Comprehensive advanced forecasting method and system for water inrush and mud inrush disaster sources

Abstract

The invention provides a comprehensive advanced forecasting method and system for water inrush and mud inrush disaster sources. Belongs to the field of tunnel construction exploration. The system comprises a data collection module, a feature extraction module, a space-time graph neural network (ST-GNN) model, a model training module and an early warning module, a large amount of geological, climate and hydrological data are collected, feature extraction is performed by using a self-encoding neural network, the ST-GNN model is constructed, model training is performed, and a prevention and early warning decision is made. The system is helpful for improving the prevention and coping capacity for water inrush and mud inrush disasters. Particularly, all data are stored on the block chain data sharing platform, so that the security and transparency of the data are ensured, and the reliability and credibility of the data are greatly improved. Generally speaking, the system and the method can accurately predict the time and the place where sudden disasters may occur, give out early warning in time, avoid or reduce the influence of the disasters on life and property of people, and have important practical value.

Description

Technical Field

[0001] The invention belongs to the field of tunnel construction exploration, and more specifically relates to a comprehensive advance prediction method and system for water and mud inrush disaster sources. Background Art

[0002] Water and mud inrush disasters are common and serious geological disasters in the construction of underground projects such as tunnels and mines. They not only cause huge losses of manpower and material resources, but also have a serious impact on the progress of project construction. Such disasters usually occur suddenly and are highly harmful, posing a great threat to people's lives and property safety.

[0003] According to existing research, the causes of sudden water and mud disasters are multifaceted, including natural factors such as geological structure, lithology, rock formation inclination, groundwater conditions, and human factors such as construction methods, excavation speed, and support methods. However, the current prediction and early warning of sudden water and mud disasters that may occur during construction mainly rely on geological surveys and on-site monitoring, combined with experience for analysis, which has problems such as low prediction accuracy and long response time.

[0004] In addition, the comprehensive processing and analysis of complex data information such as groundwater pressure conditions, distribution of fractured zones, and various engineering parameters in actual projects requires a lot of manpower and material resources, and is prone to misjudgment or omission. Some researchers have also used physical methods such as seismology, acoustic emission, and electromagnetic waves to detect tunnel surrounding rocks, but the timeliness and accuracy of these technologies for disaster warnings still need to be improved.

[0005] In addition, in the early warning research of sudden geological disasters, we have not seen the use of high-tech such as blockchain to store early warning information to ensure the accuracy and security of data. In the process of disaster early warning, the accuracy and objectivity of information are crucial. The distributed storage and tamper-proof characteristics of blockchain make it potential in ensuring the accuracy and reliability of early warning information.

[0006] Therefore, how to use information technology to conduct in-depth prediction and early warning of water and mud sudden disasters is a scientific problem that needs to be solved urgently. Summary of the invention

[0007] The technical problem to be solved by the present invention is how to use information technology, physical detection methods and data analysis models to accurately warn and predict sudden water and mud disasters, thereby improving the timeliness and accuracy of the warning, helping relevant departments and personnel to make emergency preparations and avoid risks in advance, and reducing casualties and property losses.

[0008] In order to achieve the above object, the present invention is implemented by adopting the following technical scheme: the method comprises: Data collection: using seismic wave detection, electromagnetic field detection, and underground radar to collect construction tunnel geology, previous rainfall, and flood data; Use blockchain technology to establish a data sharing platform to ensure data security and transparency; Feature extraction: collect geological, climate, and hydrological data to find key factors and patterns for predicting disasters; Model construction: Based on the results of data processing, the occurrence patterns of water and mud inrush disasters are derived and a prediction model is constructed; Model training: Train the model using historical data, including both normal and abnormal data; Implement early warning: After the model is trained, forecasts and early warnings are carried out.

[0009] In one embodiment, the data collection includes: Quantify data related to the sources of water and mud inrush disasters, including geological conditions, climate conditions, and monitoring data; Use seismic wave detection and electromagnetic field detection to conduct geological surveys in various places to obtain underground structure images and stratigraphic distribution characteristics.

[0010] In one embodiment, the establishment of a data sharing platform includes: Determine the platform architecture: a platform framework that uses blockchain technology to record and share data; define and decompose the functions of each part; A data structure designed to store and process data. The data structure includes a block header, which contains the hash value of the previous block, a timestamp, and the difficulty information of the block; a block body, which contains a list of transactions; each block is added to the blockchain in chronological order; Data encryption protection uses hash functions to ensure data integrity.

[0011] In one embodiment, the feature extraction uses an autoencoder neural network, including: Build the model: determine the structure of the autoencoder neural network, including the input layer, hidden layer, and output layer; the number of nodes in the hidden layer is less than that in the input layer to achieve data compression and abstract representation; Input data: Input data on water and mud inrush disasters, including geological, climate and hydrological data; Encoding process: During the encoding process, the input data is converted into a low-dimensional encoding through the hidden layer. This process is achieved through the following formula: ; Among them, σ is the activation function, W is the weight matrix, b is the bias, and x is the input data; Decoding process: In the decoding process, the code is converted back into the representation of the original data, which can be achieved by the following formula: ; Among them, σ is the activation function, W' is the decoding weight matrix, b' is the decoding bias, h is the encoding, and y is the decoded data.

[0012] The model is trained by minimizing the difference between the original data and the decoded data, using the mean square error as the loss function: Find key factors and patterns: A trained model can convert input data into low-dimensional codes, which represent abstract representations or key features of the data. By analyzing the codes, the key factors and patterns for predicting water and mud sudden disasters can be found.

[0013] In one embodiment, the model building includes: By improving the deep neural network (DNN) and combining it with the idea of ​​graph neural network (GNN), a model suitable for processing spatial and temporal features is designed. This model can better capture the complex relationships in water and mud inrush disasters.

[0014] Improved model: Spatial-Temporal Graph Neural Network (ST-GNN); (1) Model architecture; GNN is used to process spatial features and model geographic location and underground structure information as a graph structure; Combined with LSTM to capture time features and process time series data such as rainfall and groundwater level changes; (2) Model implementation steps; The water and mud disaster area is modeled as a graph. ,in is a set of nodes (monitoring points), is the edge set (geological connections between adjacent nodes); assign a feature vector to each node (geological features, historical data); Message passing mechanism: Message passing is performed on the graph to update node features; ; ; in, Is a node The set of neighbor nodes of Represents the neighbor nodes of node i Aggregate the information obtained; Represents the final feature of generating node i at layer k; represents the hidden feature vector of node j at the k−1 layer; is the feature representation of node i in the k-1th layer of the graph neural network; AGGREGATE(): is an aggregation function used to summarize the feature information of neighboring nodes; UPDATE(): is an update function used to update the hidden features of the node; Time Series Processing: Using LSTM to process time series data: ,in, is the input feature at time t, is the hidden state, represents the hidden state of the LSTM at time step t−1; it contains a compressed representation of all the input information from time step 1 to t−1 and is used as part of the input at time step t; Feature fusion and prediction: Fusion of the output features of the graph neural network and LSTM: ; in, Output features of the graph neural network layer, Output features of the LSTM layer, It is the fused feature vector. CONCAT(): feature concatenation operation, which concatenates feature vectors from different sources. Finally, prediction is made through the fully connected layer: ; in, The model’s predicted output; Activation function, The weight matrix of the fully connected layer, The bias vector of the fully connected layer; Model Validation and Evaluation: Evaluate the performance of the model using cross-validation and performance metrics.

[0015] In one embodiment, the model training includes: The training process of the spatial-temporal graph neural network (ST-GNN) includes data preparation, model initialization, forward propagation, loss calculation, back propagation, and parameter update: (1) Data preparation; (2) Model initialization Initialize the parameters of the graph neural network (GNN) and long short-term memory network (LSTM), including weights and biases; (3) Forward propagation Message passing and updating of graph neural network layer (GNN): updating the embedding vector of the node through the graph neural network layer; ; ; in, Is a node The set of neighbor nodes of Represents the neighbor nodes of node i Aggregate the information obtained; Represents the final feature of generating node i at layer k; represents the hidden feature vector of node j at the k−1 layer; is the feature representation of node i in the k-1th layer of the graph neural network; AGGREGATE(): is an aggregation function used to summarize the feature information of neighboring nodes; UPDATE(): is an update function used to update the hidden features of the node; The aggregation function (AGGREGATE) includes sum, average or maximum, and the update function (UPDATE) is the nonlinear ReLU transformation; Time series processing (LSTM); LSTM unit calculation: For each time step t, calculate the hidden state of the LSTM; ; ; ; ; ; ; in, are the forget gate, input gate, and output gate, is the cell state, is a hidden state; , , , They are the weight matrices of the forget gate, input gate, output gate, and candidate memory content; , , , They are the forget gate, input gate, output gate and bias vector of candidate memory content; Candidate memory content, representing the new candidate memory vector, σ(): sigmoid activation function, used to calculate the activation value of the gate; tanh(): hyperbolic tangent activation function, used to calculate the candidate memory content and hidden state; (4) Feature fusion and output: Fusion of the output features of GNN and LSTM; ; in, Output features of the graph neural network layer, The output features of the LSTM layer, It is the fused feature vector, CONCAT(): feature concatenation operation, which concatenates feature vectors from different sources; The final output is calculated through the fully connected layer; ; in, The model’s predicted output; Activation function, The weight matrix of the fully connected layer, The bias vector of the fully connected layer; (5) Loss calculation: Use the loss function to calculate the error between the predicted value and the true value; ; in, Loss function, Sample size, The loss value of a single sample; (6) Backpropagation: Calculate the gradient of the loss function with respect to the model parameters and use the chain rule to backpropagate the error through each layer of the network; (7) Parameter update: update model parameters; ; in, is the learning rate, is the loss function with respect to the parameters The gradient of Updated model parameters; Through the above training process, the ST-GNN model can effectively learn spatial and temporal features and improve the prediction ability of water and mud inrush disasters.

[0016] In one embodiment, the implementation of early warning includes: Make decisions based on the prediction results. If the prediction results exceed the set threshold, it is determined that there is a possibility of sudden water and mud disasters. This threshold is dynamically adjusted according to the actual situation. If the system decides that a sudden water or mud disaster will occur, the system will directly issue early warning information through the Internet, radio or other communication methods, including the time, location and predicted level of the disaster.

[0017] On the other hand, a comprehensive advance prediction system for water and mud sudden disaster sources is provided, the system is applicable to the method described, and the system comprises: The data collection module uses seismic wave detection, electromagnetic field detection, and underground radar to collect construction tunnel geology, previous rainfall, and flood data, and quantifies the collected geological, climate, and hydrological data; The blockchain data sharing platform ensures the security and transparency of the collected data; the feature extraction module uses an autoencoder neural network to compress and abstract data to find the key factors and patterns for predicting disasters; The space-time graph neural network (ST-GNN) model builds a prediction model based on the feature extraction results, combines the geographical location, underground structure information and historical data to capture spatial and temporal features, and captures the complex relationships in predicting water and mud inrush disasters; the model training module trains and updates model parameters through historical data; The early warning module makes decisions based on the model prediction results. If the set threshold is exceeded, it is determined that there is a possibility of sudden water and mud disasters, and early warning information is issued through communication.

[0018] Beneficial effects of the present invention: Data is collected using a variety of technologies such as seismic wave detection, electromagnetic field detection, and underground radar to ensure the comprehensiveness and authenticity of the data. Data is then processed and predicted using autoencoder neural networks and space-time graph neural network (ST-GNN) models, improving the accuracy of early warnings for sudden water and mud disasters.

[0019] By training and optimizing the model, we can ensure that we receive early warning information before a disaster occurs, so that we can make emergency preparations in advance and avoid or reduce losses.

[0020] The system can monitor in real time, automatically execute early warning judgment and information release. Users only need to understand the early warning information and act accordingly, which greatly reduces the difficulty of operation.

[0021] It is suitable for tunnel construction, mining and other underground projects, and has wide practicality in preventing and reducing sudden water and mud disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of the method of the present invention; Figure 2 Constructing a flow chart for the prediction model of the present invention; Figure 3 This is a flow chart of the prediction model training of the present invention. DETAILED DESCRIPTION

[0023] In order to facilitate understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0024] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used in the present invention in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0025] Embodiment 1: Figure 1 As shown, a comprehensive advance prediction method for water and mud inrush disaster sources is characterized in that the method comprises: S1. Data collection: Collect construction tunnel geology, previous rainfall, and flood data. Add technologies such as seismic wave detection, electromagnetic field detection, and underground radar to gain in-depth understanding and real-time monitoring of possible disaster geology. Use intelligent sensing devices to monitor the actual situation and provide real-time updated data.

[0026] Quantify the data related to the sources of sudden water and mudslide disasters, including geological conditions (such as stratigraphic structure, lithology, rock formation inclination, etc.), climatic conditions (including previous rainfall, floods), monitoring data, etc.

[0027] Use technologies such as seismic wave detection, electromagnetic field detection, and underground radar to conduct geological surveys in various locations to obtain underground structure images and stratigraphic distribution characteristics, thereby further understanding possible geological disasters.

[0028] Intelligent sensing devices and related monitoring equipment are deployed in high-risk areas. These hardware devices can collect data on geological conditions, climate change, etc. in real time.

[0029] S2. Establish a sharing platform: Use blockchain technology to establish a data sharing platform to ensure data security and transparency, and respond to and prevent disasters quickly and accurately.

[0030] Determine the platform architecture: Based on the characteristics of decentralization, determine the platform framework for data recording and sharing using blockchain technology. Define and decompose the functions of each part, such as data submission, recording, verification, sharing, etc.

[0031] Design consensus mechanism: According to the system requirements, the consensus mechanism ensures the consistency of data among all nodes and is also the key to maintaining the security of the blockchain.

[0032] A data structure designed to store and process data. A typical blockchain data structure includes a block header (containing the hash value of the previous block, timestamp, difficulty of the block, etc.) and a block body (containing a list of transactions). Each block is added to the blockchain in chronological order.

[0033] Security assurance: including data encryption protection and using hash functions to ensure data integrity. Typical hash functions include SHA-256.

[0034] like Figure 2 As shown, S3, feature extraction: Use deep learning neural networks to process the collected nonlinear, high-dimensional and various types of geological, climate, hydrological and other data to find the key factors and patterns for predicting disasters.

[0035] S301, constructing a model: determining the structure of the autoencoder neural network, which generally includes an input layer, a hidden layer, and an output layer. The number of nodes in the hidden layer is less than that in the input layer, so as to achieve data compression and abstract representation.

[0036] S302. Input data: input the pre-processed and integrated data related to water and mud sudden disasters, including geological, climate, hydrological and other data.

[0037] S303, encoding process: In the encoding process, the input data is converted into low-dimensional encoding through the hidden layer. This process can be achieved by the following formula: h = σ(Wx + b); Among them, σ is the activation function, W is the weight matrix, b is the bias, and x is the input data.

[0038] S304, decoding process: In the decoding process, the code is converted back into the representation of the original data, which can be achieved by the following formula: y = σ(W'h + b'); Among them, σ is the activation function, W' is the decoding weight matrix, b' is the decoding bias, h is the encoding, and y is the decoded data.

[0039] S305. Train the model by minimizing the difference between the original data and the decoded data, using mean square error as the loss function: L(x, y) = ||x - y||^2; Among them, L(x, y) is the loss function, which represents the difference between the input data x and the decoded data y; ||x - y|| 2 This is the mean squared error between x and y.

[0040] Finding key factors and patterns: A trained model can convert input data into low-dimensional codes, which represent abstract representations or key features of the data. By analyzing these codes, key factors and patterns for predicting water and mud inrush disasters can be found.

[0041] S4. Model construction: Based on the results of data processing, the occurrence patterns that may cause sudden water and mud disasters are derived and a prediction model is constructed.

[0042] The model construction includes: By improving the deep neural network (DNN) and combining it with the idea of ​​graph neural network (GNN), a model suitable for processing spatial and temporal features is designed. This model can better capture the complex relationships in water and mud inrush disasters.

[0043] Improved model: Spatial-Temporal Graph Neural Network (ST-GNN); Model architecture,Graph Neural Network (GNN): GNN is used to process spatial features and model information such as geographic location and underground structure as a graph structure.Long Short-Term Memory Network (LSTM): LSTM is combined to capture temporal features and process time series data such as rainfall and groundwater level changes.

[0044] The model implementation steps are as follows: (1) Graph construction; The water and mud disaster area is modeled as a graph. ,in is a set of nodes (such as monitoring points), is the edge set (e.g., geological connections between adjacent nodes). Each node is assigned a feature vector (e.g. geological features, historical data).

[0045] (2) Graph neural network layer; Message passing mechanism: Message passing is performed on the graph to update node features; ; ; in, Is a node The set of neighbor nodes of Represents the neighbor nodes of node i Aggregate the information obtained; Represents the final feature of generating node i at layer k; represents the hidden feature vector of node j at the k−1 layer; is the feature representation of node i in the k-1th layer of the graph neural network; AGGREGATE(): is an aggregation function used to summarize the feature information of neighboring nodes; UPDATE(): is an update function used to update the hidden features of the node.

[0046] (3) Time series processing - Use LSTM to process time series data: ; in, is the input feature at time t, is a hidden state; represents the hidden state of the LSTM at time step t−1; it contains a compressed representation of all the input information from time step 1 to t−1, and is used as part of the input at time step t.

[0047] (4) Feature fusion and prediction: Fusion of the output features of the graph neural network and LSTM: ; in, Output features of the graph neural network layer, Output features of the LSTM layer, It is the fused feature vector. CONCAT(): feature concatenation operation, which concatenates feature vectors from different sources. Finally, prediction is made through the fully connected layer: ; in, The model’s predicted output; Activation function, The weight matrix of the fully connected layer, The bias vector of the fully connected layer; Model Validation and Evaluation: Use cross-validation and performance metrics (such as accuracy, F1 score) to evaluate the performance of the model.

[0048] S5. Model training: Verify and train the model through historical data, including data of normal and abnormal situations.

[0049] like Figure 3 As shown in Figure 1, the training process of the spatial-temporal graph neural network (ST-GNN) involves multiple steps, including data preparation, model initialization, forward propagation, loss calculation, back propagation, and parameter update. The following is a detailed training process, including relevant mathematical formulas: S501, model initialization: Initialize the parameters of the graph neural network (GNN) and long short-term memory network (LSTM), including weights and biases.

[0050] S502, forward propagation (1) Graph Neural Network Layer (GNN); Message passing and updating: updating the embedding vector of a node through the graph neural network layer; ; ; in, Is a node The set of neighbor nodes of Represents the neighbor nodes of node i Aggregate the information obtained; Represents the final feature of generating node i at layer k; represents the hidden feature vector of node j at the k−1 layer; is the feature representation of node i in the k-1th layer of the graph neural network; AGGREGATE(): is an aggregation function used to summarize the feature information of neighboring nodes; UPDATE(): is an update function used to update the hidden features of the node.

[0051] Aggregation functions include sum, average, or maximum. Update functions are usually nonlinear transformations, such as ReLU.

[0052] (2) Time series processing (LSTM); LSTM unit calculation: For each time step t, calculate the hidden state of the LSTM; ; ; ; ; ; ; in, are the forget gate, input gate, and output gate, is the cell state, is a hidden state; , , , They are the weight matrices of the forget gate, input gate, output gate, and candidate memory content; , , , They are the forget gate, input gate, output gate and bias vector of candidate memory content; Candidate memory content, representing the new candidate memory vector, σ(): sigmoid activation function, used to calculate the activation value of the gate; tanh(): hyperbolic tangent activation function, used to calculate the candidate memory content and hidden state.

[0053] (3) Feature fusion and output; Feature fusion: Fusion of the output features of GNN and LSTM; ; in, Output features of the graph neural network layer, Output features of the LSTM layer, It is the fused feature vector, CONCAT(): feature connection operation, which concatenates feature vectors from different sources.

[0054] Prediction output: Calculate the final output through the fully connected layer; ; in, The model’s predicted output; Activation function, The weight matrix of the fully connected layer, Bias vector for the fully connected layer.

[0055] S503, loss calculation: Use an appropriate loss function (such as mean square error or cross entropy) to calculate the error between the predicted value and the true value.

[0056] ; in, Loss function, Sample size, The loss value for a single example.

[0057] S504, back propagation, calculate the gradient of the loss function with respect to the model parameters. Use the chain rule to back propagate the error through each layer of the network; S505: Parameter update Update model parameters using an optimization algorithm such as Adam or SGD; ;

[0058] in, is the learning rate, is the loss function with respect to the parameters The gradient of Updated model parameters;

[0059] Evaluate model performance on the validation set and adjust hyperparameters to improve the generalization ability of the model.

[0060] Through the above training process, the ST-GNN model can effectively learn spatial and temporal features, thereby improving the prediction ability of water and mud inrush disasters.

[0061] S6. Implement early warning: After the model is trained, forecasts and early warnings are carried out. If a sudden water or mud disaster is predicted to occur, relevant personnel or departments are notified in a timely manner to take measures in advance to prevent or reduce the impact of the disaster.

[0062] Decision-making stage: After training and evaluating the model, the system will make a decision based on the prediction results. If the prediction results exceed the set threshold, it is determined that there is a possibility of sudden water and mud disasters. In actual applications, this threshold needs to be dynamically adjusted according to the actual situation to ensure the accuracy and timeliness of the warning.

[0063] Warning release: If the system decides that a sudden water or mud disaster may occur, the system will directly release warning information through the Internet, radio or other communication methods, including the time, location and predicted disaster level of the possible disaster.

[0064] Receiving early warning: The personnel who receive early warning include workers at the tunnel construction site, as well as geological engineers, safety engineers, etc. Upon receiving the early warning signal, these personnel shall immediately stop working and evacuate the construction site.

[0065] Response to warning: After receiving the warning, the responsible personnel at the construction site should immediately organize the evacuation of personnel and report the warning information to the relevant management department. The management department will take corresponding emergency measures based on the warning information, such as activating the emergency response mechanism, mobilizing relevant equipment and personnel, and preparing for emergency repairs.

[0066] The technical solution of the present invention integrates multiple monitoring technologies such as seismic wave detection, electromagnetic field detection, underground radar, etc., and finds out the key factors and patterns of water and mud sudden disasters by collecting and analyzing multiple data such as construction tunnel geology, rainfall and flood data. In addition, blockchain technology is used to ensure the security and transparency of various types of data, greatly improving the utilization efficiency and value of data, and also ensuring the authenticity and credibility of the data source to a certain extent.

[0067] The main innovative points of the present invention are: 1. Use auto-encoding neural network to extract data features, ensuring the comprehensiveness and effectiveness of feature extraction; 2. Use the space-time graph neural network (ST-GNN) model for model training, which can better capture the spatial and temporal characteristics of water and mud inrush disasters and improve the accuracy of prediction; 3. Data collection and data sharing use blockchain technology, which improves data security and transparency and ensures data authenticity; 4. By setting dynamic thresholds for early warning, the timeliness and accuracy of disaster warnings are improved.

[0068] Embodiment 2: Based on the method described in Embodiment 1, Embodiment 2 is constructed: A comprehensive advance prediction system for water and mud sudden disaster sources, comprising: a data collection module, which uses seismic wave detection, electromagnetic field detection, and underground radar to collect construction tunnel geology, previous rainfall, and flood data, and quantifies the collected geological, climate, and hydrological data; a blockchain data sharing platform, which ensures the security and transparency of the collected data; a feature extraction module, which uses an autoencoder neural network to compress and abstractly represent data, and find out the key factors and patterns for predicting disasters; a space-time graph neural network (ST-GNN) model, which constructs a prediction model based on feature extraction results, combines geographic location, underground structure information, and historical data to capture spatial and temporal features, and captures complex relationships in predicting water and mud sudden disasters; a model training module, which trains and updates model parameters through historical data; an early warning module, which makes decisions based on model prediction results, and determines that a water and mud sudden disaster is likely to occur if the set threshold is exceeded, and issues early warning information through communication.

[0069] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0070] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A comprehensive advance prediction method for water and mud inrush disaster sources, characterized by: The method includes: Data collection: using seismic wave detection, electromagnetic field detection, and underground radar to collect construction tunnel geology, previous rainfall, and flood data; Use blockchain technology to establish a data sharing platform to ensure data security and transparency; Feature extraction: collect geological, climate, and hydrological data to find key factors and patterns for predicting disasters; Model construction: Based on the results of data processing, the occurrence patterns of water and mud inrush disasters are derived and a prediction model is constructed; Model training: Train the model using historical data, including both normal and abnormal data; Implement early warning: After the model is trained, forecasts and early warnings are carried out.

2. The comprehensive advance prediction method for water and mud inrush disaster sources according to claim 1 is characterized by: The data collection includes: Quantify data related to the sources of water and mud inrush disasters, including geological conditions, climate conditions, and monitoring data; Use seismic wave detection and electromagnetic field detection to conduct geological surveys in various places to obtain underground structure images and stratigraphic distribution characteristics.

3. The comprehensive advance prediction method for water and mud inrush disaster sources according to claim 1 is characterized by: The establishment of the data sharing platform includes: Determine the platform architecture: a platform framework that uses blockchain technology to record and share data; define and decompose the functions of each part; A data structure designed to store and process data. The data structure includes a block header, which contains the hash value of the previous block, a timestamp, and the difficulty information of the block; a block body, which contains a list of transactions; each block is added to the blockchain in chronological order; Data encryption protection uses hash functions to ensure data integrity.

4. The comprehensive advance prediction method for water and mud inrush disaster sources according to claim 1 is characterized by: The feature extraction adopts an autoencoder neural network, including: Build the model: determine the structure of the autoencoder neural network, including the input layer, hidden layer, and output layer; the number of nodes in the hidden layer is less than that in the input layer to achieve data compression and abstract representation; Input data: Input data on water and mud inrush disasters, including geological, climate and hydrological data; Encoding process: During the encoding process, the input data is converted into a low-dimensional encoding through the hidden layer. This process is achieved through the following formula: ； Among them, σ is the activation function, W is the weight matrix, b is the bias, and x is the input data; Decoding process: In the decoding process, the code is converted back into the representation of the original data, which can be achieved by the following formula: ； Among them, σ is the activation function, W' is the decoding weight matrix, b' is the decoding bias, h is the code, and y is the decoded data; The model is trained by minimizing the difference between the original data and the decoded data, using the mean square error as the loss function: Find key factors and patterns: A trained model can convert input data into low-dimensional codes, which represent abstract representations or key features of the data. By analyzing the codes, the key factors and patterns for predicting water and mud sudden disasters can be found.

5. The comprehensive advance prediction method for water and mud inrush disaster sources according to claim 1 is characterized by: The model construction includes: Improve the deep neural network (DNN) and combine it with the idea of ​​graph neural network (GNN) to design a model suitable for processing spatial and temporal features. This model can better capture the complex relationships in water and mud inrush disasters. Improved model: Spatial-Temporal Graph Neural Network (ST-GNN); (1) Model architecture; GNN is used to process spatial features and model geographic location and underground structure information as a graph structure; Combined with LSTM to capture time features and process time series data such as rainfall and groundwater level changes; (2) Model implementation steps; The water and mud disaster area is modeled as a graph. in is a set of nodes (monitoring points), is the edge set (geological connections between adjacent nodes); assign a feature vector to each node (geological features, historical data); Message passing mechanism: Message passing is performed on the graph to update node features; ； ； in, Is a node The set of neighbor nodes of Represents the neighbor nodes of node i Aggregate the information, represents the final feature of generating node i at layer k, represents the hidden feature vector of node j at the k−1 layer, is the feature representation of node i in the k-1th layer of the graph neural network; AGGREGATE(): is an aggregation function used to summarize the feature information of neighboring nodes; UPDATE(): is an update function used to update the hidden features of the node; Time Series Processing: Using LSTM to process time series data: ,in, is the input feature at time t, is the hidden state, represents the hidden state of the LSTM at time step t−1; it contains a compressed representation of all the input information from time step 1 to t−1 and is used as part of the input at time step t; Feature fusion and prediction: Fusion of the output features of the graph neural network and LSTM: ； in, Output features of the graph neural network layer, The output features of the LSTM layer, It is the fused feature vector. CONCAT(): feature concatenation operation, which concatenates feature vectors from different sources. Finally, prediction is made through the fully connected layer: ； in, The model’s predicted output; Activation function, The weight matrix of the fully connected layer, The bias vector of the fully connected layer; Model Validation and Evaluation: Evaluate the performance of the model using cross-validation and performance metrics.

6. The comprehensive advance prediction method for water and mud inrush disaster sources according to claim 1 is characterized by: The model training includes: The training process of the spatial-temporal graph neural network (ST-GNN) includes data preparation, model initialization, forward propagation, loss calculation, back propagation, and parameter update: (1) Data preparation; (2) Model initialization; Initialize the parameters of the graph neural network (GNN) and long short-term memory network (LSTM), including weights and biases; (3) Forward propagation; Message passing and updating of graph neural network layer (GNN): updating the embedding vector of the node through the graph neural network layer; ； ； in, Is a node The set of neighbor nodes of Represents the neighbor nodes of node i Aggregate the information, represents the final feature of generating node i at layer k, represents the hidden feature vector of node j at the k−1 layer, It is the feature representation of node i in the k-1th layer of the graph neural network; the aggregation function AGGREGATE is used, including summation, average or maximum value, and the update function UPDATE is a nonlinear ReLU transformation; Time series processing LSTM unit calculation: For each time step t, calculate the hidden state of the LSTM; ； ； ； ； ； ； in, are the forget gate, input gate, and output gate, is the cell state, is a hidden state; , , , They are the weight matrices of the forget gate, input gate, output gate, and candidate memory content; , , , They are the forget gate, input gate, output gate and bias vector of candidate memory content; Candidate memory content, representing the new candidate memory vector, σ(): sigmoid activation function, used to calculate the activation value of the gate; tanh(): hyperbolic tangent activation function, used to calculate the candidate memory content and hidden state; (4) Feature fusion and output: Fusion of the output features of GNN and LSTM; ； in, Output features of the graph neural network layer, The output features of the LSTM layer, It is the fused feature vector. CONCAT(): feature concatenation operation, which concatenates feature vectors from different sources. The final output is calculated through the fully connected layer; ； in, The model’s predicted output; Activation function, The weight matrix of the fully connected layer, The bias vector of the fully connected layer; (5) Loss calculation: Use the loss function to calculate the error between the predicted value and the true value; ； in, Loss function, Sample size, The loss value of a single sample; (6) Backpropagation: Calculate the gradient of the loss function with respect to the model parameters and use the chain rule to backpropagate the error through each layer of the network; (7) Parameter update: update model parameters; ； in, is the learning rate, is the loss function with respect to the parameters The gradient of Updated model parameters; Through the above training process, the ST-GNN model can effectively learn spatial and temporal features and improve the prediction ability of water and mud inrush disasters.

7. The comprehensive advance prediction method for water and mud inrush disaster sources according to claim 1 is characterized by: The implementation of early warning includes: Make decisions based on the prediction results. If the prediction results exceed the set threshold, it is determined that there is a possibility of sudden water and mud disasters. This threshold is dynamically adjusted according to the actual situation. If the system decides that a sudden water or mud disaster will occur, the system will directly issue early warning information through the Internet, radio or other communication methods, including the time, location and predicted level of the disaster.

8. A comprehensive advance prediction system for water and mud inrush disaster sources, the system being applicable to the method according to any one of claims 1 to 7, characterized in that: The system comprises: The data collection module uses seismic wave detection, electromagnetic field detection, and underground radar to collect construction tunnel geology, previous rainfall, and flood data, and quantifies the collected geological, climate, and hydrological data; The blockchain data sharing platform ensures the security and transparency of the collected data; the feature extraction module uses an autoencoder neural network to compress and abstract data to find the key factors and patterns for predicting disasters; The space-time graph neural network (ST-GNN) model builds a prediction model based on the feature extraction results, combines the geographical location, underground structure information and historical data to capture spatial and temporal features, and captures the complex relationships in predicting water and mud inrush disasters; the model training module trains and updates model parameters through historical data; The early warning module makes decisions based on the model prediction results. If the set threshold is exceeded, it is determined that there is a possibility of sudden water and mud disasters, and early warning information is issued through communication.

Images