation of coastal data, such as the consistent salinity change trends of adjacent sensors and similar historical wind speed patterns, the K-nearest neighbor algorithm is used to find the most similar spatiotemporal samples, and the missing values ​​are filled in with weighted averages to reduce errors.

[0037] For long-term missing data, such as environmental data with strong periodicity like tides and temperature exceeding 6 hours, the Transformer's self-attention mechanism captures long-period dependencies (such as bi-monthly tidal cycles and seasonal salinity changes) and predicts values ​​for the missing periods.

[0038] Due to inherent deficiencies in data acquisition and processing for the environmental adaptability assessment of coastal buildings, the assessment data coverage is low, and the monitoring of key components of coastal buildings (such as splash zones and foundation piles) is difficult. Reliance on manual inspection leads to numerous data gaps. Through the data processing module of this application, this invention can process outliers according to data type, fill in missing data, and convert the original data into usable samples, laying the foundation for subsequent data analysis.

[0039] In one embodiment of the present invention, the step of extracting features from multidimensional preprocessed data and historical disease text data according to data type to obtain multidimensional basic features includes: extracting multi-frequency features from time-series data to obtain multi-frequency time-series features; using the CLAHE algorithm to enhance image data, extracting image features using a pre-trained ResNet50 model, and compressing image features through PCA dimensionality reduction to obtain key visual features; using the BERT model to perform NER recognition on historical disease text data to extract key entity features, mapping key entity features to key numerical features, and concatenating key numerical features and sensor data from multidimensional preprocessed data to obtain disease-related features.

[0040] Preferably, the multidimensional basic features include multi-frequency temporal features, key visual features, and disease-related features.

[0041] Specifically, the process of acquiring multidimensional basic features includes: extracting multi-frequency features from the time-series data in the multidimensional preprocessed data to obtain multi-frequency time-series features; and extracting multi-domain features from the time-series data in the multidimensional preprocessed data, such as stress, wind speed, and salinity. Taking weekly stress data as an example, features are extracted in the time domain, frequency domain, and time-frequency domain combined. The mean, peak factor (ratio of peak value to mean), and trend slope of the stress for each week are calculated to obtain the stress time-domain features; the stress signal is decomposed into five scales using wavelet transform, and the energy proportion of each scale is extracted to obtain the stress frequency-domain features; and the time difference between the peak wind speed and the peak stress is calculated to obtain the stress joint features.

[0042] Visual features are extracted from the image data in the multidimensional preprocessed data to obtain key visual features, including: using the CLAHE algorithm to enhance image contrast and remove grayscale unevenness caused by salt stains; then using a pre-trained ResNet50 model to extract 2048-dimensional image features, and then using PCA to reduce the dimensions to 50 dimensions, focusing on retaining features such as crack length, width, and number of branches to obtain key visual features.

[0043] The analysis and feature mapping of historical disease text data yields disease-related features, including: using the BERT model to perform NER identification on historical disease text data, extracting key entity features, mapping key entity features to key numerical features, and splicing key numerical features and sensor data from multidimensional preprocessed data to obtain disease-related features.

[0044] Through the configuration method of this embodiment, the present invention can extract features from multidimensional preprocessed data and historical disease text data according to data type, obtain multidimensional basic features, and provide a high-quality data foundation for subsequent analysis steps.

[0045] In one embodiment of the present invention, the multi-factor collaborative processing of multi-dimensional basic features to obtain coupled features includes: calculating the mutual information value between multi-frequency time-series features, key visual features, and disease-related features using a mutual information analysis method; selecting basic features corresponding to highly correlated factor pairs with mutual information values ​​greater than a preset information value to obtain factor-related features; performing coupling processing on the factor-related features according to preset coupling feature construction rules to obtain initial coupled features; and using an XGBoost model to rank the feature importance of the initial coupled features and filtering them based on preset conditions to obtain coupled features.

[0046] Preferably, the preset information value is 0.3. The larger the mutual information value, the more information is shared between the two features.

[0047] The XGBoost model is used to rank the initial coupled features by importance and then filter them based on preset conditions to obtain coupled features. The process is as follows: XGBoost is used to handle the nonlinear correlation between features, and the feature importance calculated through gain directly reflects the feature's contribution to reducing model error. The initial coupled features are then ranked from highest to lowest importance. Next, coupled features are filtered based on preset conditions. Specifically, the preset conditions are: 1. Whether the importance of the coupled feature is higher than that of its constituent individual features. If it is higher, it indicates that coupling has indeed enhanced the correlation; if it is lower, it indicates that coupling has no gain and is redundant, so it is discarded. 2. Features whose importance ranks among the top 10 of the total features. 3. Adjusting the selection of coupled features based on the regional characteristics of the project. For example, in northern projects, the average number of freeze-thaw cycles in the past three months multiplied by the lowest temperature of the current month, even if ranked 11th, can be retained because freeze-thaw damage is a core issue in the north.

[0048] Through the coupling method of this embodiment, the present invention can solve the problem of complex and highly coupled environmental action mechanisms, such as the superposition of salt spray corrosion, typhoon load, storm surge intrusion, etc. in coastal environments, forming nonlinear damage effects. It deeply analyzes the coupling effect and further solves the problem that traditional assessment methods mostly use single-factor analysis, which makes it difficult to quantify the cumulative damage of multi-field coupling.

[0049] In one embodiment of the present invention, the formula for calculating the mutual information value includes: Mutual information values ​​of discrete basic features The calculation formula is: , in, For a discrete basic feature in a multidimensional basic feature set, This is another discrete basic feature among multidimensional basic features. for The corresponding value, for The corresponding value, for and Joint probability quality at time, for Marginal probability quality at time, for Marginal probability quality at that time.

[0050] Mutual information value of continuous basic features The calculation formula is: , in, For example, a continuous basic feature in a multidimensional set of basic features. It is another continuous basic feature among multidimensional basic features. for The corresponding value, for The corresponding value, for and Joint probability density at time, for Marginal probability density at time, for Marginal probability density at time.

[0051] Through the configuration method of this embodiment, the present invention can identify nonlinear correlations between environmental and structural features, such as the coupling effect of salinity and humidity; and by screening factor correlation features with mutual information values ​​greater than preset information values, it can avoid spurious correlations, so as to provide high-quality factor correlation features for subsequent construction of coupling features and improve the accuracy of subsequent analysis and evaluation results.

[0052] In one embodiment of the present invention, the step of coupling the factor-related features according to the preset coupling feature construction rules to obtain the initial coupling features includes: performing product-type coupling on salinity features and humidity features based on building damage mechanisms to obtain product-type features; performing power-type coupling on wind speed features and building height features based on fluid mechanics and structural mechanics laws to obtain power-type features; performing product processing on the average salinity value of a first preset time period and the peak humidity value of a second preset time period to obtain time-series features; and integrating the product-type features, power-type features, and time-series features to obtain the initial coupling features.

[0053] Specifically, based on the building damage mechanism, a product-type coupling of salinity and humidity characteristics is performed to obtain the principle of the product-type characteristics: the essence of steel corrosion in coastal buildings is an electrochemical reaction, the rate of which is jointly determined by the electrolyte concentration (salinity) and the humidity of the reaction environment, and the two have a synergistic accelerating effect rather than a simple additive effect. Salinity (S): provides electrolytes such as Cl⁻, which destroys the passivation film on the surface of the steel reinforcement. The higher the salinity, the higher the electrolyte concentration, and the more abundant the raw materials for the corrosion reaction. Humidity (H): determines the ion migration rate. When the humidity is below 60%, the inside of the concrete is dry, ions are difficult to move, and corrosion almost stops. When the humidity is above 80%, the pores of the concrete are filled with water, ions migrate rapidly, and the corrosion rate increases sharply. When salinity > 3.0% and humidity > 80%, the corrosion rate is not a simple sum of the contributions from salinity and humidity, but rather a product of the two. For example, the synergistic effect is significant: for every 0.5% increase in salinity and every 10% increase in humidity, the corrosion rate increases by approximately (0.5% / 3.0%) × (10% / 80%) × 100% = 20.8%, which is much higher than the additive effect (0.5% + 10% = 10.5%).

[0054] Based on the laws of fluid mechanics and structural mechanics, a power-law coupling of wind speed characteristics and building height characteristics is performed. The principle of the power-law characteristic is as follows: the typhoon load on coastal buildings follows the laws of fluid mechanics and structural mechanics. The load magnitude increases nonlinearly with wind speed and building height. The source of the power-law of wind speed is that the essence of wind load is the impact of wind kinetic energy on the building. According to the kinetic energy formula, the kinetic energy of the wind is proportional to the square of the wind speed. At the same time, in the building structural code, the wind pressure height variation coefficient in the wind load standard value calculation formula is related to the height, while the basic wind pressure is directly related to the power-law of wind speed. Under the influence of typhoons, the higher the building, the greater the wind speed in the gradient wind zone. That is, the wind speed is low near the ground due to friction, but the wind speed increases significantly after the building height is >100m. The wind vibration effect of high-rise structures is more obvious, and the cumulative effect of load transfer to the foundation is stronger. For example, when the wind speed increases from 20m / s to 30m / s, the square of the wind speed increases from 400 to 900, an increase of 125%. If the building height increases from 50m to 100m, the product of the square of the wind speed and the building height increases by about 160%, which is highly consistent with the actual monitored displacement increase of 155% of high-rise structures, and is far better than the linear combination of the product of wind speed and building height.

[0055] Based on the long-term cumulative effect and the short-term impact effect, the average salinity of the first preset time period and the peak humidity of the second preset time period are multiplied to obtain the time-series characteristics; wherein, the first preset time period is the most recent 3 months and the second preset time period is the most recent month.

[0056] Corrosion damage in coastal buildings is a combination of long-term cumulative effects and short-term impact effects; characteristics of a single time scale cannot fully reflect the damage process. Steel reinforcement corrosion is a slow process; salinity fluctuations within one month have a limited impact on corrosion depth, while the average salinity over three months reflects the intensity of continuous electrolyte erosion. If the three-month average is >3.5%, it indicates continuous Cl⁻ accumulation inside the concrete, and the passivation film has already suffered localized damage. When long-term accumulation weakens the passivation film, extreme humidity in a given month, such as continuous rainfall from typhoons during the rainy season with peak humidity >95%, becomes a triggering factor, accelerating ion migration and electrochemical reactions, causing a significant increase in corrosion depth in the short term. If the average salinity over the past three months is low (less than 3.0%), corrosion is relatively slow even with a high peak humidity in that month. If the average salinity is high (>3.5%) and the peak is high (>95%), the corrosion depth will show a step-like increase.

[0057] The XGBoost model is then used to rank the initial coupling features by feature importance and to filter them based on preset conditions to obtain the coupling features.

[0058] Through the configuration method of this embodiment, the present invention can provide key feature data for subsequent steps, which helps to solve the problems of high dimensionality, weak correlation and strong coupling of multidimensional preprocessed data, and can reduce the amount of computation in subsequent steps.

[0059] In one embodiment of the present invention, the step of analyzing and processing the core feature matrix using a preset environmental adaptability assessment model to obtain a building environmental adaptability assessment report that includes building status assessment results, environmental adaptability assessment results, and maintenance decisions includes: analyzing the core feature matrix using a steel reinforcement corrosion depth prediction model to obtain corrosion depth prediction results; analyzing the core feature matrix using a typhoon load-based structural vertex displacement prediction model to obtain structural vertex displacement time-series prediction results; using a multi-factor assessment model to assess the durability, disaster resistance, and ecological properties of the core feature matrix to obtain multi-factor assessment results; and integrating the corrosion depth prediction results, structural vertex displacement time-series prediction results, and multi-factor assessment results to obtain a building environmental adaptability assessment report.

[0060] Through the configuration of this embodiment, the present invention provides a comprehensive assessment of the environmental adaptability evaluation system for coastal buildings. By using long-term corrosion prediction, multimodal displacement prediction, and multi-factor comprehensive evaluation, it solves the shortcomings of traditional methods (such as empirical formulas and single models) in terms of accuracy, generalization, and engineering practicality.

[0061] Furthermore, in one embodiment of the present invention, the step of using a steel reinforcement corrosion depth prediction model to analyze the core feature matrix and obtain corrosion depth prediction results includes: obtaining core temporal features in the core feature matrix, using LSTM to capture the long-term dependencies of the core temporal features to obtain initial prediction results; and filtering the initial prediction results based on physical constraints to obtain corrosion depth prediction results.

[0062] Preferably, the core temporal features include salinity, humidity, the product coupling features of salinity and humidity, and the thickness of the concrete protective layer.

[0063] The process of filtering initial prediction results based on physical constraints includes determining the loss function of the steel corrosion depth prediction model. loss function The calculation formula is: in, The MSE loss function is... To predict corrosion depth, This is the actual detection depth. The penalty coefficient is... The thickness of the protective layer is used. A penalty coefficient is applied to ensure the predicted corrosion depth does not exceed the protective layer thickness. If the predicted corrosion depth exceeds the protective layer thickness, an additional 10 times the MSE penalty is applied to optimize the predicted corrosion depth and ensure the accuracy and precision of the prediction results.

[0064] Specifically, this invention uses the gating mechanism of LSTM (input gate, forget gate, output gate) to remember key temporal information in the long term. For example, the peak salinity (3.8%) during the 2021 rainy season is selectively retained by the forget gate and continuously affects the corrosion calculation in subsequent years through the input gate. Ultimately, the accuracy of capturing the temporal correlation of corrosion depth over 5 years is improved to 94%, which is far higher than that of traditional time series models.

[0065] The steel corrosion depth prediction model can accurately identify the cumulative impact of slow variables such as the gradual increase in salinity year by year, thus avoiding maintenance delays caused by ignoring long-term trends.

[0066] The main constraints include: adding a penalty term to the loss function: if the predicted corrosion depth is greater than the protective layer thickness, an additional 10 times MSE loss is added to force the model to learn the physical boundary; adding a threshold cutoff to the output layer: if the final predicted value exceeds the protective layer thickness, it is automatically corrected to the protective layer thickness. Through the dual constraints formed by the loss function and physical rules, the prediction results are ensured to remain within the actual engineering scenario.

[0067] Furthermore, in one embodiment of the present invention, the method of using a typhoon-loaded structural vertex displacement prediction model to analyze the core feature matrix and obtain the temporal prediction result of the structural vertex displacement includes: obtaining cloud map features, wind speed features, and displacement features in the core feature matrix; using a CNN layer to extract the spatial structure features of the cloud map features; using an LSTM layer to capture the temporal variation features of the wind speed features; and performing a fusion correlation analysis on the spatial structure features, temporal variation features, and displacement features to obtain the temporal prediction result of the structural vertex displacement.

[0068] Preferably, the cloud image features are derived from satellite cloud images, the wind speed features are derived from wind speed sensors on the building roof, and the displacement features are derived from displacement monitoring instruments at the building apex. The time-series prediction results of the structural apex displacement are presented in the form of displacement prediction curves and textual descriptions.

[0069] Specifically, the CNN layer extracts spatial structure features from satellite cloud images, such as eye area and cloud top temperature gradient, to quantify the inherent intensity of the typhoon. For example, the clearer the eye, the higher the spatial feature score, and the stronger the corresponding load. The LSTM layer extracts temporal dynamic features of wind speed, such as hourly wind speed growth rate and peak duration, to quantify the typhoon's action. The fusion layer automatically assigns higher weights to the spatial features and long-lasting temporal features of strong typhoons through feature concatenation and attention weight allocation. For example, for a Category 12 typhoon, the spatial feature weight is 0.6, and the temporal feature weight is 0.4. Then, a fully connected layer matches and predicts the concatenated features, their corresponding weights, and displacement features to obtain the temporal prediction results of the structural apex displacement.

[0070] The structural vertex displacement prediction model under typhoon load efficiently integrates spatial and temporal features, solving the problem that traditional single models, such as LSTM using only wind speed time series and CNN using only cloud maps, cannot fully characterize the relationship between typhoon intensity, load, and displacement.

[0071] Furthermore, in one embodiment of the present invention, the method of using a multi-factor evaluation model to assess the durability, disaster resistance, and ecological properties of the core feature matrix to obtain multi-factor evaluation results includes: acquiring the core features of each indicator; mapping each indicator to a first interval using the efficacy coefficient method to obtain a quantitative score for each indicator; assigning weights to each indicator using a combined weighting method; analyzing the correlation between each indicator based on the core features, the quantitative scores and weights of each indicator using the Pearson correlation coefficient to construct an indicator correlation graph; and analyzing the indicator correlation graph using a graph neural network model to capture the synergistic effects between each indicator and outputting multi-factor evaluation results containing each indicator.

[0072] Preferably, the indicators of the multi-factor evaluation model include durability, disaster resistance and ecology. The core characteristics of durability include the depth of steel corrosion and the area of ​​concrete spalling; the core characteristics of disaster resistance include typhoon displacement, storm surge flood resistance height and seismic bearing capacity; and the core characteristics of ecology include the sewage discharge compliance rate.

[0073] In this embodiment, the first interval is 0-100. The combined weighting method for assigning weights to each indicator is as follows: Weights of each indicator are calculated using AHP and entropy methods; based on core features, quantitative scores, and weights of each indicator, Pearson correlation coefficients are used to analyze the correlation between indicators. When constructing the indicator correlation graph, each node contains the indicator value, indicator threshold, and indicator weight; the edge weight between nodes is the product of the Pearson correlation coefficient and the engineering correlation degree between the two indicators. Furthermore, a graph neural network model is used to process the node and edge information of the indicator correlation graph, capturing the synergistic effects between indicators and outputting multi-factor evaluation results for durability, disaster resistance, and ecology.

[0074] The multi-factor evaluation model of this invention solves the problems of traditional weighted summation methods and single classification models being unable to handle the mutual influence between indicators and weight bias by using complex correlations of multiple indicators, fusion of subjective and objective weights, and graph neural networks.

[0075] In one embodiment of the present invention, to address the problem of limited sample sizes for extreme events such as typhoons and storm surges, data is expanded through sample generation and transfer learning to improve the model's generalization ability. The main process includes: A deep convolutional generative adversarial network (GAN) is used to generate adapted typhoon cloud images and storm surge water level images / time series data. Existing extreme samples (such as 500 cloud images of a Category 10 typhoon and 100 storm surge water level time series) are used as training data. The generator processes the 100-dimensional random noise input to output simulated Category 12 typhoon cloud images (64×64) or simulated storm surge water level time series (72 hours) for a once-in-a-century storm. A discriminator is used to distinguish between real samples and generated samples, and the two are trained adversarially until the discriminator accuracy approaches 50%. When the accuracy rate approaches 50%, the generated 1,000 Category 12 typhoon cloud images and 500 simulated storm surge time series data are added to the extreme sample library.

[0076] A wind resistance assessment model for coastal bridges (similar to the wind resistance principle of buildings) is adopted. The bottom layer of the pre-trained model (such as the first 10 layers of CNN and the first 5 layers of LSTM) is frozen to retain the general feature extraction capability. The fully connected layers of the wind resistance assessment model are then fine-tuned to obtain an adjusted wind resistance assessment model. The adjusted wind resistance assessment model is then trained with extreme samples of coastal buildings to adapt to differences in building height and structural form. Wind tunnel test data is used for verification. If the error between the predicted displacement and the experimental value is greater than the set threshold, the proportion of fine-tuned samples is adjusted. When the verification conditions are met, simulated data for transfer learning extension is obtained.

[0077] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0078] It should be noted that in the description of this invention, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0079] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0080] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0081] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0082] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0083] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0084] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based coastal building environmental adaptability assessment system, characterized in that, include: The data acquisition module acquires multi-dimensional monitoring data and historical disease text data; The data processing module uses edge nodes to preprocess multi-dimensional monitoring data to obtain multi-dimensional preprocessed data. Based on a pre-defined 3D building model, spatial mapping and temporal alignment are performed on the multi-dimensional preprocessed data to obtain spatiotemporally aligned multi-dimensional preprocessed data. The feature extraction module extracts features from multidimensional preprocessed data and historical disease text data according to data type to obtain multidimensional basic features; Multi-factor collaborative processing is performed on the multi-dimensional basic features to obtain coupled features; a recursive feature elimination method is used to remove features from the multi-dimensional basic features and coupled features to obtain the core feature matrix; The comprehensive assessment module uses a pre-set environmental adaptability assessment model to analyze and process the core feature matrix, resulting in a building environmental adaptability assessment report that includes building status assessment results, environmental adaptability assessment results, and maintenance decisions.

2. The coastal building environmental adaptability assessment system according to claim 1, characterized in that, The aforementioned preprocessing of multi-dimensional monitoring data using edge nodes to obtain multi-dimensional preprocessed data includes: using the DBSCAN clustering algorithm to process outliers in spatially analyzed sensor data, and using the Isolation Forest algorithm to process outliers in time-series environmental data, in order to remove outlier data; when the missing time of a data item is less than a first preset time, using the K-nearest neighbor algorithm to analyze and select similar monitoring data to supplement the data; when the missing time of a data item is greater than or equal to the first preset time, using a preset Transformer prediction model to analyze historical monitoring data, using the predicted data as supplementary data.

3. The coastal building environmental adaptability assessment system according to claim 1, characterized in that, The aforementioned method involves extracting features from multidimensional preprocessed data and historical disease text data according to data type to obtain multidimensional basic features. This includes: extracting multi-frequency features from time-series data to obtain multi-frequency time-series features; using the CLAHE algorithm to enhance image data and extracting image features using a pre-trained ResNet50 model, as well as compressing image features through PCA dimensionality reduction to obtain key visual features; using the BERT model to perform NER recognition on historical disease text data to extract key entity features, mapping key entity features to key numerical features, and concatenating key numerical features with sensor data from multidimensional preprocessed data to obtain disease-related features.

4. The coastal building environmental adaptability assessment system according to claim 1, characterized in that, The process of performing multi-factor collaborative processing on multi-dimensional basic features to obtain coupled features includes: calculating the mutual information value between multi-frequency time-series features, key visual features, and disease-related features using mutual information analysis; selecting basic features corresponding to highly correlated factor pairs with mutual information values ​​greater than preset information values ​​to obtain factor-related features; performing coupling processing on the factor-related features according to preset coupling feature construction rules to obtain initial coupled features; and using the XGBoost model to rank the feature importance of the initial coupled features and filtering them based on preset conditions to obtain coupled features.

5. The coastal building environmental adaptability assessment system according to claim 4, characterized in that, The formula for calculating the mutual information value includes: Mutual information values ​​of discrete basic features The calculation formula is: ， in, For a discrete basic feature in a multidimensional basic feature set, This is another discrete basic feature among multidimensional basic features. for The corresponding value, for The corresponding value, for and Joint probability quality at time, for Marginal probability quality at time, for Marginal probability quality at that time; Mutual information value of continuous basic features The calculation formula is: ， in, For example, a continuous basic feature in a multidimensional set of basic features. It is another continuous basic feature among multidimensional basic features. for The corresponding value, for The corresponding value, for and Joint probability density at time, for Marginal probability density at time, for Marginal probability density at time.

6. The coastal building environmental adaptability assessment system according to claim 4, characterized in that, The process of coupling factor-related features according to preset coupling feature construction rules to obtain initial coupling features includes: performing product-type coupling on salinity and humidity features based on building damage mechanisms to obtain product-type features; performing power-type coupling on wind speed and building height features based on fluid mechanics and structural mechanics laws to obtain power-type features; performing product processing on the average salinity value of the first preset time period and the peak humidity value of the second preset time period to obtain time-series features; and integrating product-type features, power-type features, and time-series features to obtain initial coupling features.

7. The coastal building environmental adaptability assessment system according to claim 1, characterized in that, The aforementioned method of analyzing and processing the core feature matrix using a pre-set environmental adaptability assessment model to obtain a building environmental adaptability assessment report that includes building condition assessment results, environmental adaptability assessment results, and maintenance decisions includes: analyzing the core feature matrix using a steel reinforcement corrosion depth prediction model to obtain corrosion depth prediction results; analyzing the core feature matrix using a typhoon load-based structural vertex displacement prediction model to obtain structural vertex displacement time-series prediction results; using a multi-factor assessment model to assess the durability, disaster resistance, and ecological properties of the core feature matrix to obtain multi-factor assessment results; and integrating the corrosion depth prediction results, structural vertex displacement time-series prediction results, and multi-factor assessment results to obtain the building environmental adaptability assessment report.

8. The coastal building environmental adaptability assessment system according to claim 7, characterized in that, The method of using a steel reinforcement corrosion depth prediction model to analyze the core feature matrix and obtain corrosion depth prediction results includes: obtaining the core temporal features in the core feature matrix, using LSTM to capture the long-term dependencies of the core temporal features to obtain initial prediction results; and filtering the initial prediction results based on physical constraints to obtain corrosion depth prediction results.

9. The coastal building environmental adaptability assessment system according to claim 7, characterized in that, The method described above uses a typhoon-loaded structural vertex displacement prediction model to analyze the core feature matrix and obtain the temporal prediction results of the structural vertex displacement. This includes: obtaining cloud map features, wind speed features, and displacement features from the core feature matrix; extracting the spatial structure features of the cloud map features using a CNN layer; capturing the temporal variation features of the wind speed features using an LSTM layer; and performing a fusion and correlation analysis on the spatial structure features, temporal variation features, and displacement features to obtain the temporal prediction results of the structural vertex displacement.

10. The coastal building environmental adaptability assessment system according to claim 7, characterized in that, The aforementioned multi-factor evaluation model is used to assess the durability, disaster resistance, and ecological properties of the core feature matrix, yielding multi-factor evaluation results. This includes: acquiring the core features of each indicator; mapping each indicator to a first interval using the efficacy coefficient method to obtain a quantitative score for each indicator; assigning weights to each indicator using a combined weighting method; analyzing the correlation between indicators based on the core features, the quantitative scores and weights of each indicator using Pearson correlation coefficients to construct an indicator correlation graph; and analyzing the indicator correlation graph using a graph neural network model to capture the synergistic effects between indicators, outputting multi-factor evaluation results containing all indicators.

Images