Risk situation automatic evaluation method based on multi-source data access

Abstract

The present application relates to the technical field of automatic risk situation assessment, and discloses a risk situation automatic assessment method based on multi-source data access. The method accesses multi-source heterogeneous data streams, generates a uniform spatiotemporal reference risk event set after standardization processing. The cause-and-effect chain between events is established based on the relationship between time and space, and a composite risk trajectory is integrated and generated. The diffusion energy level, influence range and structural stability of the trajectory are calculated in real time, and the trajectory is dynamically assigned a stage label of dormancy, activity or outbreak. By continuously monitoring the stage labels and physical influence ranges of different trajectories, when multiple trajectories simultaneously turn to the active or outbreak state and the spatial ranges overlap, the overlapping area is automatically determined as a high-risk conflict domain and an early warning is generated. The method realizes continuous tracking of the risk evolution process and automatic perception of multi-risk interaction conflicts.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic risk situation assessment, and particularly to a risk situation automatic assessment method based on multi-source data access. BACKGROUND

[0002] The existing risk situation assessment technology mainly relies on the monitoring of a single or limited data source, and lacks the access and fusion capability of multi-source heterogeneous risk perception data. The system usually uses static rules or fixed models to make independent judgments on events, and it is difficult to extract risk information with internal logical association from different sources of dynamic and continuous data streams. This leads to a one-sided and isolated description of complex risks, which cannot reflect the real evolution process of risks in the time and space dimensions.

[0003] The existing technology mainly uses simple time or space proximity principles when dealing with the correlation between risk events, and lacks the ability to automatically trace and dynamically build the causal relationship between events. The evaluation results are usually snapshot presentations of discrete events, and cannot form continuous trajectories describing the whole process of risk generation, diffusion and evolution. In addition, the existing scheme usually evaluates a single risk entity, lacks real-time monitoring and judgment mechanism for the interaction mode between different risk evolution entities, and cannot warn the regional strong conflict effect caused by the convergence and superposition of multiple risks in space and time.

[0004] A technology is needed that can automatically build and continuously track risk evolution trajectories from multi-source data. At the same time, a technology scheme is needed that can identify the interaction between multiple independent risk trajectories in real time and automatically warn potential high-risk conflict areas. SUMMARY

[0005] The purpose of the present application is to provide a risk situation automatic assessment method based on multi-source data access to solve the problems raised in the background art.

[0006] To achieve the above purpose, the present application provides a risk situation automatic assessment method based on multi-source data access, which comprises:

[0007] Continuously access heterogeneous risk perception data streams from at least three independent sources, perform meta-information stripping and ontology alignment on each data stream, and generate a standardized risk event set with a unified space-time coordinate baseline;

[0008] Perform risk event tracing operations on the standardized risk event set, which includes establishing causal relationship chains between multiple standardized risk events according to the time sequence and physical proximity of event occurrence, and integrating multiple standardized risk events with causal correlation into a composite risk trajectory;

[0009] Tracking the evolution state of each composite risk trajectory in real time, the evolution state including diffusion energy level, influence range and internal structure stability, and dynamically assigning a running phase label of dormant, active or outbreak stage to each composite risk trajectory according to the current evolution state of the composite risk trajectory;

[0010] Continuously comparing the interaction mode between different composite risk trajectories, when the running phase labels of two or more composite risk trajectories are simultaneously converted to active or outbreak, and there is an overlapping area in their physical influence range, the overlapping area is determined as a high-risk conflict domain, and a risk field domain conflict warning is generated.

[0011] Preferably, the operation of performing meta-information stripping and ontology alignment on each data stream includes parsing the native protocol format of each data stream, extracting the payload containing timestamp, geographic coordinates, risk type code and intensity reading, and stripping the attached information including transmission path and device identification;

[0012] The extracted payload is semantically mapped according to the preset risk ontology knowledge graph, and the different descriptions of the same risk entity in different sources are uniformly mapped to the standard concept node in the knowledge graph, so as to complete the ontology alignment;

[0013] The unified space-time coordinate baseline is realized by introducing a virtual global space-time grid, and all the payloads after ontology alignment are normalized and mapped to a specific space-time voxel in the global space-time grid according to their timestamp and geographic coordinates, forming a standardized risk event set, and each space-time voxel can accommodate at most one standardized risk event.

[0014] Preferably, the risk event tracing operation is performed on the global space-time grid, taking the standardized risk event at the previous time as the potential cause and the standardized risk event at the next time as the potential result;

[0015] Iterative search for a pair of standardized risk events that are continuous in time and adjacent in space, if the two continuous standardized risk events have inheritance or derivation relationship in risk type code, or the intensity reading shows a logical growth trend, a directed causal relationship edge is established between them;

[0016] By iteratively connecting multiple causal relationship edges, multiple scattered standardized risk events are connected into a complete composite risk trajectory, and each composite risk trajectory is assigned a unique trajectory identifier.

[0017] Preferably, the diffusion energy level of the composite risk trajectory is obtained by calculating the weighted moving average of the intensity readings of all standardized risk events contained in the composite risk trajectory, and the weight value decays with the remoteness of the event occurrence time;

[0018] The influence range of the composite risk trajectory is obtained by calculating the minimum convex hull area of the geographic coordinates of all the standardized risk events contained in the composite risk trajectory;

[0019] The internal structure stability of the composite risk trajectory is obtained by calculating the fluctuation variance of the number and rate of newly added causal relationship edges of the composite risk trajectory in the latest time window.

[0020] Preferably, the dynamic allocation of the running phase label is based on a set of adaptive thresholds that are dynamically adjusted according to the statistical characteristics of the composite risk trajectory in historical data;

[0021] When the diffusion energy level of the composite risk trajectory is lower than the active threshold, the influence area is smaller than the range threshold, and the internal structure stability is higher than the stability threshold, it is assigned a dormant label;

[0022] When the diffusion energy level exceeds the active threshold but does not reach the outbreak threshold, or the influence area exceeds the range threshold but the internal structure stability is lower than the stability threshold, it is assigned an active label;

[0023] When the diffusion energy level exceeds the outbreak threshold, and at the same time, the influence area exceeds the range threshold and the internal structure stability is lower than the instability threshold, it is assigned an outbreak label.

[0024] Preferably, the comparison of the interaction mode is realized by continuously calculating the spatio-temporal correlation degree between any two different composite risk trajectories;

[0025] The spatio-temporal correlation degree is calculated by the event occurrence coincidence degree of the two composite risk trajectories on the time axis, the influence range overlap area proportion in space, and the semantic similarity of the running phase labels of the two trajectories;

[0026] When the spatio-temporal correlation degree of the two composite risk trajectories exceeds the preset strong correlation threshold, and the running phase labels of each are active or outbreak, it is determined that there is a strong interaction between them.

[0027] Preferably, the determination of the high-risk conflict domain is immediately executed when a strong interaction is monitored, and the geographic boundary of the high-risk conflict domain is defined by the intersection area of the convex hulls of the influence ranges of all the composite risk trajectories that interact;

[0028] A separate conflict domain archive is generated for each determined high-risk conflict domain, which records the identifiers of all the composite risk trajectories involved in the conflict, the spatial boundary coordinates of the conflict domain, the conflict triggering time, and the initial conflict intensity;

[0029] The risk field conflict early warning is generated based on the conflict field archive, and the early warning content at least includes the position of the high-risk conflict field, the number of involved risk trajectories, and the estimated conflict duration.

[0030] Preferably, after the risk field conflict early warning is generated, evolution intervention logic for the affected composite risk trajectory is started;

[0031] The evolution intervention logic first performs trajectory toughness evaluation on each composite risk trajectory in strong interaction, and the trajectory toughness is measured by the average duration that the composite risk trajectory maintains its structure without collapse in a similar conflict situation in history.

[0032] According to the trajectory toughness evaluation result, an intervention priority is generated for each composite risk trajectory involved in the conflict, and the trajectory with lower toughness is given higher intervention priority.

[0033] Preferably, according to the intervention priority, a virtual isolation operation is performed on the selected composite risk trajectory starting from the high priority.

[0034] The virtual isolation operation is realized by inserting a virtual neutralization event node in the causal chain of the composite risk trajectory, and the neutralization event node is marked as “inhibition” in the risk type code, and the strength reading is a negative value, which is used to offset the strength growth of the subsequent events in the causal chain.

[0035] After the virtual isolation operation, the diffusion energy level, influence range and internal structure stability of the intervened composite risk trajectory are recalculated, and the running phase label is updated accordingly.

[0036] Preferably, after the virtual isolation operation on all composite risk trajectories higher than the specified priority is completed, the state of the original high-risk conflict field is re-evaluated.

[0037] The spatiotemporal correlation degree between all composite risk trajectories remaining in the region is recalculated, and if the updated spatiotemporal correlation degree is below the strong correlation threshold, it is determined that the original high-risk conflict field has been eliminated, and a conflict field elimination notice is generated.

[0038] If part of the spatiotemporal correlation degree is higher than the strong correlation threshold, it is determined whether a new high-risk conflict field is formed according to the updated running phase label, and the whole process from risk event tracing to evolution intervention is iteratively executed until no new risk field conflict early warning is generated.

[0039] Compared with the prior art, the present application has the following advantages:

[0040] Through event-sourced operations, discrete events are integrated into continuous composite risk trajectories according to the causal chain established by the time sequence and physical proximity of standardized events. Then the diffusion level, influence range and internal structural stability of the trajectory are quantified in real time, and based on these dynamic parameters, the trajectory is assigned a running phase label. This makes the evolution of risk be characterized as a life cycle entity with state, rather than a series of isolated events, realizing the cognitive transformation of risk situation from static snapshot to dynamic process.

[0041] By continuously comparing the running phase labels of different composite risk trajectories, and when multiple trajectories are found to simultaneously jump to active or outbreak, and their physical influence ranges overlap in space, the overlapping area is automatically determined as a high-risk conflict domain and a warning is generated. This realizes the transition from independent monitoring of a single risk to systematic perception of the spatio-temporal interaction and conflict effect between multiple risk entities, and the triggering condition of the warning changes from a single threshold to a composite logic based on the coordinated changes of multiple entity states and spatial relationships.

[0042] The whole process constructs an evaluation model that can describe the dynamic evolution of risk and its interaction by standardizing, tracking and state processing of multi-source heterogeneous data. The result of situation assessment is upgraded from a list of discrete events to a spatial network graph reflecting the evolution phase of risk entities and their interaction, improving the systematicness and forward-looking of risk perception. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The working principle diagram of the risk situation automatic evaluation method based on multi-source data access described in the application;

[0044] Figure 2 The flowchart for meta-information stripping and ontology alignment and generation of standardized risk event set;

[0045] Figure 3 The flowchart for calculating the evolution state parameters of composite risk trajectory;

[0046] Figure 4 The running phase distribution statistical graph of different time intervals of composite risk trajectory;

[0047] Figure 5 The conflict intensity and resolution state comparison graph before and after the intervention of high-risk conflict domain. DETAILED DESCRIPTION

[0048] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0049] With reference to Figure 1 The present application provides a risk situation automatic evaluation method based on multi-source data access, which comprises: continuously accessing heterogeneous risk perception data streams from at least three independent sources, performing meta-information stripping and ontology alignment operations on each data stream, and generating a standardized risk event set with a unified space-time coordinate baseline. The risk event tracing operation is performed on the standardized risk event set. According to the time sequence and physical proximity of the event occurrence, a causal relationship chain is established between multiple standardized risk events, and multiple standardized risk events with causal correlation are integrated into a composite risk track. The evolution state of each composite risk track is tracked in real time, including the diffusion energy level, the influence range and the internal structure stability, and a running phase label in the dormant, active or outbreak stage is dynamically assigned to the composite risk track according to the current evolution state of the composite risk track. The interaction mode between different composite risk tracks is continuously compared. When the running phase labels of two or more composite risk tracks are simultaneously converted to active or outbreak, and there is an overlapping area in their physical influence range, the overlapping area is determined as a high-risk conflict domain, and a risk field conflict warning is generated.

[0050] In an embodiment of the present application, with reference to Figure 2 Performing meta-information stripping and ontology alignment on each data stream comprises analyzing the native protocol format of each data stream, extracting the payload containing timestamp, geographic coordinates, risk type code and intensity reading, and stripping the attached information including transmission path and device identification. The extracted payload is semantically mapped according to the preset risk ontology knowledge graph, and the different descriptions of the same risk entity in different sources are uniformly mapped to the standard concept node in the knowledge graph, so as to complete the ontology alignment. The unified space-time coordinate baseline is realized by introducing a virtual global space-time grid. All the payloads that have been ontology-aligned are normalized and mapped to a specific space-time voxel in the global space-time grid according to their timestamp and geographic coordinates, forming a standardized risk event set, and each space-time voxel can accommodate at most one standardized risk event.

[0051] In a specific implementation, performing meta-information stripping and ontology alignment for each data stream includes parsing native protocol format of each data stream and extracting payload. For example, the accessed heterogeneous risk-aware data streams can include data streams from weather monitoring sites, data streams from traffic flow sensors, and data streams from social media public opinion analysis platforms. The weather monitoring site data stream adopts a binary protocol, and a data frame thereof contains wind speed, rainfall intensity, air pressure value, site geographic coordinates, and millisecond-level timestamp. The parsing operation identifies the protocol frame header and frame tail, extracts the wind speed reading as the intensity reading, extracts the rainfall type code as the risk type code, extracts the geographic coordinates and timestamp, and strips the site device serial number and signal strength auxiliary information in the data frame. The traffic flow sensor data stream adopts a JSON format message, and the message body contains lane number, average vehicle speed, vehicle density, latitude and longitude coordinates, and second-level timestamp. The parsing operation reads the JSON key-value pairs, extracts the vehicle density reading as the intensity reading, extracts the traffic congestion event code as the risk type code, extracts the latitude and longitude coordinates and timestamp, and strips the sensor model and network address auxiliary information in the message header. The social media public opinion analysis platform data stream adopts an XML format push, and the effective content includes topic text, sentiment polarity score, geographic location label, and posting time. The parsing operation parses the XML tags, extracts the sentiment polarity score as the intensity reading, extracts the social panic event code as the risk type code, extracts the coordinates parsed from the geographic location label and the posting time, and strips the push source user identifier and forwarding path auxiliary information. Through the above operations, the payload containing the timestamp, geographic coordinates, risk type code, and intensity reading is extracted from each data stream.

[0052] In some embodiments, the extracted payloads are semantically mapped against a preset risk ontology knowledge graph to achieve ontology alignment. The risk ontology knowledge graph defines a standardized risk concept hierarchy and relationships, for example, setting "meteorological disasters" as the top-level concept, which includes "strong winds", "heavy rain" and other sub-concepts; setting "traffic incidents" as the top-level concept, which includes "congestion", "accidents" and other sub-concepts; setting "public safety incidents" as the top-level concept, which includes "social panic", "rumor spread" and other sub-concepts. The semantic mapping operation encodes and associates the risk types extracted from different source data streams to the corresponding standard concept nodes in the risk ontology knowledge graph. The rainfall type code "HEAVY_RAIN" in the meteorological monitoring station data stream is mapped to the "heavy rain" concept node in the knowledge graph; the traffic congestion event code "JAM_LEVEL_3" in the traffic flow sensor data stream is mapped to the "congestion" concept node in the knowledge graph; the social panic event code "PANIC_REPORT" in the social media public opinion analysis platform data stream is mapped to the "social panic" concept node in the knowledge graph. Through mapping, the different descriptions of the same or similar risk entities in different sources are unified to the standard concept nodes in the knowledge graph, thereby completing ontology alignment, for example, the "wind speed value" and "wind level" described by different sensors for wind are unified and associated to "strong wind" concept.

[0053] It can be understood that the unified space-time coordinate baseline is realized by introducing a virtual global space-time grid. The global space-time grid is a reference system that is discretely divided in the spatial dimension and the time dimension. In the spatial dimension, the grid divides the geographical area into square cells with a side length of L meters; in the time dimension, the grid divides the time axis into time slices with a fixed length of T seconds. Each cell determined by a spatial cell and a time slice is called a space-time voxel. All payloads that have undergone ontology alignment are normalized and mapped into specific space-time voxels of the global space-time grid according to their timestamps and geographical coordinates. For a payload, the index of the spatial cell to which it belongs is determined by dividing the geographical coordinates by L and taking the integer part downward, and the index of the time slice to which it belongs is determined by dividing the timestamp by T and taking the integer part downward. The mapping operation ensures that each space-time voxel contains at most one standardized risk event, and if multiple payloads appear in the same space-time voxel, the payload with the highest intensity reading or the latest time is selected to generate a standardized risk event according to the preset rules. In the standardized risk event set formed after mapping, each event has a unified space-time coordinate based on the global space-time grid. The calculation of the space-time voxel index can use the following relationship:

[0054]

[0055] wherein: represents the space-time voxel index, and is a geographical coordinate component of the payload, is a timestamp of the payload, is a spatial cell edge length, is a time slice length, symbol denotes a floor operation, subscript , , denote a spatial X-direction index, a spatial Y-direction index and a time-direction index, respectively.

[0056] In one embodiment of the present application, referring to Figure 3 , the risk event provenance operation is performed on a global spatio-temporal grid, with a standardized risk event at a previous time instant as a potential cause and a standardized risk event at a later time instant as a potential result. A traversal search is performed to find pairs of standardized risk events that are consecutive in time and adjacent in space. If the two consecutive standardized risk events have an inheritance or derivation relationship in risk type encoding, or exhibit a logical increasing trend in intensity reading, a directed causal relationship edge is established between them. By iteratively connecting multiple causal relationship edges, multiple dispersed standardized risk events are concatenated into a complete composite risk trajectory, and each composite risk trajectory is assigned a unique trajectory identifier. The diffusion energy level of a composite risk trajectory is obtained by calculating a weighted moving average of intensity readings of all standardized risk events contained in the composite risk trajectory, with the weight decaying with the remoteness of the event occurrence time. The influence range of a composite risk trajectory is obtained by calculating the minimum convex hull area formed by the geographical coordinates of all standardized risk events contained in the composite risk trajectory. The internal structural stability of a composite risk trajectory is obtained by calculating the fluctuation variance of the number and rate of newly added causal relationship edges within a recent time window.

[0057] In practice, risk event tracing is performed on a global spatiotemporal grid. The operational logic is that a standardized risk event from a previous moment is the potential cause, and a standardized risk event from a subsequent moment is the potential result. The system traverses and searches for pairs of standardized risk events that are temporally continuous and spatially adjacent. Continuous means that the time slice index values ​​of two standardized risk events differ by 1, and adjacent means that the spatial cells of the two standardized risk events are spatially contiguous or diagonally adjacent in the global spatiotemporal grid. If, in a searched pair of standardized risk events, the two consecutive standardized risk events have an inheritance or derivation relationship in risk type encoding, or the intensity readings show a logical increasing trend, then a directed causal relationship edge is established between them. The inheritance or derivation relationship is determined based on the risk ontology knowledge graph. For example, if the concept of "strong wind" in the knowledge graph may derive the concept of "trees falling," then a standardized risk event with a risk type encoding mapped to "strong wind" and a slightly later, spatially adjacent standardized risk event with a risk type encoding mapped to "trees falling" can be determined to have a derivation relationship. Logical growth trends are determined when the intensity reading of a subsequent event is greater than that of a preceding event. By iteratively connecting multiple causal relationship edges, multiple dispersed standardized risk events are chained together into a complete composite risk trajectory, and a unique trajectory identifier is assigned to each composite risk trajectory.

[0058] In some embodiments, the diffusion level of a composite risk trajectory is obtained by calculating a weighted moving average of the intensity readings of all normalized risk events contained in the composite risk trajectory. For a composite risk trajectory, the normalized risk events are arranged in ascending order by time slice index, and the intensity reading of each event is... The corresponding time slice index is The current time slice index is Calculate the diffusion energy level The relationship is:

[0059]

[0060] in: It is the total number of standardized risk events included in the composite risk trajectory. It is the attenuation factor, a constant between 0 and 1, and the weight. The decay factor is set to be proportional to the time difference between the event occurrence time and the current time, i.e. the older the event, the smaller the weight of its intensity reading on the current diffusion level. The impact range of a compound risk trajectory is calculated as the minimum convex hull area of all the standardized risk events contained in the trajectory. The convex hull algorithm is used to calculate the minimum convex polygon that encloses all the input points, and the area of the polygon is taken as the impact range. The internal structure stability of a compound risk trajectory is calculated as the variance of the number of newly added causal relationship edges in a recent time window. A time window of M consecutive time slices is set, and the number of newly added edges in each time slice is counted to form a sequence. The variance of the sequence is taken as the measure of the internal structure stability, and a higher variance indicates a less stable internal structure.

[0061] It can be understood that the traversal search of causal relationships and the construction of compound risk trajectories are a continuous process. In the global spatio-temporal grid, as new time slice data arrives, new standardized risk events are generated. The traceability operation will take these new events as potential results and search for spatially adjacent events in the previous time slice as potential causes, trying to establish new causal relationship edges. The newly established causal relationship edges may extend an existing compound risk trajectory, may connect two independent trajectories to merge them, or may create a new trajectory containing only a single causal relationship edge based on a new event. After each successful establishment of a causal relationship edge or merging of trajectories, the diffusion level, impact range, and internal structure stability of the affected compound risk trajectories need to be recalculated. Optionally, when establishing a causal relationship edge, in addition to checking the inheritance and derivation relationship of the risk type code, it is also possible to check whether the growth of the intensity reading exceeds a minimum threshold to avoid false associations caused by noise. Optionally, the trajectory identifier assigned to each compound risk trajectory can contain the spatio-temporal voxel index of its initial event occurrence and a random sequence code to ensure uniqueness.

[0062] In one embodiment of the application, the dynamic assignment of the running phase label is based on a set of adaptive thresholds, which are dynamically adjusted according to the statistical characteristics of the composite risk trajectories in the historical data. When the diffusion energy level of a composite risk trajectory is lower than the active threshold, the influence area is smaller than the range threshold, and the internal structure stability is higher than the stability threshold, it is assigned with the dormant label. When the diffusion energy level exceeds the active threshold but does not reach the outbreak threshold, or the influence area exceeds the range threshold but the internal structure stability is lower than the stability threshold, it is assigned with the active label. When the diffusion energy level exceeds the outbreak threshold, and at the same time the influence area exceeds the range threshold and the internal structure stability is lower than the instability threshold, it is assigned with the outbreak label. The comparison of interaction patterns is realized by continuously calculating the spatiotemporal correlation degree between any two different composite risk trajectories. The spatiotemporal correlation degree is calculated by the coincidence degree of event occurrence on the time axis, the proportion of the overlapping area of the influence range in space, and the semantic similarity of the running phase labels of the two trajectories. When the spatiotemporal correlation degree of two composite risk trajectories exceeds the preset strong correlation threshold, and the running phase labels of each are active or outbreak, it is determined that there is strong interaction between them.

[0063] In specific implementation, the dynamic assignment of the running phase label is based on a set of adaptive thresholds, which are dynamically adjusted according to the statistical characteristics of the composite risk trajectories in the historical data. When the diffusion energy level of a composite risk trajectory is lower than the active threshold, the influence area is smaller than the range threshold, and the internal structure stability is higher than the stability threshold, it is assigned with the dormant label. When the diffusion energy level exceeds the active threshold but does not reach the outbreak threshold, or the influence area exceeds the range threshold but the internal structure stability is lower than the stability threshold, it is assigned with the active label. When the diffusion energy level exceeds the outbreak threshold, and at the same time the influence area exceeds the range threshold and the internal structure stability is lower than the instability threshold, it is assigned with the outbreak label. The dynamic adjustment of the adaptive thresholds is based on the historical values of the indicators of all composite risk trajectories in a sliding time window, for example, the active threshold can be set to the 70th percentile of the historical values of the diffusion energy level, the outbreak threshold is set to the 90th percentile, the range threshold is set to the 65th percentile of the historical values of the influence area, the stability threshold is set to the 30th percentile of the historical values of the internal structure stability, and the instability threshold is set to the 15th percentile. These percentiles are recalculated regularly according to new historical data to reflect the latest distribution characteristics of the situation.

[0064] In some embodiments, the comparison of interaction patterns is achieved by continuously calculating the spatiotemporal correlation between any two different composite risk trajectories. The spatiotemporal correlation is calculated from the coincidence of event occurrence on the time axis, the proportion of overlapping area of influence range in space, and the semantic similarity of the running phase labels of the two trajectories. The coincidence of event occurrence on the time axis is obtained by comparing the sets of time slice indexes of the standardized risk events contained in the two composite risk trajectories, and calculating the ratio of the intersection size to the union size of the two sets of time slice indexes. The proportion of overlapping area of influence range in space is obtained by calculating the ratio of the intersection area of the convex hull polygons of the influence ranges of the two composite risk trajectories to the smaller convex hull area. The semantic similarity of the running phase labels is mapped to numerical values according to a preset mapping relationship, for example, dormancy is mapped to 1, active is mapped to 2, and outbreak is mapped to 3, and then the inverse of the difference between the corresponding numerical values of the two labels is calculated. The calculation relationship of the spatiotemporal correlation R is:

[0065]

[0066] wherein: represents the coincidence of event occurrence on the time axis, represents the proportion of overlapping area of influence range in space, represents the semantic similarity of the running phase labels, , , is a preset weighting coefficient, and satisfies . The weighting coefficient can be configured according to the emphasis of time, space, and label factors in different application scenarios.

[0067] It can be understood that when the spatiotemporal correlation of the two composite risk trajectories exceeds a preset strong correlation threshold, and the running phase labels of each are active or outbreak, it is determined that there is strong interaction between the two. The strong correlation threshold is a preset constant between 0 and 1, used to distinguish the strength of the correlation. The determination logic is that only when the running phase labels of the two trajectories indicate non-dormant state at the same time, and the calculated spatiotemporal correlation R value is greater than the strong correlation threshold, the determination of strong interaction is triggered. Optionally, when calculating the semantic similarity of the running phase labels, a more detailed mapping rule can be set, for example, dormancy is mapped to 1, active is mapped to 2, and outbreak is mapped to 4, so that the difference between the outbreak label and the active label is greater than the difference between the active label and the dormancy label, thereby reflecting the stronger conflict potential of the outbreak state in the calculation. Optionally, the coincidence of event occurrence on the time axis can be calculated in a weighted form, giving higher weight to the event coincidence closer to the current time.

[0068] ​In one embodiment of the present application, the determination of high-risk conflict domain is performed immediately upon the detection of strong interaction, and the geographical boundary of the high-risk conflict domain is defined by the intersection area of the convex hulls of the influence ranges of all the compound risk trajectories involved in the interaction. A separate conflict domain profile is generated for each determined high-risk conflict domain, which records the identifiers of all the compound risk trajectories involved in the conflict, the spatial boundary coordinates of the conflict domain, the conflict trigger time, and the initial conflict intensity. The generation of risk field conflict warning is based on the conflict domain profile, and the warning content at least includes the location of the high-risk conflict domain, the number of risk trajectories involved, and the estimated conflict duration. After the generation of risk field conflict warning, the evolution intervention logic for the affected compound risk trajectories is automatically started. The evolution intervention logic first performs trajectory resilience assessment on each compound risk trajectory involved in the strong interaction, and the trajectory resilience is measured by the average duration of maintaining the structure of the compound risk trajectory without collapse in a similar conflict situation in history. According to the trajectory resilience assessment results, an intervention priority is generated for each compound risk trajectory involved in the conflict, and the lower the resilience of the trajectory, the higher the intervention priority.

[0069] In a specific implementation, the determination of high-risk conflict domain is performed immediately upon the detection of strong interaction, and the geographical boundary of the high-risk conflict domain is defined by the intersection area of the convex hulls of the influence ranges of all the compound risk trajectories involved in the interaction. Specifically, the minimum convex polygons of each compound risk trajectory involved in the strong interaction are obtained, and the total intersection area of these polygons in the two-dimensional plane is calculated, which is the range of the high-risk conflict domain. A separate conflict domain profile is generated for each determined high-risk conflict domain, which records the identifiers of all the compound risk trajectories involved in the conflict, the spatial boundary coordinates of the conflict domain, the conflict trigger time, and the initial conflict intensity. The initial conflict intensity can be obtained by calculating the sum of the diffusion energy levels of all the compound risk trajectories involved in the conflict at the current time. The generation of risk field conflict warning is based on the conflict domain profile, and the warning content at least includes the location of the high-risk conflict domain, the number of risk trajectories involved, and the estimated conflict duration. The conflict duration can be estimated according to the average duration of similar conflict cases in history or the current evolution trend of the involved trajectories. After the generation of risk field conflict warning, the system automatically starts the evolution intervention logic for the affected compound risk trajectories. The evolution intervention logic first performs trajectory resilience assessment on each compound risk trajectory involved in the strong interaction, and the trajectory resilience is measured by the average duration of maintaining the structure of the compound risk trajectory without collapse in a similar conflict situation in history. Structure collapse refers to the situation where the compound risk trajectory does not disintegrate due to a large number of causal relationship edges breaking or a large number of standardized risk events disappearing during the conflict.

[0070] In some embodiments, the trajectory resilience assessment requires querying the historical database, retrieving all instances in history where the current composite risk trajectory was recorded to have engaged in a strong interaction, and extracting the duration of each instance from the onset of the conflict to the collapse of the trajectory structure or the end of the conflict, the trajectory resilience is calculated as:

[0071]

[0072] wherein: is the total number of retrieved historical similar conflict scenarios, is the duration of the current evaluated composite risk trajectory maintaining its structure from collapse in the th historical conflict scenario. According to the trajectory resilience assessment result, an intervention priority is generated for each composite risk trajectory involved in the conflict, the generation rule is that the lower the resilience, the higher the intervention priority is assigned, a direct mapping way is to set the intervention priority as a negative correlation with the trajectory resilience , for example, using the relationship , wherein is a very small positive number to prevent division by zero error, the calculation result value the larger, the higher the intervention priority. Referring to Table 1, a record table of a conflict domain profile is shown.

[0073] Table 1: Record table of conflict domain profile

[0074]

[0075] It can be understood that the conflict domain profile can serve as a complete data source for risk field conflict early warning, the early warning message can directly extract the conflict domain identifier as the event number, extract the spatial boundary coordinates to describe the location, and count the length of the list of participating trajectory identifiers to obtain the number of risk trajectories involved. The start of the evolution intervention logic is automatic and immediate, once the early warning is generated, the intervention logic starts the resilience assessment and priority calculation process for all composite risk trajectories associated with the early warning in parallel. Optionally, the "historical similar conflict scenarios" in the trajectory resilience assessment can be filtered by comparing the similarity of the number of participating trajectories, the combination of running phase labels, the conflict domain area, etc. between the historical conflict and the current conflict, only the historical records with similarity exceeding the threshold are selected for calculation. Optionally, in addition to the trajectory resilience, the generation of the intervention priority can also consider the current diffusion energy level of the composite risk trajectory, and the highest intervention priority is assigned to the trajectory with high diffusion energy level and low resilience.

[0076] Referring to Figure 4In the dynamic monitoring of the running stage of the composite risk trajectory, the number distribution of the risk trajectory in the dormant, active, and outbreak stages in each time interval from 07:00 to 14:00 is intuitively presented. Specifically, each time interval corresponds to the stacked column structure of the three types of stages: the bottom gray block represents the number of dormant stage trajectories, the middle yellow block represents the number of active stage trajectories, and the top red block represents the number of outbreak stage trajectories. For example, in the 10:00-11:00 interval, there are 11 outbreak stage trajectories, 16 active stage trajectories, which are the peak interval of the number of outbreak and active stages in the monitoring period; and in the 07:00-08:00 interval, there are 15 dormant stage trajectories, which are the relative high point of the number of this stage. This distribution characteristic can assist in locating the concentrated active / outbreak period of the risk trajectory and provide quantitative basis in the time dimension for the prediction of high-risk conflict domains.

[0077] In an embodiment of the present application, according to the intervention priority, the selected composite risk trajectory is executed with a virtual isolation operation starting from high priority. The virtual isolation operation is implemented by inserting a virtual neutralization event node in the causal chain of the composite risk trajectory, which is marked as "inhibition" in the risk type code, and its strength reading is a negative value, which is used to offset the strength growth of the events in the subsequent causal chain. After the virtual isolation operation, the diffusion energy level, influence range, and internal structure stability of the intervened composite risk trajectory are recalculated, and the running stage label is updated accordingly. After completing the virtual isolation operation of all composite risk trajectories above the specified priority, the state of the original high-risk conflict domain is re-evaluated. The spatio-temporal correlation between all composite risk trajectories remaining in the region is recalculated, and if the updated spatio-temporal correlation is below the strong correlation threshold, it is determined that the original high-risk conflict domain has been dissolved, and a conflict domain dissolution notification is generated. If some of the spatio-temporal correlations are still above the strong correlation threshold, it is determined whether a new high-risk conflict domain is formed according to the updated running stage label, and the whole process from risk event tracing to evolution intervention is iteratively executed until no new risk field conflict warning is generated.

[0078] In a specific implementation, according to the intervention priority, a virtual isolation operation is performed on the selected compound risk trajectory starting from the highest priority, which is implemented by inserting a virtual neutralization event node in the causal chain of the compound risk trajectory. The neutralization event node is an artificially constructed standardized risk event, which is marked as "inhibition" in the risk type code, and its intensity reading is a negative value, which is used to offset the intensity growth of the events in the subsequent causal chain. The insertion operation occurs at a specific position in the causal relationship chain of the selected compound risk trajectory, which is usually after the last standardized risk event with a higher positive intensity reading. The inserted neutralization event node establishes new causal relationship edges with the previous and subsequent events, thereby becoming part of the compound risk trajectory. After the virtual isolation operation, the diffusion energy level, influence range, and internal structure stability of the intervened compound risk trajectory need to be recalculated, and the running phase label is updated accordingly.

[0079] In some embodiments, the absolute value of the negative intensity reading of the neutralization event node is related to the current state of the intervened compound risk trajectory. One way to calculate it is to make its absolute value equal to the weighted sum of the intensity readings of the events near the insertion point of the intervened trajectory, so as to produce sufficient offset effect. The calculation relationship of the negative intensity reading of the neutralization event node is:

[0080]

[0081] Wherein: is the number of standardized risk events selected from the intervened compound risk trajectory immediately before the insertion position, is the original positive intensity reading of the th selected event, is the weight given to the th selected event. The weight can be allocated according to the time distance of the event occurrence, and the weight of the event closer to the insertion time is larger. After inserting the neutralization event node, in the subsequent calculation of the diffusion energy level of the compound risk trajectory, this negative intensity reading will participate in the weighted moving average calculation, thereby lowering the overall diffusion energy level value. The calculation of the influence range may cause the minimum convex hull area to change due to the addition of a new geographic coordinate point (the neutralization event node needs to be assigned a reasonable geographic coordinate, such as the mean value of the selected event coordinates). The fluctuation variance value of the internal structure stability will also be recalculated due to the increase of the causal relationship edge and the change of the event sequence. Based on these updated indicator values, the system reassigns a running phase label to the intervened compound risk trajectory according to the dynamic allocation rules of the running phase label, and the result may be downgraded from "outbreak" to "active", or from "active" to "dormant".

[0082] ​It can be understood that after completing the virtual isolation operation on all composite risk trajectories higher than the specified priority, it is necessary to reevaluate the status of the original high-risk conflict domain. The reevaluation first recalculates the spatiotemporal correlation between all composite risk trajectories remaining in the region, which refers to the composite risk trajectories that still exist near or inside the geographical boundary of the original conflict domain after the virtual isolation intervention. If all updated spatiotemporal correlations are below the strong correlation threshold, it is determined that the original high-risk conflict domain has been resolved, and a conflict domain resolution notification is generated, which should include the original conflict domain identifier and the resolution time. If some spatiotemporal correlations are still above the strong correlation threshold, it is determined whether a new high-risk conflict domain is formed according to the updated running phase label. The judgment logic is to check whether the running phase labels of the trajectory pairs with spatiotemporal correlations still above the strong correlation threshold are both active or explosive. If the condition is met, these trajectory pairs constitute a new strong interaction, and thus a new high-risk conflict domain may be formed. The system will iteratively execute the whole process from risk event tracing to evolution intervention, i.e., re-tracing based on the latest standardized risk event set, trajectory state tracking, label assignment, interaction comparison, conflict domain determination, and warning generation, until no new risk field conflict warning is generated. Optionally, the specified priority can be set as an absolute priority value, for example, only performing virtual isolation operation on composite risk trajectories with intervention priority higher than 5. Optionally, when iteratively executing the whole process, the system can enter a special monitoring and processing loop, which processes the real-time data of the affected area at a higher frequency until the risk situation of the area returns to stable.

[0083] Referring to Figure 5In the intervention effect evaluation of high-risk conflict domains, the initial conflict intensity, the post-intervention conflict intensity, and the conflict duration are integrated as three core indicators. The multivariate visual correlation analysis is realized with the conflict domain ID as the dimension. Specifically, the red column in the figure represents the initial conflict intensity of each conflict domain, the green column corresponds to the post-intervention conflict intensity, and the black line node marks the conflict duration (unit: hour). At the same time, the final state of each conflict domain is explicitly indicated by labeling “resolved” and “partially resolved”. From the data presentation logic, the initial conflict intensity of each conflict domain is different (such as conflict domains 01 and 02 with an initial intensity higher than 0.9), and the post-intervention intensity decreases to varying degrees: the post-intervention intensity of conflict domains 03, 04, and 05 decreases significantly, combined with the “resolved” label, indicating that the virtual isolation and other intervention measures in the corresponding region effectively reduce the diffusion energy level of the composite risk trajectory; while the post-intervention intensity of conflict domains 01, 02, and 06 is still at a high level, matching the “partially resolved” state, and their conflict duration is also relatively longer (such as conflict domain 06 with a duration close to 6 hours). The parameters and state labeling system of the figure (such as conflict intensity scale and resolution state label) realize the intuitive quantification of the influence of intervention measures on the evolution state (running phase label) and the spatiotemporal correlation degree of conflict domains, providing multi-dimensional visual support for the effectiveness verification of intervention strategies in high-risk conflict domains.

[0084] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between such entities or operations. Moreover, the terms “include”, “contain” or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0085] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for automatic risk assessment based on multi-source data access, characterized in that, The method includes: Heterogeneous risk perception data streams are continuously accessed from at least three independent sources. Meta-information is stripped and ontology is aligned for each data stream to generate a standardized set of risk events with a unified spatiotemporal coordinate baseline. A risk event tracing operation is performed on a standardized risk event set. The risk event tracing operation includes establishing a causal relationship chain among multiple standardized risk events based on the order of events and their physical proximity, and integrating multiple standardized risk events with causal relationships into a composite risk trajectory. The evolution status of each composite risk trajectory is tracked in real time. The evolution status includes diffusion energy level, influence range and internal structural stability. Based on the current evolution status of the composite risk trajectory, an operational stage label of dormant, active or explosive stage is dynamically assigned to it. By continuously comparing the interaction patterns between different composite risk trajectories, when the operational phase labels of two or more composite risk trajectories are simultaneously turned active or erupted, and their physical influence ranges overlap, the overlapping area is determined to be a high-risk conflict domain, and a risk field conflict warning is generated. The risk event tracing operation is performed on a global spatiotemporal grid, with the standardized risk event of the previous moment as the potential cause and the standardized risk event of the next moment as the potential result. Traverse and search for standardized risk event pairs that are consecutive in time and adjacent in space. If two consecutive standardized risk events have a successor or derivative relationship in risk type coding, or if the intensity readings show a logical increasing trend, then establish a directed causal relationship edge between them. By iteratively connecting multiple causal relationship edges, multiple dispersed standardized risk events are linked together into a complete composite risk trajectory, and a unique trajectory identifier is assigned to each composite risk trajectory. The diffusion level of the composite risk trajectory is obtained by calculating the weighted moving average of the intensity readings of all standardized risk events included in the composite risk trajectory, with the weights decreasing as the time since the event occurred. The influence range of the composite risk trajectory is obtained by calculating the minimum convex hull area formed by the geographic coordinates of all standardized risk events included in the composite risk trajectory; The internal structural stability of the composite risk trajectory is obtained by calculating the variance of the number and rate of newly added causal edges within the recent time window.

2. The automatic risk assessment method based on multi-source data access according to claim 1, characterized in that, The process of performing metadata stripping and ontology alignment for each data stream includes parsing the native protocol format of each data stream, extracting the payload containing timestamps, geographic coordinates, risk type codes, and intensity readings, and stripping ancillary information including transmission paths and device identifiers. The extracted payload is semantically mapped according to the preset risk ontology knowledge graph. The different descriptions of the same risk entity from different sources are uniformly mapped to the standard concept nodes in the knowledge graph, thereby completing the ontology alignment. The unified spatiotemporal coordinate baseline is achieved by introducing a virtual global spatiotemporal grid. All payloads aligned with the ontology are normalized and mapped to the spatiotemporal voxels of the global spatiotemporal grid according to their timestamps and geographic coordinates, forming a standardized risk event set. Each spatiotemporal voxel can hold at most one standardized risk event.

3. The automatic risk assessment method based on multi-source data access according to claim 2, characterized in that, The dynamic allocation of the operation phase labels is based on a set of adaptive thresholds, which are dynamically adjusted according to the statistical characteristics of composite risk trajectories in historical data. When the diffusion energy level of a composite risk trajectory is lower than the active threshold, the area of ​​influence is smaller than the range threshold, and the internal structural stability is higher than the stable threshold, a dormant label is assigned to it. When the diffusion energy level exceeds the active threshold but does not reach the burst threshold, or the area of ​​influence exceeds the range threshold but the internal structural stability is lower than the stability threshold, an active label is assigned to it. When the diffusion energy level exceeds the burst threshold, and simultaneously the affected area exceeds the range threshold and the internal structural stability is below the instability threshold, a burst label is assigned to it.

4. The automatic risk assessment method based on multi-source data access according to claim 3, characterized in that, The comparison of the interaction patterns is achieved by continuously calculating the spatiotemporal correlation between any two different composite risk trajectories; The spatiotemporal correlation is calculated by the overlap of events on the time axis of the two composite risk trajectories, the proportion of overlapping area of ​​their influence range in space, and the semantic similarity of the operation stage labels of the two trajectories. When the spatiotemporal correlation of two composite risk trajectories exceeds the preset strong correlation threshold, and their respective operational stage labels are both active or outbreak, it is determined that there is a strong interaction between the two.

5. The automatic risk assessment method based on multi-source data access according to claim 4, characterized in that, The determination of high-risk conflict domains is executed immediately upon detection of strong interactions. The geographical boundary of a high-risk conflict domain is defined by the intersection of the convex hulls of the influence ranges of all the composite risk trajectories that have interacted. For each identified high-risk conflict domain, an independent conflict domain file is generated. The conflict domain file records the identifiers of all composite risk trajectories involved in the conflict, the spatial boundary coordinates of the conflict domain, the conflict trigger time, and the initial conflict intensity. The generation of the risk field conflict early warning is based on the conflict field archive. The early warning content includes at least the location of the high-risk conflict field, the number of risk trajectories involved, and the estimated duration of the conflict.

6. The automatic risk assessment method based on multi-source data access according to claim 5, characterized in that, After generating a risk field conflict warning, the evolutionary intervention logic for the affected composite risk trajectory is initiated; The evolutionary intervention logic first assesses the resilience of each composite risk trajectory in a strong interaction, which is measured by the average duration for which the composite risk trajectory maintains its structure without disintegration in similar historical conflict scenarios. Based on the trajectory resilience assessment results, an intervention priority is generated for each composite risk trajectory involved in the conflict, with the trajectory with lower resilience being assigned a higher intervention priority.

7. The automatic risk assessment method based on multi-source data access according to claim 6, characterized in that, Based on intervention priorities, starting with the highest priority, virtual isolation operations are performed on the selected composite risk trajectories; The virtual isolation operation is achieved by inserting a virtual neutralizing event node into the causal chain of the composite risk trajectory. The neutralizing event node is marked as "suppressed" in the risk type encoding and has a negative intensity reading to cancel out the intensity increase of events in its subsequent causal chain. After the virtual isolation operation, the diffusion level, impact range and internal structural stability of the intervened composite risk trajectory are recalculated, and its operational stage label is updated accordingly.

8. The automatic risk assessment method based on multi-source data access according to claim 7, characterized in that, After completing the virtual isolation operation for all composite risk trajectories with higher than the specified priority, reassess the status of the original high-risk conflict domain. Recalculate the spatiotemporal correlation between all composite risk trajectories remaining in the region. If the updated spatiotemporal correlation is lower than the strong correlation threshold, it is determined that the original high-risk conflict domain has been resolved, and a conflict domain resolution notification is generated. If some spatiotemporal correlations are still higher than the strong correlation threshold, then based on the updated operational stage labels, it is determined whether a new high-risk conflict domain has been formed, and the entire process from tracing the source of risk events to evolutionary intervention is iteratively executed until no new risk domain conflict warnings are generated.

Images