A method and system for intelligent risk assessment of a hydropower station based on multi-source heterogeneous threat situation awareness

Abstract

This invention discloses an intelligent risk assessment method and system for hydropower stations based on multi-source heterogeneous threat situation awareness. The method includes: collecting threat data in real time through multi-source sensors deployed around the hydropower station to generate a multi-source heterogeneous threat dataset; deeply fusing the multi-source data based on graph neural networks to extract threat situation features and construct a five-dimensional threat space model; applying an improved threat potential field theory for accurate threat quantification, calculating the potential energy of individual threats and the synergistic effect of multiple threats; combining an environment-operating condition coupling model to achieve accurate risk level assessment; and introducing an adaptive adjustment mechanism to monitor changes in the threat situation in real time and dynamically optimize the assessment strategy. Through multi-dimensional threat awareness, graph neural network fusion, threat potential field quantification, and environmental coupling modeling, accurate perception and scientific assessment of the complex threat situation of hydropower stations are achieved, providing reliable technical support for ensuring the safe operation of hydropower stations.

Description

Technical Field

[0001] This invention relates to the field of hydropower station safety risk assessment technology, specifically to a fusion space sky water land The intelligent risk assessment method for hydropower stations based on five-dimensional threat sources achieves accurate perception and dynamic risk assessment of the complex threat situation of hydropower stations through multi-sensor data fusion and optimized risk quantification algorithms. Background Technology

[0002] As critical national infrastructure, hydropower stations face multi-dimensional and multi-type security threats during long-term operation, including drones from the air, unmanned surface vessels (USVs) on the water, underwater vehicles, intrusions from reservoir banks, and cyberattacks. These threats are highly complex and dynamic, and often exhibit a trend of coordinated attacks, seriously threatening the safe operation of hydropower stations. Therefore, establishing a risk assessment system capable of comprehensively perceiving and accurately evaluating multi-source heterogeneous threats is of great significance for ensuring the safe operation of hydropower stations.

[0003] Currently, existing hydropower station risk assessment technologies mostly employ single-dimensional assessments or simple overlay methods, making it difficult to accurately perceive and assess complex multi-source threat situations. Specifically, they have the following shortcomings: Lack of multi-dimensional threat perception capability: Existing systems mostly use independent threat detection modules, failing to achieve unified perception of multi-dimensional threats such as those in the air, on the surface, underwater, on the shore, and on the network, resulting in blind spots in threat identification.

[0004] Ignoring the synergistic effects of threats: Traditional assessment methods do not fully consider the synergistic effects between threats from different domains, especially the coupling effect between cyberattacks and physical attacks, and cannot accurately assess the real risks of composite attack patterns such as "network paralysis + physical breakthrough".

[0005] The algorithm fails to take into account the unique operating environment of hydropower stations: the existing algorithm does not systematically consider the dynamic impact of environmental factors such as meteorology, hydrology, and operating conditions on the threat implementation and protection capabilities, resulting in a discrepancy between the assessment results and the actual risks.

[0006] Therefore, there is an urgent need to develop an intelligent assessment method that can integrate multi-source heterogeneous threat data, possess high-precision situational awareness capabilities, and accurately quantify complex threat risks, so as to improve the safety protection capabilities and operational reliability of hydropower stations in complex threat environments. Summary of the Invention

[0007] The purpose of this invention is to address the aforementioned problems by providing a method and system for intelligent risk assessment of hydropower stations based on multi-source heterogeneous threat situation awareness. By constructing a five-dimensional threat space model, a graph neural network (GNN) is used to fuse multi-source threat data, generating a comprehensive threat situation feature Ψ. Based on this, an improved threat potential field algorithm is used for risk quantification, and a dynamic risk assessment is performed by combining an environment-operating condition coupling model. Ultimately, this achieves accurate perception and scientific assessment of the complex threat situation of hydropower stations.

[0008] The technical solution of the present invention is as follows: A method for intelligent risk assessment of hydropower stations based on multi-source heterogeneous threat situation awareness is proposed. This method performs real-time risk assessment of hydropower stations by fusing five-dimensional threat source data from air, space, water, ground, and network. The hydropower station is equipped with radar, sonar, photoelectric systems, underwater acoustic detectors, and network monitoring equipment. The method includes the following steps: Threat data from the air, water surface, underwater, reservoir bank, and cyberspace are collected by multi-source sensors and combined with meteorological, hydrological, and operational environmental information obtained from the hydropower station's environmental monitoring system. The collected raw data is then filtered, standardized, and spatiotemporally aligned to generate a multi-source heterogeneous threat dataset. This paper utilizes graph neural networks to fuse multi-source heterogeneous data, defining each threat source as a graph node. Edge relationships are constructed based on spatial distance, temporal difference, and cooperation coefficient. Message passing and feature fusion are performed through a multi-layer graph neural network to extract threat situation features, construct a five-dimensional threat space model, and generate comprehensive threat situation features. ; Threat quantification is performed using threat potential field theory to calculate the potential energy of individual threats and the synergistic effect of multiple threats. By combining the environment-condition coupling model, the threat risk level can be dynamically assessed; Monitor changes in the threat landscape in real time and dynamically adjust assessment strategies based on risk assessment results.

[0009] This invention employs a graph neural network (GNN) to effectively integrate heterogeneous threat data collected from multiple sensors, including radar, sonar, and optoelectronic systems, thereby improving the hydropower station's ability and accuracy in perceiving complex threat situations. By using GNN for deep data fusion and employing filtering, denoising, and spatiotemporal alignment processes, the completeness and reliability of the generated comprehensive threat situation feature Ψ are ensured.

[0010] Furthermore, the fusion processing of the multi-source heterogeneous data based on graph neural networks specifically includes: node initial feature vector The structure is as follows: , in, : Three-dimensional spatial coordinates; : Three-dimensional velocity vector; : Three-dimensional acceleration vector; One-hot encoded vector of threat type; Capability vector; : Feature vector; : Behavior vector; Edge relationship construction algorithm: for node pairs edge weight The calculation formula is: , in, The spatial Euclidean distance; For the detection time difference; The spatial attenuation constant; The time decay constant; These are the normalized weighting coefficients; Coordination coefficient The calculation formula is: , in, Basic synergy coefficient; This is an indicator function; it returns 1 if the condition is true, and 0 otherwise. For behavioral similarity (cosine similarity); This is a communication association indicator function; , , These are the corresponding weighting coefficients; A multi-layer graph neural network is applied to fuse features of the constructed graph, and the comprehensive threat situation features are finally generated through a global readout operation. .

[0011] Furthermore, the formula for calculating the threat potential energy of a single unit is as follows: , in, For a moment Source of threat The position vector; , The potential field attenuation coefficient; To prevent small constants with a denominator of 0; For spatial points and the source of the threat Euclidean distance.

[0012] Furthermore, the calculation of multi-threat synergy includes: Constructing the basic synergy matrix Define the basic coordination matrix for five types of threats (e.g., air threats, surface threats, underwater threats, reservoir shore threats, and cyber threats): , in, Indicates threats in the base state With threats The coefficient of synergy; Calculate the dynamic synergy coefficient Dynamic correction considering the consistency of time, space, velocity, and objective: , in, , , These are the weighting coefficients for each correction term; For spatiotemporal proximity, For speed coordination, For consistency of goals; Overall momentum The total potential field is the sum of the potential energy of a single threat and the combined potential energy of multiple threats: , in, The total number of threat sources, As a threat and The synergistic enhancement of potential energy.

[0013] By applying an improved threat potential field theory, this invention can accurately quantify individual threats and the synergistic effects of multiple threats in the complex operating environment of hydropower stations. This is crucial for identifying complex attack patterns such as "network paralysis + physical breach." Optimized potential energy calculation and synergy coefficient matrix ensure the scientific accuracy of threat quantification, significantly enhancing the security protection capabilities of hydropower stations.

[0014] Furthermore, the combined environment-condition coupling model specifically includes: Define the meteorological influence matrix Hydrological influence function Operating condition state matrix ; Constructing a third-order coupled tensor Its elements Indicates the first Threats in the Under such circumstances, the first The coupling coefficient for each working condition; ,in, Based on the coupling coefficient, For coupling functions; Choquet integrals were used to calculate the nonlinear coupling effect values ​​of meteorological, hydrological, and operational factors. ,in , , These are the normalized meteorological, hydrological, and operational impact values, respectively. Environmental Coupling Risk Value ,in Based on the basic risk value.

[0015] Designed based on an environment-operational condition coupled model, this system can dynamically assess threats and risks under different meteorological, hydrological, and operational conditions. This design not only improves the accuracy of system assessments under extreme weather and special operating conditions (such as flood discharge and maintenance), but also ensures the real-time nature and adaptability of risk assessments, further guaranteeing the safe and stable operation of hydropower stations in complex environments.

[0016] Furthermore, the dynamic assessment of threat risk levels specifically includes: Final risk value Normalization to Interval obtained ; Based on the dynamic threshold vector Risk levels are categorized, and thresholds are adjusted adaptively. Based on the system false alarm rate Dynamically adjust according to the current operating condition; The risk level is determined as follows:

[0017] in, This is the normalized risk value.

[0018] Furthermore, the filtering, standardization, and spatiotemporal alignment processing of the collected raw data specifically includes: Data preprocessing: The raw data is smoothed using a Gaussian filter. The filtering process can be described as follows: , in, The standard deviation is used to control the smoothness of the filter kernel. The filtered data is then standardized. ,in, The original data, The mean, The standard deviation is denoted as .

[0019] Spatiotemporal alignment: , in, For sensor timestamps, This is the calibration deviation.

[0020] Furthermore, the application of threat potential field theory for threat quantification also includes: Based on the shortest expected effective arrival time among all threats The basic time pressure is calculated using a piecewise function; If there are multiple waves of attacks, then use time intervals. The base time pressure is adjusted to obtain the final time pressure factor. .

[0021] Furthermore, the dynamic adjustment of the assessment strategy based on the risk assessment results specifically includes: Real-time monitoring of changes in the number of threats Threat distance changes and changes in environmental characteristics ; When satisfied or or Under any of these conditions, a reassessment of the risk assessment model will be triggered; Update model parameters using gradient descent: ,in For model parameters, For learning rate, This is the loss function.

[0022] An assessment strategy based on an adaptive dynamic adjustment mechanism is adopted. This involves first monitoring changes in threat situation parameters in real time, and then dynamically optimizing the assessment model based on the risk assessment results. This approach helps address the dynamic evolution of multi-source threats, ensuring the system's continuous and efficient operation in complex threat environments. It also provides standardized and comparable risk assessment results, offering a reliable scientific basis for hydropower station safety decisions.

[0023] This application also includes a hydropower station intelligent risk assessment system based on multi-source heterogeneous threat situation awareness. The system includes a central processing unit, a data fusion module and a risk assessment module, and is equipped with radar, sonar, photoelectric system, underwater acoustic detector and network monitoring equipment. The central processing unit is used to execute a hydropower station intelligent risk assessment method based on multi-source heterogeneous threat situation awareness.

[0024] Compared with existing technologies, the advantages of this invention are: 1. This invention employs a graph neural network (GNN), which can effectively integrate multi-source heterogeneous threat data, improving the ability to perceive complex threat situations. Through data fusion and feature extraction, the accuracy and completeness of threat situation characteristics are ensured; 2. By applying the improved threat potential field theory, this invention can accurately quantify the synergistic effects of individual threats and multiple threats, and can effectively identify real threats. 3. Based on the environment-operating condition coupling model design, it can dynamically assess the threat risks under different environments and operating conditions, improving the adaptability and accuracy of the assessment; 4. By adopting a risk quantification method based on mathematical theory, standardized and comparable risk assessment results are provided, offering a scientific basis for decision support. Attached Figure Description

[0025] Figure 1 This is a flowchart of the hydropower station threat risk assessment algorithm based on multi-source heterogeneous data of the present invention. Detailed Implementation

[0026] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0027] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0028] Please see Figure 1 This paper presents an intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness. The method aims to improve the ability of hydropower stations to perceive and assess complex threat situations. The method includes: collecting data in a five-dimensional threat space through multi-source sensors; and using a graph neural network (GNN) to fuse the multi-source heterogeneous data to generate comprehensive threat situation features. Threat quantification is performed based on an improved threat potential field theory, calculating the potential energy of individual threats and the synergistic effects of multiple threats. Dynamic risk assessment is conducted using an environment-condition coupling model to ensure the accuracy and real-time nature of the assessment results. An adaptive adjustment mechanism is introduced to monitor changes in the threat situation in real time and dynamically optimize the assessment strategy. Specifically, the following steps are included: Multi-source heterogeneous data acquisition and preprocessing: By deploying a multi-source sensor system around the hydropower station, data from the five-dimensional threat space can be collected simultaneously: Data collection: Aerial threats: Information such as the position, speed, and altitude of aerial targets is collected through radar and electro-optical tracking systems; Surface threats: Data such as the trajectory and speed of surface targets are collected through surface radar and infrared imaging systems; Underwater threats: Information such as depth and heading of underwater targets is collected through sonar and underwater acoustic detectors; Threats to the reservoir bank: Information on personnel and vehicle activity is collected through video surveillance and perimeter intrusion detection systems; Cyber ​​threats: Data such as abnormal traffic and supply characteristics are collected through network traffic monitoring and intrusion detection systems.

[0029] Data preprocessing: The raw data is smoothed using a Gaussian filter to reduce noise interference. The filtering process can be described as follows: , in, The standard deviation is used to control the smoothness of the filter kernel. The filtered data is then standardized. , in, The original data, The mean, The standard deviation is denoted as .

[0030] Spatiotemporal alignment: Different sensors may have different sampling frequencies and timestamps, requiring spatiotemporal alignment. , in, For sensor timestamps, This is the calibration deviation.

[0031] Threat situation feature extraction based on graph neural networks: Detailed graph structure construction process: Node definition and initialization. Each detected threat source is defined as a node in the graph. The initial feature vectors are constructed as follows: , The components are defined in detail as follows: : Three-dimensional spatial coordinates; : Three-dimensional velocity vector; : Three-dimensional acceleration vector; One-hot encoded vector of threat type; Capability vectors (load equivalent, maneuverability, stealth level, autonomy level); : Feature vectors (RCS value, infrared feature, acoustic feature, electromagnetic feature); : Behavior vector (proximity pattern, formation type, cooperation identifier).

[0032] Edge relation construction algorithm: for any pair of nodes edge weight The calculation formula is: , in, The spatial Euclidean distance; For the detection time difference; The spatial attenuation constant; The time decay constant; These are the normalized weighting coefficients; Coordination coefficient The calculation formula is: , in, Basic synergy coefficient; This is an indicator function; it returns 1 if the condition is true, and 0 otherwise. For behavioral similarity (cosine similarity); This is a communication association indicator function; , , These are the corresponding weighting coefficients; edge set The construction principles are: , in, , , This is the grouping decision function.

[0033] Detailed algorithm for GNN feature fusion: Message passing mechanism: No. The message passing process at a layer consists of three phases: Phase 1: Message Generation node To the node The message being transmitted is defined as: , in, This is a vector concatenation operation; For the first The message weight matrix of the layer; It is the bias vector; For activation functions; Phase 2: Message Aggregation Node Aggregated messages: , Among them, aggregation operation Available options: Weighted average: ； Attention mechanism: Attention coefficient: Max pooling: .

[0034] Phase 3: Feature Update Node feature update formula: , in, This is a self-loop weight matrix; The neighbor weight matrix; This is the bias vector.

[0035] Multi-layer stacking and residual connections: , Among them, the feature increment: , For layer normalization operation: , in, and These are the mean and variance, respectively. and For learnable parameters, It is a small constant.

[0036] Global feature readout: after After layer processing, a graph-level threat situation feature is generated: , Numerical representation of readout operations: Mean pooling: ; Max pooling: ; Summation pooling: ; Hierarchical pooling: ; Attention pooling: Among them, the global attention weights are: .

[0037] Improved Threat Potential Field Theory and Quantification: Threat parameter extraction and standardization: for each threat source Threat type coefficient The determination is made using a piecewise function: Aerial threat targets: , in, Mass of the drone (unit: kg).

[0038] Surface threat targets: , in, The tonnage of the unmanned surface vessel (unit: tons).

[0039] Underwater threat targets: ;in, The depth at which underwater threats occur (unit: meters).

[0040] Cyber ​​threat targets: ;in, This represents the severity level of a cyber threat.

[0041] Threat intensity The dynamic calculation of threat intensity is a function of time, defined as: , The kinetic energy factor is calculated using the following formula: , in, As a source of threat velocity vector (norm) (for speed magnitude) As a source of threat acceleration vector (norm) (This refers to the magnitude of the acceleration).

[0042] The ability factor (piecewise function) is calculated using the following formula: , in, The equivalent of TNT in terms of explosive threat; For the interference power of electronic threats, This represents the maximum interference power. The number of vulnerabilities representing cyber threats. The impact level of a cyber threat.

[0043] The time-varying factor (piecewise function) is calculated using the following formula: , in, As a source of threat Current distance to the target Maximum range for threat detection.

[0044] Calculation of spatial potential distribution: Assume a three-dimensional spatial domain is ( , , They are space in , , The range of axes), discretized into grid points ( , , (These represent the number of grid cells for each axis).

[0045] Basic threat potential For any point in space Threat source The potential energy generated at this point is: , in, For a moment Source of threat The position vector; , The potential field attenuation coefficient; To prevent small constants with a denominator of 0; For spatial points and the source of the threat Euclidean distance.

[0046] Effective potential energy considering terrain shielding Introducing a shading function to correct terrain shading effects: , Among them, the masking function For piecewise functions: , This is the partial shading coefficient (the value is determined based on the degree of shading).

[0047] Dynamic generation of the synergy coefficient matrix: Basic Synergy Matrix Define the basic coordination matrix for five types of threats (e.g., air threats, surface threats, underwater threats, reservoir shore threats, and cyber threats): , in, Indicates threats in the base state With threats The synergy coefficient.

[0048] Dynamic Coordination Coefficient Dynamic correction considering the consistency of time, space, velocity, and objective: , in, , , These are the weighting coefficients for each correction term; For spatiotemporal proximity, For speed coordination, For consistency of goals.

[0049] Spacetime proximity The calculation formula is: , , in, The spatiotemporal proximity attenuation coefficient (unit: meters).

[0050] Speed ​​Coordination The calculation formula is: , indicating the consistency of the speed directions of the two threats (value range [-1, 1], 1 is completely in the same direction, -1 is completely in opposite directions).

[0051] Goal Consistency The calculation formula is: ,in, , Threats , The target position vector, The maximum detection distance for the target (value range [0,1], 1 means the target is completely identical).

[0052] Collaborative potential superposition algorithm: Overall momentum The total potential field is the sum of the potential energy of a single threat and the combined potential energy of multiple threats: , in, The total number of threat sources, As a threat and The synergistic enhancement of potential energy.

[0053] Synergistic enhancement of potential energy The calculation formula is: , When the cooperative potential energy exceeds the threshold, nonlinear amplification is employed: , in, For the cooperative potential threshold, This is a non-linear amplification factor.

[0054] Threat arrival time calculation: Threat of uniformly accelerated motion ( ): , in, As a source of threat Radial velocity (velocity component along the "threat-target" line): , As a source of threat Radial acceleration (acceleration component along the "threat-target" line): , As a source of threat Initial distance to the target; This is the target position vector.

[0055] Threat of uniform motion ( ): .

[0056] Valid arrival time: Considering maneuver path and environmental delay correction: , in, This is the path correction factor (the maneuvering path is longer than the straight line). This refers to the environmental delay factor (such as the impact of wind speed and water flow on threatening movements).

[0057] Shortest arrival time : This refers to the earliest time among all threats to reach the target.

[0058] Calculation of time-pressure function: Base S Type of pressure function : , Multi-wave attack correction pressure function : , in, The time interval between two adjacent waves of threats (in seconds) is the time between them. The shorter the interval, the larger the pressure correction factor.

[0059] Environmental-operating condition coupling risk assessment: Refined modeling of meteorological impacts: Meteorological Influence Matrix Define the meteorological influence matrix ,element Indicates the first The type of meteorological conditions on the first Impact coefficients for threat categories. Rows in the matrix correspond to threat types (air, surface, underwater, reservoir bank, cyber), and columns correspond to weather conditions (sunny, light rain, heavy rain, fog, storm). , Degradation matrix of detection system Define the degradation matrix of the detection system , element represents the first The type of meteorological conditions on the first Performance degradation coefficient of the detection system. Rows in the matrix correspond to the detection system (radar, optical, infrared, sonar), and columns correspond to weather conditions (sunny, light rain, heavy rain, fog, storm): .

[0060] Comprehensive meteorological influence function : , in, The total number of detection systems (here) ), For the first The weather pattern on the first Impact coefficient of threat class For the first The product of the performance of all detection systems under various weather conditions.

[0061] Additional wind speed effect correction : , in, Actual wind speed (unit: m / s).

[0062] Modeling the dynamic impact of hydrological conditions: Hydrological influence function : , in, The water level affects the weight. For the flow rate to affect the weight, The component affected by the flood discharge status is reduced by 2 to avoid excessive amplification of the impact due to the superposition of multiple factors.

[0063] Water level affects the quantity : , in, This is the actual water level. This is the normal water level.

[0064] Flow rate affects components : , in, This represents the actual traffic volume. This represents normal traffic volume; the absolute value indicates that "both excessively high and low traffic volumes will have an impact".

[0065] The impact of flood discharge status : , Differential Impact Matrix of Hydrology on Threats : , The matrix rows correspond to "types of hydrological impacts," and the columns correspond to "sub-threats" (which need to be defined in conjunction with specific scenarios). The differences in coefficients reflect the different sensitivities of different threats to hydrological conditions.

[0066] Comprehensive impact assessment of operating conditions: Operating condition influence vector : Define the influence vector of operating conditions The meanings of each element are as follows: Target vulnerability coefficient (the higher the vulnerability, the lower the vulnerability). The larger); System response capability coefficient (the stronger the response capability, the better). The larger); Sensor availability coefficient (the higher the availability, the better) The larger); Personnel readiness coefficient (the higher the readiness, the better) The larger (the larger).

[0067] Operating condition state matrix : Define the operating condition state matrix The row corresponds to the "dimensionality of the working condition influence vector". List the corresponding operating conditions (normal, flood discharge, maintenance, emergency, startup): .

[0068] Gaussian membership function (detailed evaluation of operating conditions): Fuzzy attributes used to quantify the state of a chemical process: , in, The center value of the "ideal working condition" The standard deviation (to control the rate of membership decay).

[0069] Comprehensive operating conditions : , in, and This means that "the lower the responsiveness / sensor availability, the stronger the amplification effect on risk."

[0070] Tensor product coupled calculation: Third-order coupled tensor : Constructing a third-order coupled tensor ,element Indicates "the Threats in the Under such circumstances, the first The coupling coefficient for "various working conditions" is calculated using the following formula: , in, Based on the coupling coefficient, This is the coupling function.

[0071] Coupling function : , in, (Meteorological impact) (Hydrological impact) (Operating conditions).

[0072] Calculation of nonlinear coupling effects: Fuzzy measure Let the factor set ( Meteorological factors Hydrological factors (Operating condition factor), fuzzy measure Defined in the power set Above, the coupling weights of different factor combinations are quantified: .

[0073] Choquet integral (nonlinear coupling modeling): Calculation of multi-factor nonlinear coupling effects using Choquet integration: , in: (Total number of factors); Factor value The results sorted in descending order (i.e.) ); (including the first) A set of one or more sorting factors); (Initial value).

[0074] Comprehensive Risk Value Calculation: Basic Risk The fundamental risk is directly determined by the total threat field: , in, The calculated "multi-threat total potential field".

[0075] Environmental coupling risk Introducing corrections for nonlinear coupling effects between the environment and operating conditions: , in, The coupling effect value calculated for Choquet integrals.

[0076] Time modulation risk Adjusted by time pressure factor: , in, for S Type-time pressure function.

[0077] Uncertainty correction factor Considering the uncertainties in data quality and sensor coverage: .

[0078] Data quality factor : .

[0079] in, For data integrity, To adjust the parameters; Sensor coverage factor : , in, For sensor coverage, To adjust the parameters.

[0080] Final risk value : .

[0081] Risk normalization Normalize the risk value to the range [0, 100] for easier risk assessment: , in, This is the preset "maximum expected risk value".

[0082] Dynamic risk level classification: Dynamic threshold vector : Define risk level threshold vector The initial value is These correspond to the critical values ​​for "low, medium, high, and extremely high" risks.

[0083] Threshold adaptive adjustment Thresholds are dynamically adjusted based on false alarm rate and operating conditions: , in: : System false alarm rate, when At that time, take (Increase the threshold to reduce false positives); : Operating condition indication function, during flood discharge ,Pick (Lower the threshold to increase vigilance).

[0084] Risk level determination : , in, (Normalized risk value).

[0085] Adaptive dynamic adjustment mechanism: Status monitoring monitors threat posture parameters in real time: Changes in the number of threats: ; Threat distance changes: ; Changes in environmental characteristics: .

[0086] Triggering conditions: A re-evaluation is triggered when the following conditions are met: ; ; .

[0087] The model parameters are updated using gradient descent. , in, For model parameters, For learning rate, This is the loss function.

[0088] Example: Example 1: Multi-target cooperative threat assessment: Scenario setup: 3 unmanned surface vessels + 2 drones + network DDoS attack; Input parameters: Unmanned surface vessel: speed 15m / s, range 1000m, payload 500kg; Drone: Speed ​​30m / s, altitude 100m, payload 25kg; DDoS attack: Strength 0.7, estimated duration 300 seconds; Environment: Heavy rain (visibility 200m), flood discharge period.

[0089] Calculation process: (1) Individual threat potential: ; ; .

[0090] (2) Synergistic effect: (Surface-air coordination); (Network-Physical Collaboration); .

[0091] (3) Environmental coupling: (Impact of heavy rain); (Impact of flood discharge); .

[0092] (4) Time pressure: Shortest arrival time: Second; .

[0093] (5) Overall risk value: .

[0094] (6) Risk level: Extremely high risk (red alert).

[0095] Example 2: Algorithm Complexity Analysis: The computational complexity analysis of the method of this invention is as follows: Data preprocessing stage: ,in Number of sensor data points; Graph Neural Network Fusion: ,in For the number of nodes, The number of sides; Threat field calculation: ,in Number of threat sources; Environmental Coupling Calculation: ,in From the perspective of environmental factors; Overall algorithm complexity: ; For typical scenarios (number of threat sources <50, environmental factors <10), the algorithm can complete calculations in sub-seconds, meeting the needs of real-time assessment.

[0096] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. A method for intelligent risk assessment of hydropower stations based on multi-source heterogeneous threat situation awareness, characterized in that, The method performs real-time risk assessment of hydropower stations by fusing five-dimensional threat source data from air, space, water, land, and network. The hydropower stations are equipped with radar, sonar, photoelectric systems, underwater acoustic detectors, and network monitoring equipment. The method includes the following steps: Threat data from the air, water surface, underwater, reservoir bank, and cyberspace are collected by multi-source sensors and combined with meteorological, hydrological, and operational environmental information obtained from the hydropower station's environmental monitoring system. The collected raw data is then filtered, standardized, and spatiotemporally aligned to generate a multi-source heterogeneous threat dataset. This paper utilizes graph neural networks to fuse multi-source heterogeneous data, defining each threat source as a graph node. Edge relationships are constructed based on spatial distance, temporal difference, and cooperation coefficient. Message passing and feature fusion are performed through a multi-layer graph neural network to extract threat situation features, construct a five-dimensional threat space model, and generate comprehensive threat situation features. ; Threat quantification is performed using threat potential field theory to calculate the potential energy of individual threats and the synergistic effect of multiple threats. By combining the environment-condition coupling model, the threat risk level can be dynamically assessed; Monitor changes in the threat landscape in real time and dynamically adjust assessment strategies based on risk assessment results.

2. The intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness as described in claim 1, characterized in that, The specific steps of fusing the multi-source heterogeneous data based on graph neural networks include: node initial feature vector The structure is as follows: ， in, : Three-dimensional spatial coordinates; : Three-dimensional velocity vector; : Three-dimensional acceleration vector; One-hot encoded vector of threat type; Capability vector; : Feature vector; : Behavior vector; Edge relationship construction algorithm: for node pairs edge weight The calculation formula is: ， in, The spatial Euclidean distance; For the detection time difference; The spatial attenuation constant; The time decay constant; These are the normalized weighting coefficients; Coordination coefficient The calculation formula is: ， in, Basic synergy coefficient; This is an indicator function; it returns 1 if the condition is true, and 0 otherwise. For behavioral similarity (cosine similarity); This is a communication association indicator function; , , These are the corresponding weighting coefficients; A multi-layer graph neural network is applied to fuse features of the constructed graph, and the comprehensive threat situation features are finally generated through a global readout operation. .

3. The intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness as described in claim 1, characterized in that, The formula for calculating the threat potential energy of a single unit is as follows: ， in, For a moment Source of threat The position vector; , The potential field attenuation coefficient; To prevent small constants with a denominator of 0; For spatial points and the source of the threat Euclidean distance.

4. A method for intelligent risk assessment of hydropower stations based on multi-source heterogeneous threat situation awareness, as described in claim 1 or 3, is characterized in that... The calculation of multi-threat synergistic effects includes: Constructing the basic synergy matrix Define the basic coordination matrix for five types of threats (e.g., air threats, surface threats, underwater threats, reservoir shore threats, and cyber threats): ， in, Indicates threats in the base state With threats The coefficient of synergy; Calculate the dynamic synergy coefficient Dynamic correction considering the consistency of time, space, velocity, and objective: ， in, , , These are the weighting coefficients for each correction term; For spatiotemporal proximity, For speed coordination, For the sake of goal consistency; Overall momentum The total potential field is the sum of the potential energy of a single threat and the combined potential energy of multiple threats: ， in, The total number of threat sources, As a threat and The synergistic enhancement of potential energy.

5. The intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness as described in claim 1, characterized in that, The combined environment-operating condition model specifically includes: Define the meteorological influence matrix Hydrological influence function Operating condition state matrix ; Constructing a third-order coupled tensor Its elements Indicates the first Threats in the Under such circumstances, the first The coupling coefficient for each working condition; ,in, Based on the coupling coefficient, For coupling functions; Choquet integrals were used to calculate the nonlinear coupling effect values ​​of meteorological, hydrological, and operational factors. ,in , , These are the normalized meteorological, hydrological, and operational impact values, respectively. Environmental Coupling Risk Value ,in Based on the basic risk value.

6. The intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness as described in claim 1, characterized in that, The dynamic assessment of threat risk levels specifically includes: Final risk value Normalization to Interval ; Based on the dynamic threshold vector Risk levels are categorized, and thresholds are adjusted adaptively. Based on the system false alarm rate Dynamically adjust according to the current operating condition; The risk level is determined as follows: in, This is the normalized risk value.

7. The intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness as described in claim 1, characterized in that, The filtering, standardization, and spatiotemporal alignment processing of the collected raw data specifically includes: Data preprocessing: The raw data is smoothed using a Gaussian filter. The filtering process can be described as follows: ， in, The standard deviation is used to control the smoothness of the filter kernel. The filtered data is then standardized. ,in, The original data, The mean, Standard deviation; Spatiotemporal alignment: ， in, For sensor timestamps, This is the calibration deviation.

8. The intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness as described in claim 1, characterized in that, The application of threat potential field theory for threat quantification also includes: Based on the shortest expected effective arrival time among all threats The basic time pressure is calculated using a piecewise function; If there are multiple waves of attacks, then use time intervals. The base time pressure is adjusted to obtain the final time pressure factor. .

9. The intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness as described in claim 1, characterized in that, The specific provisions for dynamically adjusting the assessment strategy based on the risk assessment results include: Real-time monitoring of changes in the number of threats Threat distance changes and changes in environmental characteristics ; When satisfied or or Under any of these conditions, a reassessment of the risk assessment model will be triggered; Update model parameters using gradient descent: ,in For model parameters, For learning rate, This is the loss function.

10. A smart risk assessment system for hydropower stations based on multi-source heterogeneous threat situation awareness, characterized in that, The system includes a central processing unit, a data fusion module, and a risk assessment module, and is equipped with radar, sonar, photoelectric system, underwater acoustic detector, and network monitoring equipment. The central processing unit is used to execute the intelligent risk assessment method for hydropower stations based on multi-source heterogeneous threat situation awareness as described in any one of claims 1-9.

Images