Regional security monitoring system based on computer networking
By combining digital twin models and LSTM-AE health baseline models, the problems of fault location and linkage strategy optimization in sub-health states of security systems are solved, enabling accurate monitoring and predictive linkage of camera and link status, and improving the effectiveness of system fault warning and linkage strategies.
Patent Information
- Application Number
- CN202611205981.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-10
- Publication Date
- 2026-11-17
AI Technical Summary
When existing security systems encounter sub-optimal conditions such as dust accumulation on camera lenses and degraded link bandwidth, their linkage strategies fail, making it impossible to locate faults accurately and promptly. This leads to a decline in the quality of monitored images, a decrease in the accuracy of intelligent analysis systems, and the creation of security risks.
A regional security monitoring system based on computer networking is constructed. Device tags and performance parameters are bound through a digital twin model. The reconstruction error is calculated by combining the LSTM-AE health baseline model to generate a health vector, thereby realizing fault location and predictive linkage strategy optimization.
It enables early detection of sub-health conditions and accurate fault location, improves the lead time and accuracy of fault warnings, ensures the effectiveness of linkage strategies, and avoids security mission failures due to equipment or link failures.
Smart Images

Figure CN123123557A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of security monitoring and management technology, and in particular to a regional security monitoring system based on computer networking. Background Technology
[0002] With the construction of large-scale security systems such as safe cities and smart parks, the deployment density of surveillance cameras has increased exponentially. Traditional operation and maintenance management models rely heavily on manual inspections and obvious fault alarms of the equipment itself (such as equipment offline, video signal loss, etc.). However, many hidden dangers that lead to the decline or even failure of surveillance images are not sudden disconnections, but are in a gradually deteriorating "sub-healthy" state. This seriously weakens the clarity of the surveillance images and the confidence of target detection algorithms, directly reducing the accuracy of the back-end intelligent analysis system and constituting a major hidden security vulnerability in the security system.
[0003] However, the linkage strategies of existing security systems are usually statically preset, such as "perimeter alarm triggers corresponding area lighting and camera preset positions." This type of solution has fundamental flaws: First, the linkage strategy is based on the assumption that "all devices are in ideal working condition," and does not consider the impact of sub-health conditions such as dust accumulation on camera lenses, degraded link bandwidth, and degraded node performance on the linkage effect in the actual deployment environment. This means that the preset strategy may fail at critical moments due to device malfunction. Second, maintenance personnel need to rush to the site to check each lens, adjust the focus, or test network equipment. This is extremely inefficient and highly dependent on personal experience. It is impossible to accurately, timely, and effectively distinguish common faults, and it is even more difficult to transform vague fault phenomena into quantifiable decision support information. As a result, it is impossible to form a closed-loop linkage from "discovery-diagnosis-quantification-handling."
[0004] Therefore, how to construct a complete technical solution encompassing "sub-health perception, fault location, linkage strategy optimization, and closed-loop feedback" has become a key challenge that urgently needs to be addressed in this field. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a regional security monitoring system based on computer networking, which can realize early detection of the sub-health status of the security monitoring system, accurate inversion and location of root cause failures, quantitative assessment of equipment health, and optimization and closed-loop feedback of predictive linkage strategies for security incident handling.
[0006] The objective of this invention can be achieved through the following technical solution: a regional security monitoring system based on computer networking, comprising:
[0007] The data association module builds a 1:1 digital twin model based on the collected static environmental information of the entire target monitoring domain, binds exclusive coordinate labels and virtual device proxy components to all hardware devices, encapsulates performance parameters and establishes communication topology relationships, and finally constructs a snapshot of the current state of the entire domain through real-time data channels.
[0008] The health assessment module extracts a three-dimensional time-varying feature sequence based on the collected network transmission data packets and application data packets, combines it with the LSTM-AE health baseline model, calculates the reconstruction error E(t), and makes a judgment based on the preset reconstruction error threshold, outputting an inversion positioning command or continuous routine monitoring.
[0009] When the inversion positioning command is generated, the network tomography module performs cosine similarity matching between the real-time reconstructed feature vector and the preset exclusive fault feature knowledge base, and outputs the discrimination processing result to construct the health vector.
[0010] When a sub-health abnormality exists and an external alarm event is triggered, the sub-health-linkage module triggers a predictive linkage strategy deduction process and outputs the optimal predictive linkage strategy under the sub-health abnormality.
[0011] The closed-loop feedback module continuously tracks the status of the front-end device based on the output optimal predictive linkage strategy. When the output execution condition changes abruptly, it selects the suboptimal alternative strategy and sends it out for execution.
[0012] Preferably, the target monitoring full-domain static environment information includes building walls, floor structure, and access roads; the performance parameters include maximum resolution, adjustable bit rate range, and upper limit of bandwidth for each communication port; and the current status snapshot includes read-only spatial environment data, real-time operating status of all devices, network link parameters, alarm queues, and target object coordinate trajectory.
[0013] Preferably, the three-dimensional time-varying feature sequence includes an image shallow feature sequence, a bit rate fluctuation feature sequence, and a network transmission feature sequence;
[0014] The shallow features of the image include the proportion of high-frequency coefficient components and the overall diffusion measure of motion vectors; the bit rate fluctuation features include the average bit rate of encoding and transmission within a unit time window and the variance of the bit rate of encoding and transmission within a unit time window; and the network transmission features include the TCP packet retransmission rate, the average end-to-end round-trip delay, and the latency jitter amplitude.
[0015] Preferably, the analysis process of the LSTM-AE health baseline model is as follows:
[0016] Historical time-series data under normal operating conditions of each video stream and each network link are collected as normal samples. LSTM-AE is used as the core network for baseline modeling. The normal samples are trained to obtain a healthy baseline model and output the reconstructed feature vector Ek(t) under normal operating conditions, where k represents the dimension feature, k is a natural number greater than zero, and t represents the sampling time, t>0.
[0017] The real-time collected three-dimensional time-varying feature sequence is continuously input into the corresponding exclusive health baseline model, and the real-time reconstructed feature vector EXk(t) is output.
[0018] Calculate the real-time reconstruction error E(t), retrieve the preset reconstruction error threshold for discrimination, and output the sub-health abnormality inversion positioning command or continuous routine monitoring.
[0019] Preferably, the analysis process of the health vector is as follows:
[0020] The cosine similarity YXg between the real-time reconstructed feature vector EXk of sub-health abnormalities and each exclusive fault feature in the preset exclusive fault feature knowledge base is calculated using cosine similarity. Here, g represents the exclusive fault number and g is a natural number greater than zero. The cosine similarity YXg is discriminated against the preset cosine similarity threshold and outputs the inverted root cause fault / inverted concurrent fault / atypical weak state / inverted unknown root cause. For the inverted root cause fault and the inverted concurrent fault, the degradation quantification index L is calculated. Based on the degradation quantification index L, the current effective performance percentage η is calculated, and the estimated remaining available time Ty is also calculated.
[0021] A health vector is constructed based on the discrimination processing results, the degradation quantification index L, the current effective performance percentage η, and the estimated remaining available time Ty.
[0022] Preferably, the degradation quantification index L is calculated as follows: if it is an inverted root cause fault, then the degradation quantification index = 1 - cosine similarity YXgmax; if it is an inverted concurrent fault, then the degradation quantification index = the sum of the cosine similarity YXg corresponding to the inverted concurrent fault - the product value of the corresponding cosine similarity YXg.
[0023] The current effective performance percentage η = (1 - L / preset critical degradation degree) × 100%, η ∈ [0%, 100%]. Obtain the historical G degradation quantification index L, and obtain the degradation time series evolution slope a by fitting the least squares method. When a > 0, Ty = (preset critical degradation degree - L) / a.
[0024] Preferably, the analysis process of the predictive linkage strategy deduction procedure is as follows:
[0025] Starting from external alarm events and using the target's current speed, direction of movement, and scene access constraints as boundary conditions, a data twin model is used to generate all feasible movement paths of the attacker / intrusion target within a future period. Each feasible movement path is independently bound to exclusive temporal spatial data.
[0026] For each feasible movement path, a corresponding candidate linkage strategy combination is generated by combining the preset device-area mapping rules and the linkage action template library. Each candidate strategy contains a set of linkage action sequences.
[0027] Obtain devices and links with a clear estimated remaining failure time Ty, and establish a time-series quantitative verification mechanism for the entire lifecycle of the strategy: set the complete execution time Tz of a single candidate linkage strategy combination, retain candidate linkage strategy combinations in the entire candidate linkage strategy combination where Tz is all < Ty, and construct a candidate linkage strategy combination pool;
[0028] For a single candidate linkage strategy combination in the candidate linkage strategy combination pool, the comprehensive decision score is calculated based on S1×a1+S2×a2+S3×a3, where a1, a2 and a3 are preset weight factors and are all greater than zero.
[0029] The candidate linkage strategy combination with the highest comprehensive decision score is selected from the pool of candidate linkage strategy combinations as the optimal predictive linkage strategy under the condition of sub-health abnormalities.
[0030] Preferably, S1 represents the imaging quality of key nodes, S2 represents the percentage of effective target coverage time, and S3 represents the linkage response delay.
[0031] S1 = Detection confidence × w1 + Image clarity × w2 + Recognition accuracy × w3, where w1, w2, and w3 are all preset weight coefficients and are all greater than zero; S2 = Target effective coverage time / Target effective coverage total time; S3 = max(0, 1 - Total policy delay / Policy preset maximum tolerable linkage baseline delay).
[0032] Preferably, the output execution condition change process is as follows: Set change trigger conditions: If a sudden abnormality occurs, such as a complete interruption of the link or a complete failure of the device imaging, or if the change in parameters such as the real-time effective performance percentage of the link / device or the reconstruction error exceeds the preset floating range, it is determined to be an execution condition change.
[0033] Based on the candidate linkage strategy combination pool, candidate linkage strategy combinations involving failed links / devices are removed from the pool. The remaining candidate linkage strategy combinations are sorted in descending order, and the candidate linkage strategy combination ranked first is selected as the second-best alternative strategy.
[0034] The beneficial effects of this invention are as follows:
[0035] (1) This invention relies on three-dimensional time-varying feature sequences and combines LSTM-AE health baseline model to calculate real-time reconstruction error, which can keenly identify the "sub-healthy" state, greatly improve the advance and accuracy of fault warning, and accurately distinguish single faults through feature cosine similarity matching, and can also effectively identify concurrent faults and unknown root causes. At the same time, it constructs a health vector to transform the ambiguous abnormal state into an intuitive and quantifiable assessment of the degree of deterioration and development trend.
[0036] (2) When sub-health abnormalities and external alarm events occur concurrently, this invention generates a multi-path probability situation distribution map of the target based on a digital twin model. On this basis, combined with the device health vector, a time-series quantitative verification mechanism for the entire life cycle of the strategy is introduced to ensure the effectiveness of the linkage strategy. The optimal linkage strategy is dynamically deduced from the candidate pool through comprehensive consideration, which fundamentally avoids the risk of security task failure due to device or link failure during the linkage process. In the process of executing the optimal linkage strategy, by continuously tracking the status of the front-end device, once a sudden change in the execution condition is detected, the suboptimal alternative strategy that does not depend on the fault link can be immediately selected from the candidate strategy pool and activated, so as to realize dynamic strategy switching and seamless execution. Attached Figure Description
[0037] The invention will now be further described with reference to the accompanying drawings;
[0038] Figure 1 This is a flowchart of the system of the present invention;
[0039] Figure 2 This is a schematic diagram of the health assessment module analysis of the present invention;
[0040] Figure 3 This is a reference diagram for the health vector analysis of the present invention. Detailed Implementation
[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0042] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments;
[0043] Example 1:
[0044] Please see Figures 1 to 3 As shown, the present invention is a regional security monitoring system based on computer networking, including a main control platform, a data association module, a health assessment module, a network tomography module, a sub-health linkage module, and a closed-loop feedback module;
[0045] Among them, the main control platform and the data association module have a two-way communication connection, the main control platform and the health assessment module and the network tomography module have a one-way communication connection, the health assessment module and the sub-health-linkage module have a one-way communication connection, the sub-health-linkage module and the network tomography module have a one-way communication connection, and the network tomography module and the closed-loop feedback module have a one-way communication connection.
[0046] The data association module builds a 1:1 digital twin model based on the collected static environmental information of the entire target monitoring area. It binds exclusive coordinate labels and virtual device proxy components to all hardware devices, encapsulates performance parameters, and establishes communication topology relationships. Finally, it constructs a snapshot of the current state of the entire area through a real-time data channel, specifically including:
[0047] Static environmental information of the entire target monitoring area is collected through methods such as on-site surveying, laser scanning, aerial surveying, and importing architectural CAD drawings, including building walls, floor structures, and access roads;
[0048] Three-dimensional solid modeling is carried out based on static environmental information, and a 1:1 digital twin model is built according to the actual spatial dimensions, relative orientation, and elevation ratio.
[0049] Within the digital twin model, each front-end hardware device is bound to a unique coordinate label, and various physical parameters are entered: three-dimensional installation coordinates, installation height above the ground, lens field of view range, adjustable focal length range, etc.
[0050] Divide the area into monitoring sub-zones, warning boundary lines, high-risk key areas, and personnel evacuation routes. Complete the area marking within the digital twin model and establish a spatial partition index table.
[0051] At the same time, all security hardware devices (such as PTZ cameras, network switching equipment, and lighting devices) within the target monitoring area are classified into three categories: image acquisition, network transmission, and linkage execution. Each security hardware device instantiates a corresponding virtual device proxy component, and the virtual proxy encapsulates rated performance parameters such as maximum resolution, adjustable bit rate range, upper limit of bandwidth for each communication port, data packet buffer capacity, dimming level, and response latency.
[0052] Establish communication topology relationships between all virtual device agents and save wired / wireless links and hierarchical control relationships between devices;
[0053] Establish a real-time synchronous data channel between the digital twin model and security hardware devices, continuously receive inputs from security hardware devices regarding device operating status (such as network latency), network topology status (such as online / offline), and external events (such as device failure, human touch, and intrusion of moving targets), and construct a snapshot of the current status of the entire target monitoring domain. The snapshot data structure includes read-only data of the spatial environment, real-time operating status of all devices, network link parameters, alarm queues that have occurred, and coordinate trajectory of the target object, etc.
[0054] The main control module is used to display the current status snapshot of the digital twin model and the entire target monitoring domain;
[0055] The health assessment module extracts a three-dimensional time-varying feature sequence based on the collected network transmission data packets and application data packets. Combined with the LSTM-AE health baseline model, it calculates the reconstruction error E(t), and uses a preset reconstruction error threshold for discrimination. The module then outputs an inversion positioning command or continuous routine monitoring, specifically including:
[0056] The network transmission data packets and application data packets of each video stream are continuously collected by the deployed traffic acquisition probes. The network transmission data packets include source / destination IP, TCP segments (such as sequence number, acknowledgment number), etc., and the application data packets include high-frequency / low-frequency transformation coefficient distribution data, video stream minimum transmission unit, etc.
[0057] For each video stream, extract three-dimensional time-varying feature sequences from network transmission data packets and application data packets: including shallow image feature sequences, bitrate fluctuation feature sequences, and network transmission feature sequences;
[0058] Image shallow feature sequence extraction: proportion of encoded high-frequency coefficient components, overall diffusion measure of motion vectors.
[0059] High-frequency coefficient component ratio: After a video frame is processed by existing technologies (such as DCT discrete cosine transform), the ratio of the total energy of the high-frequency transform coefficients to the total energy of the global coefficients. The high-frequency coefficient component ratio is used to characterize the edge details and texture richness of the image, and directly corresponds to the lens cleanliness and image sharpness.
[0060] Motion vector global diffusion metric: Suppose a single frame image is divided into N (N>0) coded macroblocks, and the motion vector corresponding to the i-th macroblock is... dxi is the horizontal motion component, and dyi is the vertical motion component;
[0061] Calculate the vector deviation magnitude ei between the motion vector of each macroblock and the global average vector;
[0062] Based on the statistical dispersion of the deviation modulus ei of all macroblock vectors, a motion vector overall diffusion metric is defined. The motion vector overall diffusion metric is used to characterize the uniformity and rationality of dynamic changes in the image, and can accurately reflect hidden equipment faults such as lens defocus, image blur, and imaging abnormalities.
[0063] Bitrate fluctuation feature extraction: Set a fixed time-series statistics window, continuously collect the encoding and transmission bitrate data of a single video stream, and construct the bitrate fluctuation time-series features, including the average encoding and transmission bitrate within a unit time window, the variance of encoding and transmission bitrate within a unit time window, etc.
[0064] Network transmission feature extraction: For the dedicated TCP transmission link corresponding to a single video stream, core timing parameters of the transmission layer are continuously collected to construct a time-varying feature sequence of network status, including TCP packet retransmission rate, end-to-end round-trip delay average, delay jitter amplitude, etc.
[0065] In the initial stage of operation, historical time-series data of each video stream and each network link under normal operating conditions are collected as normal samples. The normal sample collection needs to cover normal operation data under different lighting conditions (day / night / rain) and different scene busy levels (peak / idle time). The existing Long Short-Term Memory Autoencoder (LSTM-AE) technology is used as the core network for baseline modeling. The normal samples are trained to obtain a healthy baseline model, and the reconstructed feature vector Ek(t) under normal operating conditions is output, where k represents the dimension feature, k is a natural number greater than zero, and t represents the sampling time, t>0.
[0066] The real-time acquired three-dimensional time-varying feature sequence is continuously input into the corresponding dedicated long short-term memory autoencoder health baseline model, and the real-time reconstructed feature vector EXk(t) is output. All dimensional features are normalized to the same dimension range.
[0067] Calculate the real-time reconstruction error E(t). Where M is the total number of dimensional features, M > 0;
[0068] The real-time reconstruction error E(t) is compared with the preset reconstruction error threshold. If the real-time reconstruction error E(t) > the preset reconstruction error threshold, it is determined that there is a sub-health abnormality and the inversion positioning command is triggered.
[0069] If the real-time reconstruction error E(t) ≤ the preset reconstruction error threshold, it is determined that there is no sub-health abnormality and routine monitoring continues.
[0070] Example 2:
[0071] When generating the inversion localization command, the network tomography module performs cosine similarity matching between the real-time reconstructed feature vector and the preset dedicated fault feature knowledge base, and outputs the discrimination processing result to construct a health vector, specifically including:
[0072] Obtain the real-time reconstructed feature vector EXk(t) corresponding to sub-health abnormalities;
[0073] The real-time reconstructed feature vector EXk(t) is matched with the preset exclusive fault feature knowledge base. The cosine similarity YXg between the real-time reconstructed feature vector EXk of sub-health abnormality and each exclusive fault feature in the preset exclusive fault feature knowledge base is calculated by cosine similarity, where g represents the exclusive fault number and g is a natural number greater than zero.
[0074] The exclusive fault feature knowledge base is constructed by injecting calibrated faults (such as using attenuating filters with different transmittance to simulate lens contamination, fine-tuning the focal length to simulate out-of-focus, and using a network damage analyzer to simulate link degradation) to collect three-dimensional feature sequences and extracting their reconstructed feature vectors. The exclusive faults include lens contamination, camera out-of-focus, and link performance degradation.
[0075] The cosine similarity YXg is compared with the retrieved preset cosine similarity threshold for discrimination:
[0076] If the cosine similarity YXg is greater than the preset cosine similarity threshold, then the specific fault corresponding to the maximum value in the cosine similarity YXg is set as the inversion root cause fault.
[0077] If the cosine similarity YXg is greater than the preset cosine similarity threshold, and the maximum value of cosine similarity YXg is not unique, then it is determined to be an inversion concurrency fault (e.g., lens contamination + slight blurring, abnormal camera imaging + slight network link degradation).
[0078] If the cosine similarity YXg is less than or equal to the preset cosine similarity threshold, and the difference between the real-time reconstruction error E(t) and the preset reconstruction error threshold is less than the preset difference threshold, then it is judged as an atypical weak state, and no fault is marked or maintenance alarm is triggered. Only the time-series status is recorded for baseline incremental update.
[0079] If the cosine similarity YXg is less than or equal to the preset cosine similarity threshold, and the difference between the real-time reconstruction error E(t) and the preset reconstruction error threshold is greater than or equal to the preset difference threshold, then it is judged as an unknown root cause of inversion.
[0080] For both root cause and concurrent failures, the degradation quantification index L is calculated:
[0081] If the root cause of the fault is inverted, then the degradation quantification index = 1 - cosine similarity YXgmax;
[0082] If it is an inverted concurrent fault, then the degradation quantification index = the sum of the cosine similarity YXg corresponding to the inverted concurrent fault - the product value of the corresponding cosine similarity YXg;
[0083] The current effective performance percentage η is calculated based on the degradation quantification index L. The current effective performance percentage η = (1-L / preset failure critical degradation degree) × 100%, η∈[0%, 100%];
[0084] Obtain G (G>0) historical degradation quantification indicators L, and fit the degradation time series evolution slope a using the least squares method. When a>0, calculate the estimated remaining usable time Ty, Ty = (preset failure critical degradation degree - L) / a;
[0085] A health vector is constructed based on the discrimination processing results, degradation quantification index L, current effective performance percentage η, estimated remaining available time Ty, device / link identifier, and other information.
[0086] The Sub-health Linkage Module triggers a predictive linkage strategy deduction process when sub-health abnormalities exist and external alarm events are triggered, outputting the optimal predictive linkage strategy under sub-health abnormalities, specifically including:
[0087] When sub-health abnormalities exist and trigger external alarm events (such as perimeter intrusion detection signals, abnormal behavior recognition alarms, etc.), a predictive linkage strategy simulation process is triggered, which specifically includes:
[0088] Starting from external alarm events and using the target's current speed, direction of movement, and scene access constraints as boundary conditions, a data twin model is used to generate all feasible movement paths of the attacker / intrusion target within a future period of time (e.g., 1 minute). Each feasible movement path is independently bound to exclusive temporal spatial data, including continuous spatial coordinate sequences, corresponding timestamp sequences, attributes of the areas traversed by the path, coverage status of key security points, and path termination probability weights, forming a multi-path probability situational distribution map.
[0089] For each feasible movement path, the system combines a spatial partition index table, preset device-area mapping rules, and a linkage action template library. This template library defines the device types and action ranges that can be invoked for each alarm type, generating corresponding candidate linkage strategy combinations. Each candidate strategy contains a set of linkage action sequences, such as the preset position and tracking parameters of the specified camera, the opening range and brightness of the lighting device, the beam pointing and transmission parameters of the directional acoustic wave generator, the lock / unlock command of the access controller, the rise / lower command of the roadblock device, and the scheduling target position of the mobile node, etc.
[0090] For example: One feasible movement path, Action Sequence Number - Trigger Time - Action Template ID - Target Device ID - Device Location Area - Action Parameter Summary (Emergency lighting fully on, covering all areas from WP00 to WP04, Level 3, High Brightness);
[0091] Obtain devices and links with a clear estimated remaining failure time Ty, and establish a time-series quantitative verification mechanism for the entire lifecycle of the strategy: set the complete execution time Tz of a single candidate linkage strategy combination, retain candidate linkage strategy combinations in the entire candidate linkage strategy combination where Tz is all < Ty, and construct a candidate linkage strategy combination pool;
[0092] For a single candidate linkage strategy combination in the candidate linkage strategy combination pool, obtain the key node imaging quality S1, the target effective coverage time ratio S2, and the linkage response delay S3, and then normalize the values to [0, 100].
[0093] The comprehensive decision score is calculated based on S1×a1+S2×a2+S3×a3, where a1, a2 and a3 are preset weight factors and are all greater than zero.
[0094] The candidate linkage strategy combination with the highest comprehensive decision score is selected from the candidate linkage strategy combination pool as the optimal predictive linkage strategy under the condition of sub-health abnormality.
[0095] Among them, the imaging quality of key nodes S1: based on the health vector, the real-time detection confidence, image clarity and recognition accuracy are calculated;
[0096] Wherein, detection confidence = η × standard target detection confidence, image clarity = η × standard image clarity, and recognition accuracy = η × standard target recognition accuracy;
[0097] The imaging quality S1 of the key node is calculated based on the combination of detection confidence × w1 + image clarity × w2 + recognition accuracy × w3, where w1, w2 and w3 are preset weight coefficients and are all greater than zero.
[0098] The percentage of effective coverage time for the target is S2, where S2 = effective coverage time for the target / total effective coverage time for the target;
[0099] The linkage response delay S3 is calculated as follows: S3 = max(0, 1 - total policy delay / maximum tolerable linkage baseline delay preset by the policy). The total policy delay is the cumulative delay of the entire link from the triggering of the event to the execution of all linkage devices for a single candidate linkage policy combination, including instruction scheduling delay, network transmission delay, device response delay, etc.
[0100] Example 3:
[0101] The closed-loop feedback module continuously tracks the status of the front-end devices based on the optimal predictive linkage strategy output. When the output execution condition changes abruptly, it selects the suboptimal alternative strategy and issues it for execution, specifically including:
[0102] Based on the optimal predictive linkage strategy, the status of the front-end devices (such as cameras and transmission links) involved in this linkage task is continuously tracked, and the three-dimensional time-varying feature sequence is updated in real time.
[0103] Set mutation trigger conditions: If a sudden anomaly occurs, such as a complete link interruption or complete failure of device imaging, or if the change in parameters such as the real-time effective performance percentage of the link / device or the reconstruction error exceeds the preset fluctuation range, it is determined to be a mutation in the execution condition. When a mutation in the execution condition is output, the optimal predictive linkage strategy becomes invalid, and incremental simulation operation is initiated, as follows:
[0104] Based on the candidate linkage strategy combination pool, candidate linkage strategy combinations involving failed links / devices are removed from the pool. The remaining candidate linkage strategy combinations are sorted in descending order. The candidate linkage strategy combination with the highest ranking is selected as the second-best alternative strategy and immediately sent to the physical device to complete the dynamic strategy switching and continuation execution.
[0105] In summary, this invention, relying on three-dimensional time-varying feature sequences and combined with an LSTM-AE health baseline model, calculates real-time reconstruction errors. It can keenly identify "sub-healthy" states such as lens contamination, out-of-focus areas, and link degradation that have not yet triggered explicit alarms, significantly improving the lead time and accuracy of fault warnings. Furthermore, by performing cosine similarity matching between the real-time reconstructed feature vector and a pre-set dedicated fault feature knowledge base, it accurately distinguishes single faults and effectively identifies concurrent faults and unknown root causes. Simultaneously, it constructs a health vector, transforming ambiguous abnormal states into intuitive and quantifiable assessments of degradation levels and development trends. Moreover, when sub-health abnormalities and external alarm events occur concurrently, it utilizes a digital twin model... The system generates a multi-path probability situational distribution map of the target. Based on this, and combined with the device health vector, it introduces a time-series quantitative verification mechanism for the entire lifecycle of the strategy. This ensures that the execution time of the linkage strategy is strictly less than the estimated remaining time of device failure. The system also dynamically derives the optimal linkage strategy from the candidate pool through comprehensive consideration. This fundamentally avoids the risk of security task failure due to device or link failure during the linkage process. Furthermore, during the execution of the optimal linkage strategy, the system continuously tracks the status of the front-end devices. Once a sudden change in the execution condition is detected, the system can immediately select and activate a suboptimal alternative strategy that does not depend on the faulty link from the candidate strategy pool, thereby achieving dynamic strategy switching and seamless execution.
[0106] The threshold is set for comparative analysis of results to determine whether they are good or bad. The value of the threshold is determined by a combination of large-scale model analysis of sample data and human experience. It can also be adjusted appropriately based on seasonal or common-sense influencing factors.
[0107] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A regional security monitoring system based on computer networking, characterized in that, include: The data association module builds a 1:1 digital twin model based on the collected static environmental information of the entire target monitoring domain, binds exclusive coordinate labels and virtual device proxy components to all hardware devices, encapsulates performance parameters and establishes communication topology relationships, and finally constructs a snapshot of the current state of the entire domain through real-time data channels. The health assessment module extracts a three-dimensional time-varying feature sequence based on the collected network transmission data packets and application data packets, combines it with the LSTM-AE health baseline model, calculates the reconstruction error E(t), and makes a judgment based on the preset reconstruction error threshold, outputting an inversion positioning command or continuous routine monitoring. When the inversion positioning command is generated, the network tomography module performs cosine similarity matching between the real-time reconstructed feature vector and the preset exclusive fault feature knowledge base, and outputs the discrimination processing result to construct the health vector. When a sub-health abnormality exists and an external alarm event is triggered, the sub-health-linkage module triggers a predictive linkage strategy deduction process and outputs the optimal predictive linkage strategy under the sub-health abnormality. The closed-loop feedback module continuously tracks the status of the front-end device based on the output optimal predictive linkage strategy. When the output execution condition changes abruptly, it selects the suboptimal alternative strategy and sends it out for execution.
2. The regional security monitoring system based on computer networking according to claim 1, characterized in that, The target monitoring full-domain static environment information includes building walls, floor structure, and access roads. The performance parameters include maximum resolution, adjustable bit rate range, and upper limit of bandwidth for each communication port. The current status snapshot includes read-only data of the spatial environment, real-time operating status of all devices, network link parameters, alarm queues, and target object coordinate trajectory.
3. The regional security monitoring system based on computer networking according to claim 1, characterized in that, The three-dimensional time-varying feature sequence includes a shallow image feature sequence, a bitrate fluctuation feature sequence, and a network transmission feature sequence. The shallow features of the image include the proportion of high-frequency coefficient components and the overall diffusion measure of motion vectors; the bit rate fluctuation features include the average bit rate of encoding and transmission within a unit time window and the variance of the bit rate of encoding and transmission within a unit time window; and the network transmission features include the TCP packet retransmission rate, the average end-to-end round-trip delay, and the latency jitter amplitude.
4. The regional security monitoring system based on computer networking according to claim 3, characterized in that, The analysis process of the LSTM-AE healthy baseline model is as follows: Historical time-series data under normal operating conditions of each video stream and each network link are collected as normal samples. LSTM-AE is used as the core network for baseline modeling. The normal samples are trained to obtain a healthy baseline model and output the reconstructed feature vector Ek(t) under normal operating conditions, where k represents the dimension feature, k is a natural number greater than zero, and t represents the sampling time, t>0. The real-time collected three-dimensional time-varying feature sequence is continuously input into the corresponding exclusive health baseline model, and the real-time reconstructed feature vector EXk(t) is output. Calculate the real-time reconstruction error E(t), retrieve the preset reconstruction error threshold for discrimination, and output the sub-health abnormality inversion positioning command or continuous routine monitoring.
5. The regional security monitoring system based on computer networking according to claim 4, characterized in that, The analysis process of the health vector is as follows: The cosine similarity YXg between the real-time reconstructed feature vector EXk of sub-health abnormalities and each exclusive fault feature in the preset exclusive fault feature knowledge base is calculated using cosine similarity. Here, g represents the exclusive fault number and g is a natural number greater than zero. The cosine similarity YXg is discriminated against the preset cosine similarity threshold and outputs the inverted root cause fault / inverted concurrent fault / atypical weak state / inverted unknown root cause. For the inverted root cause fault and the inverted concurrent fault, the degradation quantification index L is calculated. Based on the degradation quantification index L, the current effective performance percentage η is calculated, and the estimated remaining available time Ty is also calculated. A health vector is constructed based on the discrimination processing results, the degradation quantification index L, the current effective performance percentage η, and the estimated remaining available time Ty.
6. The regional security monitoring system based on computer networking according to claim 5, characterized in that, The calculation process of the degradation quantification index L is as follows: If it is an inverted root cause fault, then the degradation quantification index = 1 - cosine similarity YXgmax; if it is an inverted concurrent fault, then the degradation quantification index = the sum of cosine similarity YXg corresponding to the inverted concurrent fault - the product value of the corresponding cosine similarity YXg. The current effective performance percentage η = (1 - L / preset critical degradation degree) × 100%, η ∈ [0%, 100%]. Obtain the historical G degradation quantification index L, and obtain the degradation time series evolution slope a by fitting the least squares method. When a > 0, Ty = (preset critical degradation degree - L) / a.
7. The regional security monitoring system based on computer networking according to claim 1, characterized in that, The analysis process of the predictive linkage strategy deduction procedure is as follows: Starting from external alarm events and using the target's current speed, direction of movement, and scene access constraints as boundary conditions, a data twin model is used to generate all feasible movement paths of the attacker / intrusion target within a future period. Each feasible movement path is independently bound to exclusive temporal spatial data. For each feasible movement path, a corresponding candidate linkage strategy combination is generated by combining the preset device-area mapping rules and the linkage action template library. Each candidate strategy contains a set of linkage action sequences. Obtain devices and links with a clear estimated remaining failure time Ty, and establish a time-series quantitative verification mechanism for the entire lifecycle of the strategy: set the complete execution time Tz of a single candidate linkage strategy combination, retain candidate linkage strategy combinations in the entire candidate linkage strategy combination where Tz is all < Ty, and construct a candidate linkage strategy combination pool; For a single candidate linkage strategy combination in the candidate linkage strategy combination pool, the comprehensive decision score is calculated based on S1×a1+S2×a2+S3×a3, where a1, a2 and a3 are preset weight factors and are all greater than zero. The candidate linkage strategy combination with the highest comprehensive decision score is selected from the pool of candidate linkage strategy combinations as the optimal predictive linkage strategy under the condition of sub-health abnormalities.
8. The regional security monitoring system based on computer networking according to claim 7, characterized in that, S1 represents the imaging quality of key nodes, S2 represents the percentage of effective target coverage time, and S3 represents the linkage response delay. S1 = Detection confidence × w1 + Image clarity × w2 + Recognition accuracy × w3, where w1, w2, and w3 are all preset weight coefficients and are all greater than zero; S2 = Target effective coverage time / Target effective coverage total time; S3 = max(0, 1 - Total policy delay / Policy preset maximum tolerable linkage baseline delay).
9. The regional security monitoring system based on computer networking according to claim 1, characterized in that, The process of sudden changes in the output execution condition is as follows: Set the sudden change trigger condition: If there is an instantaneous abnormality such as a complete interruption of the link or a complete failure of the device imaging, or if the change in parameters such as the real-time effective performance percentage of the link / device or the reconstruction error exceeds the preset floating range, it is determined to be a sudden change in the execution condition. Based on the candidate linkage strategy combination pool, candidate linkage strategy combinations involving failed links / devices are removed from the pool. The remaining candidate linkage strategy combinations are sorted in descending order, and the candidate linkage strategy combination ranked first is selected as the second-best alternative strategy.
