Hierarchical collaborative management and control method and system for regional situation monitoring

Abstract

The application belongs to the technical field of data processing, and provides a layered cooperative management and control method and system for regional situation monitoring. The weighted fusion based on node credibility is used to realize the fusion processing of heterogeneous data sources, and obtain credible fusion data. The situation of the region is evaluated based on the credible fusion data, and a regional cooperative strategy is generated according to the situation evaluation result. A cross-regional scheduling scheme is generated with the global risk and resource utilization rate as the optimization target, and the execution instruction of the monitoring node is generated after the constraint correction of the regional cooperative strategy, so as to realize the layered management and control from the global optimization target to the node execution. The effect deviation is determined by comparing the evaluation result after execution with the situation evaluation result before execution, and the parameters of the node credibility and the situation evaluation are reversely adjusted based on the effect deviation, so that the parameters of the node credibility and the situation evaluation can be adaptively calibrated according to the actual operation effect.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a hierarchical collaborative management and control method and system for regional situation monitoring. Background Technology

[0002] Regional situational awareness is a technology that uses heterogeneous data collected by multiple types of sensors to comprehensively perceive the environmental status, target activities, and resource conditions of a specific area, and generates monitoring and control strategies accordingly. It is widely used in border security monitoring, smart city management, and environmental monitoring.

[0003] Existing multi-source data fusion methods lack the ability to dynamically evaluate the quality of data sources when facing complex scenarios where some data sources may be abnormal. The fusion process cannot adaptively adjust the weights of each data source according to the actual performance of the data, resulting in abnormal data sources having an inappropriate impact on the fusion results. Furthermore, existing methods lack the ability to adaptively calibrate based on actual execution results, and the evaluation accuracy cannot be continuously improved when the monitoring environment or system status changes.

[0004] In view of the above-mentioned shortcomings of the existing technology, the technical problem to be solved by this application is how to achieve hierarchical, collaborative and precise control of regional situational monitoring in complex scenarios. Summary of the Invention

[0005] The purpose of this application is to overcome the shortcomings of the prior art and to provide a hierarchical collaborative management and control method and system for regional situation monitoring.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, a hierarchical collaborative management and control method for regional situation monitoring is provided. This method includes: acquiring heterogeneous data from monitoring nodes in each region, generating node credibility based on the status information of each monitoring node, weighting and fusing the spatiotemporally aligned heterogeneous data with the node credibility as the weight, and outputting credible fused data. Based on the trusted fusion data, a situation assessment is performed on the region, and a regional collaborative strategy is generated based on the situation assessment results. Based on the situation assessment results of multiple regions, with the optimization objectives of global risk and resource utilization, a cross-regional scheduling scheme is generated. The regional coordination strategy is constrained and modified according to the cross-regional scheduling scheme, and the execution instructions of the corresponding monitoring nodes are generated and issued for execution. Based on the monitoring data after execution, a new situation assessment is performed to obtain the post-execution assessment result. The post-execution assessment result is compared with the situation assessment result. After determining the effect deviation, the node credibility and situation assessment parameters of each monitoring node in the corresponding area are adjusted in reverse.

[0007] Secondly, a hierarchical collaborative management and control system for regional situation monitoring is provided, which includes: a fusion module, an evaluation module, a scheduling module, and a feedback module. The fusion module is used to acquire heterogeneous data from monitoring nodes in each region, generate node credibility based on the status information of each monitoring node, and perform weighted fusion of the spatiotemporally aligned heterogeneous data with the node credibility as the weight to output credible fused data. The assessment module performs a situation assessment of the area based on the trusted fusion data and generates a regional collaborative strategy based on the situation assessment results. The scheduling module, based on the situation assessment results of multiple regions, generates a cross-regional scheduling scheme with the optimization objectives of global risk and resource utilization. It then modifies the regional coordination strategy according to the cross-regional scheduling scheme, generates execution instructions for the corresponding monitoring nodes, and issues them for execution. The feedback module is used to reassess the situation based on the monitoring data after execution, obtain the post-execution assessment result, compare the post-execution assessment result with the situation assessment result, determine the effect deviation, and then adjust the node credibility and situation assessment parameters of each monitoring node in the corresponding area in reverse.

[0008] Compared with existing technologies, the beneficial effects of this application are as follows: weighted fusion based on node credibility enables the fusion processing of heterogeneous data sources, ensuring the quality of credible fused data from the data source; situation assessment of the region is performed based on credible fused data, and regional collaborative strategies are generated based on the situation assessment results, realizing a structured link from data perception to strategy generation, ensuring that strategy generation is based on evidence; a cross-regional scheduling scheme is generated with global risk and resource utilization as optimization objectives, and the cross-regional scheduling scheme is used to constrain and modify the regional collaborative strategies before generating execution instructions for monitoring nodes, realizing hierarchical control from global optimization objectives to node execution; the post-execution evaluation results are compared with the pre-execution situation assessment results to determine the effect deviation, and the parameters of node credibility and situation assessment are adjusted in reverse based on the effect deviation, so that the parameters of node credibility and situation assessment can be adaptively calibrated according to the actual operation effect. Attached Figure Description

[0009] Figure 1 This is a flowchart of a hierarchical collaborative control method for regional situational monitoring provided in an embodiment of this application. Figure 2 This is a flowchart illustrating the logic of outputting trusted fusion data provided in the embodiments of this application. Figure 3 This is a flowchart illustrating the logic of the regional collaboration strategy provided in this application embodiment. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more pairs. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0011] Example 1 This embodiment takes a multimodal monitoring scenario in a certain area as an example. Radar monitoring nodes, photoelectric monitoring nodes and acoustic monitoring nodes are deployed in this area, and each node generates heterogeneous data of different data types.

[0012] refer to Figure 1 This is a flowchart of a hierarchical collaborative management and control method for regional situation monitoring provided in this application embodiment. The method includes: S1, obtaining heterogeneous data from monitoring nodes in each region, generating node credibility based on the status information of each monitoring node, weighting and fusing the spatiotemporally aligned heterogeneous data with the node credibility as the weight, and outputting credible fused data.

[0013] Furthermore, such as Figure 2 As shown, the output trusted fusion data includes: After spatiotemporal alignment, heterogeneous data is divided into multiple modal groups according to data type. The heterogeneous data in the same modal group are weighted and fused within the group with node credibility as the weight to obtain the modal fusion value of each modal group and determine the modal uncertainty. Based on the differences between modal fusion values, the modal conflict degree between different modal groups is determined. When the modal conflict degree is greater than the conflict threshold, the consistency center in the modal fusion value is determined as the modal consensus. Determine the degree of deviation between each modality fusion value and the modality consensus, correct the node credibility of the monitoring nodes within the modality group based on the degree of deviation, and re-execute the weighted fusion within the group; Based on modal uncertainty, cross-modal fusion of modal fusion values ​​is performed to output reliable fusion data.

[0014] The sampling frequency, acquisition time, and spatial location of each monitoring node in the region are different, resulting in heterogeneous data with differences in timestamps and spatial coordinates, which cannot be directly fused. In the time dimension, interpolation compensation is used for time alignment. With a unified time reference, for data collected by each node at a non-reference time, the equivalent data value of the node at the reference time is calculated by linear interpolation based on the values ​​of the two adjacent times. For example, if the data value of a radar node at timestamp t=0.3s is x1 and the data value at timestamp t=1.2s is x2, then the equivalent data value linearly interpolated to the reference time t=1s is x1+(x2-x1)×(1-0.3) / (1.2-0.3).

[0015] In the spatial dimension, based on the known deployment coordinates and detection direction of each node, the target position pointed to by the data of each node is transformed from the node's local coordinate system to the unified regional coordinate system, eliminating spatial deviations caused by different node positions. After spatiotemporal alignment, the heterogeneous data of all nodes have a unified time reference and spatial coordinates. Subsequently, the heterogeneous data is divided into multiple mode groups according to data type. The data output by all radar nodes is assigned to the radar mode group, denoted as mode group A; the data output by all optoelectronic nodes is assigned to the optoelectronic mode group, denoted as mode group B; and the data output by all acoustic nodes is assigned to the acoustic mode group, denoted as mode group C. Data within the same mode group has the same physical meaning and data dimensions. Different mode groups remain independent of each other, and their respective fusion processes do not interfere with each other.

[0016] In this embodiment, mode group A contains data from three radar nodes R1, R2, and R3; mode group B contains data from two optoelectronic nodes O1 and O2; and mode group C contains data from one acoustic node S1. Spatiotemporal alignment and mode group partitioning are the basis for all subsequent fusion steps.

[0017] A single modality group contains data from multiple monitoring nodes. Due to differences in equipment status, deployment location, and working environment, the reliability of the data varies among nodes. Directly taking a simple average of the data from each node within the group would treat the data from high-reliability nodes and low-reliability nodes equally, reducing the quality of the fusion result. Furthermore, during weighted fusion, the degree of dispersion among the data from each node within the group reflects the reliability of the fusion result for that modality group. Taking modality group A as an example, let the spatiotemporally aligned data values ​​of three radar nodes R1, R2, and R3 within modality group A be d1, d2, and d3, respectively, and their corresponding node confidence levels be C1, C2, and C3, respectively. The modality fusion value D... A =(C1×d1+C2×d2+C3×d3) / (C1+C2+C3); The physical meaning of this formula is that the credibility of each node is used as the weight of the corresponding data. The higher the credibility of the node, the greater the proportion of its data in the fusion result.

[0018] Modal uncertainty is determined based on the weighted dispersion of the data at each node within the mode group, and is measured using weighted variance, σ. A 2 =[C1×(d1-D A ) 2 +C2×(d2-D A ) 2 +C3×(d3-D A ) 2 ] / (C1+C2+C3); σ ALet Variance be the modal uncertainty of modal group A. A larger variance indicates greater divergence among the data points of each node within the group, resulting in greater uncertainty in the fusion result; a smaller variance indicates greater convergence of the data within the group, resulting in more reliable fusion results. For example, suppose the data values ​​and confidence levels of three radar nodes are as follows: node R1 data value d1=100, confidence level C1=0.9; node R2 data value d2=102, confidence level C2=0.8; node R3 data value d3=96, confidence level C3=0.5; modal fusion value D... A =(0.9×100+0.8×102+0.5×96) / (0.9+0.8+0.5)≈99.82; The weighted squared deviations between the data of each node and the modal fusion value are: R1 is 0.9×(100-99.82) 2 =0.029; R² is 0.8×(10²-99.82) 2 =3.802; R3 is 0.5×(96-99.82) 2 =7.296; weighted variance σ A 2 =(0.029+3.802+7.296) / 2.2≈5.058, modal uncertainty σ A ≈2.25.

[0019] The calculation results show that node R3 has low confidence and its data value deviates far from the other two nodes. Its weighted contribution to the modal fusion value is only 48 / 219.6≈21.9%, which is lower than its theoretical average contribution (33.3%) in terms of the number of nodes. At the same time, the data deviation of R3 is the main factor leading to the increase of modal uncertainty. If R3 is removed and only R1 and R2 are fused, the modal uncertainty will be significantly reduced.

[0020] The same intra-group weighted fusion operation is performed on modal groups B and C respectively to obtain their respective modal fusion values ​​and modal uncertainties. After intra-group fusion, each modal group has a modal fusion value representing the overall perception conclusion of that modality, and a modal uncertainty that measures the reliability of the fusion result, providing input for subsequent cross-modal conflict detection and cross-modal fusion.

[0021] In actual operation, due to sensor failure, environmental interference, or target camouflage, the fusion results of some modal groups may differ significantly from those of other modal groups, resulting in intermodal conflicts. If cross-modal fusion is performed directly without detecting and processing these conflicts, erroneous modal data will be mixed into the final result, reducing the quality of reliable fusion data. The modal conflict degree is obtained based on the difference between the fusion values ​​of each modality. Here, the coefficient of variation (i.e., the ratio of the standard deviation to the mean of each fusion value) is taken as the modal conflict degree. The larger the coefficient of variation, the more serious the divergence between the fusion values ​​of each modality, and the higher the degree of conflict. The smaller the coefficient of variation, the closer the fusion values ​​of each modality are to each other, and the lower the degree of conflict.

[0022] When the modal conflict level exceeds the conflict threshold, the consensus center among the modal fusion values ​​is determined as the modal consensus. This involves performing cluster analysis on the modal fusion values ​​and taking the cluster center containing the majority of modal fusion values ​​as the modal consensus. When there is a clear majority and minority, the cluster center of the majority is the direction of the modal consensus. For example, suppose the results of three modal groups after intra-group fusion are: modal fusion value D of modal group A. A =99.82, modal uncertainty σ A =2.25; Modal fusion value D of mode group B B =100.15, modal uncertainty σ B =3.1; Modal fusion value D of mode group C C =115.6, modal uncertainty σ C =8.5.

[0023] The mean of the three fusion values ​​is (99.82 + 100.15 + 115.6) / 3 ≈ 105.19, the standard deviation is ≈ 7.36, and the coefficient of variation is 7.36 / 105.19 ≈ 0.07. If the conflict threshold is 0.05, then 0.07 > 0.05, the modal conflict degree exceeds the threshold, and conflict handling is triggered. In cluster analysis, D A =99.82 and D B =100.15 is close to the value, while D C =115.6 differs from the previous two by 15.78 and 15.45 respectively, clearly belonging to the same cluster as an outlier; therefore, D A and D B The majority cluster is formed, and its cluster center is (99.82+100.15) / 2≈99.99, which means the modal consensus is about 100. The fusion value of modality group C, 115.6, deviates from the consensus by about 15.6, which belongs to the abnormal modality that needs to be processed.

[0024] Once intermodal conflicts are detected, it is necessary to identify which data causes the modality group to deviate from the consensus direction of the majority modalities. If no credibility correction is made, these anomalous nodes will still affect the fusion result with their original credibility weights in the next round of intra-group weighted fusion, causing the modality conflicts to persist. The first step is to determine the modality deviation, that is, to calculate the absolute difference between each modality fusion value and the modality consensus, as the degree of deviation of the modality group. Then, the modality deviation is attributed to each monitoring node in the modality group. That is, for each node in the modality group, it is determined whether the node's data value pushes the modality fusion value toward or away from the modality consensus. If a node's data value deviates from the modality consensus in the same direction as the modality fusion value, that is, the node's data is one of the factors causing the modality deviation, and the degree of attribution of the deviation to the deviation is relatively large. Conversely, if the node's data value is close to the modality consensus, the degree of attribution of the deviation to the deviation to the deviation is relatively small. For nodes with a large degree of attribution of deviation, their node credibility is reduced.

[0025] The following provides specific explanations using two scenarios. The first scenario involves a modality group with only one node. Modality group C (acoustic modality group) contains only node S1, with a data value d6 = 115.6 and a current confidence level C6 = 0.7. The fusion value D of modality group C... C =115.6 deviates from the modal consensus by approximately 15.6, which is a significant deviation. Since there is only one node S1 in the group, the degree of attribution of the deviation by this node is 100%, that is, the deviation of modal group C is entirely caused by S1. The credibility of S1 is reduced from C6=0.7 to C6'=0.4. After the credibility is reduced, the influence of this node in the next round of cross-modal fusion will be significantly compressed.

[0026] The second scenario involves multiple nodes within a modality group, some of which are abnormal. If the data values ​​of three radar nodes within modality group A are: R1 = d1=100, R2 = d2=102, and R3 = d3=112; and the current modality fusion value D... A =105.1, because the data of R3 deviates significantly, the fusion value is pulled away from the modal consensus of 100.

[0027] The data value of R1 is consistent with the modal consensus, and the data value of R2 is close to the modal consensus. The data from these two nodes are the factors that push the modal fusion value towards the consensus direction. The data value of R3, 112, is far from the modal consensus and is the main factor that pulls the modal fusion value away from the consensus direction. Therefore, R3 has the greatest attribution degree to the deviation. The confidence of R3 is reduced from C3=0.5 to C3'=0.3, while the confidence of R1 and R2 remains unchanged. After the correction, the intra-group weighted fusion is re-executed, i.e., D A =(0.9×100+0.8×102+0.3×112) / (0.9+0.8+0.3)=102.6; Uncorrected fusion value D A=105.1, deviating from modal consensus by approximately 5.1 (100), corrected fusion value D A =102.6 deviates from the modal consensus by only about 2.6. After the credibility of R3 is reduced, its contribution to the fusion result decreases from (0.5×112) / (0.9+0.8+0.5)×112≈25.5% before the correction to (0.3×112) / (0.9+0.8+0.3)×112≈16.4%, and the influence of its outlier data is effectively suppressed. At the same time, the credibility of R1 and R2 remains unchanged, and their contribution to the fusion increases accordingly, making the fusion result closer to the data direction of the majority of nodes.

[0028] The corrected modal fusion values ​​and modal uncertainties provide a more reliable basis for subsequent cross-modal fusion; the corrected node credibility also affects the intra-group weighted fusion in the next round of data fusion.

[0029] After the aforementioned steps, each modal group has completed intra-group weighted fusion and confidence correction, obtaining its own modal fusion value and modal uncertainty. The fusion results of different modal groups represent the perception conclusions of different types of sensors for the same monitoring target. Since the types, quantities, and working environments of the sensors used in each modal group are different, the reliability of their fusion results is also different, which is manifested as differences in modal uncertainty. If the fusion values ​​of each modality are simply averaged, high-reliability and low-reliability modes will be treated equally.

[0030] Let the modal fusion values ​​of each modal group after correction (or without correction) be D1', D2', and D3', and the corresponding modal uncertainties be σ1, σ2, and σ3, respectively. Then the cross-modal fusion weight h of the i-th modal group is... i =(1 / σ i ) / Σ j (1 / σ j Trusted fused data = Σ i (h i ×D i The smaller the modal uncertainty, the larger its reciprocal, and the higher the fusion weight of that modal group; conversely, the larger the modal uncertainty, the smaller its reciprocal, and the lower the fusion weight of that modal group. This ensures that reliable modes dominate the final fusion result, effectively compressing the influence of unreliable modes. For example, the corrected data for each modal group are: the modal fusion value D of modal group A. A =102.6, modal uncertainty σ A =2.5; Modal fusion value D of mode group B B =100.15, Modal uncertainty σ B =3.1; Modal fusion value D of mode group C C =115.6, modal uncertainty σ C =10.2.

[0031] The cross-modal fusion weights for each mode are calculated as follows: 1 / σ A =1 / 2.5=0.4, 1 / σ B =1 / 3.1=0.323, 1 / σ C =1 / 10.2=0.098, weighted normalized denominator=0.4+0.323+0.098=0.821, h A =0.4 / 0.821≈0.487, h B =0.323 / 0.821≈0.393, h C =0.098 / 0.821≈0.12; Trusted fused data =0.487×102.6+0.393×100.15+0.12×115.6≈103.2.

[0032] The fusion results allow us to analyze the contributions of each modality group. Modality groups A and B, with their low modal uncertainties, receive approximately 48.7% and 39.3% of the fusion weights, respectively, totaling about 88%, making them the dominant contributors to the credible fusion data. Modality group C, with its high uncertainty and reduced credibility, receives only about 12% of the fusion weight, effectively suppressing its impact on the credible fusion data. Without the aforementioned conflict detection and credibility correction, modality group C would participate in the fusion with uncorrected credibility and uncertainty, significantly increasing the impact of its anomalous data on the final fusion result and causing the credible fusion data to deviate from the true value.

[0033] Specifically, the credibility of the generated nodes includes: Extract equipment operating status and historical fault records from the status information of monitoring nodes, determine the status score based on the equipment operating status, determine the fault score based on the historical fault records in a time-series decay manner, and weight and fuse the status score and fault score to obtain the health component of each monitoring node. The data components of each monitoring node are determined based on the degree of consistency between the heterogeneous data of each monitoring node and the heterogeneous data of neighboring nodes in the same mode group. The health component and the data component are normalized and weighted and fused to obtain the node credibility of each monitoring node.

[0034] The generation of node credibility first requires extracting the device operating status and historical fault records from the status information of each monitoring node. The device operating status describes the current real-time working status of the node, including but not limited to real-time parameters reflecting the current working status of the node such as signal strength, signal-to-noise ratio, communication packet loss rate, device temperature, remaining power or power supply stability, and computing load rate. The historical fault records describe the faults that have occurred on the node since its deployment, including the fault timestamp, fault type (such as sensor failure, communication interruption and data anomaly), fault duration and repair status, and other historical data.

[0035] Taking node R1 as an example, the extracted device operating status is as follows: signal strength -45dBm, signal-to-noise ratio 28dB, communication packet loss rate 0.2%, device temperature 42℃, and remaining power 85%. The extracted historical fault records are: a communication interruption occurred once in January 2024, lasting 2 hours, which has been repaired; a sensor calibration offset occurred once in June 2024, lasting 4 hours, which has been repaired; there are no fault records after that. The device operating status and historical fault records assess the health of the node from the current and historical time dimensions, respectively.

[0036] After normalizing all parameters of the equipment's operating status to the same dimension, a comprehensive evaluation is performed to obtain a status score s reflecting the current working condition of the equipment. The value ranges from 0 to 1, with a value closer to 1 indicating a better current equipment status. Taking node R1 as an example, the normalization process of each parameter is as follows: signal strength of -45dBm is normalized to 0.583 within the preset effective range (-80dBm to -20dBm); signal-to-noise ratio of 28dB is normalized to 0.6 within the preset range (10dB to 40dB); communication packet loss rate of 0.2% is normalized to 0.96 within the preset range (0% to 5%); equipment temperature of 42℃ is normalized to 0.633 within the preset range (20℃ to 80℃); and remaining battery power of 85% is normalized to 0.85 within the preset range (0% to 100%).

[0037] After weighted averaging of the normalized values, the overall status score s = 0.725 is obtained; this score reflects that the overall equipment working condition of node R1 at the current moment is at a medium to high level.

[0038] For each fault in the historical fault record, an exponential decay weight is applied based on the time interval between the fault occurrence and the current time. That is, the longer the time interval, the more severe the decay, and the smaller the impact of the fault on the current fault score. The physical meaning of this time-series decay mechanism is that a fault that occurred a long time ago and has been repaired should have less reference significance for the reliability of the current equipment than a fault that occurred recently.

[0039] Specifically, for the k-th fault, let the time interval between its occurrence and the current time be Δt. kThe attenuation coefficient is λ (λ=0.1 in this embodiment, in units of per month), and the time-series attenuation weight for this fault is e^(-λ×Δt). k The severity of the fault (ranging from 0 to 1, with larger values ​​indicating more severe faults) is multiplied by the time-series decay weight to obtain the current impact value of the fault. The current impact values ​​of all historical faults are summed to obtain the comprehensive impact value of historical faults. This value is then subtracted from 1 and truncated to above 0 to obtain the fault score f, which ranges from 0 to 1. The closer the value is to 1, the smaller the impact of the historical fault.

[0040] Taking node R1 as an example, the communication interruption failure in January 2024 is currently in December 2024, so the time interval Δt1 = 11 months, the time series decay weight e^(-0.1×11) ≈ 0.333, the severity of this failure is taken as 0.6, and the current impact value = 0.333×0.6 = 0.2; the sensor calibration offset failure in June 2024 has a time interval Δt2 = 6 months, the time series decay weight e^(-0.1×6) ≈ 0.549, the severity of this failure is taken as 0.4, and the current impact value = 0.549×0.4 = 0.22; historical The overall impact of the fault is 0.2 + 0.22 = 0.42, and the fault score is f = max(1 - 0.42, 0) = 0.58. This calculation shows that the fault in January 2024, being 11 months old, only drags down the current credibility by about 0.2, while the fault in June, being only 6 months old, has a drag of about 0.22, slightly higher than the fault in January. Time decay naturally diminishes the impact of long-standing faults, while recent faults are still fully reflected in the fault score. If no fault occurs at this node recently, its fault score will gradually recover as the impact of historical faults continues to decay.

[0041] The health component is obtained by weighted fusion of the status score and the fault score, reflecting the device health level of the node from two dimensions: current status and historical performance. The health component H = α × s + (1 - α) × f, where α is the weight coefficient of the status score and (1 - α) is the weight coefficient of the fault score. The value of α is (0, 1), which is dynamically adjusted according to the focus of the monitoring scenario. The larger α is, the more the health component is biased towards the current device status; the smaller α is, the more the health component is biased towards historical fault records. In the long-term deployment monitoring scenario of this embodiment, α = 0.6 is taken, so that both the current device status and historical fault records have a significant impact on the health component. Taking node R1 as an example, the health component H = 0.6 × 0.725 + 0.4 × 0.58 = 0.667. This value indicates that the device health level of node R1 is at a medium level, that is, the current device status is acceptable, but the two historical faults still have a certain impact on its health component.

[0042] In contrast, if there exists a node R2 whose current device state is similar to that of R1 (s=0.72), but has no historical fault records (f=1), then its health component H=0.6×0.72+0.4×1=0.832, which is significantly higher than that of R1. This difference is entirely caused by historical fault records, reflecting the importance that the health component places on historical reliability.

[0043] A node's good device status does not necessarily mean that its current output data is accurate. The node's data may deviate from that of similar sensors due to the special environment of its location. By comparing the node's data with the data of neighboring nodes in the same mode group, the degree of anomaly in the node's data can be identified. The higher the degree of consistency, the larger the data component; the lower the degree of consistency, the smaller the data component. Specifically, for a node in a mode group, take several of its closest spatially distant neighboring nodes in the same mode group and calculate the normalized difference between the node's data value and the data values ​​of each neighboring node. The smaller the difference, the more consistent the node's data is with other nodes in the same mode group, and the higher the data component.

[0044] Taking node R3 in mode group A as an example, let the data value of R3 be d3=96, the data value of its neighboring nodes R1 in the same mode group be d1=100, and the data value of R2 be d2=102. The absolute difference between R3 and R1 is |96-100|=4, the absolute difference between R3 and R2 is |96-102|=6, and the average difference between R3 and its neighboring nodes is (4+6) / 2=5. Within the preset consistency evaluation range (0 to 15), the average difference of 5 is normalized to form the data component D. data =1-5 / 15≈0.667; In comparison, the average difference between the data of node R1 and R2 and R3 is (|100-102|+|100-96|) / 2=(2+4) / 2=3, and the normalized data component D data =1-3 / 15≈0.8; R1's data has higher consistency with neighboring nodes than R3, thus obtaining a higher data component.

[0045] This assessment method does not rely on the historical performance of the equipment, but judges the data quality based on the actual performance of the node and whether it is consistent with the statements of neighboring nodes in the same modality group, so as to capture situations where the equipment is in good condition but the current data is deviated due to environmental factors.

[0046] The health component assesses the long-term reliability of a node from a device perspective, while the data component assesses the quality of the node's current output from a data perspective. These two components characterize the node's trustworthiness from different angles and need to be merged into a unified node trustworthiness value. Since the calculation methods and value ranges of the two components may differ, they are first normalized to the same dimension before being weighted and merged. That is, node trustworthiness C = β × H + (1 - β) × Ddata , where β is the weighting coefficient of the health component and (1-β) is the weighting coefficient of the data component. The value of β controls the contribution ratio of the two dimensions to the node credibility.

[0047] Taking node R1 as an example, with β=0.5, the node reliability C1=0.5×0.667+0.5×0.8=0.734; taking node R3 as an example, let its health component H=0.52 (lower than R1, because its historical failures are more frequent or its current equipment condition is worse), and its data component D... data =0.667, then the node credibility C3 = 0.5 × 0.52 + 0.5 × 0.667 = 0.594; Node R1 has a higher credibility of 0.734 because its device status is good and its data is highly consistent with neighboring nodes; Node R3 has a lower credibility of 0.594 because its device status and data consistency are not as good as R1's. This credibility value is used as the weight of the corresponding node data in the subsequent intra-group weighted fusion. Nodes with higher credibility contribute more in the fusion, and nodes with lower credibility contribute less.

[0048] The value of β ranges from (0,1). In this embodiment, β=0.5 is used as the initial value, indicating that the health dimension and the data dimension contribute equally to the node credibility. The choice of the weight coefficient β has a directional impact on the generation of node credibility. In scenarios with high equipment reliability requirements, such as long-term deployed monitoring networks, β can be increased to focus on the health dimension, and β=0.7 is taken, so that nodes with good equipment status for a long time but slightly deviated data at individual moments can still maintain high credibility. In scenarios with high data accuracy requirements (such as instantaneous high-precision monitoring tasks), β can be decreased to focus on the data dimension, and β=0.3 is taken, so that nodes with good equipment history but serious inconsistencies with neighboring nodes in current data are promptly reduced in credibility. The specific value of β is dynamically adjusted through the effect deviation of subsequent feedback links to achieve adaptive optimization of node credibility generation.

[0049] The generated node credibility will be used as the weight of each node's data in the weighted fusion within the group. At the same time, if inter-modal conflicts are detected during the fusion process, the credibility will also be dynamically corrected based on the degree of deviation. In the S4 feedback loop, the deviation of the execution effect will further drive the adjustment of the node credibility.

[0050] S2. Based on the trusted fusion data, conduct a situation assessment of the region and generate a regional collaborative strategy based on the situation assessment results.

[0051] Situation assessment specifically includes: The threat situation in the region is obtained by evaluating the abnormal characteristics of the trusted fusion data, the coverage situation in the region is quantitatively evaluated based on the coverage of each modality group, and the resource situation in the region is obtained by evaluating the remaining working capacity of each monitoring node. The threat situation, coverage situation and resource situation are weighted and fused with dimensional weights to obtain the situation assessment result.

[0052] The first dimension of regional situation assessment is the threat situation, reflecting the level of security threats currently existing in the region. Trusted fusion data is generated by cross-modal fusion of monitoring data from multiple modal groups, including the comprehensive perception results of target features from various sensors. When certain target features deviate from the normal baseline features of the region, they are considered anomalous features, which may indicate a potential threat. Threat situation assessment is the quantitative analysis of these anomalous features. Specifically, target feature vectors are extracted from the trusted fusion data and compared with the baseline feature vectors of the region under normal conditions. During the comparison, the degree of deviation of each feature dimension is weighted and evaluated using a threat sensitivity parameter. The threat sensitivity parameter controls the sensitivity to anomalous features; that is, the larger the parameter, the higher the threat level of the same anomalous feature; the smaller the parameter, the lower the threat level of the same anomalous feature.

[0053] For example, the target feature vector extracted from the trusted fusion data of a certain region contains three dimensions: the target density feature value is 0.72, the target motion feature value is 0.85, and the target signal feature value is 0.6; the corresponding dimensions of the normal baseline feature vector of this region are 0.3, 0.4, and 0.35, respectively; the deviations of each dimension are as follows: target density deviation |0.72-0.3|=0.42, target motion deviation |0.85-0.4|=0.45, and target signal deviation |0. 6-0.35|=0.25; Let the threat sensitivity parameter be 1.2. A parameter greater than 1 indicates that the target is more sensitive to anomalies. Then the threat assessment values ​​for each dimension are: target density dimension 0.42×1.2=0.504, target movement dimension 0.45×1.2=0.54, and target signal dimension 0.25×1.2=0.3; The comprehensive threat situation value is taken as the weighted average (each dimension is equally weighted), and the threat situation value Th=(0.504+0.54+0.3) / 3≈0.448.

[0054] The threat sensitivity parameter is adjustable, and its initial value is determined based on historical threat data for the region, specifically the mean μ of the deviation of historical anomalous characteristics in the region during normal operation. e and standard deviation σ e Set the threat sensitivity parameter to p / μ ep is a preset anomaly detection multiplier factor, ranging from 0.5 to 2. When p is larger, the threat sensitivity is higher, and even minor anomalies can generate a higher threat assessment. When p is smaller, the threat sensitivity is lower, and a larger deviation is required to trigger a higher threat assessment. In areas with high security requirements, such as border monitoring areas, this parameter is initially set higher, so that even minor anomalies can generate a higher threat assessment. In areas with low security requirements, such as general environmental monitoring areas, this parameter is initially set lower, such as 0.8, to avoid overreacting to normal fluctuations. This parameter can be dynamically adjusted in subsequent feedback stages based on the deviation of the assessment effect. The threat situation value is directly used for subsequent regional collaborative strategy generation. The higher the threat situation value, the higher the risk level, the more stringent the target coverage, and the larger the resulting monitoring gap, the stronger the driving force for strategy generation.

[0055] The second dimension of regional situation assessment is coverage status, which reflects the completeness of coverage of various types of monitoring within the region. Coverage status describes the proportion of the space covered by the monitoring data of a certain modality group within the region. If the sensors of a certain modality group cover most of the space within the region, the coverage status of that modality group is high; if there are large monitoring blind spots, the coverage status is low. The region is divided into several grid cells, and the proportion of grid cells that can be effectively covered by the monitoring data of each modality group is counted out to the total number of grid cells. This proportion represents the coverage status of that modality group. Coverage status is quantitatively assessed using a coverage threshold parameter. The coverage threshold parameter defines the standard for sufficient coverage; that is, when the coverage status of a certain modality group is lower than the threshold, the coverage of that modality group is considered insufficient, and there is a monitoring gap.

[0056] For example, the area is divided into 100 grid cells. Mode group A (radar) has an effective coverage grid of 82 cells with a coverage level of 0.82; mode group B (electro-optical) has an effective coverage grid of 65 cells with a coverage level of 0.65; and mode group C (acoustic) has an effective coverage grid of 40 cells with a coverage level of 0.4. Assuming a coverage threshold parameter of 0.7, then mode group A's coverage level of 0.82 is greater than the coverage threshold parameter, indicating sufficient coverage; mode group B's coverage level of 0.65 is less than the coverage threshold parameter. The coverage is insufficient; the coverage of mode group C is 0.4, which is significantly less than the coverage threshold parameter, indicating severe coverage insufficiency; the coverage of each mode group is weighted and averaged to obtain the coverage situation value Cov. The coverage of each mode group is assigned different weights according to its strategic importance in the region. For example, the radar mode group has a weight of 0.4, the electro-optical mode group has a weight of 0.35, and the acoustic mode group has a weight of 0.25. Then Cov = 0.4 × 0.82 + 0.35 × 0.65 + 0.25 × 0.4 = 0.656.

[0057] The coverage status value and the coverage degree of each modality group are used together in the subsequent strategy generation stage. The more accurate the coverage status assessment, the more accurate the gap calculation, and the more targeted the generated strategy. The initial value of the coverage threshold parameter is determined according to the monitoring task level of the area. The monitoring task level is divided into three levels: basic monitoring, key monitoring, and high-risk monitoring. The coverage threshold parameter is 0.5 for the basic monitoring level, 0.7 for the key monitoring level, and 0.9 for the high-risk monitoring level. In this embodiment, the coverage threshold parameter is 0.7, corresponding to the key monitoring level. If the actual monitoring scenario of the area cannot provide sufficient sensor resources to meet the high coverage threshold, the coverage threshold parameter can be appropriately reduced, but it should not be lower than 0.5, otherwise it will lead to an excessively large monitoring blind zone.

[0058] The third dimension of regional situation assessment is resource situation, which reflects the remaining working capacity of each monitoring node in the region. The working capacity of a monitoring node is constrained by various factors, including remaining power, computing load, communication bandwidth, and sensor lifespan. When a node has sufficient remaining working capacity, it can undertake more tasks in strategy execution. When a node has insufficient remaining working capacity, forcibly assigning too many tasks will lead to node overload or even failure, affecting the stability of the entire monitoring system. Specifically, the remaining working capacity index of each monitoring node is assessed. This index comprehensively considers factors such as the proportion of remaining power, the current computing load rate, the proportion of available communication bandwidth, and the expected remaining lifespan of the sensors. After normalizing each factor to the same dimension, a weighted sum is calculated. The value of the remaining working capacity index ranges from 0 to 1, with a value closer to 1 indicating a greater workload that the node can undertake.

[0059] For example, there are 6 monitoring nodes in this area, with their remaining working capacity indices as follows: node R1 is 0.85, node R2 is 0.78, node R3 is 0.42, node O1 is 0.6, node O2 is 0.7, and node S1 is 0.55. The resource status value Res of the area is a comprehensive index of the remaining working capacity indices of each node, taking the minimum and average values ​​as a reference. The average value is (0.85+0.78+0.42+0.6+0.7+0.55) / 6≈0.65, and the minimum value is 0.42 (node ​​R3). The average value reflects the overall resource sufficiency of the area, while the minimum value reflects the resource bottleneck within the area. The resource status value comprehensively reflects the overall resource status of the area and is used to determine resource conflicts in subsequent strategy generation. When a task allocation instruction requires a node with a lower remaining working capacity to undertake more tasks, it is determined that a resource conflict exists.

[0060] Threat posture, coverage posture, and resource posture each characterize the current situation of the region from different dimensions. They are weighted and fused together to obtain a comprehensive situation assessment result. The dimension weights are adjustable parameters that control the relative importance of each dimension in the comprehensive assessment. Let the dimension weights of the three dimensions be: threat dimension weight w1=0.45, coverage dimension weight w2=0.35, and resource dimension weight w3=0.2. Then the situation assessment result = w1×Th+w2×Cov+w3×Res=0.45×0.448+0.35×0.656+0.2×0.65=0.562.

[0061] The selection of dimension weights reflects the current control objectives of the region. The sum of the three weights is 1. In a control scenario aimed at preventing security threats, w1=0.45, w2=0.35, w3=0.2; in a control scenario aimed at optimizing resource management, w1=0.2, w2=0.35, w3=0.45 is recommended; in a control scenario aimed at ensuring coverage integrity, w1=0.25, w2=0.5, w3=0.25 is recommended. The baseline weights of each dimension are dynamically adjusted in the subsequent feedback process based on the dimensional deviations of each dimension assessment. In this embodiment, the threat dimension has the highest weight, reflecting the control strategy of prioritizing the prevention of security threats in the region.

[0062] Specifically, such as Figure 3 As shown, the regional cooperation strategy includes: Based on the threat situation assessment results, the risk level is determined to determine the target coverage of each modality group. The difference between the target coverage and the coverage of the corresponding modality group in the coverage situation is taken as the monitoring gap of the corresponding modality group and constitutes the gap vector. The urgency of the gap in each modal group is determined based on the magnitude and direction of the gap vector. The modal groups are sorted according to the urgency of the gap, and a set of monitoring actions is generated for the monitoring gap of each modal group according to the sorting result and mapped to the task allocation instructions of the monitoring nodes. Based on the resource situation, determine whether there are resource conflicts, and assign instructions for tasks with resource conflicts, prioritizing and coordinating them with the urgency of the gap as the weight, and generating regional coordination strategies.

[0063] Regional coordination strategies transform threat posture into actionable monitoring targets. Threat posture values ​​are continuous numerical indicators that need to be mapped to discrete risk levels. Based on these risk levels, the target coverage that each modality group should achieve is determined. The risk level classification uses a segmented threshold method. The preset risk level classification rules are as follows: threat status values ​​in [0, 0.2) are low risk (Level 1), [0.2, 0.4) are low to medium risk (Level 2), [0.4, 0.6) are medium risk (Level 3), [0.6, 0.8) are medium to high risk (Level 4), and values ​​above 0.8 are high risk (Level 5). In this embodiment, the threat status value Th = 0.448, which falls within the medium risk range (Level 3). The higher the risk level, the more stringent the monitoring coverage requirements for each modality group in the area, i.e., the higher the target coverage. The target coverage rules corresponding to each risk level are as follows: Level 1 corresponds to a target coverage of 0.5 for each modality group, Level 2 corresponds to 0.6, Level 3 corresponds to 0.7, Level 4 corresponds to 0.85, and Level 5 corresponds to 0.95. The physical meaning of this mapping relationship is that a larger monitoring blind spot is allowed during low-risk periods, while as comprehensive a monitoring coverage as possible is required during high-risk periods.

[0064] In this embodiment, the risk level is 3, so the target coverage of each modality group is 0.7. It should be noted that the target coverage can also be set differently according to the strategic importance of the modality group. For modality groups that play a key role in the security monitoring of the area (such as radar modality groups), their target coverage can be higher than that of other modality groups. The determination of the target coverage is the input for subsequent monitoring gap calculation. The higher the target coverage is set, the larger the monitoring gap of each modality group, the more monitoring actions need to be arranged in the generated strategy, and the greater the demand for resources. The mapping relationship between risk level and target coverage is a dynamic adjustment process driven by the threat situation. That is, when the threat situation changes due to changes in the environment or the effect of execution, the risk level changes accordingly, the target coverage is adjusted accordingly, and the generated strategy is updated accordingly.

[0065] After determining the target coverage level for each modality group, it is compared with the actual coverage level obtained from the coverage status assessment. The difference is the monitoring gap for each modality group. The monitoring gap is a non-negative value; when the actual coverage level has reached or exceeded the target coverage level, the gap is zero, representing the amount of coverage deficiency that the modality group needs to compensate for to achieve the target coverage level. The monitoring gaps of each modality group are combined to form a gap vector. Each component of the gap vector corresponds to the gap value of a modality group, and the dimension of the vector is equal to the number of modality groups. For example, the target coverage of each modality group... The coverage levels and actual coverage are as follows: Mode A (radar): target coverage level 0.7, actual coverage level 0.82, monitoring gap = max(0.7-0.82,0) = 0 (meets the standard, no gap); Mode B (electro-optical): target coverage level 0.7, actual coverage level 0.65, monitoring gap = 0.7-0.65 = 0.05; Mode C (acoustic): target coverage level 0.7, actual coverage level 0.4, monitoring gap = 0.7-0.4 = 0.3; gap vector = (0, 0.05, 0.3).

[0066] The gap vector shows that the coverage of mode group A has met the standard and no additional strategy is needed; mode group B has a slight gap and needs to be appropriately reinforced; mode group C has a significant gap and is the direction that the current strategy needs to focus on; the gap vector provides a quantitative basis for subsequent priority ranking and strategy generation.

[0067] The gap vector contains gap information for each modality group. However, the gap sizes of different modality groups can vary significantly. A comprehensive assessment of the urgency of each modality group's gap is necessary to determine the priority of strategy implementation. Gap urgency depends not only on the gap size of a single modality group (amplitude dimension) but also on the position of that modality group within the overall gap vector structure (direction dimension). Specifically, the amplitude of the gap vector reflects the overall scale of the gaps in each modality group; a larger amplitude indicates a more severe overall monitoring coverage deficiency. The amplitude is calculated using the norm of the gap vector (e.g., the Euclidean norm). The direction of the gap vector reflects the relative distribution of gaps in each modality group; that is, the larger the projection of a gap component of a particular modality group onto the vector direction, the higher the proportion of that modality group's gap in the overall gap, and the greater its urgency.

[0068] The urgency of the gap for each mode group is determined by the proportion of the gap value in the gap vector amplitude. Mode groups with larger gap values ​​have higher urgency; mode groups with zero gap values ​​have zero urgency. For example, in the gap vector (0, 0.05, 0.3), the gap vector amplitude is... The urgency of the gap in each modality group is calculated based on its proportion of the total gap. The urgency of modality group A is 0 / (0+0.05+0.30)=0; the urgency of modality group B is 0.05 / 0.35≈0.143; the urgency of modality group C is 0.3 / 0.35≈0.857. The modality groups are sorted according to their urgency: Modality group C (urgency 0.857, highest priority) > Modality group B (urgency 0.143, second priority) > Modality group A (urgency 0, no priority requirement). This sorting result determines the processing order of subsequent strategy generation, prioritizing the monitoring actions for the modality group with the largest gap and the highest urgency.

[0069] Based on the priority ranking, starting with the modality group with the highest urgency, a corresponding set of monitoring actions is generated for each modality group with a gap, and the action set is mapped to the task allocation instructions of specific monitoring nodes. The generation of the monitoring action set depends on the nature of the gap and the characteristics of the modality group. Taking modality group C (acoustic, gap 0.3) as an example, this modality group currently has only one acoustic node S1 with a coverage level of 0.4, which is significantly different from the target coverage level of 0.7. The monitoring actions to make up for this gap may include: activating the backup acoustic node in the dormant state in this area, adjusting the detection direction and sampling frequency of the existing acoustic node S1 to expand the effective coverage range, or coordinating acoustic nodes in the neighboring area to provide support coverage for this area. If there is a backup acoustic node S2 in the dormant state in this area, the monitoring action set is to wake up node S2 and deploy it to the center of the acoustic coverage blind spot.

[0070] The monitoring action set is mapped to node-level task allocation instructions. The task allocation instruction for node S2 is to switch from dormant to active state, adjust the detection direction to the specified coverage blind zone, and set the sampling frequency to the standard frequency. The task allocation instruction for node S1 is to maintain the current task unchanged. Taking mode group B (photoelectric, gap 0.05) as an example, the gap of this mode group is small and can be compensated by fine-tuning the parameters of the existing nodes. The monitoring action set is to adjust the detection angle of node O2 to cover the blind zone of the boundary area. The task allocation instruction for node O2 is to shift the detection angle clockwise by a certain number of degrees. Taking mode group A (radar, gap 0) as an example, this mode group has met the standard and no additional monitoring actions need to be arranged. The existing task allocation remains unchanged.

[0071] The task allocation instruction is the core output of the strategy, but before it is finally issued, it still needs to undergo resource conflict detection and priority coordination.

[0072] In the aforementioned steps, the task allocation instructions for each modal group are generated independently without considering the resource constraints of the monitoring nodes. In actual operation, a node may be required by task allocation instructions from multiple modal groups simultaneously, or the remaining working capacity of a node may be insufficient to handle the assigned task volume. In this case, it is necessary to determine whether there is a resource conflict and prioritize coordination based on the urgency of the gap. Specifically, based on the assessed resource status (the remaining working capacity index of each node), it is checked whether the target node corresponding to each task allocation instruction has sufficient remaining working capacity to execute the task. If the total number of tasks assigned to a node exceeds its remaining working capacity, it is determined that there is a resource conflict.

[0073] For example, the monitoring action of the aforementioned mode group C requires waking up the backup acoustic node S2 and requiring S2 to operate at the highest sampling frequency, with a required working capacity index of 0.7. However, according to the resource situation assessment results, the current remaining working capacity index of node S2 is only 0.55 (due to its poor battery condition after a long period of dormancy), which is insufficient to meet the requirement of 0.7. At this time, it is determined that there is a resource conflict. The coordination method is to use the urgency of the gap in the corresponding mode group as the weight to determine the priority of the allocation of the node's limited resources. The urgency of mode group C is 0.857 (the highest), so it enjoys the highest priority in resource allocation. The coordination result is to downgrade the task standard of node S2 from the highest sampling frequency to the standard sampling frequency (the required working capacity index is reduced to 0.5, which can be met by the remaining capacity of 0.55 of node S2). At the same time, the coverage loss caused by the downgrade of the sampling frequency is partially compensated by the fine adjustment of the detection direction of node S1.

[0074] In another scenario, node O2 is simultaneously required by both the task of modal group B (adjusting the detection angle, requiring 0.3 working capacity) and the task of modal group C (assisting acoustic coverage, requiring 0.25 working capacity). However, O2's remaining working capacity is 0.4, which is insufficient to simultaneously meet the total requirement of 0.55 for both tasks. In this case, coordination is performed based on the urgency of the gap between the two modal groups. The urgency of modal group C (0.857) is much higher than that of modal group B (0.143). Therefore, node O2 prioritizes executing the task of modal group C (consuming 0.25 working capacity). The remaining working capacity of 0.15 is insufficient to complete the task of modal group B (requiring 0.3). Thus, the part of the task of modal group B that node O2 performs is changed to be undertaken by node O1 (if O1's remaining capacity allows), or this part of the task is downgraded.

[0075] After resource conflict coordination, all task allocation instructions are within the capabilities of each node, and the resource allocation priority is consistent with the urgency of the gap. The higher the urgency of the gap, the higher the priority of the resources it receives, ensuring that limited resources flow to the most urgent monitoring needs first. The coordinated set of task allocation instructions is the final regional coordination strategy. This strategy may be further constrained and modified in the subsequent cross-regional scheduling stage of S3 according to the cross-regional scheduling scheme.

[0076] S3. Based on the situation assessment results of multiple regions, with global risk and resource utilization as optimization objectives, generate a cross-regional scheduling scheme, modify the regional coordination strategy according to the cross-regional scheduling scheme, generate execution instructions for the corresponding monitoring nodes, and issue them for execution.

[0077] Furthermore, generating a cross-regional scheduling scheme includes: The risk level of each region is determined based on the threat and coverage situation in the situation assessment results. The overall risk is obtained after integration. The resource surplus or deficit of each region is determined based on the resource situation and coverage situation. The region is divided into deficit regions and surplus regions based on resource surplus and deficit. The scheduling priority of each deficit region is determined based on the expected risk reduction value after resource replenishment in the deficit region, with the goal of reducing overall risk. Based on scheduling priority, surplus regions are matched with deficit regions in order and the amount of resources transferred is determined to generate a cross-regional scheduling scheme with the goal of improving resource utilization.

[0078] The expected risk reduction value is obtained by determining the amount of supplementary resources based on the resource gap in the deficit area, and by subtracting the risk level before supplementing resources from the expected risk level after supplementing resources.

[0079] First, determine the risk level of each region, and from this, obtain the overall risk. The risk level is the driving factor for cross-regional scheduling; that is, regions with higher risks need to be prioritized in resource allocation. The risk level is determined based on the threat situation and coverage situation in each region's situation assessment results. The threat situation reflects the degree of security threat currently faced by the region, while the coverage situation reflects the completeness of the region's monitoring coverage. Regions with high threat situation and low coverage situation have the highest risk level because they face a large threat, but their monitoring coverage is insufficient to address it. Regions with low threat situation and high coverage situation have the lowest risk level because they face a small threat and their monitoring coverage is sufficient.

[0080] Areas with both high threat and coverage status have a moderate to high risk level; although coverage is sufficient, the threat remains. Areas with both low threat and coverage status have a moderate risk level; although coverage is insufficient, the threat itself is not significant. Specifically, the risk level is defined as the weighted product of threat status and coverage gap rate. Coverage gap rate = 1 - coverage status value, reflecting the degree of inadequacy in monitoring coverage. Risk level = threat status × (1 - coverage status), indicating that risk is jointly determined by the degree of threat and coverage deficiency, and neither can be lacking. If there is only a threat but sufficient coverage, the risk is controllable; if there is insufficient coverage but no threat, the risk is limited. Only when the threat is high and the coverage is insufficient is the risk truly significant.

[0081] After the risk levels of each region are determined, they are integrated to obtain the global risk. The global risk reflects the comprehensive risk level faced by all regions as a whole. The global risk is the sum of the risk levels of each region, i.e., global risk = Σ (risk levels of each region). This ensures that the global risk remains sensitive to any increase in risk in any region, and that cross-regional scheduling is not ignored because the risk in a certain region is diluted. For example, the situation assessment results of three regions are as follows: Region A: Threat situation Th A =0.45, coverage situation Cov A =0.8, Risk level = 0.45 × (1 - 0.8) = 0.09; Region B: Threat posture Th B =0.62, coverage situation Cov B =0.55, Risk Level = 0.62 × (1 - 0.55) = 0.279; Region C: Threat Situation Th C =0.38, coverage situation Cov C =0.48, Risk level = 0.38 × (1 - 0.48) = 0.198, Global risk = 0.09 + 0.279 + 0.198 = 0.567.

[0082] The risk level distribution across the three regions shows that region B has the highest risk level because of its high threat profile and low coverage profile. The combined effect of these two deficiencies makes its risk significantly higher than other regions. Region A has the lowest risk level because of its high coverage profile. Even though the threat profile is not low, the adequacy of the coverage effectively suppresses the risk. Region C has a moderate risk level. Although its threat profile is not high, its coverage profile is the lowest. The coverage deficiencies make its risk higher than that of region A. Global risk serves as the benchmark for subsequent optimization to reduce global risk.

[0083] The resource surplus / deficit is determined for each region to classify them into deficit and surplus regions. The resource surplus / deficit measures the extent to which each region can meet its monitoring needs under the current resource level. That is, regions with abundant resources have the ability to export resources, while regions with insufficient resources need to obtain resources from external sources. The resource surplus / deficit is determined based on the resource status and coverage status. The resource status reflects the remaining working capacity of the monitoring nodes in each region, i.e., the amount of available resources. The coverage status reflects the current monitoring coverage gap in each region, i.e., the amount of resource demand. Resource surplus / deficit = available resources - resource demand. A positive difference indicates that the region still has a surplus of resources after meeting its own needs, while a negative difference indicates that the region's resources are insufficient to meet its own needs, resulting in a deficit.

[0084] For example, the resource status and coverage status of the three regions are as follows: Resource status of region A (average remaining working capacity of each node) Res A =0.82, coverage situation Cov A =0.8, coverage gap rate = 1 - 0.8 = 0.2, resource demand = 0.2 (normalized), resource surplus / deficit = 0.82 - 0.2 = 0.62, the surplus / deficit is positive, and region A is classified as a surplus region; the resource status Res of region B B =0.48, coverage situation Cov B =0.55, coverage gap rate = 1 - 0.55 = 0.45, resource demand = 0.45, resource surplus / deficit = 0.48 - 0.45 = 0.03, the surplus / deficit is close to zero, the resources in region B are barely self-sufficient, with no obvious surplus or deficit; the resource situation Res in region C C =0.35, coverage situation Cov C =0.48, coverage gap rate =1-0.48=0.52, resource demand =0.52, resource surplus / deficit =0.35-0.52=-0.17, the surplus / deficit is negative, and region C is classified as a deficit region.

[0085] The distribution of resource surplus and deficit shows that Region A has a significant resource surplus, and the remaining working capacity of its monitoring nodes far exceeds the current coverage demand, enabling it to export resources to other regions. Region B has a basically balanced resource situation, with neither a significant surplus to export nor a significant deficit to be supplemented. Region C has a resource deficit, and the remaining working capacity of its monitoring nodes is insufficient to make up for the current coverage gap, requiring external resource supplementation.

[0086] The expected risk reduction value after resource replenishment is determined for each deficit area. Based on this, the scheduling priority of each deficit area is determined with the goal of reducing overall risk. The expected risk reduction value is the amount of resources that need to be replenished based on the current resource gap of the deficit area. The risk level of the area under the current resource level and the expected risk level after resource replenishment are calculated separately, and the difference between the two is taken as the expected risk reduction value of the area. When a deficit area receives a certain amount of resource replenishment, the available resources in the area increase, the effective monitoring capability to make up for the coverage gap is enhanced, and the coverage status is improved accordingly. After the coverage status is improved, the coverage gap rate decreases, and the risk level decreases accordingly. Expected risk level = threat status × (1 - expected coverage status), where the expected coverage status is determined based on the degree of coverage gap reduction and changes in the threat status.

[0087] Taking the deficit region C as an example, the current status of region C is risk level = 0.198, resource surplus / deficit = -0.17 (0.17 units of resource deficit), and coverage status = 0.48; the amount of resources that region C needs to replenish is 0.17 units; if the replenished 0.17 units of resources are used entirely to fill the coverage gap, then the expected improvement in coverage status is related to the ratio of the replenished resources to the current gap size; the current coverage gap rate = 1 - 0.48 = 0.52, and the expected gap ratio that can be filled after replenishing 0.17 units of resources is approximately 0.17 / 0.52 ≈ 0.327, that is, filling about 32.7% of the existing gap, and the expected coverage status = 0.48 + 0.327 × 0.52 = 0.65.

[0088] It should be noted that the improvement in coverage is not strictly linearly related to the amount of supplementary resources in all cases. In the initial stage when the coverage gap is large, a small amount of resource input can bring about a significant improvement in coverage (increasing marginal returns); when the coverage gap is small, further improvement in coverage requires more resource input (decreasing marginal returns); the expected risk level = 0.38 × (1 - 0.65) = 0.133, and the expected risk reduction value = current risk level - expected risk level = 0.198 - 0.133 = 0.065. This value means that if region C receives 0.17 units of resource supplementation, its risk level will decrease from 0.198 to 0.133, a reduction of 0.065, and the global risk will also decrease by 0.065 accordingly; if more resources are invested, the expected coverage will be further improved, and the expected risk reduction value will be greater, but more resources need to be transferred from the surplus region, which may affect the security of the surplus region itself; when determining the actual amount of resource transfer, the scheduling plan needs to strike a balance between the risk reduction effect in the deficit region and the safety margin in the surplus region.

[0089] If multiple deficit regions exist, their expected risk reduction values ​​need to be calculated separately, and scheduling priorities should be determined based on these values. With the goal of reducing global risk, the region with the larger the expected risk reduction value will contribute more to the reduction of global risk after receiving resource replenishment, and therefore has a higher scheduling priority. If another deficit region D exists at the same time, and its expected risk reduction value is 0.042 (less than region C's 0.065), then the scheduling priority is region C > region D. Cross-region scheduling should prioritize matching resource sources for region C to maximize the reduction effect of global risk. In this embodiment, only region C is a deficit region, so region C has the highest and unique scheduling priority and does not need to be compared and ranked with other deficit regions.

[0090] Once the scheduling priority is determined, resources are matched from surplus regions and the amount of resources transferred is determined in descending order of priority for each deficit region. The matching process needs to consider that the amount of resources transferred should meet the resource gap of the deficit region as much as possible, which serves to reduce overall risk, and that the amount of resources transferred should not exceed the transferable capacity of the surplus region, which serves to improve resource utilization and ensure that surplus resources are used effectively rather than left idle. For the deficit region with the highest priority, the region with the most abundant transferable resources from the surplus region is selected as the resource source, and the amount of resources transferred is the smaller of the following two values: the resource gap of the deficit region and the transferable resources of the surplus region. If the transferable resources of a single surplus region are insufficient to make up for the entire gap of the deficit region, matching continues from other surplus regions until the gap of the deficit region is completely made up or the transferable resources of all surplus regions are exhausted.

[0091] For example, region C has a resource deficit of 0.17, while region A has a resource surplus of 0.62. The amount of resources that can be transferred out should be a reasonable proportion of the surplus, such as 50%, to avoid excessive output that could increase the risk for region A. Therefore, the amount of resources that can be transferred out from region A is 0.62 × 0.5 = 0.31. Since the amount of resources that can be transferred out from region A (0.31) is greater than the deficit of region C (0.17), all resources should be transferred from region A to region C. The amount of resources transferred is min(0.17, 0.31) = 0.17. The cross-regional scheduling scheme is... 0.17 units of resources are transferred from region A to region C. This scheduling scheme achieves two optimization objectives simultaneously. In terms of global risk, after region C acquires the resources, its risk level decreases from 0.198 to 0.133, and the global risk decreases from 0.567 to 0.567-0.065=0.502. In terms of resource utilization, 0.17 units of the 0.62 units of surplus resources in region A are converted from idle to effective use, thus improving resource utilization. Region A still retains a surplus of 0.62-0.17=0.45 units as a safety margin.

[0092] Specifically, the execution instructions for generating the corresponding monitoring nodes include: Map the resource transfer volume in the cross-regional scheduling scheme to the transfer relationship of monitoring nodes, and determine the monitoring nodes after the transfer in each region; Based on the monitoring nodes after each regional transfer, the task allocation instructions in the regional collaboration strategy are constrained and modified, and the constrained and modified task allocation instructions are used to generate execution instructions for the corresponding monitoring nodes and sent out for execution.

[0093] The cross-regional scheduling scheme determines the amount of resources transferred between regions, but this scheme is at the regional level and needs to be mapped to the transfer relationships of specific monitoring nodes before it can be executed at the node level. Specifically, the resource transfer amount is converted into which specific monitoring nodes are transferred from surplus regions to deficit regions according to preset mapping rules. The choice of mapping rules depends on the needs of the application scenario. In surplus regions, nodes that have the least impact on the current tasks in the surplus region are prioritized for transfer, that is, nodes with low task priority in the surplus region or those that can be replaced by other nodes. At the same time, the distance or communication reachability between the node and the deficit region are considered, and nodes with lower transfer costs are prioritized.

[0094] The mapping rule is to first calculate the transition score Z for each candidate node m in the surplus region. m =w t ×(1-P m )+w r ×R m +w d ×(1-D m / D max ), P m R is the normalized value of the task priority of node m in the surplus region (ranging from 0 to 1, with higher values ​​indicating higher priority). m D represents the remaining working capacity of node m (values ​​from 0 to 1). m Let D be the communication distance between node m and the deficient region. max w represents the maximum communication distance between all candidate nodes in the surplus region and the deficit region; t w r w d To assign weights to the transfer score, the relative importance of task priority, remaining capacity, and transfer cost are controlled respectively, satisfying w t +w r +w d =1; In this embodiment, w is taken as 1. t =0.5, w r =0.3, w d =0.2 indicates that minimizing the task impact is the primary consideration, followed by residual capacity and transfer costs. Then, all candidate nodes are assigned a transfer score Z. mNodes are sorted from highest to lowest resource density, and resources are transferred sequentially until the cumulative transferred resources meet the resource gap in the deficit region. The transferred resource contribution of each node is equal to the node's remaining working capacity R. m .

[0095] In this embodiment, the transition scores of the five nodes in region A are calculated as follows: Node A1: P=0.9, R=0.9, D / D max =0.8, Z=0.5×(1-0.9)+0.3×0.9+0.2×(1-0.8)=0.36; Node A4: P=0.2, R=0.75, D / D max =0.3, Z=0.5×(1-0.2)+0.3×0.75+0.2×(1-0.3)=0.765; Node A5: P=0.2, R=0.7, D / D max =0.5, Z=0.5×(1-0.2)+0.3×0.7+0.2×(1-0.5)=0.71; the sorting result is A4(0.765)>A5(0.71)>A1(0.36), so node A4 is selected first for transfer, and its transfer resource contribution of 0.75 satisfies the gap demand of region C.

[0096] For example, a cross-regional scheduling scheme requires transferring 0.17 units of resources from region A to region C. Region A currently has 5 monitoring nodes, with their remaining working capacity and current task priorities as follows: Node A1 has 0.9 units of remaining capacity and high task priority (responsible for core monitoring tasks); Node A2 has 0.85 units of remaining capacity and high task priority; Node A3 has 0.8 units of remaining capacity and medium task priority (responsible for auxiliary monitoring tasks); Node A4 has 0.75 units of remaining capacity and low task priority (responsible for redundancy backup tasks); Node A5 has 0.7 units of remaining capacity and low task priority. According to the mapping rules, nodes with low task priority and moderate remaining capacity are selected first. Node A4 (low task priority, 0.75 units of remaining capacity) and Node A5 (low task priority, 0.7 units of remaining capacity) are candidate nodes. Assuming that the resource contribution of a single node is approximately a certain proportion of its remaining capacity (e.g., 0.1 to 0.15 units), Node A4 (low transfer cost and good communication reachability with region C) is selected for transfer. The resource contribution of Node A4 is approximately 0.17 units, which exactly meets the shortfall requirement of region C.

[0097] The transfer relationship is determined as node A4 is transferred from region A to region C. The node set after the transfer is updated as follows: the monitoring nodes after the transfer in region A are {A1, A2, A3, A5}, a total of 4 nodes; the monitoring nodes after the transfer in region C are the original node set in region C ∪ {A4}. This mapping process is the key bridge connecting the region-level scheduling scheme and the node-level execution instructions.

[0098] After the node transfer relationship is determined, the original regional coordination strategies of each relevant region need to be constrained and modified. The core principle of constraint modification is to minimize the adjustment of the task allocation instructions affected by the original strategy after the node set changes, so that the strategy matches the new node set. Constraint modification is divided into two cases: surplus regions and deficit regions. For surplus regions (regions that have lost nodes), the tasks originally undertaken by the transferred nodes need to be redistributed to the remaining nodes, or the tasks should be downgraded or abandoned if they cannot be replaced. Node A4 originally undertook a redundant backup task (low task priority) in region A. The essential role of this task was to provide redundancy guarantee for the main monitoring task of region A. After node A4 is transferred, the redundant backup task is partially taken over by node A5, which has the lowest task priority among the remaining nodes. After the takeover, the workload of node A5 increases, but it is still within its remaining working capacity. The task allocation instructions after the constraint modification of region A are as follows: nodes A1 and A2 maintain their original core monitoring tasks, node A3 maintains its auxiliary monitoring tasks, and node A5 adds some of the redundant backup tasks of the original A4 to its original tasks.

[0099] For the deficient areas (areas that acquire nodes), the newly added nodes need to be incorporated into the existing task allocation system to supplement the coverage gaps or replace some of the tasks of existing overloaded nodes. For example, in area C, node A4 is acquired. In the original task allocation of area C, there are some coverage blind spots, i.e., monitoring gaps. Node A4 is assigned to the center of the coverage blind spot to perform monitoring tasks to make up for the coverage gaps. At the same time, some of the tasks of the original overloaded nodes in area C (such as node C3, with only 0.3 remaining working capacity) are released to node A4 to alleviate the overload of C3. The task allocation instructions after the constraint correction for area C are that the original nodes C1 and C2 maintain their existing tasks, and some of the overloaded tasks of node C3 are transferred to node A4. Node A4 undertakes the monitoring tasks of the coverage blind spots and some of the original tasks of C3.

[0100] The task allocation instructions, after constraint correction, have fully considered the changes brought about by node migration, and have been converted into execution instructions for the corresponding monitoring nodes and issued. The execution instructions include the specific actions and parameters that each monitoring node needs to perform. Taking node A4 as an example, its execution instructions include the following: switch from the network domain of region A to the network domain of region C, adjust the monitoring coverage direction to the center of the specified coverage blind zone, set the sampling frequency to the standard frequency, and set the task priority to the normal monitoring priority of region C. Taking node A5 as an example, its execution instructions include the following: while maintaining the original tasks, increase the coverage of the redundant backup tasks originally undertaken by A4, and switch the working mode from standard mode to high-load mode. Taking node C3 as an example, its execution instructions include the following: mark some overloaded tasks as migrated and stop the execution of these tasks, and restore the working mode from high-load mode to standard mode.

[0101] The execution instructions are sent to each monitoring node through the communication network of the regional control center. After receiving the execution instructions, each node adjusts its working status, coverage direction, sampling frequency and other parameters according to the instructions, and starts to execute the new task allocation in the next sampling cycle. After the execution instructions are issued, each node executes the monitoring tasks according to the new instructions. The generated monitoring data will be re-weighted and fused in the next iteration cycle to output new reliable fused data, and enter a new round of situation assessment, strategy generation and cross-regional scheduling cycle. At the same time, it is compared with the situation assessment results of this round. After determining the effect deviation, the node reliability and situation assessment parameters are adjusted to ensure that the next round of assessment and strategy generation are calibrated and optimized from data quality to assessment.

[0102] S4. Based on the monitoring data after execution, re-evaluate the situation to obtain the post-execution evaluation result. Compare the post-execution evaluation result with the situation evaluation result, determine the effect deviation, and then adjust the node credibility and situation evaluation parameters of each monitoring node in the corresponding area in reverse.

[0103] The reverse adjustment of the node reliability and situation assessment parameters for each monitoring node in the corresponding area includes: The effect deviation between the post-implementation evaluation results and the situation assessment results is attributed to the contribution of each monitoring node in the corresponding area to the trusted fusion data, thus obtaining the attribution deviation for each monitoring node. Simultaneously, the effect deviation is decomposed into the evaluation dimensions in the situation assessment results to obtain the dimensional deviation of each evaluation dimension; Based on the attribution bias of each monitoring node, the node credibility of the corresponding monitoring node is adjusted, and based on the dimensional bias of each assessment dimension, the corresponding parameters in the situation assessment are adjusted.

[0104] The starting point for reverse adjustment is the determination of effect deviation. Effect deviation measures the change between the situation assessment results before and after strategy implementation, reflecting the gap between the actual and expected effects of this round of strategy implementation. If the strategy implementation is effective, the post-implementation assessment results should be significantly better than the pre-implementation assessment results, with a small effect deviation. If the strategy implementation is ineffective, the improvement between the post-implementation assessment results and the pre-implementation assessment results should be insufficient or even worse, with a large effect deviation. After the execution command is issued and a complete monitoring cycle is completed, the monitoring nodes in each region collect monitoring data according to the new task allocation instructions. This monitoring data is weighted and fused to output new reliable fused data, which is then used for situation assessment to obtain the post-implementation assessment results. The post-implementation assessment results are then compared with the pre-implementation situation assessment results dimension by dimension to determine the effect deviation.

[0105] For example, the situation assessment result for a certain region before strategy implementation is: Threat Situation Th before=0.448, coverage situation Cov before =0.656, Resource Status Res before =0.65, overall assessment value =0.562, the situation assessment result after strategy implementation is: Threat Situation Th after =0.38, coverage situation Cov after =0.72, Resource Status Res after =0.68, overall assessment value =0.555; the effect deviation for each dimension is calculated as follows: threat posture deviation = |Th after -Th before |=|0.38-0.448|=0.068 (Threat decreasing, trend positive); Coverage bias=|Cov after -Cov before |=|0.72-0.656|=0.064 (Coverage improved, positive trend); Resource situation deviation =|Res after -Res before |=|0.68-0.65|=0.03 (Resource balance has improved slightly, and the trend is positive).

[0106] Overall performance deviation = |Post-execution overall assessment value -Pre-execution overall assessment value| = |0.555 - 0.562| = 0.007. This small value indicates that the overall assessment result has only changed slightly. However, the analysis of each dimension shows that both the threat posture and coverage posture have improved, but the improvement has not reached the expected level. If the strategy objective is to reduce the threat posture to below 0.3, then the gap between the actual value of 0.38 and the target is the main source of performance deviation. Performance deviation is the driving signal for subsequent dual-path reverse tracking. That is, the larger the deviation, the more the current data quality or assessment needs to be calibrated; the smaller the deviation, the better the current mechanism is operating, and the smaller the adjustment should be.

[0107] Once the performance deviation is determined, it is necessary to trace back along the generation path of the trusted fusion data to attribute the overall performance deviation to the contribution of each monitoring node in the corresponding area to the trusted fusion data of that area. This yields the attribution deviation for each monitoring node. The attribution deviation reflects the degree to which the data quality of that node contributes to the final assessment deviation. The larger the attribution deviation, the greater the impact of the node's data on the trusted fusion data, and thus the greater the impact on the situation assessment results. Therefore, it is one of the main factors leading to performance deviation.

[0108] The basic logic of attribution is that credible fused data is generated by weighted fusion of data from each node. The contribution of each node's data to the fusion result is determined by its node credibility. The higher the node credibility, the greater the influence of that node's data on the fusion result. When the fusion result leads to evaluation bias, the node with the greater contribution is also more responsible for the bias. Therefore, attribution bias is related to the node's contribution to the credible fused data and the degree of consistency between that node's data and the data from other nodes.

[0109] Specifically, a node-by-node impact analysis method is used for attribution. This involves sequentially assuming that a node is removed from the fusion process, recalculating the reliable fusion data and situation assessment results without that node, and comparing the changes in effect deviation before and after removal. If the effect deviation significantly decreases after removing a node, it indicates that the data from that node is the main source of the deviation, and its attribution deviation should be large. If the effect deviation does not change much or even increases after removal, it indicates that the data from that node contributes little to the deviation or even helps to reduce the deviation, and its attribution deviation should be small. For example, there are four monitoring nodes in this area, with node reliability and data values ​​as follows: Node N1 reliability 0.8, data value 105; Node N2 reliability 0.7, data value 102; Node N3 reliability 0.65, data value 108; Node N4 reliability 0.55, data value 135.

[0110] The weighted fusion result of the four nodes is (0.8×10⁵+0.7×10²+0.65×10⁸+0.55×135) / (0.8+0.7+0.65+0.55)≈111.06. After situation assessment, this fusion result deviates from the expected value. The node-by-node impact analysis is as follows: After removing node N1 (highest credibility, largest contribution), the fusion result is (71.4+70.2+74.25) / 1.9≈113.61, and the effect deviation changes by 2.55. The data value of N1, 105, is lower than the mean of the other three nodes, (10²+10⁸+135) / 3≈115. The fusion value increases after removing N1, indicating that the data of N1 is dragging down the fusion result. After removing node N2 (second highest credibility), ... The fusion result is approximately 114.23 (84 + 70.2 + 74.25) / 2, with a change in effect deviation of 3.17. The data value of N2, 102, deviates from the direction of most nodes, and the fusion value increases more significantly after removal. After removing node N3, the fusion result is approximately 112.02 (84 + 71.4 + 74.25) / 2.05, with a change in effect deviation of 0.96. The data value of N3, 108, is relatively neutral, and its removal has a relatively small impact. After removing node N4, the fusion result is approximately 104.93 (84 + 71.4 + 70.2) / 2.15, with a change in effect deviation of 6.13. The data value of N4, 135, deviates significantly from the range of the other three nodes and is the abnormal factor that greatly increases the fusion result. After removal, the fusion value returns to the normal range, and the effect deviation is greatly reduced.

[0111] After normalizing the variation in effect deviation for each node, the attribution bias is obtained. The sum of the variation in deviation for the four nodes is 2.55 + 3.17 + 0.96 + 6.13 = 12.81. The attribution biases are N1 = 2.55 / 12.81 ≈ 0.199, N2 = 3.17 / 12.81 ≈ 0.247, N3 = 0.96 / 12.81 ≈ 0.075, and N4 = 6.13 / 12.81 ≈ 0.479. From the attribution results, it can be seen that the attribution bias of node N4 is... The node with the largest attribution bias (0.479) has a significantly different data value of 135 compared to other nodes, making it the primary factor causing the bias. The attribution bias of node N2 is the second largest (0.247), while that of node N1 is moderate (0.199). The attribution bias of node N3 is the smallest (0.075), and its data is relatively neutral, contributing the least to the bias. This attribution result provides a precise basis for subsequent node credibility adjustments, meaning that the larger the attribution bias of a node, the greater the magnitude of its credibility adjustment.

[0112] After determining the attribution bias, the credibility of each monitoring node is adjusted. The basic principle of adjustment is that the larger the attribution bias, the greater the contribution of the node's data to the effect bias, and the greater the reduction in its credibility should be, so that the influence of the node in the next round of fusion is compressed; the smaller the attribution bias, the smaller the contribution of the node's data to the bias, and its credibility should remain basically unchanged or only slightly adjusted. Specifically, a proportional adjustment method is adopted, that is, the adjustment amount of node credibility = adjustment coefficient × attribution bias of the node, and the adjusted node credibility = current credibility × (1 - adjustment coefficient × attribution bias). The adjustment coefficient controls the sensitivity of the overall adjustment. The larger the adjustment coefficient, the greater the change in credibility caused by the same attribution bias; the smaller the adjustment coefficient, the more conservative the adjustment. To prevent excessive jumps in credibility in a single iteration, a minimum lower limit of credibility is set to ensure that even if the attribution bias is large, the node credibility will not drop to zero.

[0113] The selection of the adjustment coefficient and the minimum confidence level needs to balance adjustment sensitivity and system stability. The adjustment coefficient ranges from (0,1). A larger value results in a larger single adjustment amplitude and a faster system response, but may lead to confidence level oscillations. A smaller value results in a smoother adjustment and a more stable system, but a slower convergence speed. In this embodiment, the adjustment coefficient is set to 0.5 as the initial value for medium sensitivity. The minimum confidence level ranges from [0.05,0.2]. A value that is too small may result in abnormal nodes still having a weak but non-zero fusion weight, while a value that is too large may over-compress the recovery space of the downgraded nodes. In this embodiment, the minimum confidence level is set to 0.1.

[0114] For example, node N1 has a current confidence level of 0.8 and an attribution bias of 0.199, so its adjusted confidence level is 0.8 × (1 - 0.5 × 0.199) = 0.721; node N2 has a current confidence level of 0.7 and an attribution bias of 0.247, so its adjusted confidence level is 0.7 × (1 - 0.5 × 0.247) = 0.614; node N3 has a current confidence level of 0.65 and an attribution bias of 0.075, so its adjusted confidence level is 0.65 × (1 - 0.5 × 0.075) = 0.626; and node N4 has a current confidence level of 0.55 and an attribution bias of 0.479, so its adjusted confidence level is 0.55 × (1 - 0.5 × 0.479) = 0.419.

[0115] The adjustment results show that the confidence level of node N4, with the largest attribution bias, decreased from 0.55 to 0.419, a reduction of approximately 23.8%, which is the largest decrease among the four nodes. The confidence level of node N3, with the smallest attribution bias, decreased from 0.65 to 0.626, a reduction of only about 3.7%, remaining relatively stable. This differentiated adjustment method ensures the accuracy of the adjustment. The adjusted node confidence levels will be used as weights in the next round of weighted fusion in S1. Taking node N4 as an example, after its confidence level decreased from 0.55 to 0.419, its contribution to the fusion result in the next round of fusion will decrease from approximately 20.4% (0.55 / 2.7) to approximately 16.3% (0.419 / 2.569), effectively compressing the influence of its outlier data on the fusion result.

[0116] In the situation assessment calculation path, the effect deviation can be decomposed into three assessment dimensions: threat situation, coverage situation, and resource situation, resulting in the dimension deviation for each dimension. The dimension deviation reflects the degree of contribution of each assessment dimension to the final effect deviation. That is, the larger the dimension deviation of an assessment dimension, the greater its impact on the final result, and the more the corresponding assessment parameters need to be calibrated. The decomposition method of dimension deviation is to calculate the change of each assessment dimension before and after the strategy is implemented, and compare the change with the expected change. The expected change is the improvement that each dimension should have achieved based on the strategy design objectives. The difference between the actual improvement and the expected improvement is the dimension deviation of that dimension.

[0117] For example, the assessment values ​​and expected targets for each dimension before and after strategy implementation are as follows: Threat posture: 0.448 before implementation, 0.38 after implementation, actual improvement 0.068, expected target 0.3, expected improvement = 0.448 - 0.3 = 0.148, dimension deviation = |actual improvement - expected improvement| = |0.068 - 0.148| = 0.08. The actual improvement in threat posture (0.068) is only about 46% of the expected (0.148), indicating a significant deviation in threat posture assessment or strategy implementation; Coverage posture: 0.656 before implementation, 0.72 after implementation, actual improvement 0.064, expected... The target for the period is 0.75, the expected improvement is 0.75 - 0.656 = 0.094, the dimensional deviation is |0.064 - 0.094| = 0.03, and the actual improvement of the coverage situation (0.064) is about 68% of the expected (0.094), with a small deviation. The resource situation is 0.65 before implementation and 0.68 after implementation, with an actual improvement of 0.03. The expected target is 0.7, the expected improvement is 0.7 - 0.65 = 0.05, the dimensional deviation is |0.03 - 0.05| = 0.02, and the actual improvement of the resource situation (0.03) is about 60% of the expected (0.05), with the smallest deviation.

[0118] After normalizing the dimensional biases, the dimensional bias distribution is obtained. The sum of the biases is 0.08 + 0.03 + 0.02 = 0.13. The dimensional biases are: threat dimension = 0.08 / 0.13 ≈ 0.615, coverage dimension = 0.03 / 0.13 ≈ 0.231, and resource dimension = 0.02 / 0.13 ≈ 0.154. The dimensional bias distribution shows that the threat posture has the largest dimensional bias, indicating that there is the largest deviation between the assessment results of the threat posture and the actual effect after the strategy is implemented. The possible reason is that the threat sensitivity parameter is set improperly, either too high, which leads to an overestimation of the threat, or too low, which leads to an underestimation of the threat. This results in the strategy generated based on the assessment being either insufficient or excessive in responding to the threat. The dimensional bias of the coverage posture is in the middle, and the dimensional bias of the resource posture is the smallest, indicating that the assessment accuracy of these two dimensions is relatively good.

[0119] After determining the dimensional bias, the corresponding parameters of each assessment dimension in the situation assessment are adjusted. The threat situation, coverage situation, and resource situation in the situation assessment are weighted and fused with dimension weights to obtain the comprehensive assessment result. The dimension weights control the relative importance of each dimension in the comprehensive assessment. The larger the dimensional bias of an assessment dimension, the less accurate the assessment result of that dimension is, and the more calibrated the assessment parameters of that dimension are. For dimensions with large dimensional bias, the weights are shifted in the direction of reducing the influence of that dimension on the comprehensive assessment result, so as to reduce the contamination of the comprehensive result by the assessment error of that dimension. At the same time, the weights of dimensions with small dimensional bias are increased accordingly, so that their accurate assessment results account for a larger proportion in the comprehensive assessment.

[0120] For example, before adjustment, the dimension weights were: threat dimension weight w1 = 0.45, coverage dimension weight w2 = 0.35, resource dimension weight w3 = 0.2, and dimension deviations were: threat dimension 0.615, coverage dimension 0.231, and resource dimension 0.154. The adjustment method was to make the adjustment amount of the dimension weights inversely proportional to the dimension deviations; that is, the larger the dimension deviation, the more the weight of that dimension was reduced; the smaller the dimension deviation, the more the weight of that dimension was increased. Let the adjustment range be 0.15, so that the adjustment of the dimension weights has sufficient sensitivity while avoiding excessive jumps in weights. The adjustment process of each dimension weight is as follows.

[0121] The threat dimension (maximum deviation 0.615) has its weight reduced, with the new weight w1' = w1 - 0.15 × 0.615 = 0.358; the coverage dimension (medium deviation 0.231) has its weight moderately increased, with the new weight w2' = w2 + 0.15 × (1 - 0.615) × (0.231 / (0.231 + 0.154)) = 0.385; the resource dimension (minimum deviation 0.154) has its weight moderately increased, with the new weight w3' = w3 + 0.15 × (1 - 0.615) × (0.154 / (0.231 + 0.154)) = 0.223. After normalization, w1'' = 0.358 + w2' + w3' = 0.385 + 0.223 = 0.966. / 0.966≈0.371, w2''=0.385 / 0.966≈0.399, w3''=0.223 / 0.966≈0.231; The adjusted dimension weights are: threat dimension 0.371 (down from 0.45), coverage dimension 0.399 (up from 0.35), and resource dimension 0.231 (up from 0.20); The physical meaning of this adjustment is that the threat posture performed poorly in this round of assessment (with the largest dimensional deviation), and its assessment accuracy was questioned. Therefore, its weight was reduced in the next round of comprehensive assessment to reduce the contamination of the comprehensive results by its assessment error; the assessments of coverage posture and resource posture were relatively accurate, so their weights were increased in the next round of comprehensive assessment to make the accurate assessment dimensions account for a larger proportion in the comprehensive results.

[0122] The adjusted dimensional weights will be used in the next round of situation assessment. Taking the threat dimension as an example, after its weight is reduced from 0.45 to 0.371, its contribution to the next round of comprehensive assessment will decrease from 45% to 37.1%. Even if there are still biases in the assessment of the threat situation, its impact on the comprehensive assessment results will be effectively compressed. As the threat situation assessment parameters are continuously calibrated in subsequent iterations, the accuracy of the threat situation assessment will gradually improve, the dimensional bias will decrease accordingly, and its weight will gradually recover.

[0123] The core feature of the reverse adjustment mechanism described in claim 9 is its two-layer structure. The node credibility adjustment of the data quality layer and the dimension weight adjustment of the evaluation are carried out simultaneously and independently, but they both serve the goal of reducing the effect deviation. The adjustment of the data quality layer solves the problem of whether the input data is reliable. When the data quality of some nodes is poor, causing the credible fusion data to deviate from the true value, the credibility of these nodes is reduced, so that the influence of abnormal data in the next round of fusion is compressed and the accuracy of the credible fusion data is improved. This adjustment acts at the very beginning of the entire method chain, improving the input quality from the data source.

[0124] The adjustment of dimensional weights in the assessment addresses the issue of assessment accuracy. When there is a significant discrepancy between the results of certain assessment dimensions and the actual effects, adjusting the dimensional weights increases the influence of accurate dimensions and decreases the influence of inaccurate dimensions in the next round of assessment, thereby improving the reliability of the overall assessment results. This adjustment applies to the situation assessment stage. The adjustments at the two levels are independent yet complementary. The improved node credibility makes the credible fusion data more accurate, providing higher-quality input for situation assessment. The optimization of dimensional weights makes the situation assessment more accurately reflect the true state of each dimension. In the next round, the entire link from S1 to S4 is improved in both data quality and assessment, and the effect deviation continues to converge in subsequent iterations.

[0125] In the first iteration, the effect deviation was 0.007 (change in the overall assessment value), with the largest deviation in the threat posture dimension and the largest attribution deviation in node N4. After two-layer adjustment, the credibility of node N4 decreased from 0.55 to 0.419, and the threat dimension weight decreased from 0.45 to 0.371. In the second iteration, due to the reduction in the credibility of node N4, the influence of its anomalous data in the weighted fusion decreased from 20.4% to 16.3%, improving the accuracy of the credible fusion data. The threat posture assessed based on this fusion data is closer to the true value. At the same time, because the threat dimension weight was reduced, even if there is still a slight deviation in the threat posture assessment, its contamination of the overall assessment result is also reduced. It is expected that the effect deviation in the second round will be less than 0.007 in the first round.

[0126] As iterations continue, the node with the largest attribution bias in each cycle is continuously downweighted, and the evaluation dimension with the largest dimensional bias is continuously adjusted, resulting in a gradual convergence trend in the effect bias. When the effect bias is less than the convergence threshold (0.005 in this embodiment) for N consecutive iteration cycles, it indicates that the current data quality and evaluation are accurate enough, and the system enters a steady state. At this point, the node credibility and dimensional weights are only adjusted when environmental changes cause the effect bias to increase again, demonstrating the continuous calibration capability.

[0127] Example 2 This application provides a hierarchical collaborative management and control system for regional situational monitoring, which includes a fusion module, an evaluation module, a scheduling module, and a feedback module.

[0128] The fusion module is used to acquire heterogeneous data from monitoring nodes in each region, generate node credibility based on the status information of each monitoring node, and perform weighted fusion of the spatiotemporally aligned heterogeneous data with the node credibility as the weight to output credible fused data. The assessment module performs a situation assessment of the area based on the trusted fusion data and generates a regional collaborative strategy based on the situation assessment results. The scheduling module, based on the situation assessment results of multiple regions, generates a cross-regional scheduling scheme with the optimization objectives of global risk and resource utilization. It then modifies the regional coordination strategy according to the cross-regional scheduling scheme, generates execution instructions for the corresponding monitoring nodes, and issues them for execution. The feedback module is used to reassess the situation based on the monitoring data after execution, obtain the post-execution assessment result, compare the post-execution assessment result with the situation assessment result, determine the effect deviation, and then adjust the node credibility and situation assessment parameters of each monitoring node in the corresponding area in reverse.

[0129] The technical principle of the system in this application embodiment is similar to the method described above in this application embodiment. Therefore, the system implementation is the same as the method implementation, and the detailed technical means will not be repeated.

Claims

1. A hierarchical collaborative control method for regional situation monitoring, characterized in that, include: Heterogeneous data is obtained from monitoring nodes in each region. Node credibility is generated based on the status information of each monitoring node. The spatiotemporally aligned heterogeneous data is then weighted and fused using the node credibility as a weight to output credible fused data. Based on the trusted fusion data, a situation assessment is performed on the region, and a regional collaborative strategy is generated based on the situation assessment results. Based on the situation assessment results of multiple regions, with the optimization objectives of global risk and resource utilization, a cross-regional scheduling scheme is generated. The regional coordination strategy is constrained and modified according to the cross-regional scheduling scheme, and the execution instructions of the corresponding monitoring nodes are generated and issued for execution. Based on the monitoring data after execution, a new situation assessment is performed to obtain the post-execution assessment result. The post-execution assessment result is compared with the situation assessment result. After determining the effect deviation, the node credibility and situation assessment parameters of each monitoring node in the corresponding area are adjusted in reverse.

2. The hierarchical collaborative control method for regional situational monitoring as described in claim 1, characterized in that, The output trusted fusion data includes: After spatiotemporal alignment, heterogeneous data is divided into multiple modal groups according to data type. The heterogeneous data in the same modal group are weighted and fused within the group with node credibility as the weight to obtain the modal fusion value of each modal group and determine the modal uncertainty. Based on the differences between modal fusion values, the modal conflict degree between different modal groups is determined. When the modal conflict degree is greater than the conflict threshold, the consistency center in the modal fusion value is determined as the modal consensus. Determine the degree of deviation between each modality fusion value and the modality consensus, correct the node credibility of the monitoring nodes within the modality group based on the degree of deviation, and re-execute the weighted fusion within the group; Based on modal uncertainty, cross-modal fusion of modal fusion values ​​is performed to output reliable fusion data.

3. The hierarchical collaborative control method for regional situational monitoring as described in claim 2, characterized in that, Generating the node credibility includes: Extract the device operating status and historical fault records from the status information of the monitoring nodes, determine the status score based on the device operating status, determine the fault score based on the historical fault records in a time-series decay manner, and weight and fuse the status score and the fault score to obtain the health component of each monitoring node. Based on the degree of consistency between the heterogeneous data of each monitoring node and the heterogeneous data of neighboring nodes in the same mode group, the data components of each monitoring node are determined. The health component and the data component are normalized and weighted and fused to obtain the node credibility of each monitoring node.

4. The hierarchical collaborative control method for regional situational monitoring as described in claim 1, characterized in that, The situation assessment includes: The threat situation in the region is obtained by evaluating the abnormal characteristics of the trusted fusion data, the coverage situation in the region is quantitatively evaluated based on the coverage of each modality group, and the resource situation in the region is obtained by evaluating the remaining working capacity of each monitoring node. The threat situation, coverage situation and resource situation are weighted and fused with dimensional weights to obtain the situation assessment result.

5. The hierarchical collaborative control method for regional situational monitoring as described in claim 4, characterized in that, The generated region collaborative strategy includes: Based on the threat situation assessment results, the risk level is determined to determine the target coverage of each modality group. The difference between the target coverage and the coverage of the corresponding modality group in the coverage situation is taken as the monitoring gap of the corresponding modality group and constitutes the gap vector. The urgency of the gap in each modal group is determined based on the magnitude and direction of the gap vector. The modal groups are sorted according to the urgency of the gap, and a set of monitoring actions is generated for the monitoring gap of each modal group according to the sorting result and mapped to the task allocation instructions of the monitoring nodes. Based on the resource situation, determine whether there are resource conflicts, and assign instructions for tasks with resource conflicts, prioritizing and coordinating them with the urgency of the gap as the weight, and generating regional coordination strategies.

6. The hierarchical collaborative control method for regional situational monitoring as described in claim 4, characterized in that, The generation of the cross-regional scheduling scheme includes: The risk level of each region is determined based on the threat and coverage situation in the situation assessment results. The overall risk is obtained after integration. The resource surplus or deficit of each region is determined based on the resource situation and coverage situation. The region is divided into deficit regions and surplus regions based on resource surplus and deficit. The scheduling priority of each deficit region is determined based on the expected risk reduction value after resource replenishment in the deficit region, with the goal of reducing overall risk. Based on scheduling priority, surplus regions are matched with deficit regions in order and the amount of resources transferred is determined to generate a cross-regional scheduling scheme with the goal of improving resource utilization.

7. The hierarchical collaborative control method for regional situational monitoring as described in claim 6, characterized in that, The execution instructions for generating the corresponding monitoring node include: The resource transfer volume in the cross-regional scheduling scheme is mapped to the transfer relationship of monitoring nodes, and the monitoring nodes after the transfer in each region are determined. Based on the monitoring nodes after each regional transfer, the task allocation instructions in the regional coordination strategy are constrained and modified, and the constrained and modified task allocation instructions are used to generate execution instructions for the corresponding monitoring nodes and sent out for execution.

8. The hierarchical collaborative control method for regional situational monitoring as described in claim 6, characterized in that, The expected risk reduction value is obtained by determining the amount of supplementary resources based on the resource gap in the deficit area, and by subtracting the risk level before supplementing resources from the expected risk level after supplementing resources.

9. The hierarchical collaborative control method for regional situational monitoring as described in claim 1, characterized in that, The parameters for the node reliability and situation assessment of each monitoring node in the corresponding area that are adjusted in reverse include: The effect deviation between the post-implementation evaluation results and the situation assessment results is attributed to the contribution of each monitoring node in the corresponding area to the trusted fusion data, thus obtaining the attribution deviation for each monitoring node. The effect deviation is simultaneously decomposed into the evaluation dimensions in the situation assessment results to obtain the dimensional deviation of each evaluation dimension; Based on the attribution bias of each monitoring node, the node credibility of the corresponding monitoring node is adjusted, and based on the dimensional bias of each evaluation dimension, the corresponding parameters in the situation assessment are adjusted.

10. A hierarchical collaborative control system for regional situation monitoring, used to implement the hierarchical collaborative control method for regional situation monitoring as described in any one of claims 1-9, characterized in that, include: The module includes a fusion module, an evaluation module, a scheduling module, and a feedback module. The fusion module is used to acquire heterogeneous data from monitoring nodes in each region, generate node credibility based on the status information of each monitoring node, and perform weighted fusion of the spatiotemporally aligned heterogeneous data with the node credibility as the weight to output credible fused data. The assessment module performs a situation assessment of the area based on the trusted fusion data and generates a regional collaborative strategy based on the situation assessment results. The scheduling module, based on the situation assessment results of multiple regions, generates a cross-regional scheduling scheme with global risk and resource utilization as optimization objectives. It then modifies the regional coordination strategy according to the cross-regional scheduling scheme, generates execution instructions for the corresponding monitoring nodes, and issues them for execution. The feedback module is used to reassess the situation based on the monitoring data after execution, obtain the post-execution assessment result, compare the post-execution assessment result with the situation assessment result, determine the effect deviation, and then adjust the node credibility and situation assessment parameters of each monitoring node in the corresponding area in reverse.

Images