Geological disaster early warning method and system, electronic equipment and storage medium

Abstract

The invention relates to the technical field of geological disasters, and discloses a geological disaster early warning method and system, electronic equipment and a storage medium. The method comprises the following steps: carrying out data preprocessing on collected multi-source data to obtain preprocessed multi-source data; constructing a multi-source data set, and fusing multi-dimensional data features extracted from the multi-source data set through an attention mechanism to obtain a multi-source training data set; constructing an initial geological disaster early warning model by using a multi-layer perceptron, a residual network and a visual model, and training the initial geological disaster early warning model through the multi-source training data set to obtain a trained geological disaster early warning model; acquiring new multi-source data in real time, and performing real-time analysis and prediction on the new multi-source data by using the geological disaster early warning model to obtain geological disaster early warning result data; and when the geological disaster early-warning result data reaches a preset early-warning condition, issuing graded early-warning information through a preset channel. The method improves the early warning accuracy and timeliness of geological disasters.

Description

Technical Field

[0001] This application relates to the technical field of geological disasters, and in particular to a geological disaster warning method, system, electronic device and storage medium. Background Art

[0002] In the field of geological disaster warning, most traditional warning methods rely on a single data source or simple monitoring indicators. For example, in the early stage, only a few data such as rainfall and topography were relied on, making it difficult to comprehensively capture the complex mechanism of geological disasters. Moreover, empirical models were mostly used, which were difficult to adapt to different geological conditions and complex environmental changes. The warning accuracy and timeliness were poor, and it was often impossible to effectively warn in advance in the face of sudden disasters, easily resulting in casualties and property losses. The data sources related to geological disasters are extensive, including multi-source data such as topography, lithology, and meteorology, and the data formats are diverse and the structures are complex. Traditional data processing technologies related to geological disasters are difficult to efficiently integrate and deeply analyze them, restricting the improvement of the performance of warning models. When constructing traditional geological disaster warning models in the past, only single-type features were often considered, and problems such as gradient disappearance and overfitting were prone to occur during the training process, resulting in poor generalization ability of the models. It was difficult to accurately predict geological disasters in different regions and complex situations, affecting the reliability and practicality of the warning system. In addition, the traditional warning release and emergency response mechanisms are relatively simple. The warning release threshold is set fixed and cannot be dynamically adjusted according to real-time data and geological conditions. The warning results are inaccurate. In the emergency response stage, the allocation of rescue resources mainly relies on manual experience, lacking reasonable planning and optimization, making it difficult to achieve efficient and accurate allocation, and easily delaying the rescue time. Summary of the Invention

[0003] The main technical problem to be solved by the embodiments of this application is that the accuracy, timeliness and response efficiency of traditional geological disaster warnings need to be improved.

[0004] To solve the above technical problems, the first technical solution adopted in the embodiments of the present application is: to provide a geological disaster warning method, including: collecting multi-source data of a preset data dimension through a preset data collection path, performing data preprocessing on the collected multi-source data to obtain preprocessed multi-source data; using the preprocessed multi-source data to construct a multi-source data set including positive sample data and negative sample data, and fusing multi-dimensional data features extracted from the multi-source data set through an attention mechanism to obtain a multi-source training data set; using a multi-layer perceptron, a residual network, and a vision model to construct an initial geological disaster warning model, and training the initial geological disaster warning model through the multi-source training data set to obtain a trained geological disaster warning model; collecting new multi-source data in real time through the data collection path, using the geological disaster warning model to perform real-time analysis and prediction on the new multi-source data to obtain geological disaster warning result data; when the geological disaster warning result data reaches a preset warning condition, issuing hierarchical warning information through a preset channel, and using a pre-trained rescue resource intelligent allocation model to generate a rescue resource allocation plan according to the geological disaster warning result data.

[0005] Optionally, the step of collecting multi-source data of a preset data dimension through a preset data collection path and performing data preprocessing on the collected multi-source data to obtain preprocessed multi-source data includes: collecting geological factor correlation data of a preset data dimension through a preset data collection path, and sorting the geological factor correlation data in chronological order; calculating the mean and standard deviation of each piece of geological factor correlation data arranged in chronological order, and removing the geological factor correlation data if the geological factor correlation data exceeds a first data range, where the first data range is composed of the mean and the standard deviation; using an autoregressive integrated moving average-based missing value processing model to perform missing value processing on the geological factor correlation data, where the expression of the missing value processing model is , represents the autoregressive operator, represents the moving average operator, represents the backward shift operator, represents the order of differencing, represents the white noise sequence, represents t the geological factor correlation data at time

[0006] Optionally, the step of constructing a multi-source data set including positive sample data and negative sample data using the preprocessed multi-source data, and fusing multi-dimensional data features extracted from the multi-source data set through an attention mechanism to obtain a multi-source training data set includes: executing a preset sliding window sampling strategy, setting a preset number of grid regions as evaluation units, determining positive sample data and negative sample data according to the evaluation units, and adding the positive sample data and the negative sample data to the multi-source data set; using the multi-source data set, calculating terrain dimension feature data and geomorphic dimension feature data through a preset digital elevation model; obtaining vegetation-related data in the multi-source data set, and processing the vegetation-related data through a preset normalized difference vegetation index model to obtain vegetation coverage feature data; processing the formation lithology data in the multi-source data set through a pre-trained deep learning model to obtain formation lithology feature vector data, and performing dimensionality reduction processing on the formation lithology feature vector data to obtain formation lithology feature data; rasterizing the fracture data in the multi-source data set, calculating the intensity value of the influence of each grid unit by the fracture, and generating fracture feature data according to the intensity value; fusing multi-dimensional data features extracted from the multi-source data set through an attention mechanism, and mapping the multi-dimensional data features to the same dimensional space; calculating the attention weights of the multi-dimensional data features to obtain a multi-source training data set.

[0007] Optionally, the steps of constructing an initial geological disaster warning model using a multi-layer perceptron, a residual network, and a vision model, and training the initial geological disaster warning model with the multi-source training dataset to obtain a trained geological disaster warning model include: using a multi-layer perceptron, a residual network, and a vision model as sub-models to construct an initial geological disaster warning model, and inputting the multi-source training data in the multi-source training dataset into the initial geological disaster warning model; performing initial feature extraction and non-linear transformation on the multi-source training data through the multi-layer hidden layers of the multi-layer perceptron to obtain the first geological feature vector data; inputting the multi-source training data into the residual blocks of the residual network, performing data transformation through the residual blocks, and performing a residual link operation on the feature vectors after data change processing to obtain the second geological feature vector data; dividing the multi-source training data into image block vectors, performing self-attention mechanism calculation according to the image block vectors through the self-attention mechanism module of the vision model, and processing the calculation results with a multi-layer encoder to obtain the third geological feature vector data; performing feature fusion processing on the first geological feature vector data, the second geological feature vector data, and the third geological feature vector data based on the attention mechanism to obtain a geological feature fusion vector; performing prediction calculation according to the geological feature fusion vector, calculating the loss value between the predicted value and the true value through the binary cross-entropy loss function, and continuously optimizing the weight parameters and bias parameters of each sub-model according to the loss value until the loss value converges to obtain the trained geological disaster warning model.

[0008] Optionally, the steps of collecting new multi-source data in real time through the data collection means, and using the geological disaster warning model to perform real-time analysis and prediction on the new multi-source data to obtain geological disaster warning result data include: collecting new multi-source data in real time through the data collection means, and removing the noise data in the new multi-source data using the wavelet transform method; identifying the outliers in the new multi-source data using a preset deep learning-based anomaly detection model, and repairing the outliers through a preset machine learning-based interpolation method; sending the new multi-source data after data preprocessing to the geological disaster warning model, and performing real-time analysis and prediction through the geological disaster warning model to obtain geological disaster warning result data; performing online training and optimization on the geological disaster warning model using the new multi-source data through a preset stochastic gradient descent algorithm.

[0009] Optionally, before the step of calculating the attention weights of the multi-dimensional data features to obtain the multi-source training data set, the method further includes: obtaining position information data in the multi-source data set; calculating using the position information data through a preset position encoding formula to obtain a position encoding feature vector; and performing feature encoding fusion processing on the position encoding feature vector and the multi-dimensional data features.

[0010] Optionally, before the step of performing prediction calculation according to the geological feature fusion vector and calculating the loss value between the predicted value and the true value through a binary cross-entropy loss function, the method further includes: training for a preset first number of rounds at a first learning rate during the training process; when the training process of the first number of rounds ends, continuing to train at a second learning rate, wherein a regularization term is added to the loss function through a regularization method during the training process, the first learning rate is less than the second learning rate, and the second learning rate decreases according to a preset training round interval.

[0011] To solve the above technical problems, the second technical solution adopted in the embodiments of the present application is: providing a geological disaster warning device, including: a data acquisition preprocessing module, configured to acquire multi-source data of a preset data dimension through a preset data acquisition channel, and perform data preprocessing on the acquired multi-source data to obtain preprocessed multi-source data; a training data set module, configured to use the preprocessed multi-source data to construct a multi-source data set including positive sample data and negative sample data, and fuse multi-dimensional data features extracted from the multi-source data set through an attention mechanism to obtain a multi-source training data set; a geological disaster warning model module, configured to construct an initial geological disaster warning model using a multi-layer perceptron, a residual network, and a vision model, and train the initial geological disaster warning model through the multi-source training data set to obtain a trained geological disaster warning model; a geological disaster prediction result module, configured to acquire new multi-source data in real time through the data acquisition channel, and perform real-time analysis and prediction on the new multi-source data using the geological disaster warning model to obtain geological disaster warning result data; and a geological disaster release and rescue module, configured to, when the geological disaster warning result data reaches a preset warning condition, release hierarchical warning information through a preset channel, and generate a rescue resource allocation plan using a pre-trained rescue resource intelligent allocation model according to the geological disaster warning result data.

[0012] To solve the above technical problems, the third technical solution adopted in the embodiments of the present application is: providing an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the geological disaster warning method as described above.

[0013] To solve the above technical problems, the fourth technical solution adopted in the embodiments of this application is: to provide a non-volatile computer-readable storage medium storing computer-executable instructions, which, when executed by an electronic device, cause the electronic device to execute the geological disaster warning method as described above.

[0014] Different from the related art, this application comprehensively obtains multi-dimensional data such as topography and vegetation coverage through multi-source data collection and preprocessing and ensures data quality, breaking the limitation of traditional single data sources. An early warning system is constructed by fusing multi-layer perceptrons, residual networks, and vision models. Compared with traditional single models, the accuracy of early warning is greatly improved. New data is collected in real time and dynamically analyzed and predicted to ensure the timeliness of early warning. When an early warning is triggered, the hierarchical early warning information release mechanism and the pre-trained intelligent allocation model of rescue resources are combined to achieve a full-process closed-loop management from disaster early warning to the generation of rescue plans, effectively improving the overall efficiency and effect of geological disaster response. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] One or more embodiments are described by way of example with reference to the corresponding drawings, which do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are represented as similar elements, unless otherwise stated, and the drawings in the figures do not constitute a scale limitation.

[0016] Figure 1 is a schematic diagram of the operating environment of the geological disaster warning method provided by the embodiments of this application.

[0017] Figure 2 is a schematic diagram of the execution process of the geological disaster warning method provided by the embodiments of this application.

[0018] Figure 3 is a schematic diagram of the execution process of obtaining a multi-source training data set in the geological disaster warning method provided by the embodiments of this application.

[0019] Figure 4 is a schematic diagram of the execution process of obtaining a trained geological disaster warning model in the geological disaster warning method provided by the embodiments of this application.

[0020] Figure 5 is a schematic diagram of the system structure of the geological disaster warning device provided by the embodiments of this application.

[0021] Figure 6 is a schematic diagram of the hardware structure of an electronic device for executing the geological disaster warning method provided by the embodiments of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to make the objectives, technical solutions and advantages of this application clearer and more understandable, the following further details this application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0023] It should be noted that if there is no conflict, the various features in the embodiments of this application can be combined with each other, and all are within the protection scope of this application. In addition, although the functional modules are divided in the device schematic diagram and the logical sequence is shown in the flowchart, in some cases, the steps shown or described can be executed in a different module division from that in the device schematic diagram or a different order from that in the flowchart.

[0024] Unless otherwise defined, all the technical and scientific terms used in this specification have the same meaning as commonly understood by those skilled in the technical field to which this application belongs. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not used to limit this application. The term "and / or" used in this specification includes any and all combinations of one or more of the related listed items.

[0025] For the convenience of understanding this embodiment, first, a geological disaster warning method disclosed in the embodiments of this application will be introduced in detail. Please refer to Figure 1 , Figure 1 is a schematic diagram of the operating environment of the geological disaster warning method provided in the embodiments of this application. As shown in Figure 1 , the execution subject of the geological disaster warning method provided in the embodiments of this application is generally an electronic device with a certain computing ability, such as a computer device. In some possible implementation manners, the geological disaster warning method can be implemented by a processor calling computer-readable instructions stored in a memory. Among them, Figure 1 the computer device in Figure 1 can be a server. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. It can be understood that

[0026] Please continue to refer to Figure 2 , Figure 2 is a schematic diagram of the execution process of the geological disaster warning method provided in the embodiments of this application. As shown in Figure 2 , it includes the following steps: S1. Collect multi-source data of preset data dimensions through preset data collection channels, and perform data preprocessing on the collected multi-source data to obtain preprocessed multi-source data.

[0027] As an alternative implementation, the above step S1 may include the following steps S11 to S14.

[0028] S11. Collect geological factor-related data of preset data dimensions through preset data collection channels, and sort the geological factor-related data in chronological order.

[0029] Among them, step S11 obtains data through pre-set data collection channels, which can be various sensors (such as displacement sensors, rainfall sensors, etc.), satellite remote sensing, geological exploration reports, etc. The collected data is related to geological factors, covering multi-faceted data such as stratigraphic lithology, geological structure, topography, and hydrogeology, and must meet the requirements of preset data dimensions to ensure that the data can provide effective information for subsequent analysis. Arrange the collected geological factor-related data in chronological order because the occurrence of geological disasters is often related to the data changes in the time series. Arranging in chronological order can facilitate the subsequent analysis of the evolution trend of data over time, such as analyzing the changes in factors such as formation pressure and rainfall in different periods, so as to better discover potential geological disaster risk signals.

[0030] S12. Calculate the mean and standard deviation of each piece of geological factor-related data arranged in chronological order. If the geological factor-related data exceeds the first data range, remove the geological factor-related data, where the first data range is composed of the mean and the standard deviation.

[0031] Among them, step S12 is to perform outlier processing on the geological factor-related data arranged in chronological order. Specifically, first calculate the mean and standard deviation for each piece of geological factor-related data (such as displacement data, rainfall data, etc.). The mean reflects the average level of the data, and the standard deviation reflects the degree of dispersion of the data. Taking the mean as the center and combining the standard deviation to determine a data range, if the data exceeds this range, it is very likely to be an outlier because in a normal distribution, most data should be concentrated within this range. Removing the outliers outside the range can avoid the interference of abnormal data on subsequent geological disaster analysis and early warning model training, ensure data quality, and make the analysis results more accurate and reliable. For example, for a certain data feature , calculate its mean and standard deviation , and regard the data points outside the range as outliers and eliminate them.

[0032] S13. Use a missing value processing model based on autoregressive integrated moving average to process the missing values in the geological factor correlation data. The expression of the missing value processing model is , represents the autoregressive operator, represents the moving average operator, represents the backward shift operator, represents the order of differencing, represents the white noise sequence, represents t the geological factor correlation data at time

[0033] Among them, when actually collecting geological data, due to reasons such as sudden sensor failures and interruptions during data transmission, some data may be missing. If these missing data are not processed, it will affect the subsequent analysis and judgment of geological disasters and also reduce the accuracy of the early warning model. Using the autoregressive integrated moving average model (ARIMA, Autoregressive Integrated Moving Average Model) can utilize the already collected and complete historical data, combined with the laws and characteristics presented by the data over time, to reasonably infer and supplement the missing data, making the entire geological factor correlation data complete and continuous, and providing a data basis for more accurately analyzing geological disaster situations and building a reliable early warning model in the future. For example, by analyzing the displacement data of a mountain over a period of time, predicting the displacement situation of the mountain in the future, and thus judging whether there is a risk of landslide.

[0034] S14. Perform frequency domain analysis on the geological factor correlation data arranged in time series through Fourier transform, extract the periodic components of the geological factor correlation data, and analyze the amplitudes and phases of different frequency components to obtain the periodic change law of the geological factor correlation data.

[0035] Among them, in the research related to geological disasters, the laws of many geological factors changing over time are not simple linear changes and may have periodic fluctuations. Fourier transform can transform the geological factor correlation data in the time domain to the frequency domain and decompose the data into a combination of sine and cosine waves of different frequencies. For example, using methods such as Fourier transform to extract the periodic components of the data, the formula is: , where is the time series data, is the result of its Fourier transform, is the frequency. Through this operation, the periodic components in the geological factor-related data can be extracted. Subsequently, the amplitudes and phases of different frequency components are analyzed. The magnitude of the amplitude reflects the importance of the frequency component in the entire data, and the phase reflects the relative position of the frequency component in time. Through such analysis, the periodic change law of the geological factor-related data can be grasped. For example, the rainfall in some areas may have seasonal periodic changes, or the pressure of the strata may also have fluctuations with specific periods.

[0036] S2. Use the preprocessed multi-source data to construct a multi-source data set including positive sample data and negative sample data, and fuse the multi-dimensional data features extracted from the multi-source data set through the attention mechanism to obtain a multi-source training data set.

[0037] As an alternative implementation manner, please continue to refer to Figure 3 , Figure 3 is a schematic flowchart of the execution process for obtaining a multi-source training data set in the geological disaster warning method provided by the embodiments of the present application. As shown in Figure 3 , it includes the following steps S21 to step S27.

[0038] S21. Execute a preset sliding window sampling strategy, set a preset number of grid areas as evaluation units, determine positive sample data and negative sample data according to the evaluation units, and add the positive sample data and negative sample data to the multi-source data set.

[0039] For example, by using the sliding window sampling strategy, a grid area with a size of 16x16 is set as an evaluation unit. For the geological disaster point data, if the disaster point is located within the window unit, it is determined as a positive sample. For the negative sample, it is randomly selected in the entire area, and it is ensured that the distance between the center point of the negative sample and all known disaster points is greater than two evaluation units, so as to construct a multi-source data set.

[0040] S22. Use the multi-source data set to calculate terrain dimension feature data and geomorphic dimension feature data through a preset digital elevation model.

[0041] Among them, the digital elevation model (Digital-Elevation-Model, DEM) is a digital representation of the earth's surface topography and geomorphology. It is constructed by measuring and collecting elevation data of a series of points on the ground, and then through data processing and interpolation methods. Usually, the data is stored in the form of regular grids. Each grid unit corresponds to a specific ground position, and its value represents the elevation of this position, which can accurately reflect the undulation changes of the terrain, including various terrain features such as mountains, valleys, plains, and rivers. By analyzing and processing the DEM data, various topographic and geomorphic information can be extracted, such as slope, aspect, elevation difference, terrain roughness, etc.

[0042] For example, the slope is calculated using the following slope calculation formula: ; wherein, represents the elevation value, and are the plane coordinate directions. Through this formula, the terrain steepness of each evaluation unit can be accurately quantified, providing a key terrain basis for geological hazard risk assessment.

[0043] S23. Obtain the vegetation-related data in the multi-source dataset, and process the vegetation-related data through a preset normalized difference vegetation index (NDVI) model to obtain vegetation coverage characteristic data.

[0044] Among them, the NDVI model can calculate the vegetation coverage characteristic data according to the reflection characteristics of vegetation in different bands. Vegetation coverage is an important factor affecting the occurrence of geological disasters. A higher vegetation coverage can usually reduce soil erosion and the risk of geological disasters. Therefore, this characteristic data is of great significance for geological hazard early warning.

[0045] For example, the following NDVI calculation formula is used to calculate the vegetation index: ; wherein, NIR represents the reflectance in the near-infrared band, and RED represents the reflectance in the red band.

[0046] S24. Process the stratigraphic lithology data in the multi-source dataset through a pre-trained deep learning model to obtain stratigraphic lithology characteristic vector data, and perform dimensionality reduction processing on the stratigraphic lithology characteristic vector data to obtain stratigraphic lithology characteristic data.

[0047] Among them, first use the pre-trained deep learning model to analyze the stratigraphic lithology data to obtain stratigraphic lithology characteristic vector data. However, the obtained initial vector data contains complex characteristic information of the stratigraphic lithology and may have a relatively high dimension, which is not conducive to subsequent calculations and analyses. Therefore, dimensionality reduction processing is performed on the stratigraphic lithology characteristic vector data to remove redundant information and retain the most critical features, finally obtaining stratigraphic lithology characteristic data for more efficient use in the geological hazard early warning model.

[0048] For example, for the text description information in the stratigraphic lithology data, use the pre-trained Bert model for vectorization processing. Input the stratigraphic text T into the Bert model fine-tuned with a large amount of geological data to obtain the vector representation V(T) of each stratigraphic unit, whose dimension is d ( d determined jointly by the Bert model architecture and the characteristics of geological data), that is Meanwhile, to improve the computational efficiency and model performance, principal component analysis (PCA) is performed on the 768-dimensional vector (if the output of the original Bert model is 768-dimensional) for dimensionality reduction. Let the original vector be V(T) , and the low-dimensional vector after PCA transformation be V’(T) , and the transformation matrix be U , then we have: V’(T)=U T V(T), where U is the eigenvector matrix obtained by performing PCA analysis on all formation vectors. In the actual implementation process, the first k principal components ( k <768) can also be selected according to the principle that the cumulative contribution rate reaches a certain threshold (such as 95%) to form U, ensuring that key information is retained while reducing the data dimension and reducing the computational burden.

[0049] S25. Rasterize the fracture data in the multi-source dataset, calculate the intensity value of each grid cell affected by the fracture, and generate fracture feature data based on the intensity value.

[0050] For example, an exponential decay function is used to describe the influence of the fracture on the surrounding area, and the formula is as follows: ; In the formula, I(d) represents the influence intensity at a distance d from the fracture, λ is the initial influence intensity at the fracture, λ is the attenuation coefficient (

[0051] can be determined by fitting analysis based on geological conditions and historical disaster data, and by continuously optimizing its value, the influence of the fracture can be more accurately reflected). Using this formula to process the fracture data, the intensity value of each grid cell affected by the fracture can be obtained, which is an important feature of the comprehensive dataset.

[0052] S26. Through the attention mechanism, fuse the multi-dimensional data features extracted from the multi-source dataset and map the multi-dimensional data features to the same-dimensional space.

[0053] Among them, since the data features from different sources may have different dimensions and scales, it is not conducive to direct comprehensive analysis. Therefore, through the attention mechanism, these multi-dimensional data features are mapped to the same-dimensional space, making the different features comparable, facilitating subsequent unified processing and analysis of all features, and thus better mining the relationships between the features and the comprehensive impact on geological disaster warning. F 1 ,F 2 ,…, F k , first, map the features to the same dimensional space through their respective embedding layers to obtain E 1 , E 2 ,…, E k . Calculate the attention weight matrix A : ; wherein, Q and K are the query matrix and the key matrix calculated based on the feature vectors respectively, d k is the dimension of the key matrix. Then, perform weighted summation on the feature vectors through the attention weights to obtain the fused feature vector: .

[0054] S27. Calculate the attention weights of the multi-dimensional data features to obtain a multi-source training data set.

[0055] Among them, through the above steps S21 to S27, the sliding window sampling strategy is used to determine the positive and negative sample data and expand it into a multi-source data set, increasing the richness and representativeness of the data. Feature data are extracted from multiple dimensions such as topography, vegetation cover, formation lithology, and fault structure, comprehensively considering various factors affecting geological disasters. The terrain and landform features are calculated using a digital elevation model, the vegetation data are processed using a normalized difference vegetation index model, the formation lithology data are processed using a pre-trained deep learning model and dimensionality reduction is performed, and the fault data are rasterized and the influence intensity is calculated, comprehensively and deeply excavating the value of multi-source data. Through the attention mechanism, the multi-dimensional features are mapped to the same space, and the attention weights of each feature are calculated, highlighting the role of important features and reducing the interference of secondary features, enabling the multi-source training data set to more accurately reflect the relevant information of geological disasters, thereby improving the accuracy and reliability of the geological disaster warning model.

[0056] As another alternative implementation, before the step of calculating the attention weights of the multi-dimensional data features to obtain a multi-source training data set, it may further include: obtaining the position information data in the multi-source data set. Then, calculate using the position information data through a preset position encoding formula to obtain a position encoding feature vector. Preferably, perform feature encoding fusion processing on the position encoding feature vector and the multi-dimensional data features.

[0057] For example, in a visual model, the input data is divided into multiple image patches, each image patch is regarded as a vector (token), and positional encoding is added to retain the positional information of the data. The positional encoding uses sine-cosine positional encoding, and the formula is as follows: Wherein, pos represents the position, i is the dimension index, d model is the dimension of the input vector. Through this positional encoding method, the model can distinguish features at different positions and improve the ability to understand the spatial structure of the data.

[0058] S3. Use a multi-layer perceptron, a residual network, and a visual model to construct an initial geological disaster warning model, and train the initial geological disaster warning model with a multi-source training data set to obtain a trained geological disaster warning model.

[0059] As an alternative implementation, please continue to refer to Figure 4 , Figure 4 which is a schematic execution flow diagram of obtaining a trained geological disaster warning model in the geological disaster warning method provided by the embodiment of the present application. As shown in Figure 4 , it includes the following steps S31 to step S36.

[0060] S31. Use a multi-layer perceptron, a residual network, and a visual model as sub-models to construct an initial geological disaster warning model, and input the multi-source training data in the multi-source training data set into the initial geological disaster warning model.

[0061] Among them, a multi-layer perceptron, a residual network, and a visual model are selected as sub-models and combined to form a comprehensive model. Then, the multi-source training data in the multi-source training data set is input into this initial model, providing a data basis for subsequent feature extraction and model training. The advantages of different sub-models can be fully utilized to analyze and process the data from multiple perspectives to improve the accuracy of geological disaster warning.

[0062] S32. Perform initial feature extraction and non-linear transformation on the multi-source training data through the multi-layer hidden layers of the multi-layer perceptron to obtain the first geological feature vector data.

[0063] Among them, the multi-layer perceptron has multiple hidden layers, and the hidden layers can perform initial feature extraction and non-linear transformation on the data. Through multi-layer processing, deep-level feature information in the data can be mined, and the multi-source training data is converted into the first geological feature vector data. The non-linear transformation enables the model to learn complex patterns and relationships in the data, providing a more valuable feature representation for subsequent analysis and prediction.

[0064] For example, the calculation method of the hidden layer neurons is as follows: where, h (i) is the output of the i -th hidden layer, W (i) is the weight matrix of the i -th layer, b (i) is the bias vector, is the activation function, and the activation function can be the Relu function, that is .

[0065] S33. Input multi-source training data into the residual blocks of the residual network, perform data transformation through the residual blocks, and then perform residual connection operations on the feature vectors after data change processing to obtain the second geological feature vector data.

[0066] Among them, the residual block is the core component of the residual network and can effectively transform the data. After the data is transformed, the processed feature vectors will be subjected to residual connection operations, which can alleviate the problem of gradient disappearance, enable the network to be trained deeper, thereby learning more complex features, and finally obtain the second geological feature vector data, which contains important information about the geological situation extracted by the residual network.

[0067] For example, the output formula in the residual block of the residual network is: output=x+F(x) ; where, x is the input, F(x) is the transformation or mapping of the input.

[0068] S34. Divide the multi-source training data into image block vectors, perform self-attention mechanism calculations on the image block vectors through the self-attention mechanism module of the vision model, and process the calculation results using a multi-layer encoder to obtain the third geological feature vector data.

[0069] Among them, the above vision model can be a vision model based on Transformer. In the vision model based on Transformer, the multi-source training data is divided into image block vectors, and each image block vector contains geological feature information of a specific area. The self-attention mechanism module will deeply analyze the image block vectors and calculate the degree of association, that is, the attention weight, between each image block vector and all other image block vectors. In this way, the model can automatically focus on the important parts of the data and ignore irrelevant information. For example, in geological image data, it can accurately identify key feature areas related to geological disasters, such as faults and fractures.

[0070] For example, to better capture long-range dependencies and global features in the data, a Transformer-based vision model is incorporated into the model. The core of the Transformer is the self-attention mechanism, and its calculation method is: ; wherein, Q, K, V represent the query vector, the key vector, and the value vector respectively, d k represents the dimension of the key vector K of.

[0071] S35. Feature fusion processing is performed on the first geological feature vector data, the second geological feature vector data, and the third geological feature vector data based on the attention mechanism to obtain a geological feature fusion vector.

[0072] Among them, the attention mechanism can assign different weights according to the importance of each feature vector, effectively fuse the feature vectors from different sub-models, and obtain a geological feature fusion vector. The fusion method can make full use of the feature information extracted by each sub-model, comprehensively consider various factors, and improve the accuracy and reliability of geological disaster warning.

[0073] For example, let the feature extracted by the multi-layer perceptron be F MLP , the feature extracted by the residual network be F ResNet , and the feature extracted by the vision model be F ViT . First, these features are mapped to the same-dimensional space through their respective embedding layers to obtain E MLP , E ResNet , E ViT . Then calculate the attention weight matrix A : ; wherein, Q, K is calculated according to the fused feature vectors. Finally, the feature vectors are weighted and summed through the attention weights to obtain the fused feature vector F fusion : F fusion =A 1 E MLP +A 2 E ResNet +A 3 EViT 。

[0074] S36. Perform prediction calculations based on the geological feature fusion vector, calculate the loss value between the predicted value and the true value through the binary cross-entropy loss function, and continuously optimize the weight parameters and bias parameters of each sub-model according to the loss value until the loss value converges to obtain a trained geological disaster warning model.

[0075] Among them, the loss value reflects the deviation degree between the model prediction result and the actual situation. Continuously optimize the weight parameters and bias parameters of each sub-model according to the loss value, so that the prediction result of the model continuously approaches the true value. The optimization process will be iterated continuously until the loss value converges, that is, the performance of the model reaches a stable state. At this time, a trained geological disaster warning model is obtained.

[0076] For example, the error backpropagation algorithm is used in the training process to update the model parameters to minimize the difference between the model predicted value and the true label. Binary cross-entropy is used as the loss function, and the calculation formula is: ; ; Among them, y i is the true label, p i is the predicted probability, N is the number of samples. By continuously adjusting the weights and biases of the model, the value of the loss function is gradually reduced, thereby optimizing the performance of the model.

[0077] As another alternative implementation, before the step of calculating the loss value between the predicted value and the true value through the binary cross-entropy loss function, it may further include: training for a preset first number of rounds at a first learning rate during the training process. Then, when the training process of the first number of rounds ends, continue to train at a second learning rate, where a regularization term is added to the loss function by regularization during the training process, the first learning rate is less than the second learning rate, and the second learning rate decreases according to a preset training round interval.

[0078] For example, during the training process, the learning rate is adjusted using a warm-up strategy. First, optimize the model for several rounds (such as 6 rounds) at a lower first learning rate (such as 0.001). After the model is relatively stable, train at a preset second learning rate (such as 0.01), and during the training process, decay the second learning rate by a certain multiple (such as 0.1 times) at certain rounds (such as the 12th round and the 18th round) to promote the model to converge faster and more stably. To prevent model overfitting, the L2 regularization technique is used to add a regularization term to the loss function: 。Among them, λis the regularization coefficient, W is the set of all parameters of the model.

[0079] S4. Real-time collect new multi-source data through the data collection channel, and use the geological disaster warning model to perform real-time analysis and prediction on the new multi-source data to obtain geological disaster warning result data.

[0080] As an optional implementation manner, the process of obtaining the geological disaster warning result data in the above step S4 may include the following steps S41 to S44.

[0081] S41. Real-time collect new multi-source data through the data collection channel, and use the wavelet transform method to remove the noise data in the new multi-source data.

[0082] For example, adopt the wavelet transform technology to remove the noise in the data. The wavelet transform formula is: where, f(t) is the original data signal, ψ is the wavelet basis function, a is the scale parameter, b is the translation parameter.

[0083] S42. Use the preset anomaly detection model based on deep learning to identify the outliers in the new multi-source data, and perform outlier repair through the preset interpolation method based on machine learning.

[0084] Among them, the deep learning model has a powerful feature learning ability and can find the data points that do not conform to the normal pattern from complex data, that is, outliers. Outliers may be caused by reasons such as sensor failures, data transmission errors, or sudden changes in geological conditions. After identifying the outliers, repair them through the preset interpolation method based on machine learning. The machine learning interpolation method can reasonably estimate and replace the outliers according to the distribution law of the data and the information of the surrounding normal data points, which can ensure the integrity and continuity of the data, avoid the negative impact of outliers on the analysis and prediction of the subsequent geological disaster warning model, enable the model to work based on more accurate data, and improve the reliability of the warning.

[0085] For example, construct an autoencoder network, whose input is the monitoring data x , and obtain the low-dimensional representation through the encoding layer, and then reconstruct the data through the decoding layer. Train the autoencoder to minimize its reconstruction error, that is ( n(for data dimension). In the actual implementation process, the reconstruction error is calculated. If the reconstruction error exceeds the preset threshold, the data point is determined to be an outlier. For outliers, an interpolation method based on machine learning is used for repair, such as interpolation based on the K-Nearest Neighbor (KNN) algorithm. For missing value or outlier points x , its repaired value is : where x i is x of K the k nearest neighbor data points.

[0086] S43. Send the new multi-source data after data preprocessing to the geological disaster warning model, and perform real-time analysis and prediction through the geological disaster warning model to obtain geological disaster warning result data.

[0087] Among them, after the above step S43, the warning rule based on the threshold can also be combined. When the predicted disaster occurrence probability exceeds the preset threshold, or some key monitoring indicators (such as the displacement rate exceeds the warning value, the effective rainfall reaches a specific threshold, etc.) trigger the warning condition, the warning process is immediately started, and warning information is sent to relevant departments and personnel through multiple channels (such as text messages, APP push, radio, etc.).

[0088] S44. Use the new multi-source data to perform online training and optimization on the geological disaster warning model through the preset stochastic gradient descent algorithm.

[0089] Among them, in order to enable the geological disaster warning model to adapt to the changing geological environment and newly emerging data patterns, step S44 uses the preset stochastic gradient descent algorithm to perform online training and optimization on the model using the new multi-source data. The stochastic gradient descent algorithm calculates the gradient by randomly selecting a part of the data in each iteration, thereby updating the model parameters. By continuously training with new data, the model can learn new features and patterns, adjust its own parameters to improve the prediction ability for geological disasters. The way of online training and optimization can keep the model in good performance. As time goes by and new data accumulates, the accuracy and timeliness of geological disaster warning are continuously improved, and it can better cope with complex and changeable geological disaster situations.

[0090] For example, a variant algorithm of stochastic gradient descent is adopted, such as the Adadelta algorithm, and its parameter update formula is: ; ; where is the model parameter at the t t-th iteration,gt is the gradient at the t th iteration, and are the cumulative sum of the squares of past parameter updates and the cumulative sum of the squares of the current gradient respectively, ε is a small constant to prevent the denominator from being zero.

[0091] S5. When the geological disaster warning result data reaches the preset warning conditions, release the graded warning information through the preset channels, and use the pre-trained intelligent allocation model of rescue resources to generate a rescue resource allocation plan according to the geological disaster warning result data.

[0092] As an example, in the graded warning based on dynamic thresholds, combined with the rainfall data monitored in real time, use the improved effective rainfall calculation formula: (where R c is the effective rainfall, R 0 is the rainfall of the day, R i is the rainfall on the n th day within the past α days, is the rainfall coefficient determined according to the local geology and meteorological conditions, γ is the influence coefficient of rainfall change rate, is the time interval,

[0093] is the rainfall change rate), calculate the effective rainfall. Based on the effective rainfall, the susceptibility level of geological disasters (classified into three levels: low, medium, and high, referring to the previous evaluation results of geological disaster susceptibility) and real-time geological monitoring data (such as displacement rate, ground stress change, etc.), dynamically delimit the warning levels (such as blue, yellow, orange, and red four-level warnings). The blue warning indicates a certain risk and requires attention. The yellow warning indicates an increased risk and requires preventive preparations. The orange warning means a greater possibility of disaster and activates some emergency response measures. The red warning indicates that the disaster is imminent and fully activates the emergency response. For different warning levels, adjust the frequency and scope of information release. During the red warning, release it frequently through all channels to ensure that the information coverage is without dead ends. m is the number of rescue material storage points, n is the number of affected areas, x ij represents the quantity of materials allocated from the i th storage point to the j th affected area, c ij represents the quantity of materials allocated from the i th storage point to thej The comprehensive transportation cost of an affected area (taking into account transportation distance d ij , road conditions r ij , type of transportation vehicle t ij , urgency e ij and other factors, and calculated through the cost function c ij =d ij × r ij × t ij × e ij ). s i is the material reserve volume of the i th reserve point, d j is the material demand of the j th affected area. A mixed-integer programming model is established, and the objective function is indicating the minimization of the total transportation cost. The constraint conditions include: (Ensure that the materials dispatched from each reserve point do not exceed the reserve volume). (Meet the basic material needs of each affected area). x ij ≥ ≥ 0 and some x ij are integers (determined whether to be integers according to the actual situation of material allocation, such as the allocation quantity of large equipment). Through this model, combined with intelligent algorithms (such as genetic algorithms, particle swarm optimization algorithms), the optimal material allocation plan is determined to achieve the rapid and efficient distribution of relief materials. At the same time, considering the timeliness and importance of materials, urgently needed materials and key materials are preferentially allocated, such as medical supplies, food, and emergency rescue equipment. During the emergency response process, continuously collect on-site feedback data, such as the progress of rescue teams, newly emerged geological hazards in affected areas, casualty information, and changes in material demand. Real-time feedback these data to the early warning and emergency decision-making model to dynamically adjust the model.

[0094] The geological disaster warning method provided by the embodiments of this application effectively solves the problems that the accuracy, timeliness, and response efficiency of traditional geological disaster warnings need to be improved through a series of technical means such as multi-dimensional data collection and preprocessing, multi-source training dataset construction, multi-model fusion training, and real-time dynamic analysis and warning. In terms of data processing, wavelet transform is used to remove noise, and a deep learning model is used to identify and repair outliers, ensuring the reliability of the data and avoiding warning deviations caused by data quality problems. When constructing a multi-source training dataset, multi-dimensional features such as topography and vegetation coverage are comprehensively extracted, and an attention mechanism is used for feature fusion, significantly improving the model's analysis ability for complex geological conditions. In the model construction link, a multi-layer perceptron, a residual network, and a vision model are combined to give full play to the advantages of each model and further improve the accuracy of the warning. In practical applications, by collecting new data in real time and quickly processing and analyzing it, combined with a hierarchical warning rule based on thresholds, timely warning of geological disasters is achieved. At the same time, the model is optimized by online training using the stochastic gradient descent algorithm, enabling it to adapt to the changing geological environment and continuously maintain a high warning accuracy. In addition, an intelligent allocation model for rescue resources constructed based on big data and the principles of operations research can quickly generate a rescue plan according to the warning results, greatly improving the emergency response efficiency, forming a complete closed loop from data collection and processing to warning response, comprehensively improving the accuracy, timeliness, and response efficiency of geological disaster warnings, and providing technical support for the effective prevention and control of geological disasters.

[0095] Please continue to refer to Figure 5 , Figure 5 which is a schematic diagram of the system structure of the geological disaster warning device provided by the embodiments of this application. As Figure 5 shown, the geological disaster warning device 50 includes: a collected data preprocessing module 51, a training dataset module 52, a geological disaster warning model module 53, a geological disaster prediction result module 54, and a geological disaster release and rescue module 55.

[0096] The collected data preprocessing module 51 is specifically configured to collect multi-source data of a preset data dimension through a preset data collection path, and perform data preprocessing on the collected multi-source data to obtain preprocessed multi-source data.

[0097] The training dataset module 52 is specifically configured to use the preprocessed multi-source data to construct a multi-source dataset including positive sample data and negative sample data, and fuse multi-dimensional data features extracted from the multi-source dataset through an attention mechanism to obtain a multi-source training dataset.

[0098] The geological disaster warning model module 53 is specifically used to construct an initial geological disaster warning model using a multi-layer perceptron, a residual network, and a vision model, and train the initial geological disaster warning model through the multi-source training data set to obtain a trained geological disaster warning model.

[0099] The geological disaster prediction result module 54 is specifically used to collect new multi-source data in real time through the data collection channels, and use the geological disaster warning model to perform real-time analysis and prediction on the new multi-source data to obtain geological disaster warning result data.

[0100] The geological disaster release and rescue module 55 is specifically used to, when the geological disaster warning result data reaches a preset warning condition, release graded warning information through a preset channel, and use a pre-trained rescue resource intelligent allocation model to generate a rescue resource allocation plan according to the geological disaster warning result data.

[0101] As an optional implementation manner, the collected data preprocessing module 51 is further specifically used to collect geological factor correlation data with a preset data dimension through a preset data collection channel, and sort the geological factor correlation data in chronological order; calculate the mean and standard deviation of each chronologically arranged geological factor correlation data, and if the geological factor correlation data exceeds the first data range, remove the geological factor correlation data, where the first data range is composed of the mean and the standard deviation; use an autoregressive integrated moving average-based missing value processing model to perform missing value processing on the geological factor correlation data, where the expression of the missing value processing model is , represents the autoregressive operator, represents the moving average operator, represents the backward shift operator, represents the order of differencing, represents the white noise sequence, represents t the geological factor correlation data at time

[0102] As an alternative implementation, the training dataset module 52 is further specifically configured to execute a preset sliding window sampling strategy, set a preset number of grid regions as evaluation units, determine positive sample data and negative sample data according to the evaluation units, and add the positive sample data and the negative sample data to the multi-source dataset; use the multi-source dataset to calculate terrain dimension feature data and geomorphic dimension feature data through a preset digital elevation model; obtain vegetation association data in the multi-source dataset, and process the vegetation association data through a preset normalized difference vegetation index model to obtain vegetation coverage feature data; process the stratigraphic lithology data in the multi-source dataset through a pre-trained deep learning model to obtain stratigraphic lithology feature vector data, and perform dimensionality reduction processing on the stratigraphic lithology feature vector data to obtain stratigraphic lithology feature data; rasterize the fracture data in the multi-source dataset, calculate the intensity value of each grid cell affected by the fracture, and generate fracture feature data according to the intensity value; fuse multi-dimensional data features extracted from the multi-source dataset through an attention mechanism, and map the multi-dimensional data features to the same dimensional space; calculate the attention weights of the multi-dimensional data features to obtain a multi-source training dataset.

[0103] As an alternative implementation, the geological disaster warning model module 53 is further specifically configured to use a multi-layer perceptron, a residual network, and a vision model as sub-models to construct an initial geological disaster warning model, and input the multi-source training data in the multi-source training dataset into the initial geological disaster warning model; perform initial feature extraction and non-linear transformation on the multi-source training data through multiple hidden layers of the multi-layer perceptron to obtain first geological feature vector data; input the multi-source training data into the residual blocks of the residual network, perform data transformation through the residual blocks, and perform residual link operations on the feature vectors after data change processing to obtain second geological feature vector data; divide the multi-source training data into image block vectors, perform self-attention mechanism calculations according to the image block vectors through the self-attention mechanism module of the vision model, and process the calculation results using multiple layer encoders to obtain third geological feature vector data; perform feature fusion processing on the first geological feature vector data, the second geological feature vector data, and the third geological feature vector data based on the attention mechanism to obtain a geological feature fusion vector; perform prediction calculations according to the geological feature fusion vector, calculate the loss value between the predicted value and the true value through a binary cross-entropy loss function, and continuously optimize the weight parameters and bias parameters of each sub-model according to the loss value until the loss value converges to obtain the trained geological disaster warning model.

[0104] As an alternative implementation, the geological disaster prediction result module 54 is further specifically configured to collect new multi-source data in real time through the data collection channels, and use wavelet transform to remove noise data in the new multi-source data; use a preset anomaly detection model based on deep learning to identify outliers in the new multi-source data, and perform outlier repair through a preset interpolation method based on machine learning; send the new multi-source data after data preprocessing to the geological disaster warning model, and perform real-time analysis and prediction through the geological disaster warning model to obtain geological disaster warning result data; perform online training and optimization on the geological disaster warning model using the new multi-source data through a preset stochastic gradient descent algorithm.

[0105] As an alternative implementation, the training dataset module 52 is further specifically configured to obtain position information data in the multi-source dataset; calculate through a preset position encoding formula using the position information data to obtain a position encoding feature vector; perform feature encoding fusion processing on the position encoding feature vector and the multi-dimensional data features.

[0106] As an alternative implementation, the geological disaster warning model module 53 is further specifically configured to first train a preset first number of rounds at a first learning rate during the training process; when the training process of the first number of rounds ends, continue training at a second learning rate, where a regularization term is added to the loss function through regularization during the training process, the first learning rate is less than the second learning rate, and the second learning rate decreases according to a preset training round interval.

[0107] It should be noted that the above geological disaster warning device can execute the geological disaster warning method provided in the embodiments of the present application, and has corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in the embodiments of the geological disaster warning device, reference can be made to the geological disaster warning method provided in the embodiments of the present application.

[0108] Figure 6 is a schematic hardware structure diagram of an electronic device 600 for executing the geological disaster warning method provided in the embodiments of the present application, as Figure 6 shown, the electronic device 600 includes: One or more processors 610 and a memory 620, Figure 6 Taking one processor 610 as an example.

[0109] The processor 610 and the memory 620 can be connected through a bus or other means, Figure 6 Taking the connection through a bus as an example.

[0110] The memory 620, being a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the geological disaster warning method in the embodiments of the present application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, that is, to implement the geological disaster warning method in the above method embodiments.

[0111] The memory 620 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the geological disaster warning device, etc. In addition, the memory 620 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 620 may optionally include a memory remotely set relative to the processor 610, and these remote memories can be connected to the geological disaster warning device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0112] The one or more modules are stored in the memory 620 and, when executed by the one or more processors 610, execute the geological disaster warning method in any of the above method embodiments. For example, execute the method steps S1 to S5 described above Figure 2 in Figure 3 the method steps S21 to S27 in Figure 4 the method steps S31 to S36 in Figure 5 to implement the functions of modules 51 - 55 in

[0113] The above product can execute the method provided by the embodiments of the present application, and has corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference can be made to the method provided by the embodiments of the present application.

[0114] The embodiments of the present application provide a non-volatile computer-readable storage medium storing computer-executable instructions, which are executed by one or more processors, such as Figure 6 one of the processors 610 in Figure 2 such that the above one or more processors can execute the geological disaster warning method in any of the above method embodiments. For example, execute the method steps S1 to S5 described above Figure 3 the method steps S21 to S27 in Figure 4The method steps S31 to S36 in realize Figure 5 the functions of modules 51 - 55 in.

[0115] The embodiments of the present application provide a computer program product. The computer program product includes a computer program stored on a non - volatile computer - readable storage medium. The computer program includes program instructions. When the program instructions are executed by the electronic device, the electronic device can execute the electronic device method in any of the above - mentioned method embodiments. For example, execute the Figure 2 method steps S1 to S5 in, Figure 3 method steps S21 to S27 in, Figure 4 method steps S31 to S36 in realize Figure 5 the functions of modules 51 - 55 in.

[0116] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0117] Through the description of the above - mentioned implementation manners, those of ordinary skill in the art can clearly understand that each implementation manner can be realized by means of software plus a general - purpose hardware platform, and of course, it can also be realized by hardware. Those of ordinary skill in the art can understand that all or part of the processes of implementing the above - mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer - readable storage medium. When the program is executed, it can include the processes of the above - mentioned method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read - only memory (ROM), or a random access memory (RAM), etc.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting them; under the idea of the present application, the technical features in the above embodiments or different embodiments can also be combined, and the steps can be implemented in any order, and there are many other variations in different aspects of the present application as described above. For the sake of brevity, they are not provided in detail; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A geological disaster early warning method, characterized in that: include: Collecting multi-source data of preset data dimensions through a preset data collection approach, and performing data preprocessing on the collected multi-source data to obtain preprocessed multi-source data; Using the preprocessed multi-source data to construct a multi-source data set including positive sample data and negative sample data, fusing multi-dimensional data features extracted from the multi-source data set through an attention mechanism to obtain a multi-source training data set; Using a multi-layer perceptron, a residual network and a visual model to construct an initial geological disaster warning model, and training the initial geological disaster warning model through the multi-source training data set to obtain a trained geological disaster warning model; Collect the new multi-source data in real time through the data collection approach, use the geological disaster early warning model to perform real-time analysis and prediction on the new multi-source data, and obtain geological disaster early warning result data; When the geological disaster warning result data reaches the preset warning conditions, the graded warning information is released through the preset channels, and the pre-trained rescue resource intelligent deployment model is used to generate a rescue resource deployment plan based on the geological disaster warning result data.

2. The geological disaster early warning method according to claim 1, characterized in that: The step of collecting multi-source data of preset data dimensions through a preset data collection approach, and performing data preprocessing on the collected multi-source data to obtain preprocessed multi-source data includes: Collecting geological factor-related data of preset data dimensions through a preset data collection approach, and sorting the geological factor-related data in chronological order; Calculating the mean and standard deviation of each type of geological factor associated data arranged in time series, and removing the geological factor associated data if the geological factor associated data exceeds a first data range, wherein the first data range is composed of the mean and the standard deviation; The missing value processing model based on autoregressive integrated moving average is used to process the missing values ​​of the geological factor associated data, wherein the expression of the missing value processing model is: , represents the autoregressive operator, represents the moving average operator, represents the backward shift operator, represents the difference order, represents a white noise sequence, express t The geological factors associated data at the time; The frequency domain analysis of the geological factor associated data arranged in time series is performed through Fourier transform, the periodic components of the geological factor associated data are extracted, and the amplitudes and phases of different frequency components are analyzed to obtain the periodic variation law of the geological factor associated data.

3. The geological disaster early warning method according to claim 1, characterized in that: The step of using the preprocessed multi-source data to construct a multi-source data set including positive sample data and negative sample data, and fusing the multi-dimensional data features extracted from the multi-source data set through an attention mechanism to obtain a multi-source training data set includes: Execute a preset sliding window sampling strategy, set a preset number of grid areas as evaluation units, determine positive sample data and negative sample data according to the evaluation units, and add the positive sample data and the negative sample data to the multi-source data set; Using the multi-source data set, calculating terrain dimension feature data and landform dimension feature data through a preset digital elevation model; Acquire vegetation-related data in the multi-source data set, and process the vegetation-related data using a preset normalized vegetation index model to obtain vegetation coverage characteristic data; Processing the stratigraphic lithology data in the multi-source data set by a pre-trained deep learning model to obtain stratigraphic lithology feature vector data, and performing dimensionality reduction processing on the stratigraphic lithology feature vector data to obtain stratigraphic lithology feature data; rasterizing the fracture data in the multi-source data set, calculating the intensity value of each grid unit affected by the fracture, and generating fracture feature data according to the intensity value; fusing multi-dimensional data features extracted from the multi-source data sets through an attention mechanism, and mapping the multi-dimensional data features to the same dimensional space; The attention weight of each of the multi-dimensional data features is calculated to obtain a multi-source training data set.

4. The geological disaster early warning method according to claim 1, characterized in that: The steps of constructing an initial geological disaster warning model using a multi-layer perceptron, a residual network and a visual model, training the initial geological disaster warning model using the multi-source training data set, and obtaining a trained geological disaster warning model include: Using a multi-layer perceptron, a residual network and a visual model as sub-models to construct an initial geological disaster warning model, and inputting the multi-source training data in the multi-source training data set into the initial geological disaster warning model; Performing initial feature extraction and nonlinear transformation on the multi-source training data through the multi-layer hidden layers of the multi-layer perceptron to obtain first geological feature vector data; Inputting the multi-source training data into the residual block of the residual network, performing data transformation through the residual block, and performing residual link operation on the feature vector after data change processing to obtain second geological feature vector data; Dividing the multi-source training data into image block vectors, performing self-attention mechanism calculation according to the image block vectors through the self-attention mechanism module of the visual model, and processing the calculation results using a multi-layer encoder to obtain third geological feature vector data; Based on the attention mechanism, feature fusion processing is performed on the first geological feature vector data, the second geological feature vector data and the third geological feature vector data to obtain a geological feature fusion vector; Prediction calculation is performed based on the geological feature fusion vector, the loss value between the predicted value and the true value is calculated through the binary cross entropy loss function, and the weight parameters and bias parameters of each sub-model are continuously optimized according to the loss value until the loss value converges to obtain the trained geological disaster warning model.

5. The geological disaster early warning method according to claim 1, characterized in that: The step of collecting the new multi-source data in real time through the data collection approach, and using the geological disaster early warning model to perform real-time analysis and prediction on the new multi-source data to obtain geological disaster early warning result data includes: Collect the new multi-source data in real time through the data collection approach, and use wavelet transform to remove noise data in the new multi-source data; Using a preset deep learning-based anomaly detection model to identify outliers in the new multi-source data, and repairing the outliers using a preset machine learning-based interpolation method; The new multi-source data after data preprocessing is sent to the geological disaster early warning model, and the geological disaster early warning model is used to perform real-time analysis and prediction to obtain geological disaster early warning result data; The geological disaster early warning model is trained and optimized online using the new multi-source data through a preset stochastic gradient descent algorithm.

6. The geological disaster early warning method according to claim 3, characterized in that: Before the step of calculating the attention weights of the multi-dimensional data features to obtain a multi-source training data set, the step further includes: Acquire location information data in the multi-source data set; The position information data is used to calculate using a preset position coding formula to obtain a position coding feature vector; The position coding feature vector and the multi-dimensional data feature are subjected to feature coding fusion processing.

7. The geological disaster early warning method according to claim 4, characterized in that: Before the step of performing prediction calculation according to the geological feature fusion vector and calculating the loss value between the predicted value and the true value by using a binary cross entropy loss function, the method further includes: During the training process, a preset first number of rounds are first trained at a first learning rate; When the training process of the first number of rounds is completed, the training is continued with a second learning rate, wherein a regularization term is added to the loss function by a regularization method during the training process, the first learning rate is less than the second learning rate, and the second learning rate decreases according to a preset number of training rounds.

8. A geological disaster early warning device, characterized in that: include: A data collection preprocessing module is used to collect multi-source data of preset data dimensions through a preset data collection path, and perform data preprocessing on the collected multi-source data to obtain preprocessed multi-source data; A training data set module, used to construct a multi-source data set including positive sample data and negative sample data using the pre-processed multi-source data, and to obtain a multi-source training data set by fusing multi-dimensional data features extracted from the multi-source data set through an attention mechanism; A geological disaster warning model module is used to construct an initial geological disaster warning model using a multi-layer perceptron, a residual network and a visual model, and train the initial geological disaster warning model through the multi-source training data set to obtain a trained geological disaster warning model; A geological disaster prediction result module is used to collect the new multi-source data in real time through the data collection path, and use the geological disaster early warning model to perform real-time analysis and prediction on the new multi-source data to obtain geological disaster early warning result data; The geological disaster release and rescue module is used to release graded warning information through preset channels when the geological disaster warning result data reaches the preset warning conditions, and use the pre-trained rescue resource intelligent deployment model to generate a rescue resource deployment plan based on the geological disaster warning result data.

9. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the geological disaster warning method described in any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that: The non-volatile computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by an electronic device, the electronic device executes the geological disaster early warning method described in any one of claims 1-7.

Images