Environmental event dynamic risk assessment method and system based on Bayesian network

Abstract

The invention discloses an environmental event dynamic risk assessment method and system based on a Bayesian network, belongs to the technical field of artificial intelligence, and aims to solve the technical problems that in existing environmental risk assessment, dynamic data adaptability is poor, multi-source heterogeneous data fusion is difficult, and real-time performance is insufficient. Comprising the following steps: acquiring environment data through a sensor cluster deployed in an environment, and uploading the environment data to an edge computing node; performing data preprocessing on the environmental data through the edge computing node; extracting time series data fragments from the standardized data stream based on a predefined sliding time window; constructing a risk prediction model based on the dynamic Bayesian network; and taking the extracted time series data fragments as input, performing risk level analysis through a risk prediction model in combination with particle filter reasoning, predicting and outputting a risk level, a risk level probability value and a risk conduction path as prediction results, and constructing a visual risk conduction map based on the prediction results.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method and system for dynamic risk assessment of environmental events based on Bayesian networks. Background Technology

[0002] Static risk assessment models (such as the analytic hierarchy process) cannot handle the temporal changes in environmental parameters; machine learning models (such as SVM and random forest) rely on historical data for training and are difficult to adapt to sudden abnormal events; existing Bayesian network applications are mostly limited to static causal reasoning and have not solved the problem of real-time data stream updates.

[0003] The poor adaptability of dynamic data, the difficulty in integrating multi-source heterogeneous data, and the lack of real-time performance in existing environmental risk assessments are technical problems that need to be solved. Summary of the Invention

[0004] The technical objective of this invention is to address the above-mentioned shortcomings by providing a method and system for dynamic risk assessment of environmental events based on Bayesian networks, thereby solving the technical problems of poor adaptability of dynamic data, difficulty in fusion of multi-source heterogeneous data, and insufficient real-time performance in existing environmental risk assessments.

[0005] In a first aspect, the present invention provides a dynamic risk assessment method for environmental events based on Bayesian networks, characterized by comprising the following steps:

[0006] Data Acquisition: Environmental data is collected by a cluster of sensors deployed in the environment and uploaded to edge computing nodes. The environmental data is time-series data.

[0007] Data processing: Environmental data is preprocessed through edge computing nodes, and normalization and missing value correction are performed through data preprocessing to obtain a standardized data stream. The standardized data stream is then updated to the historical event database.

[0008] Data segmentation: Extracting time-series data segments from a standardized data stream based on a predefined sliding time window;

[0009] Network Model Construction: A risk prediction model is constructed based on a dynamic Bayesian network. The risk prediction model defines a node set, a dynamic edge set, and a conditional probability table. Environmental data is set as nodes, and the time dependencies between nodes are set as dynamic edges. The conditional probability table describes the probability distribution of a node under the condition of its parent node taking a pointer. The risk prediction model is used to analyze the risk level based on the environmental data as input, and outputs the risk level, the risk level probability value, and the risk transmission path.

[0010] Risk prediction: Taking the extracted time series data segments as input, the risk prediction model is used to analyze the risk level by combining particle filter inference. The predicted output includes the risk level, the probability value of the risk level, and the risk transmission path. Based on the prediction results, a visualized risk transmission map is constructed.

[0011] As a preferred option, environmental data includes water quality data, meteorological data, and geological displacement data.

[0012] As a preferred approach, for the risk prediction model, the time dependency between nodes is calculated using the mutual information algorithm, which is used to calculate the distance between nodes.

[0013] As a preferred approach, the EM algorithm is used to optimize parameters and a time decay factor is introduced during the training of the conditional probability table.

[0014] As a preferred approach, when performing risk level analysis using a risk prediction model combined with particle filtering inference, the probability after a node in the risk prediction model is approximated by weighted statistics of the particles. The risk value calculation formula is as follows:

[0015]

[0016] in, W represents the probability of the current state of node i. The probability of the current state of node i is the posterior probability, which can be obtained through particle filtering and Bayesian inference. i The hazard weight of node i is determined by data in the historical event database and reflects the degree to which the node contributes to the risk.

[0017] Secondly, the present invention provides a dynamic risk assessment system for environmental events based on Bayesian networks, comprising a data acquisition module, a data processing module, a data segmentation module, a network model construction module, and a risk prediction module.

[0018] The data acquisition module is used to perform the following operations: collect environmental data through a sensor cluster deployed in the environment, and upload the environmental data to the edge computing node, wherein the environmental data is time-series data;

[0019] The data processing module is used to perform the following operations: preprocess environmental data through edge computing nodes, perform normalization and missing value modification through data preprocessing to obtain a standardized data stream, and update the standardized data stream to the historical event database;

[0020] The data segmentation module is used to perform the following operations: extract time-series data segments from a standardized data stream based on a predefined sliding time window;

[0021] The network model construction module is used to perform the following operations: construct a risk prediction model based on a dynamic Bayesian network, and define a node set, a dynamic edge set, and a conditional probability table in the risk prediction model. Here, environmental data is set as nodes, the time dependency relationship between nodes is set as dynamic edges, and the conditional probability table describes the probability distribution of a node under the condition of its parent node taking a pointer. The risk prediction model is used to take environmental data as input, perform risk level analysis based on the environmental data, and output the risk level, risk level probability value, and risk transmission path.

[0022] The risk prediction module is used to perform the following operations: taking the extracted time series data segments as input, performing risk level analysis through the risk prediction model and particle filter inference, predicting and outputting the risk level, risk level probability value and risk transmission path as prediction results, and constructing a visualized risk transmission map based on the prediction results.

[0023] As a preferred option, environmental data includes water quality data, meteorological data, and geological displacement data.

[0024] As a preferred approach, for the risk prediction model, the time dependency between nodes is calculated using the mutual information algorithm, which is used to calculate the distance between nodes.

[0025] As a preferred approach, the EM algorithm is used to optimize parameters and a time decay factor is introduced during the training of the conditional probability table.

[0026] As a preferred approach, when performing risk level analysis using a risk prediction model combined with particle filtering inference, the probability after a node in the risk prediction model is approximated by weighted statistics of the particles. The risk value calculation formula is as follows:

[0027]

[0028] in, W represents the probability of the current state of node i. The probability of the current state of node i is the posterior probability, which can be obtained through particle filtering and Bayesian inference. i The hazard weight of node i is determined by data in the historical event database and reflects the degree to which the node contributes to the risk.

[0029] The dynamic risk assessment method and system for environmental events based on Bayesian networks of the present invention have the following advantages: environmental parameters (such as water quality, meteorological, and geological data) are collected through sensor networks, a dynamic Bayesian network model is constructed, node probabilities are updated in real time using a time-series inference algorithm, risk levels and transmission paths are output, dynamic evolution modeling of risks is achieved through a sliding time window mechanism, uncertain data are fused using evidence theory to improve the robustness of assessment, and a visualized risk transmission map is generated to assist in rapid decision-making. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] The invention will be further described below with reference to the accompanying drawings.

[0032] Figure 1 This is a flowchart of an environmental event dynamic risk assessment method based on Bayesian networks, as shown in Example 1. Detailed Implementation

[0033] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0034] This invention provides a method and system for dynamic risk assessment of environmental events based on Bayesian networks, which addresses the technical problems of poor adaptability of dynamic data, difficulty in fusion of multi-source heterogeneous data, and insufficient real-time performance in existing environmental risk assessments.

[0035] Example 1:

[0036] This invention provides a dynamic risk assessment method for environmental events based on Bayesian networks, comprising five steps: data acquisition, data processing, data segmentation, network model construction, and risk prediction.

[0037] Step S100 Data Acquisition: Collect environmental data through a sensor cluster deployed in the environment and upload the environmental data to the edge computing node. The environmental data is time-series data.

[0038] The environmental data includes water quality data, meteorological data, and geological displacement data.

[0039] In this embodiment, a cluster of meteorological / water quality / geological sensors is deployed to collect parameters such as temperature, pH, and vibration frequency in real time.

[0040] Step S200 Data Processing: Environmental data is preprocessed through edge computing nodes. Normalization and missing value correction are performed through data preprocessing to obtain a standardized data stream. The standardized data stream is then updated to the historical event database.

[0041] Step S300 Data Segmentation: Extract time-series data segments from the standardized data stream based on a predefined sliding time window.

[0042] Step S400 Network Model Construction: A risk prediction model is constructed based on a dynamic Bayesian network. In the risk prediction model, a node set, a dynamic edge set, and a conditional probability table are defined. Environmental data is set as nodes, and the time dependencies between nodes are set as dynamic edges. The conditional probability table describes the probability distribution of a node under the condition of its parent node taking a pointer. The risk prediction model is used to perform risk level analysis based on environmental data as input, and outputs the risk level, risk level probability value, and risk transmission path.

[0043] For the risk prediction model, the time dependency between nodes is calculated using the mutual information algorithm, which is used to calculate the distance between nodes.

[0044] The EM algorithm is used to optimize parameters and a time decay factor is introduced during the training of the conditional probability table.

[0045] The core of Bayesian networks (BN) is to update the probability distribution of nodes through observation data. Dynamic Bayesian networks (DBN), as its temporal extension, require inputs that include time-dependent observation information, specifically node observations and prior knowledge.

[0046] Node observations: Real-time / historical data of environmental risk factors (such as temperature, pH, rainfall, etc. collected by sensors), corresponding to the observations of each node in the DBN (such as "rainfall" and "heavy metal concentration") at different time steps.

[0047] Prior knowledge includes static and dynamic components. The static component includes causal relationships between nodes (e.g., the dependency between rainfall and river level) and the initial conditional probability table (CPT, initialized based on historical data or expert experience). The dynamic component (specific to DBN) includes temporal dependencies between nodes (e.g., the impact of rainfall at time t on the probability of landslide at time t+1) and time decay factors (used for dynamic updates of the CPT, such as the α parameter).

[0048] The output of a Bayesian network is the posterior probability distribution of nodes and the resulting risk assessment results. Because DBN supports temporal reasoning, its output emphasizes dynamism, including the posterior probability of nodes, risk quantification results, and causal transmission paths.

[0049] Posterior probability of a node: The probability distribution of each environmental node (such as "soil pollution level" or "flood risk") at the current time step (e.g., the probability of "flood risk = high" is 70%).

[0050] Risk quantification results: Comprehensive risk value (such as "Risk = Σ(node ​​probability × hazard weight)" in the patent) and risk level (red / orange / yellow / blue) calculated based on node probability.

[0051] Causal transmission path: The dependency chain of key risk factors (such as "heavy rainfall → rising river water level → dam collapse") is traced through the probability propagation path between nodes.

[0052] In this embodiment, the risk prediction model built based on a Bayesian network takes various "observed environmental data" as input, such as real-time sensor readings: today's rainfall, river water pH, mountain vibration amplitude, air pollutant concentration, etc. These specific values ​​or states (e.g., "rainfall > 50mm") can be understood as inputting various real-time environmental signals into an "intelligent discriminator" so that the intelligent discriminator knows the specific situation of the current environment. Its output includes two contents: the current risk result and the risk source. The current risk result is, for example, "the risk level of flash flood is red (high risk)" or "the probability of river pollution spreading is 70%"; the risk source is, for example, "because today's rainfall is too high → causing the mountain to loosen → which may trigger a landslide," that is, finding the key risk transmission path.

[0053] Simply put, this "intelligent judgment device" reads the input environmental data and outputs "how high is the current risk" and "what factors have led to this risk".

[0054] Step S500 Risk Prediction: Using the extracted time-series data segments as input, the risk level is analyzed through a risk prediction model combined with particle filter inference. The predicted output includes the risk level, risk level probability value, and risk transmission path as the prediction results. A visualized risk transmission map is then constructed based on the prediction results.

[0055] When performing risk level analysis using a risk prediction model combined with particle filtering inference, the probability after a node in the risk prediction model is approximated by weighted statistics of the particles. The risk value calculation formula is as follows:

[0056]

[0057] in, W represents the probability of the current state of node i. The probability of the current state of node i is the posterior probability, which can be obtained through particle filtering and Bayesian inference. i The hazard weight of node i is determined by data in the historical event database and reflects the degree to which the node contributes to the risk.

[0058] In this embodiment, particle filtering is the core algorithm for real-time temporal inference in dynamic Bayesian networks. Its essence is to approximate the posterior distribution of the system state through a set of particles.

[0059] The input includes an initial particle set, observation data, a state transition model, and an observation model. The initial particle set represents the possible initial states of the system (e.g., the initial probability distribution of each environmental node), with each particle corresponding to a possible state combination (e.g., "rainfall = 50 mm / h, pH = 6.5"). Observation data refers to sensor data at the current time step (e.g., "wind speed = 10 m / s" and "heavy metal concentration = 0.8 mg / L" at time t), used to adjust particle weights. The state transition model describes the evolution of the system state over time (from the dynamic edges and CPT of the DBN), such as how "rainfall at time t" affects the probability distribution of "soil moisture at time t+1". The observation model represents the probability of observation data occurring given the system state (e.g., "the probability that the sensor observes 'humidity = 90%' when soil moisture = high"), typically defined by the node conditional probabilities of the DBN.

[0060] The output includes an updated particle set, state estimation results, and uncertainty quantification. The updated particle set: After weighting and resampling the observed data, high-weight particles (representing more likely system states) are retained, approximating the posterior distribution of the state at the current time step. State estimation results: The posterior probability of each environmental node calculated based on the particle set (e.g., the probability of "flood risk = high" at time t, determined by the sum of the weights of the corresponding states in the particle set). Uncertainty quantification: The dispersion of the particle set (e.g., the variance of the particle weights), reflecting the reliability of the inference results (the smaller the variance, the more reliable the risk assessment).

[0061] In the method disclosed in this embodiment, a dynamic Bayesian network model is constructed, and the following definitions are made in this network model:

[0062] Node definition: Environmental elements (such as "rainfall" and "heavy metal concentration in rivers") are abstracted into network nodes;

[0063] Dynamic edge generation: Based on a historical event database, the mutual information algorithm is used to calculate the latency dependencies between nodes;

[0064] Conditional Probability Table (CPT) Training: The EM algorithm is used to optimize the parameters and introduces a time decay factor. When there are missing or latent variables in the data, EM can iteratively optimize the CPT parameters.

[0065] The model performs real-time risk assessment, and the specific steps are as follows:

[0066] (1) Input real-time data stream and extract time segments by sliding time windows;

[0067] (2) Update the posterior probability of nodes using approximate inference algorithms (such as particle filtering);

[0068] (3) Risk value calculation: Risk = Σ(node ​​probability × hazard weight).

[0069] (4) Risk visualization:

[0070] Generate a four-color risk heatmap in red, orange, yellow, and blue.

[0071] Mark key transmission paths (e.g., "heavy rain → landslide → river blockage").

[0072] Example 2:

[0073] This invention discloses a dynamic risk assessment system for environmental events based on Bayesian networks, comprising a data acquisition module, a data processing module, a data segmentation module, a network model construction module, and a risk prediction module.

[0074] The data acquisition module is used to perform the following operations: collect environmental data through a cluster of sensors deployed in the environment, and upload the environmental data to the edge computing node, wherein the environmental data is time-series data.

[0075] The environmental data includes water quality data, meteorological data, and geological displacement data.

[0076] In this embodiment, a cluster of meteorological / water quality / geological sensors is deployed to collect parameters such as temperature, pH, and vibration frequency in real time.

[0077] The data processing module performs the following operations: preprocesses environmental data through edge computing nodes, performs normalization and corrects missing values ​​through data preprocessing to obtain a standardized data stream, and updates the standardized data stream to the historical event database.

[0078] The data segmentation module is used to perform the following operation: extract time-series data segments from a standardized data stream based on a predefined sliding time window.

[0079] The network model construction module is used to perform the following operations: construct a risk prediction model based on a dynamic Bayesian network, and define a node set, a dynamic edge set, and a conditional probability table in the risk prediction model. Here, environmental data is set as nodes, the time dependency relationship between nodes is set as dynamic edges, and the conditional probability table describes the probability distribution of a node under the condition of its parent node taking a pointer. The risk prediction model is used to take environmental data as input, perform risk level analysis based on the environmental data, and output the risk level, risk level probability value, and risk transmission path.

[0080] For the risk prediction model, the time dependency between nodes is calculated using the mutual information algorithm, which is used to calculate the distance between nodes.

[0081] The EM algorithm is used to optimize parameters and a time decay factor is introduced during the training of the conditional probability table.

[0082] The core of Bayesian networks (BN) is to update the probability distribution of nodes through observation data. Dynamic Bayesian networks (DBN), as its temporal extension, require inputs that include time-dependent observation information, specifically node observations and prior knowledge.

[0083] Node observations: Real-time / historical data of environmental risk factors (such as temperature, pH, rainfall, etc. collected by sensors), corresponding to the observations of each node in the DBN (such as "rainfall" and "heavy metal concentration") at different time steps.

[0084] Prior knowledge includes static and dynamic components. The static component includes causal relationships between nodes (e.g., the dependency between rainfall and river level) and the initial conditional probability table (CPT, initialized based on historical data or expert experience). The dynamic component (specific to DBN) includes temporal dependencies between nodes (e.g., the impact of rainfall at time t on the probability of landslide at time t+1) and time decay factors (used for dynamic updates of the CPT, such as the α parameter).

[0085] The output of a Bayesian network is the posterior probability distribution of nodes and the resulting risk assessment results. Because DBN supports temporal reasoning, its output emphasizes dynamism, including the posterior probability of nodes, risk quantification results, and causal transmission paths.

[0086] Posterior probability of a node: The probability distribution of each environmental node (such as "soil pollution level" or "flood risk") at the current time step (e.g., the probability of "flood risk = high" is 70%).

[0087] Risk quantification results: Comprehensive risk value (such as "Risk = Σ(node ​​probability × hazard weight)" in the patent) and risk level (red / orange / yellow / blue) calculated based on node probability.

[0088] Causal transmission path: The dependency chain of key risk factors (such as "heavy rainfall → rising river water level → dam collapse") is traced through the probability propagation path between nodes.

[0089] In this embodiment, the risk prediction model built based on a Bayesian network takes various "observed environmental data" as input, such as real-time sensor readings: today's rainfall, river water pH, mountain vibration amplitude, air pollutant concentration, etc. These specific values ​​or states (e.g., "rainfall > 50mm") can be understood as inputting various real-time environmental signals into an "intelligent discriminator" so that the intelligent discriminator knows the specific situation of the current environment. Its output includes two contents: the current risk result and the risk source. The current risk result is, for example, "the risk level of flash flood is red (high risk)" or "the probability of river pollution spreading is 70%"; the risk source is, for example, "because today's rainfall is too high → causing the mountain to loosen → which may trigger a landslide," that is, finding the key risk transmission path.

[0090] Simply put, this "intelligent judgment device" reads the input environmental data and outputs "how high is the current risk" and "what factors have led to this risk".

[0091] The risk prediction module is used to perform the following operations: taking the extracted time series data segments as input, performing risk level analysis through the risk prediction model and particle filter inference, predicting and outputting the risk level, risk level probability value and risk transmission path as prediction results, and constructing a visualized risk transmission map based on the prediction results.

[0092] When performing risk level analysis using a risk prediction model combined with particle filtering inference, the probability after a node in the risk prediction model is approximated by weighted statistics of the particles. The risk value calculation formula is as follows:

[0093]

[0094] in, W represents the probability of the current state of node i. The probability of the current state of node i is the posterior probability, which can be obtained through particle filtering and Bayesian inference. i The hazard weight of node i is determined by data in the historical event database and reflects the degree to which the node contributes to the risk.

[0095] In this embodiment, particle filtering is the core algorithm for real-time temporal inference in dynamic Bayesian networks. Its essence is to approximate the posterior distribution of the system state through a set of particles.

[0096] The input includes an initial particle set, observation data, a state transition model, and an observation model. The initial particle set represents the possible initial states of the system (e.g., the initial probability distribution of each environmental node), with each particle corresponding to a possible state combination (e.g., "rainfall = 50 mm / h, pH = 6.5"). Observation data refers to sensor data at the current time step (e.g., "wind speed = 10 m / s" and "heavy metal concentration = 0.8 mg / L" at time t), used to adjust particle weights. The state transition model describes the evolution of the system state over time (from the dynamic edges and CPT of the DBN), such as how "rainfall at time t" affects the probability distribution of "soil moisture at time t+1". The observation model represents the probability of observation data occurring given the system state (e.g., "the probability that the sensor observes 'humidity = 90%' when soil moisture = high"), typically defined by the node conditional probabilities of the DBN.

[0097] The output includes an updated particle set, state estimation results, and uncertainty quantification. The updated particle set: After weighting and resampling the observed data, high-weight particles (representing more likely system states) are retained, approximating the posterior distribution of the state at the current time step. State estimation results: The posterior probability of each environmental node calculated based on the particle set (e.g., the probability of "flood risk = high" at time t, determined by the sum of the weights of the corresponding states in the particle set). Uncertainty quantification: The dispersion of the particle set (e.g., the variance of the particle weights), reflecting the reliability of the inference results (the smaller the variance, the more reliable the risk assessment).

[0098] The system in this embodiment can execute the method disclosed in Embodiment 1 to achieve quantitative assessment of environmental risks, and is applicable to dynamic risk early warning of sudden environmental events such as natural disasters and industrial pollution.

[0099] The above provides a detailed description of the dynamic risk assessment method and system for environmental events based on Bayesian networks provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A dynamic risk assessment method for environmental events based on Bayesian networks, characterized in that, Includes the following steps: Data Acquisition: Environmental data is collected by a cluster of sensors deployed in the environment and uploaded to edge computing nodes. The environmental data is time-series data. Data processing: Environmental data is preprocessed through edge computing nodes, and normalization and missing value correction are performed through data preprocessing to obtain a standardized data stream. The standardized data stream is then updated to the historical event database. Data segmentation: Extracting time-series data segments from a standardized data stream based on a predefined sliding time window; Network Model Construction: A risk prediction model is constructed based on a dynamic Bayesian network. The risk prediction model defines a node set, a dynamic edge set, and a conditional probability table. Environmental data is set as nodes, and the time dependencies between nodes are set as dynamic edges. The conditional probability table describes the probability distribution of a node under the condition of its parent node taking a pointer. The risk prediction model is used to analyze the risk level based on the environmental data as input, and outputs the risk level, the risk level probability value, and the risk transmission path. Risk prediction: Taking the extracted time series data segments as input, the risk prediction model is used to analyze the risk level by combining particle filter inference. The predicted output includes the risk level, the probability value of the risk level, and the risk transmission path. Based on the prediction results, a visualized risk transmission map is constructed.

2. The method for dynamic risk assessment of environmental events based on Bayesian networks according to claim 1, characterized in that, Environmental data includes water quality data, meteorological data, and geological displacement data.

3. The method for dynamic risk assessment of environmental events based on Bayesian networks according to claim 1, characterized in that, For the risk prediction model, the time dependency between nodes is calculated using the mutual information algorithm, which is used to calculate the distance between nodes.

4. The method for dynamic risk assessment of environmental events based on Bayesian networks according to claim 1, characterized in that, The EM algorithm is used to optimize parameters and a time decay factor is introduced during the training of the conditional probability table.

5. The method for dynamic risk assessment of environmental events based on Bayesian networks according to claim 1, characterized in that, When performing risk level analysis using a risk prediction model combined with particle filtering inference, the probability after a node in the risk prediction model is approximated by weighted statistics of the particles. The risk value calculation formula is as follows: in, W represents the probability of the current state of node i. The probability of the current state of node i is the posterior probability, which can be obtained through particle filtering and Bayesian inference. i The hazard weight of node i is determined by data in the historical event database and reflects the degree to which the node contributes to the risk.

6. A dynamic risk assessment system for environmental events based on Bayesian networks, characterized in that, It includes a data acquisition module, a data processing module, a data segmentation module, a network model building module, and a risk prediction module; The data acquisition module is used to perform the following operations: collect environmental data through a sensor cluster deployed in the environment, and upload the environmental data to the edge computing node, wherein the environmental data is time-series data; The data processing module is used to perform the following operations: preprocess environmental data through edge computing nodes, perform normalization and missing value modification through data preprocessing to obtain a standardized data stream, and update the standardized data stream to the historical event database; The data segmentation module is used to perform the following operations: extract time-series data segments from a standardized data stream based on a predefined sliding time window; The network model construction module is used to perform the following operations: construct a risk prediction model based on a dynamic Bayesian network, and define a node set, a dynamic edge set, and a conditional probability table in the risk prediction model. Here, environmental data is set as nodes, the time dependency relationship between nodes is set as dynamic edges, and the conditional probability table describes the probability distribution of a node under the condition of its parent node taking a pointer. The risk prediction model is used to take environmental data as input, perform risk level analysis based on the environmental data, and output the risk level, risk level probability value, and risk transmission path. The risk prediction module is used to perform the following operations: taking the extracted time series data segments as input, performing risk level analysis through the risk prediction model and particle filter inference, predicting and outputting the risk level, risk level probability value and risk transmission path as prediction results, and constructing a visualized risk transmission map based on the prediction results.

7. The dynamic risk assessment system for environmental events based on Bayesian networks according to claim 6, characterized in that, Environmental data includes water quality data, meteorological data, and geological displacement data.

8. The dynamic risk assessment system for environmental events based on Bayesian networks according to claim 6, characterized in that, For the risk prediction model, the time dependency between nodes is calculated using the mutual information algorithm, which is used to calculate the distance between nodes.

9. The dynamic risk assessment system for environmental events based on Bayesian networks according to claim 6, characterized in that, The EM algorithm is used to optimize parameters and a time decay factor is introduced during the training of the conditional probability table.

10. The environmental event dynamic risk assessment system based on Bayesian networks according to claim 6, characterized in that, When performing risk level analysis using a risk prediction model combined with particle filtering inference, the probability after a node in the risk prediction model is approximated by weighted statistics of the particles. The risk value calculation formula is as follows: in, W represents the probability of the current state of node i. The probability of the current state of node i is the posterior probability, which can be obtained through particle filtering and Bayesian inference. i The hazard weight of node i is determined by data in the historical event database and reflects the degree to which the node contributes to the risk.

Images