Specific scene safety risk identification method based on visual dynamic analysis

Abstract

The invention discloses a specific scene safety risk identification method based on visual dynamic analysis, and belongs to the technical field of safety monitoring, and the method specifically comprises the steps: collecting environment parameter time sequence data and cargo state time sequence data of a monitoring region; respectively extracting an environment characteristic component and a cargo response component; calculating a quantized value of correlation strength of the two through an information entropy algorithm, and marking the quantized value as a cargo response entropy value; a cargo response entropy evolution curve is monitored, and when the change rate of the evolution curve exceeds a historical threshold value and the change acceleration is not zero, it is judged that a state transition event occurs; extracting environmental incentive parameters and cargo response parameters of a state transition event, encoding the environmental incentive parameters and the cargo response parameters into risk feature tensors, inputting the risk feature tensors into the accident incentive chain knowledge base, outputting a matched critical intervention parameter set, and formulating an environmental reconstruction instruction set and a cargo intervention instruction set according to the critical intervention parameter set; according to the invention, the intelligent level and risk response capability of dangerous cargo safety early warning are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of security monitoring technology, and in particular to a method for identifying security risks in specific scenarios based on visual dynamic analysis. Background Art

[0002] Traditional approaches relying on manual inspections and static databases are unable to cope with the dynamic evolution of risks in complex environments. The industry urgently needs intelligent means to achieve early risk identification and rapid intervention. The application of artificial intelligence technology in security monitoring is gradually deepening, especially in the areas of computer vision and sensor fusion, where initial technological advancements have been achieved.

[0003] At present, the existing technologies mainly adopt the following means to identify the safety risks of dangerous goods: on the one hand, a monitoring network is built through environmental parameter sensors, such as temperature sensors, humidity sensors, gas concentration sensors, etc., to collect environmental parameters around dangerous goods to determine the impact of environmental changes on the goods; on the other hand, visual monitoring systems, such as cameras, infrared imaging equipment, etc. are used to observe the surface morphology and color changes of the goods, and combined with manual analysis or simple image recognition algorithms, a preliminary assessment of the status of the goods is made.

[0004] However, the monitoring model of existing technologies is like staring at a thermometer with a single indicator, focusing only on whether a single parameter exceeds the standard in isolation, but seriously ignoring the core mechanism of the evolution of dangerous goods risks, that is, the dynamic coupling of environmental factors and the response of the goods themselves. For example, before a chemical storage tank leaks at a port, it is often accompanied by abnormal fluctuations in the temperature field, the expansion of microcracks on the surface of the tank, the formation of vortices in the gas concentration field, and other coordinated changes in multi-dimensional characteristics. These factors push each other to cause a surge in the degree of disorder in the system, and ultimately lead to a systematic transition from an ordered stable state to a disordered collapse. However, in this process, a single indicator may never reach the traditional alarm threshold, making it impossible for existing technologies to identify in advance the system collapse caused by this multi-factor coupling. As a result, the dynamic relationship between the environment and the goods is ignored, and there is a lack of quantitative analysis of the degree of disorder in the system, resulting in the inability to identify the critical turning point of risk evolution. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying security risks in specific scenarios based on visual dynamic analysis to solve the following technical problems: The monitoring model of existing technologies ignores the dynamic relationship between the environment and cargo, lacks quantitative analysis of the system's chaos, and makes it impossible to identify the critical turning points of risk evolution.

[0006] The purpose of the present invention can be achieved through the following technical solutions: A method for identifying security risks in specific scenarios based on visual dynamic analysis includes the following steps: The environmental parameter sensor network collects the time series data of the environmental parameters of the monitoring area, and the cargo visual monitoring system collects the time series data of the cargo status; Perform feature dimensionality reduction on environmental parameter time series data to extract environmental feature components; perform response decoupling on cargo status time series data to extract cargo response components; A dynamic coupling model between environmental characteristic components and cargo response components is established. The quantitative value of the correlation strength between the two is calculated using the information entropy algorithm and marked as the cargo response entropy value. Monitor the evolution curve of the cargo response entropy value. When the rate of change of the evolution curve exceeds the historical threshold and the acceleration of the change is non-zero, a state transition event is determined to have occurred. Extract the environmental inducement parameters and cargo response parameters of the state transition event, encode the environmental inducement parameters and cargo response parameters into risk feature tensors and input them into the accident causal chain knowledge base, output the matching critical intervention parameter set, and formulate the environmental reconstruction instruction set and cargo intervention instruction set based on the critical intervention parameter set.

[0007] As a further solution of the present invention: the environmental parameter time series data includes time series data of three-dimensional distribution of temperature field, time series data of gradient evolution of humidity field, and time series data of migration trajectory of gas concentration field; the cargo status time series data includes time series data of surface morphology change rate, time series data of internal energy release spectrum, and time series data of interface reaction propagation direction.

[0008] As a further solution of the present invention: the specific process of performing the feature dimensionality reduction operation on the environmental parameter time series data is: Perform spatial cluster analysis on the three-dimensional distribution time series data of the temperature field to identify the motion trajectory of the local high-temperature extreme point, and calculate the sliding average of the trajectory tangential acceleration as the component of the temperature field extreme point; The isosurface topology of the humidity field gradient evolution time series data is reconstructed, and the root mean square derivative of the Gaussian curvature change of the isosurface in adjacent sampling periods is extracted as the humidity field isosurface component. Vortex dynamics analysis is carried out on the time series data of gas concentration field migration trajectory to locate the instantaneous rotation center of the vortex core, and the absolute value of the core's angular velocity around the axis is calculated as the vortex component of the gas concentration field.

[0009] As a further solution of the present invention, the specific process of performing the response decoupling operation on the cargo status time series data is as follows: Perform empirical mode decomposition on the surface morphology change rate time series data to separate the slowly varying intrinsic mode functions reflecting the inherent characteristics of the cargo from the rapidly varying residual components stimulated by the environment. The superposition of the first three order intrinsic mode functions is labeled as the inherent trend component, and the combination of higher order intrinsic mode functions and residual components beyond the first three is labeled as the stimulated fluctuation component. Wavelet packet transform is performed on the internal energy release spectrum time series data to extract the approximate coefficient reconstruction signal reflecting the steady-state characteristics as the fundamental frequency steady-state component, and the frequency band where the detail coefficient energy exceeds the threshold is synthesized as the transient offset component; A baseline direction prediction model is established for the time series data of interface reaction propagation direction. The model input is the mean of historical propagation direction and the regression of environmental parameters. The vector difference between the actual direction and the predicted direction is defined as the abnormal deflection component.

[0010] As a further solution of the present invention: the calculation process of the cargo response entropy value is: Constructing a mutual information matrix between the set of environmental characteristic components and the set of cargo response components, where the matrix elements represent the statistical dependence between a specific environmental characteristic component and a specific cargo response component; Calculate the singular value decomposition of the mutual information matrix and extract the geometric mean of the first three non-zero singular values ​​as the initial correlation strength; The historical significance weight of the environmental characteristic component is introduced. The weight value is calibrated according to the frequency of state transition events triggered by the environmental characteristic component in the past. The initial association strength is multiplied by the historical significance weight to obtain the weighted association strength. The negative natural logarithm of the weighted association strength is taken and normalized to the interval [0,1], which is the cargo response entropy value.

[0011] As a further solution of the present invention: the judgment criteria of the state transition event are: Calculate the standard deviation of the cargo response entropy value during the historical stable period, set 4 times the standard deviation as the historical stability threshold, calculate the first-order derivative of the cargo response entropy value as the rate of change, and the second-order derivative as the acceleration of change; When the rate of change exceeds the historical stability threshold for m consecutive time windows and the absolute value of the change acceleration is greater than zero, it is marked as a candidate event, and m is the set threshold; Verify whether the cargo response entropy value in the candidate event shows a monotonically deviating trend from the historical average baseline. If so, confirm that the candidate event is a state transition event, and record the starting window position, peak window position, and transition direction of the state transition event. The transition direction is divided into entropy-increasing phase transition and entropy-decreasing phase transition based on the sign of the change rate.

[0012] As a further solution of the present invention, the specific process of encoding the environmental inducement parameters and the cargo response parameters into the risk feature tensor is as follows: Intercept the time series segments of the environmental characteristic components during the duration of the state transition event; identify the three environmental characteristic components with the largest mutation amplitude in the time series segments, record their mutation direction signs and relative intensity levels, and encode them into a triplet of environmental inducement parameters; Intercept the time series segments of the cargo response components, calculate the stability coefficient of the inherent trend term, the energy proportion of the stimulated fluctuation term, the degree of retention of the fundamental frequency steady-state component, the intensity of the transient offset component, and the integral value of the abnormal deflection component angle, and normalize the five indicators of the cargo response components into a five-dimensional vector of cargo response parameters; The tensor product of the triplet of environmental inducement parameters and the five-dimensional vector of cargo response parameters is labeled as the risk feature tensor.

[0013] As a further solution of the present invention: the matching mechanism of the accident cause chain knowledge base is specifically as follows: The accident causal chain knowledge base includes a plurality of causal chain records, each of which includes a different environmental inducement parameter triple and a cargo response parameter five-dimensional vector, as well as a matching critical intervention parameter set; First, the causal chain records that completely match the combination of mutation components in the input environmental trigger parameter triple are retrieved; then the Euclidean distance between the retrieved causal chain records and the input five-dimensional vector of the cargo response parameters is calculated, and the causal chain records with Euclidean distance less than the similarity threshold are selected; the selected causal chain records are sorted in ascending order of Euclidean distance, and the critical intervention parameter sets corresponding to the top three causal chain records are selected for output; When there is no completely matched causal chain record, a component with opposite sign but the same intensity level is allowed to appear in the matching process, and it is also marked as a completely matched one.

[0014] As a further solution of the present invention, the specific process of formulating the environment reconstruction instruction and the cargo intervention instruction according to the critical intervention parameter set is as follows: Analyze the output critical intervention parameter set and extract the operation instructions whose operation objects, action parameters and execution time sequences completely overlap as common operation instructions; filter out the conflicting operation instructions involving the same operation object but with mutually exclusive parameter requirements or overlapping execution time sequences from the remaining operation instructions; For conflicting operation instructions with the same type of parameters, the operation instruction with a higher intensity level shall be retained first; when the energy control operation instruction conflicts with the environmental adjustment operation instruction, the energy control operation instruction shall be retained first; The common operation instructions and the reserved non-conflicting operation instructions are sorted in reverse chronological order of the risk transmission path to form a basic intervention sequence. The cargo response parameter index in the current risk characteristic tensor is obtained. If the intensity of the transient offset component exceeds the normalized safety threshold, a spectrum stabilization operation is inserted at the critical energy release phase of the basic intervention sequence. If the angle integral value of the abnormal deflection component is greater than the geometric tolerance, a spatial correction operation is inserted at the propagation path offset point of the basic intervention sequence. The environmental field operation instructions and cargo state intervention operation instructions in the adjusted basic intervention sequence are integrated to generate an environmental reconstruction instruction set and a cargo intervention instruction set.

[0015] Beneficial effects of the present invention: The present invention establishes a dynamic coupling model of environmental parameters and cargo state responses, adopts an information entropy algorithm to quantify the correlation strength between environmental characteristic components and cargo response components to generate cargo response entropy values, monitors the entropy change rate and acceleration in real time to identify state transition events, and solves the problem of gradual risk omission caused by isolated monitoring of a single parameter in the existing technology; constructs a risk characteristic tensor of triples and five-dimensional vectors by extracting event-related environmental inducement parameters and cargo response parameters, and realizes multi-dimensional precise matching in the accident cause chain knowledge base, overcoming the defect that traditional threshold warning cannot capture the co-evolution of multiple factors; generates an intervention sequence with reverse time sequence sorting based on the conflict arbitration mechanism, and dynamically inserts spectrum stabilization and spatial correction operations based on the transient offset component intensity and the integral value of the propagation deflection angle to form an environmental reconstruction instruction set and a cargo intervention instruction set, which advances the risk disposal node from passive response to the critical inflection point where the system chaos surges, providing a comprehensive and intelligent solution for the safe management of dangerous goods in ports. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 It is a schematic flow diagram of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] See also Figure 1 As shown, the present invention is a method for identifying security risks in specific scenarios based on visual dynamic analysis, comprising the following steps: Step 1: Collecting time series data of environmental parameters of the monitoring area based on the environmental parameter sensor network, and collecting time series data of cargo status of the cargo based on the cargo visual monitoring system; Step 2: Perform feature dimensionality reduction on the environmental parameter time series data to extract the environmental feature components; perform response decoupling on the cargo status time series data to extract the cargo response components; Step 3: Establish a dynamic coupling model between the environmental characteristic component and the cargo response component, calculate the quantitative value of the correlation strength between the two using the information entropy algorithm, and mark it as the cargo response entropy value; Step 4: Monitor the evolution curve of the cargo response entropy value. When the rate of change of the evolution curve exceeds the historical threshold and the acceleration of the change is non-zero, it is determined that a state transition event has occurred. Step 5: Extract the environmental inducement parameters and cargo response parameters of the state transition event, encode the environmental inducement parameters and cargo response parameters into risk feature tensors and input them into the accident causal chain knowledge base, output the matching critical intervention parameter set, and formulate the environmental reconstruction instruction set and cargo intervention instruction set based on the critical intervention parameter set.

[0020] In a preferred embodiment of the present invention, the environmental parameter time series data includes temperature field three-dimensional distribution time series data, humidity field gradient evolution time series data, and gas concentration field migration trajectory time series data. Among them, the temperature field three-dimensional distribution time series data is collected by a three-dimensional grid temperature sensor array. The grid node spacing of the array can be dynamically adjusted according to the thermal conductivity coefficient of the dangerous goods. For high-temperature sensitive substances, the corresponding grid density is set higher than that of normal temperature stable substances to achieve accurate monitoring of the temperature field. The humidity field gradient evolution time series data is obtained using a multi-band microwave hygrometer group. The distribution density of the probe is positively correlated with the hygroscopicity of the substance. For deliquescent substances, a double-layer humidity monitoring ring is specially deployed to more comprehensively capture the gradient changes of the humidity field. The gas concentration field migration trajectory time series data is generated by a tunable laser absorption spectrometer. Its scanning path can track the main axis direction of the material leakage and diffusion in real time, providing accurate data for the migration trajectory analysis of the gas concentration field.

[0021] The cargo status time series data includes surface morphology change rate time series data, internal energy release spectrum time series data, and interface reaction propagation direction time series data. The surface morphology change rate time series data is captured by a high-speed polarization camera. The camera frame rate setting meets the spatiotemporal analysis requirements of liquid surface tension fluctuations or solid brittle fracture processes, and can clearly record subtle changes in the cargo surface morphology. The internal energy release spectrum time series data is obtained by integrating infrared thermal imaging and acoustic emission sensing technology. It can synchronously record the peak wavelength of thermal radiation and the main frequency characteristics of stress waves, and realize multi-dimensional monitoring of the internal energy release of the cargo. The interface reaction propagation direction time series data is generated using a nanoparticle tracing system. By marking the motion vector field of the reaction front, the propagation direction and dynamic process of the interface reaction are accurately presented.

[0022] In a preferred embodiment of the present invention, the specific process of performing the feature dimensionality reduction operation on the environmental parameter time series data is as follows: For the three-dimensional distribution of temperature field time series data, spatial clustering analysis methods, such as density-based clustering algorithms, are first used to process the temperature data collected by the three-dimensional grid sensors to identify the dynamic motion trajectories of local high-temperature extreme points. Taking into account the time-varying nature of the temperature field and noise interference, the sliding average of the trajectory tangential acceleration is calculated to effectively smooth short-term fluctuations. This is used as the temperature field extreme point component that represents abnormal temperature field changes. This component can intuitively reflect the migration rate and acceleration characteristics of high-temperature areas.

[0023] When processing time-series data on the evolution of humidity field gradients, we first perform isosurface topological reconstruction on the data acquired by a multi-band microwave hygrometer system. This process generates isosurfaces at different humidity values ​​to characterize the spatial distribution of the humidity field. We then extract the variation in Gaussian curvature of the isosurfaces between adjacent sampling periods, and use root mean square derivatives to enhance the characteristics of regions with sudden changes in the humidity field. Ultimately, we obtain humidity field isosurface components. These components are sensitive to rapid changes in humidity field gradients and are particularly suitable for monitoring abnormal humidity evolution around deliquescent materials.

[0024] Using vortex dynamics analysis methods, we analyze the velocity vector distribution of the gas concentration field to locate the instantaneous center of rotation of the vortex core. By calculating the absolute value of the core's angular velocity around its axis, we quantify the vortex's strength and rotational characteristics, and thus determine the vortex component of the gas concentration field. This component effectively characterizes the dynamic characteristics of vortex motion during gas leakage and diffusion, providing a key parameter for analyzing the migration patterns of the gas concentration field.

[0025] In another preferred embodiment of the present invention, the specific process of performing the response decoupling operation on the cargo status time series data is as follows: For time series data on surface morphology change rates, the empirical mode decomposition method is used to decompose the time series data into multiple intrinsic mode functions and residual components. The slowly varying intrinsic mode functions, which reflect the inherent characteristics of the cargo, are primarily determined by the cargo material properties, while the rapidly varying residual components, which are stimulated by the environment, are directly related to external excitations. The first three order intrinsic mode functions are superimposed as the intrinsic trend component, representing the main inherent trend of the cargo surface morphology change. Higher-order intrinsic mode functions beyond the first three are combined with the residual components to form the stimulated fluctuation component, which is used to characterize abnormal surface morphology fluctuations caused by environmental excitation.

[0026] Wavelet packet transform technology is used to perform multi-resolution analysis on the internal energy release spectrum time series data, decomposing the spectrum data into approximate coefficients and detail coefficients. The approximate coefficients reflecting steady-state characteristics are extracted and reconstructed into a fundamental steady-state component, representing the normal, stable state of internal energy release. Frequency bands in the detail coefficients where energy exceeds a preset threshold are synthesized into a transient offset component, which characterizes abnormal mutations during energy release and effectively identifies transient changes in energy release.

[0027] When processing time-series data on the propagation direction of an interfacial reaction, a baseline direction prediction model is first established. Using the mean of historical propagation directions and environmental parameter regressors as model inputs, the baseline propagation direction of the interfacial reaction is predicted using machine learning or statistical regression methods. The vector difference between the actual and predicted directions is defined as the anomaly deflection component, which quantifies the degree of deviation from the interfacial reaction propagation direction and provides a key basis for determining whether the reaction is abnormal. This component is particularly suitable for monitoring risk scenarios such as runaway chemical reactions.

[0028] In another preferred embodiment of the present invention, the calculation process of the cargo response entropy value is: Calculating the cargo response entropy requires quantifying the coupling strength between the environment and cargo response through multi-step mathematical modeling. First, a mutual information matrix is ​​constructed between the set of environmental characteristic components and the set of cargo response components. This matrix has environmental characteristic components as rows and cargo response components as columns. Based on statistical analysis of historical data, each element in the matrix characterizes the statistical dependence between a specific environmental characteristic component (such as the temperature field extreme point component or the gas concentration field vortex component) and a specific cargo response component (such as the stimulated fluctuation term component or the transient offset component), reflecting the correlation between their changes.

[0029] Secondly, perform singular value decomposition on the mutual information matrix, map the high-dimensional correlation features to a low-dimensional space through matrix decomposition technology, and extract the first three non-zero singular values. For example, assuming that the first three singular values ​​obtained by decomposition are σ1, σ2, and σ3, calculate the geometric mean of these three values, that is, , as the initial correlation strength, this value preliminarily characterizes the overall correlation between the environment and the cargo response.

[0030] Finally, we introduce historical significance weights for environmental characteristic components. These weights are calibrated based on the frequency with which each environmental characteristic component triggers state transition events in historical data. For example, if a temperature field extreme point component appears 80 times out of 100 past risk events, its historical significance weight can be set to 0.8; if a gas concentration field vortex component appears less frequently, its weight can be set to 0.3. The initial correlation strength is multiplied by the corresponding weight to obtain the weighted correlation strength. After taking the negative natural logarithm of the weighted correlation strength, the result is normalized to the range of 0 to 1 through a linear transformation, ultimately obtaining the cargo response entropy value. This entropy quantifies the degree of disorder in the environmental and cargo response system; values ​​closer to 1 indicate a system approaching a disordered state.

[0031] In another preferred embodiment of the present invention, the judgment criteria for the state transition event are: First, the cargo response entropy values ​​for historically stable periods, such as the past 30 days without risk events, are statistically analyzed. The standard deviation σ is calculated, and 4σ, four times the standard deviation, is set as the historical stability threshold. For example, if the standard deviation of the entropy values ​​for the historically stable period is 0.1, the threshold is set to 0.4. Simultaneously, the first-order derivative of the cargo response entropy values ​​is calculated as the rate of change, which measures the speed of entropy change. The second-order derivative is calculated as the acceleration of change, which determines the trend of the rate of change.

[0032] When the entropy rate of change exceeds the historical stability threshold for m consecutive time windows, and the absolute value of the acceleration of change is greater than zero, it is marked as a candidate event. m is a threshold set based on the scenario risk level: in normal scenarios, m can be set to 5, and in high-risk scenarios, it can be set to 3 to improve detection sensitivity. For example, if the rate of change is 0.5 at a certain moment, exceeding the threshold of 0.4, and remains at this level for three consecutive time windows, and the acceleration of change is 0.2 (non-zero), then the candidate event is marked.

[0033] Candidate events require further verification: Check whether the cargo response entropy value shows a monotonic deviation from the historical average baseline. If the entropy value continuously increases within a continuous time window, such as from 0.3 to 0.7, or continuously decreases, such as from 0.6 to 0.2, and the deviation exceeds the historical fluctuation range (i.e., ±2σ of the historical mean), it is confirmed as a state transition event. The system records the event's starting window position, peak window position, and transition direction. If the rate of change is positive, it is determined to be an entropy-increasing phase transition, indicating that the system is developing towards disorder; if it is negative, it is determined to be an entropy-decreasing phase transition, which may be accompanied by the risk of concentrated energy release.

[0034] In another preferred embodiment of the present invention, the specific process of encoding the environmental inducement parameters and the cargo response parameters into the risk feature tensor is as follows: First, for state transition events, a time series segment of environmental characteristic components within the duration of the event is automatically captured. This time period covers key data indicating the significant impact of environmental parameters on the cargo state. By analyzing the magnitude of change in each environmental characteristic component within the time series segment, the three environmental characteristic components with the largest mutation amplitudes are identified, such as the temperature field extreme point component, the humidity field isosurface component, or the gas concentration field vortex component. For each selected component, the sign of the mutation direction is recorded, with a positive sign indicating an increase in the parameter value and a negative sign indicating a decrease in the parameter value. The relative intensity level is also determined. The relative intensity level can be divided into three levels: level 1 represents a small mutation amplitude, level 2 represents a medium mutation amplitude, and level 3 represents a large mutation amplitude. This encoding forms a triplet of environmental inducement parameters. For example, for the temperature field extreme point component, if its mutation direction is positive and its intensity level is 3, the component in the triplet is recorded as having a positive mutation direction and an intensity level of 3.

[0035] Secondly, a time series segment of the cargo response component is intercepted and multiple indicators are calculated for the data within the segment. The stability coefficient of the inherent trend term is calculated, which is used to measure the stability of the cargo's inherent characteristics; the energy proportion of the stimulated fluctuation term is calculated, reflecting the proportion of the cargo response fluctuation energy caused by environmental excitation in the total energy; the degree of retention of the fundamental frequency steady-state component is calculated, representing the maintenance of the steady-state characteristics of the cargo's internal energy release; the intensity of the transient offset component is calculated to quantify the degree of abnormal mutation during the energy release process; and the integral value of the abnormal deflection component angle is calculated to reflect the cumulative deviation of the interface reaction propagation direction. These five indicators are normalized to the range of 0 to 1 through linear transformation, forming a five-dimensional vector of cargo response parameters. For example, if the normalized values ​​of the five indicators are 0.8, 0.6, 0.7, 0.9, and 0.5, respectively, the five-dimensional vector is represented as (0.8, 0.6, 0.7, 0.9, 0.5).

[0036] Finally, a tensor product operation is performed on the triplet of environmental trigger parameters and the five-dimensional vector of cargo response parameters to generate a risk signature tensor. This tensor integrates the key mutation characteristics of environmental triggers and the multi-dimensional quantitative indicators of cargo response. It can comprehensively characterize the risk characteristics of state transition events and provide a structured feature vector for subsequent input into the accident causal chain knowledge base for matching analysis, thereby enabling accurate risk identification and the generation of intervention strategies.

[0037] In a preferred embodiment of this invention, the matching mechanism of the accident cause chain knowledge base is as follows: The accident causal chain knowledge base stores a large number of historical causal chain records. Each record contains a triplet of environmental inducing parameters, a five-dimensional vector of cargo response parameters, and a corresponding set of critical intervention parameters, forming a structured risk case database. The environmental inducing parameter triplet records the key environmental factors that trigger the risk, such as the direction and intensity of the temperature field extreme point component mutation. The five-dimensional vector of cargo response parameters quantifies the multi-dimensional response of cargo to risk events, providing a basis for subsequent intervention strategies.

[0038] The matching process begins with a precise search of environmental trigger parameters. The input environmental trigger parameter triple is parsed, and the combination of mutation components, namely the three environmental characteristic components with the largest mutation amplitudes, their mutation directions, and intensity levels, is extracted. These components are then compared with the environmental trigger parameter triples recorded in all causal chains in the knowledge base. A "complete match" here is defined as the complete consistency of the input triple's mutation component type, mutation direction sign, and intensity level with those of the environmental trigger parameters recorded in a particular causal chain. For example, if the input triple is a temperature field extreme point component with a positive mutation direction and intensity level of 3, a humidity field isosurface component with a negative mutation direction and intensity level of 2, and a gas concentration field vortex component with a positive mutation direction and intensity level of 2, then a complete match is considered only if the environmental trigger parameters of a record in the knowledge base fully conform to the type, direction, and intensity of the aforementioned three components.

[0039] If a fully matched causal chain record is retrieved, the Euclidean distance between the five-dimensional vectors of the cargo response parameters of these records and the input five-dimensional vector is further calculated. The Euclidean distance is an indicator that measures the degree of difference between two high-dimensional vectors. It is obtained by taking the square root of the sum of the squares of the differences in the components of each dimension. The smaller the distance, the greater the similarity between the two. A similarity threshold is preset, and only causal chain records with a Euclidean distance less than the threshold are retained to ensure the relevance of the matching cases. Subsequently, the filtered records are sorted from small to large by Euclidean distance, and the top three records with the smallest distance are selected, and their corresponding critical intervention parameter sets are extracted as candidate outputs. This process ensures that the output intervention plan is highly similar to the current risk characteristics, thereby improving the effectiveness of the intervention.

[0040] When there is no causal chain record in the knowledge base that fully matches the input environmental inducement parameter triple, the fault-tolerant matching mechanism is activated. This mechanism allows the situation where the mutation direction of a component in the matching process has the opposite sign but the same intensity level, and it will still be marked as a "perfect match". For example, if the mutation direction of a component in the input triple is positive and the intensity level is 3, and the corresponding component direction of a record in the knowledge base is negative and the intensity level is 3, and the other two components are fully matched, it is considered a valid match. This design takes into account that in actual applications, environmental parameters may fluctuate in opposite directions but with the same intensity due to different conditions, which improves the applicability of the knowledge base and avoids matching failures due to subtle directional differences. It is especially suitable for similar risk scenarios where environmental parameters fluctuate in opposite directions but with the same intensity.

[0041] In another preferred embodiment of the present invention, the specific process of formulating the environment reconstruction instruction and the cargo intervention instruction according to the critical intervention parameter set is as follows: First, the output critical intervention parameter set is parsed to extract the operational instructions. These instructions contain key information such as the operation object, action parameters, and execution sequence. The operation object can be a temperature field or cargo interface, while the action parameters include temperature reset amplitude and stabilizer dosage. The execution sequence includes the operation start time and duration. During this process, operational instructions with completely overlapping operation objects, action parameters, and execution sequences are identified and designated as common operational instructions. These instructions are consistent across multiple candidate intervention plans and can be retained directly, ensuring the consistency and reliability of the intervention measures.

[0042] For the remaining operations, we screen out conflicting instructions involving the same operation object but with mutually exclusive parameter requirements or overlapping execution time periods. For example, if two operations for the same temperature field require one to raise the temperature to 60°C and the other to lower it to 40°C, these instructions have mutually exclusive parameters. If the execution time intervals of the two operations overlap, this is a time period overlap conflict. If these conflicting instructions are not addressed, they may lead to conflicting intervention measures and affect the effectiveness of risk control. Therefore, priority arbitration is required.

[0043] Conflicting operational instructions are arbitrated according to pre-set priority rules. For conflicting operational instructions involving the same type of parameters, such as temperature control parameters, the higher-intensity instruction is retained first. For example, an instruction with an intensity level of 3 takes precedence over an instruction with an intensity level of 2, ensuring that more forceful intervention measures are adopted. When energy control operational instructions conflict with environmental control operational instructions, based on the core principle of risk control, the energy control instruction is retained first, such as the operation that terminates the chemical reaction chain, to curb the evolution of risks at the source and prevent further escalation.

[0044] Public operation instructions and non-conflicting operation instructions retained after arbitration are sorted to form a basic intervention sequence. This sequencing rule follows the reverse chronological order of the risk transmission path, that is, from the risk consequences to the risk source. For example, operations that directly suppress the development of an accident, such as terminating the reaction, are executed first, followed by auxiliary operations such as environmental adjustments. This ensures that intervention measures can quickly control risks and prevent the situation from escalating.

[0045] Subsequently, the cargo response parameter indicators in the current risk feature tensor are obtained and dynamic operations are inserted. If the intensity of the transient offset component exceeds the pre-set normalized safety threshold, it indicates that the energy release inside the cargo has abnormal fluctuations. During the critical phase of energy release in the basic intervention sequence, such as the peak period of energy release, a spectrum stabilization operation is inserted to adjust the energy release rhythm through vibrations of a specific frequency, so that the energy distribution tends to be stable and accidents caused by concentrated energy release are avoided. If the angle integral value of the abnormal deflection component is greater than the geometric tolerance, it indicates that the propagation direction of the interface reaction has significantly deviated. At the propagation path offset point, that is, the location where the reaction interface begins to deviate, a spatial correction operation is inserted to guide the reaction direction by setting physical barriers, etc., to ensure that the reaction proceeds as expected and to avoid further spread of risks.

[0046] The adjusted basic intervention sequences are categorized and integrated to generate an environmental reconstruction instruction set and a cargo intervention instruction set. The environmental reconstruction instruction set includes a temperature field reset command, which specifies the reset amplitude and duration to restore the normal temperature environment; a humidity field balance command, which specifies the balance target and execution time to ensure that the humidity is within a safe range; a gas concentration field purification command, which determines the scope and intensity of the purification action to eliminate the impact of harmful gases, aiming to eliminate risk inducements by adjusting environmental parameters. The cargo intervention instruction set includes a chemical bond stabilizer injection scheme, which specifies the stabilizer chemical type and injection dosage to stabilize the chemical structure of the cargo; an interface energy barrier construction scheme, which specifies the energy barrier space coordinates and construction method to control the reaction interface; and a mechanical stress relief scheme, which sets the relief activation sequence and release amount to release internal stress in the cargo, directly intervening in the cargo state and suppressing risk evolution.

[0047] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for identifying security risks in specific scenarios based on visual dynamic analysis, characterized in that: The following steps are involved: The environmental parameter sensor network collects the time series data of the environmental parameters of the monitoring area, and the cargo visual monitoring system collects the time series data of the cargo status; Perform feature dimensionality reduction on the environmental parameter time series data to extract environmental feature components; Perform response decoupling operations on the cargo status time series data to extract cargo response components; A dynamic coupling model between environmental characteristic components and cargo response components is established. The quantitative value of the correlation strength between the two is calculated using the information entropy algorithm and marked as the cargo response entropy value. Monitor the evolution curve of the cargo response entropy value. When the rate of change of the evolution curve exceeds the historical threshold and the acceleration of the change is non-zero, a state transition event is determined to have occurred. Extract the environmental inducement parameters and cargo response parameters of the state transition event, encode the environmental inducement parameters and cargo response parameters into risk feature tensors and input them into the accident causal chain knowledge base, output the matching critical intervention parameter set, and formulate the environmental reconstruction instruction set and cargo intervention instruction set based on the critical intervention parameter set.

2. The method for identifying security risks in specific scenarios based on visual dynamic analysis according to claim 1, characterized in that: The environmental parameter time series data includes the three-dimensional distribution time series data of the temperature field, the gradient evolution time series data of the humidity field, and the migration trajectory time series data of the gas concentration field; the cargo status time series data includes the surface morphology change rate time series data, the internal energy release spectrum time series data, and the interface reaction propagation direction time series data.

3. The method for identifying security risks in specific scenarios based on visual dynamic analysis according to claim 2 is characterized in that: The specific process of performing feature dimensionality reduction operation on environmental parameter time series data is as follows: Perform spatial cluster analysis on the three-dimensional distribution time series data of the temperature field to identify the motion trajectory of the local high-temperature extreme point, and calculate the sliding average of the trajectory tangential acceleration as the component of the temperature field extreme point; The isosurface topology of the humidity field gradient evolution time series data is reconstructed, and the root mean square derivative of the Gaussian curvature change of the isosurface in adjacent sampling periods is extracted as the humidity field isosurface component. Vortex dynamics analysis is carried out on the time series data of gas concentration field migration trajectory to locate the instantaneous rotation center of the vortex core, and the absolute value of the core's angular velocity around the axis is calculated as the vortex component of the gas concentration field.

4. The method for identifying security risks in specific scenarios based on visual dynamic analysis according to claim 2 is characterized in that: The specific process of performing response decoupling operations on cargo status time series data is as follows: Perform empirical mode decomposition on the surface morphology change rate time series data to separate the slowly varying intrinsic mode functions reflecting the inherent characteristics of the cargo from the rapidly varying residual components stimulated by the environment. The superposition of the first three order intrinsic mode functions is labeled as the inherent trend component, and the combination of higher order intrinsic mode functions and residual components beyond the first three is labeled as the stimulated fluctuation component. Wavelet packet transform is performed on the internal energy release spectrum time series data to extract the approximate coefficient reconstruction signal reflecting the steady-state characteristics as the fundamental frequency steady-state component, and the frequency band where the detail coefficient energy exceeds the threshold is synthesized as the transient offset component; A baseline direction prediction model is established for the time series data of interface reaction propagation direction. The model input is the mean of historical propagation direction and the regression of environmental parameters. The vector difference between the actual direction and the predicted direction is defined as the abnormal deflection component.

5. The method for identifying security risks in specific scenarios based on visual dynamic analysis according to claim 1 is characterized in that: The calculation process of the cargo response entropy value is: Constructing a mutual information matrix between the set of environmental characteristic components and the set of cargo response components, where the matrix elements represent the statistical dependence between a specific environmental characteristic component and a specific cargo response component; Calculate the singular value decomposition of the mutual information matrix and extract the geometric mean of the first three non-zero singular values ​​as the initial correlation strength; The historical significance weight of the environmental characteristic component is introduced. The weight value is calibrated according to the frequency of state transition events triggered by the environmental characteristic component in the past. The initial association strength is multiplied by the historical significance weight to obtain the weighted association strength. The negative natural logarithm of the weighted association strength is taken and normalized to the interval [0,1], which is the cargo response entropy value.

6. The method for identifying security risks in specific scenarios based on visual dynamic analysis according to claim 1, characterized in that: The criteria for determining the state transition event are: Calculate the standard deviation of the cargo response entropy value during the historical stable period, set 4 times the standard deviation as the historical stability threshold, calculate the first-order derivative of the cargo response entropy value as the rate of change, and the second-order derivative as the acceleration of change; When the rate of change exceeds the historical stability threshold for m consecutive time windows and the absolute value of the change acceleration is greater than zero, it is marked as a candidate event, and m is the set threshold; Verify whether the cargo response entropy value in the candidate event shows a monotonically deviating trend from the historical average baseline. If so, confirm that the candidate event is a state transition event, and record the starting window position, peak window position, and transition direction of the state transition event. The transition direction is divided into entropy-increasing phase transition and entropy-decreasing phase transition based on the sign of the change rate.

7. The method for identifying security risks in specific scenarios based on visual dynamic analysis according to claim 4 is characterized in that: The specific process of encoding environmental inducement parameters and cargo response parameters into risk feature tensors is as follows: Intercept the time series segments of the environmental characteristic components during the duration of the state transition event; identify the three environmental characteristic components with the largest mutation amplitude in the time series segments, record their mutation direction signs and relative intensity levels, and encode them into a triplet of environmental inducement parameters; Intercept the time series segments of the cargo response components, calculate the stability coefficient of the inherent trend term, the energy proportion of the stimulated fluctuation term, the degree of retention of the fundamental frequency steady-state component, the intensity of the transient offset component, and the integral value of the abnormal deflection component angle, and normalize the five indicators of the cargo response components into a five-dimensional vector of cargo response parameters; The tensor product of the triplet of environmental inducement parameters and the five-dimensional vector of cargo response parameters is labeled as the risk feature tensor.

8. The method for identifying security risks in specific scenarios based on visual dynamic analysis according to claim 7 is characterized in that: The matching mechanism of the accident cause chain knowledge base is specifically as follows: The accident causal chain knowledge base includes a plurality of causal chain records, each of which includes a different environmental inducement parameter triple and a cargo response parameter five-dimensional vector, as well as a matching critical intervention parameter set; First, the causal chain records that completely match the combination of mutation components in the input environmental trigger parameter triple are retrieved; then the Euclidean distance between the retrieved causal chain records and the input five-dimensional vector of the cargo response parameters is calculated, and the causal chain records with Euclidean distance less than the similarity threshold are selected; the selected causal chain records are sorted in ascending order of Euclidean distance, and the critical intervention parameter sets corresponding to the top three causal chain records are selected for output; When there is no completely matched causal chain record, a component with opposite sign but the same intensity level is allowed to appear in the matching process, and it is also marked as a completely matched one.

9. The method for identifying security risks in specific scenarios based on visual dynamic analysis according to claim 8, characterized in that: The specific process of formulating environmental reconstruction instructions and cargo intervention instructions based on the critical intervention parameter set is as follows: Analyze the output critical intervention parameter set and extract the operation instructions whose operation objects, action parameters and execution time sequences completely overlap as common operation instructions; filter out the conflicting operation instructions involving the same operation object but with mutually exclusive parameter requirements or overlapping execution time sequences from the remaining operation instructions; For conflicting operation instructions with the same type of parameters, the operation instruction with a higher intensity level shall be retained first; when the energy control operation instruction conflicts with the environmental adjustment operation instruction, the energy control operation instruction shall be retained first; The common operation instructions and the reserved non-conflicting operation instructions are sorted in reverse chronological order of the risk transmission path to form a basic intervention sequence. The cargo response parameter index in the current risk characteristic tensor is obtained. If the intensity of the transient offset component exceeds the normalized safety threshold, a spectrum stabilization operation is inserted at the critical energy release phase of the basic intervention sequence. If the angle integral value of the abnormal deflection component is greater than the geometric tolerance, a spatial correction operation is inserted at the propagation path offset point of the basic intervention sequence. The environmental field operation instructions and cargo state intervention operation instructions in the adjusted basic intervention sequence are integrated to generate an environmental reconstruction instruction set and a cargo intervention instruction set.

Images