Hidden disaster-causing factor disaster-causing risk evaluation method based on multi-source data fusion

Abstract

The invention discloses a hidden disaster-causing factor disaster-causing risk evaluation method based on multi-source data fusion, and relates to the technical field of geographic information systems. The method comprises the following steps of S01, data acquisition, S02, data preprocessing, S03, feature extraction, S04, risk assessment model construction, S05, model training and optimization, S06, risk assessment, S07, result analysis and verification and S08, result output. According to the scheme, by adopting an advanced multi-source data fusion technology, various data sources such as geological data, remote sensing images, elevation maps and hydrological data are effectively integrated; the integration not only widens the coverage range of the data, but also improves the precision and timeliness of the data. Through an automatic data preprocessing process, it is ensured that data of different sources and types can be effectively fused, and a data consistency relation is constructed; the data integration strategy can more comprehensively reflect the complexity and dynamic change of hidden disaster-causing factors, so that the accuracy of disaster-causing risk assessment is remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of geographic information systems, and in particular to a disaster risk assessment method for hidden disaster-causing factors based on multi-source data fusion. Background Art

[0002] With the rapid development of industrialization and urbanization, the impact of human activities on the geological environment has become increasingly severe, resulting in hidden disaster factors such as goafs, faults, and collapse columns, which pose a serious threat to human safety and engineering stability. Therefore, it is particularly important to accurately assess the disaster risks of these hidden disaster factors. In recent years, with the development of remote sensing technology, geographic information system (GIS) and big data processing technology, multi-source data fusion technology has provided a new perspective for the disaster risk assessment of hidden disaster factors. Multi-source data fusion technology can integrate data from different sources, types and resolutions, such as geological data, remote sensing images, elevation maps and hydrological data, to improve the accuracy and comprehensiveness of the assessment. However, the existing technology still has the following shortcomings in the application of multi-source data fusion to disaster risk assessment;

[0003] The generalization ability of existing assessment models is limited. Existing disaster risk assessment models often lack sufficient training data and optimization methods, resulting in insufficient generalization ability of the models in new areas or new situations.

[0004] In view of the shortcomings of the prior art, the present invention provides a disaster risk assessment method for hidden disaster factors based on multi-source data fusion, which aims to improve the accuracy, efficiency and usability of disaster risk assessment through automated data preprocessing, feature extraction of deep learning networks, construction and optimization of machine learning models, and visualized result output. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a disaster risk assessment method for hidden disaster factors based on multi-source data fusion, which solves the problem that the existing assessment model proposed in the background technology has limited generalization ability. The existing disaster risk assessment model often lacks sufficient training data and optimization methods, resulting in insufficient generalization ability of the model in new areas or new situations.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for disaster risk assessment of hidden disaster factors based on multi-source data fusion, the steps of which are as follows:

[0007] S01. Data collection: Collect multi-source heterogeneous data that affect hidden disaster-causing factors, including geological data, remote sensing images, elevation maps, and hydrological data;

[0008] S02. Data preprocessing: preprocess the collected data, including data cleaning, format conversion, and resolution unification, to build a data consistency relationship;

[0009] S03. Feature extraction: Use deep learning networks to extract features from preprocessed data, including geological structure features, hydrological features, and topographic features;

[0010] S04. Risk assessment model construction: Based on the extracted features, a disaster risk assessment model is constructed; the model uses machine learning methods such as support vector machines, random forests or neural networks;

[0011] S05. Model training and optimization: Use labeled data to train the risk assessment model and use cross-validation methods to optimize the model to improve the generalization ability of the model;

[0012] S06. Risk assessment: Input the extracted features into the risk assessment model to obtain the disaster risk level of hidden disaster factors;

[0013] S07. Result analysis and verification: Analyze the evaluation results and verify their accuracy through on-site verification or expert review;

[0014] S08. Result output: The assessment results are output in a visual manner, including risk level maps, risk distribution maps, and risk prevention and control recommendations.

[0015] Preferably, the geological data collection: the collection scope should cover all potential hidden disaster-causing areas, including goafs, faults, and collapse columns; the data accuracy should reach centimeter-level accuracy to ensure that key geological features can be identified; the data types include but are not limited to geological exploration reports, drilling data, and geophysical exploration results; the data sources include geological survey bureaus, mining companies, and scientific research institutions;

[0016] The remote sensing image acquisition: the resolution should be at least 1 meter to ensure that surface features and changes can be identified; the spectral range should cover visible light to infrared spectrum to obtain surface temperature and vegetation index information; the acquisition cycle should be at least once a quarter to monitor the dynamic changes of the surface; cloud cover limits the selection of the best shooting period to reduce the impact of cloud cover on data quality;

[0017] The elevation map data is collected by using airborne laser radar (LiDAR) or high-precision GPS high-precision measurement equipment; data types include digital elevation model (DEM), slope map, and aspect map; the update frequency is at least once every two years to reflect the latest changes in the terrain;

[0018] The hydrological data collection: data types include groundwater level, water quality, flow and flow direction information; the monitoring frequency of groundwater level and water quality is at least once a month, and the flow and flow direction are monitored at least once a quarter; data sources may include hydrogeological surveys, monitoring well data, and river flow monitoring stations.

[0019] Preferably, the data cleaning in S02 includes processing missing values ​​of each data point in the data set:

[0020] Mean Fill: X filled =μ, where X filled represents the value after filling, μ represents the mean;

[0021] Median fill: X filled = median(X), where X filled represents the value after filling, and median(X) represents the median of X;

[0022] Majority fill: X filled =mode(X), where X filled represents the value after filling, mode(X) represents the mode of X;

[0023] Outlier handling (using Z-score) is also included: Among them, Z represents the Z-score value, X is the original value, μ is the mean, σ is the standard deviation, and if |Z|>k (k is the threshold, such as 3), X is considered to be an outlier.

[0024] Preferably, the geological structure feature extraction: using a deep learning network, such as a convolutional neural network (CNN), to automatically learn the features of the geological structure from seismic data or geological exploration data, wherein the features include faults, folds, and lithology changes. In CNN, feature extraction can be expressed by the following formula:

[0025] f(x)=max(0,x*Wb)

[0026] Where x is the input data, W is the weight, b is the bias, * represents the convolution operation, and f(x) is the extracted feature;

[0027] The terrain and geomorphic feature extraction uses a deep learning network autoencoder (AE) to extract terrain and geomorphic features from DEM (digital elevation model) data, including slope, slope direction, and terrain relief. The formula of the autoencoder is as follows:

[0028]

[0029] Among them, x is the input data, h is the hidden layer representation, is the reconstructed input data, W x and W h are the weight matrices for encoding and decoding respectively;

[0030] The hydrological characteristics are extracted through a long short-term memory network (LSTM), which is used to learn the long-term laws of the hydrological process, and its formula is:

[0031]

[0032] Among them, h t is the hidden state, C t is the cell state, f t ,i t ,o t are the activation vectors of the forget gate, input gate and output gate respectively, W and b are the weight and bias parameters; LSTM can capture the long-term dependencies in the hydrological time series and is used to predict hydrological characteristics including water level and flow.

[0033] Preferably, the disaster risk assessment model structure in S04 includes:

[0034] Input layer: Receives features extracted from the data preprocessing step, such as geological structure features, hydrological features, and topographic features. The received features are key input data for assessing disaster risks. These data need to go through the data preprocessing step to ensure their quality and consistency.

[0035] Hidden layer: one or more hidden layers, each layer consists of multiple neurons, responsible for extracting higher-level feature representations; each neuron is a nonlinear function, which is used to weight the data of the input layer and introduce nonlinearity through an activation function (such as ReLU or Sigmoid) so that the network can learn complex patterns and extract high-level features useful for risk assessment by learning the complex relationships between input features;

[0036] Output layer: The last layer is the output layer, which is used to convert the output of the hidden layer into the final risk assessment result. For disaster risk assessment, the number of neurons in the output layer corresponds to the number of risk level categories. The output of the output layer directly corresponds to the probability or score of different risk levels. The output is used for decision support to determine the areas where preventive measures need to be taken.

[0037] Preferably, in S05, the labeled data set is divided into a training set, a validation set and a test set, with a ratio of 70% training set, 15% validation set and 15% test set, to ensure the performance of the model on unseen data.

[0038] Preferably, in S05, K-fold cross validation (such as 5-fold or 10-fold) is used to evaluate the stability and generalization ability of the model, and one fold is set aside as a validation set each time, and the rest are used as training sets, and the process is repeated K times and the average performance index is taken.

[0039] Preferably, the model training in S05 uses gradient descent and its variants (such as stochastic gradient descent, mini-batch gradient descent) to optimize model parameters, with the goal of minimizing the loss function.

[0040] Preferably, in S05, a learning rate decay strategy is implemented to gradually reduce the learning rate as the training progresses, or an adaptive learning rate optimization algorithm, such as Adam, is used.

[0041] The present invention provides a disaster risk assessment method for hidden disaster factors based on multi-source data fusion. It has the following beneficial effects:

[0042] (1) This solution effectively integrates multiple data sources such as geological data, remote sensing images, elevation maps and hydrological data by adopting advanced multi-source data fusion technology. This integration not only broadens the data coverage, but also improves the accuracy and timeliness of the data. Through the automated data preprocessing process, including data cleaning, format conversion and resolution unification, it ensures that data from different sources and types can be effectively integrated and a data consistency relationship is established. Such a data integration strategy can more comprehensively reflect the complexity and dynamic changes of hidden disaster-causing factors, thereby significantly improving the accuracy of disaster risk assessment. In addition, this solution also automatically extracts key features such as geological structure features, hydrological features and topographic features through deep learning networks, further improving the efficiency and accuracy of feature extraction.

[0043] (2) By constructing a disaster risk assessment model based on machine learning and using a large amount of labeled data to train the model, the generalization ability of the model is significantly improved. In the process of model training, a cross-validation method is used, such as K-fold cross-validation, in which one fold is set aside as a validation set each time and the rest are used as training sets. This is repeated multiple times and the average performance index is taken to ensure the performance of the model on unseen data. In addition, gradient descent and its variants are used to optimize model parameters, and a learning rate decay strategy or an adaptive learning rate optimization algorithm, such as Adam, is implemented to further improve the stability and generalization ability of the model. These measures enable the evaluation model of this scheme to not only perform well on training data, but also maintain stable evaluation performance in new areas or new situations, greatly improving the practicality and reliability of the model.

[0044] (3) By outputting the assessment results in a visual manner, such as risk level maps and risk distribution maps, this solution makes the assessment results more intuitive and easy to understand, making it easier for decision makers to understand and apply them. This visual presentation method not only improves the usability of the assessment results, but also enhances the decision makers' confidence in the assessment results. In addition, this solution also includes on-site verification or expert review steps to verify the accuracy of the assessment results, further enhancing the credibility of the assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the steps of the method of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] See also Figure 1 The embodiment of the present invention provides a technical solution: a method for disaster risk assessment of hidden disaster-causing factors based on multi-source data fusion, the steps of which are as follows:

[0048] S01. Data collection: Collect multi-source heterogeneous data that affect hidden disaster-causing factors, including geological data, remote sensing images, elevation maps, and hydrological data; S02. Data preprocessing: Preprocess the collected data, including data cleaning, format conversion, and resolution unification to build data consistency relationships; S03. Feature extraction: Use deep learning networks to extract features from preprocessed data, including geological structure features, hydrological features, and topographic features; S04. Risk assessment model construction: Based on the extracted features, a disaster risk assessment model is constructed; the model uses machine learning methods such as support vector machines, random Forest or neural network; S05. Model training and optimization: Use labeled data to train the risk assessment model, and use cross-validation method to optimize the model to improve the generalization ability of the model; S06. Risk assessment: Input the extracted features into the risk assessment model to obtain the disaster risk level of hidden disaster factors; S07. Result analysis and verification: Analyze the assessment results, and verify the accuracy of the assessment results through on-site verification or expert review; S08. Result output: Output the assessment results in a visual way, including risk level map, risk distribution map, and provide risk prevention and control suggestions.

[0049] Geological data collection: The collection scope should cover all potential hidden disaster-causing areas, including goafs, faults, and collapse columns; the data accuracy should reach centimeter-level accuracy to ensure that key geological features can be identified; data types include but are not limited to geological exploration reports, drilling data, and geophysical exploration results; data sources include geological survey bureaus, mining companies, and scientific research institutions;

[0050] Remote sensing image acquisition: The resolution should be at least 1 meter to ensure that surface features and changes can be identified; the spectral range should cover visible light to infrared spectrum to obtain surface temperature and vegetation index information; the acquisition cycle should be at least once a quarter to monitor the dynamic changes of the surface; cloud cover limits the selection of the best shooting period to reduce the impact of cloud cover on data quality;

[0051] Elevation map data collection: Use airborne laser radar (LiDAR) or high-precision GPS high-precision measurement equipment; data types include digital elevation model (DEM), slope map, aspect map; update frequency is at least once every two years to reflect the latest changes in terrain;

[0052] Hydrological data collection: Data types include groundwater level, water quality, flow and flow direction information; monitoring frequency: groundwater level and water quality are monitored at least once a month, and flow and flow direction are monitored at least once a quarter; data sources may include hydrogeological surveys, monitoring well data, and river flow monitoring stations.

[0053] Data cleaning in S02 includes processing missing values ​​for each data point in the data set:

[0054] Mean Fill: X filled =μ, where X filled represents the value after filling, μ represents the mean;

[0055] Median fill: X filled = median(X), where X filled represents the value after filling, and median(X) represents the median of X;

[0056] Majority fill: X filled =mode(X), where X filled represents the value after filling, mode(X) represents the mode of X;

[0057] Outlier handling (using Z-score) is also included: Among them, Z represents the Z-score value, X is the original value, μ is the mean, σ is the standard deviation, and if |Z|>k (k is the threshold, such as 3), X is considered to be an outlier.

[0058] Geological structure feature extraction: Using deep learning networks, such as convolutional neural networks (CNN), the features of geological structures are automatically learned from seismic data or geological exploration data. The features include faults, folds, and lithology changes. In CNN, feature extraction can be expressed by the following formula:

[0059] f(x)=max(0,x*Wb)

[0060] Where x is the input data, W is the weight, b is the bias, * represents the convolution operation, and f(x) is the extracted feature;

[0061] The topographic and geomorphic features are extracted by using a deep learning network autoencoder (AE) to extract topographic and geomorphic features from DEM (digital elevation model) data, including slope, slope aspect, and terrain relief. The formula of the autoencoder is:

[0062]

[0063] Among them, x is the input data, h is the hidden layer representation, is the reconstructed input data, W x and W h are the weight matrices for encoding and decoding respectively;

[0064] The hydrological characteristics are extracted through the long short-term memory network (LSTM). LSTM is used to learn the long-term laws of the hydrological process. The formula is:

[0065]

[0066]

[0067] Among them, h t is the hidden state, C t is the cell state, f t ,i t ,o t are the activation vectors of the forget gate, input gate and output gate respectively, W and b are the weight and bias parameters; LSTM can capture the long-term dependencies in the hydrological time series and is used to predict hydrological characteristics including water level and flow.

[0068] The disaster risk assessment model structure in S04 includes:

[0069] Input layer: Receives features extracted from the data preprocessing step, such as geological structure features, hydrological features, and topographic features. The received features are key input data for assessing disaster risks. These data need to go through the data preprocessing step to ensure their quality and consistency.

[0070] Hidden layer: one or more hidden layers, each layer consists of multiple neurons, responsible for extracting higher-level feature representations; each neuron is a nonlinear function used to weight the data of the input layer and introduce nonlinearity through an activation function (such as ReLU or Sigmoid) so that the network can learn complex patterns and extract high-level features useful for risk assessment by learning the complex relationships between input features;

[0071] Output layer: The last layer is the output layer, which is used to convert the output of the hidden layer into the final risk assessment result. For disaster risk assessment, the number of neurons in the output layer corresponds to the number of risk level categories. The output of the output layer directly corresponds to the probability or score of different risk levels. The output is used for decision support to determine the areas where preventive measures need to be taken.

[0072] In S05, the labeled dataset is divided into training set, validation set and test set, with a ratio of 70% training set, 15% validation set and 15% test set to ensure the performance of the model on unseen data.

[0073] In S05, K-fold cross validation (such as 5-fold or 10-fold) is used to evaluate the stability and generalization ability of the model. Each time, one fold is set aside as a validation set, and the rest are used as training sets. This is repeated K times and the average performance index is taken.

[0074] Model training in S05 uses gradient descent and its variants (such as stochastic gradient descent and mini-batch gradient descent) to optimize model parameters, with the goal of minimizing the loss function.

[0075] In S05, a learning rate decay strategy is implemented to gradually reduce the learning rate as training progresses, or an adaptive learning rate optimization algorithm, such as Adam, is used.

[0076] In order to comprehensively evaluate the effectiveness of this solution, a series of experiments were designed to verify the application of multi-source data fusion strategy and machine learning model in disaster risk assessment of hidden disaster factors. The following are two key tables, which summarize the results of data collection and feature extraction, as well as detailed data of model training and evaluation:

[0077] Table 1: Data collection and feature extraction

[0078]

[0079] Table 2: Model training and evaluation

[0080]

[0081] Based on the data in the above two tables, the following conclusions can be drawn:

[0082] Table 1: Effectiveness of data fusion strategy: The experimental results show that the data fusion strategy of "multi-sensor data fusion", "time series analysis" and "spatial data analysis" can significantly improve the accuracy of feature extraction. This shows that this solution effectively improves the comprehensiveness and accuracy of data by integrating multi-source data, providing a solid data foundation for subsequent risk assessment.

[0083] Improved feature extraction accuracy: The average feature extraction accuracy of all experiments reached 92.33%, with the highest reaching 95%. This result shows that the deep learning network used in this scheme can effectively extract key features from multi-source data and provide accurate input for disaster risk assessment.

[0084] Table 2 Conclusion: Applicability of model types: The experiment covers three different model types: support vector machine (SVM), random forest and neural network. The results show that all model types can achieve high accuracy (87.67% on average), indicating that this scheme can select the appropriate model type according to different data characteristics to achieve the best evaluation effect.

[0085] The important role of optimization algorithms: By using optimization algorithms such as stochastic gradient descent, gradient descent, and Adam, the training efficiency and accuracy of the model have been significantly improved. In particular, the Adam algorithm has shown excellent performance in the neural network model, with a relatively short training time and an accuracy rate of up to 91%.

[0086] Verification of model generalization ability: All experiments passed the cross-validation generalization ability evaluation, indicating that the model constructed by this scheme has good generalization ability and can maintain stable evaluation performance on unseen data.

[0087] In summary, this scheme significantly improves the accuracy and generalization ability of disaster risk assessment of hidden disaster factors by adopting advanced data fusion strategies, feature extraction of deep learning networks, and construction and optimization of machine learning models. These conclusions verify the effectiveness of this scheme and provide strong support for practical applications.

[0088] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0089] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A disaster risk assessment method for hidden disaster factors based on multi-source data fusion, characterized in that: The steps are as follows: S01. Data collection: Collect multi-source heterogeneous data that affect hidden disaster-causing factors, including geological data, remote sensing images, elevation maps, and hydrological data; S02. Data preprocessing: preprocess the collected data, including data cleaning, format conversion, and resolution unification, to build a data consistency relationship; S03. Feature extraction: Use deep learning networks to extract features from preprocessed data, including geological structure features, hydrological features, and topographic features; S04. Risk assessment model construction: Based on the extracted features, a disaster risk assessment model is constructed; the model uses machine learning methods such as support vector machines, random forests or neural networks; S05. Model training and optimization: Use labeled data to train the risk assessment model and use cross-validation methods to optimize the model to improve the generalization ability of the model; S06. Risk assessment: Input the extracted features into the risk assessment model to obtain the disaster risk level of hidden disaster factors; S07. Result analysis and verification: Analyze the evaluation results and verify their accuracy through on-site verification or expert review; S08. Result output: The assessment results are output in a visual manner, including risk level maps, risk distribution maps, and risk prevention and control recommendations.

2. The method for disaster risk assessment of hidden disaster factors based on multi-source data fusion according to claim 1 is characterized by: The geological data collection mentioned above: the collection scope should cover all potential hidden disaster-causing areas, including goafs, faults, and collapse columns; the data accuracy should reach centimeter-level accuracy to ensure that key geological features can be identified; the data types include but are not limited to geological exploration reports, drilling data, and geophysical exploration results; the data sources include geological survey bureaus, mining companies, and scientific research institutions; The remote sensing image acquisition: the resolution should be at least 1 meter to ensure that surface features and changes can be identified; the spectral range should cover visible light to infrared spectrum to obtain surface temperature and vegetation index information; the acquisition cycle should be at least once a quarter to monitor the dynamic changes of the surface; cloud cover limits the selection of the best shooting period to reduce the impact of cloud cover on data quality; The elevation map data is collected by using airborne laser radar (LiDAR) or high-precision GPS high-precision measurement equipment; data types include digital elevation model (DEM), slope map, and aspect map; the update frequency is at least once every two years to reflect the latest changes in the terrain; The hydrological data collection: data types include groundwater level, water quality, flow and flow direction information; the monitoring frequency of groundwater level and water quality is at least once a month, and the flow and flow direction are monitored at least once a quarter; data sources may include hydrogeological surveys, monitoring well data, and river flow monitoring stations.

3. The method for disaster risk assessment of hidden disaster-causing factors based on multi-source data fusion according to claim 1 is characterized by: The data cleaning in S02 includes processing the missing values ​​of each data point in the data set: Mean Fill: X filled =μ, where X filled represents the value after filling, μ represents the mean; Median fill: X filled = median(X), where X filled represents the value after filling, and median(X) represents the median of X; Majority fill: X filled =mode(X), where X filled represents the value after filling, mode(X) represents the mode of X; Outlier handling (using Z-score) is also included: Among them, Z represents the Z-score value, X is the original value, μ is the mean, σ is the standard deviation, and if |Z|>k (k is the threshold, such as 3), X is considered to be an outlier.

4. The method for disaster risk assessment of hidden disaster-causing factors based on multi-source data fusion according to claim 1 is characterized by: The geological structure feature extraction: using a deep learning network, such as a convolutional neural network (CNN), to automatically learn the features of the geological structure from seismic data or geological exploration data. The features include faults, folds, and lithology changes. In CNN, feature extraction can be expressed by the following formula: f(x)=max(0,x*Wb) Where x is the input data, W is the weight, b is the bias, * represents the convolution operation, and f(x) is the extracted feature; The terrain and geomorphic feature extraction uses a deep learning network autoencoder (AE) to extract terrain and geomorphic features from DEM (digital elevation model) data, including slope, slope direction, and terrain relief. The formula of the autoencoder is as follows: Among them, x is the input data, h is the hidden layer representation, is the reconstructed input data, W x and W h are the weight matrices for encoding and decoding respectively; The hydrological characteristics are extracted through a long short-term memory network (LSTM), which is used to learn the long-term laws of the hydrological process, and its formula is: Among them, h t is the hidden state, C t is the cell state, f t ,i t ,o t are the activation vectors of the forget gate, input gate and output gate respectively, W and b are the weight and bias parameters; LSTM can capture the long-term dependencies in the hydrological time series and is used to predict hydrological characteristics including water level and flow.

5. The method for disaster risk assessment of hidden disaster-causing factors based on multi-source data fusion according to claim 1 is characterized by: The disaster risk assessment model structure in S04 includes: Input layer: Receives features extracted from the data preprocessing step, such as geological structure features, hydrological features, and topographic features. The received features are key input data for assessing disaster risks. These data need to go through the data preprocessing step to ensure their quality and consistency. Hidden layer: one or more hidden layers, each layer consists of multiple neurons, responsible for extracting higher-level feature representations; each neuron is a nonlinear function, which is used to weight the data of the input layer and introduce nonlinearity through an activation function (such as ReLU or Sigmoid) so that the network can learn complex patterns and extract high-level features useful for risk assessment by learning the complex relationships between input features; Output layer: The last layer is the output layer, which is used to convert the output of the hidden layer into the final risk assessment result. For disaster risk assessment, the number of neurons in the output layer corresponds to the number of risk level categories. The output of the output layer directly corresponds to the probability or score of different risk levels. The output is used for decision support to determine the areas where preventive measures need to be taken.

6. The method for disaster risk assessment of hidden disaster-causing factors based on multi-source data fusion according to claim 1 is characterized by: In S05, the labeled data set is divided into a training set, a validation set and a test set, with a ratio of 70% training set, 15% validation set and 15% test set, to ensure the performance of the model on unseen data.

7. The method for disaster risk assessment of hidden disaster factors based on multi-source data fusion according to claim 1 is characterized by: In S05, K-fold cross validation (such as 5-fold or 10-fold) is used to evaluate the stability and generalization ability of the model. Each time, one fold is set aside as a validation set, and the rest are used as training sets. This is repeated K times and the average performance index is taken.

8. The method for disaster risk assessment of hidden disaster factors based on multi-source data fusion according to claim 1 is characterized by: The model training in S05 uses gradient descent and its variants (such as stochastic gradient descent and mini-batch gradient descent) to optimize model parameters, with the goal of minimizing the loss function.

9. The method for disaster risk assessment of hidden disaster factors based on multi-source data fusion according to claim 1 is characterized by: In S05, a learning rate decay strategy is implemented to gradually reduce the learning rate as the training progresses, or an adaptive learning rate optimization algorithm, such as Adam, is used.

Images