A security monitoring method and system based on big data
By employing big data security monitoring methods and utilizing edge data preprocessing and cloud-based collaborative analysis, a dynamic reliability model is constructed, enabling cross-device feature correlation analysis. This reduces false alarm rates and maintenance costs, adapts to various security scenarios, and optimizes system resource utilization.
Patent Information
- Application Number
- CN202611223054.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-13
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-08-13
AI Technical Summary
Traditional single-point security systems have a high false alarm rate under complex operating conditions, cannot achieve cross-device feature correlation analysis, have high labor costs, cannot be optimized independently, have idle computing power on edge devices and huge computing power load on the cloud, and cannot meet the needs of refined security in large-scale scenarios.
A big data-based security monitoring method is adopted. Through edge data preprocessing, multi-dimensional time-series alignment and data normalization, a three-dimensional dynamic adaptive credibility calculation model is constructed, a credibility adjustment instruction list is generated, credibility is corrected step by step, and linkage is carried out in conjunction with a four-dimensional intelligent security strategy library, and iterative optimization is performed.
It significantly reduces the false alarm rate of security systems, enables refined self-learning in multiple scenarios, optimizes bandwidth and latency, reduces operation and maintenance costs, adapts to various security scenarios, and avoids resource waste.
Smart Images

Figure CN122734422B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent security technology, and in particular relates to a security monitoring method and system based on big data. Background Technology
[0002] Traditional single-point security systems rely on a single camera or independent sensor for monitoring, with data from each device isolated from each other, forming typical information silos. They can only trigger alarms based on fixed thresholds and cannot achieve cross-device feature correlation analysis. False alarm rates remain high under complex operating conditions such as sudden changes in lighting, rain, snow, environmental noise, and equipment aging. At the same time, the system can only push alarms and lacks the ability for hierarchical handling, automatic linkage, and post-event self-optimization. It relies entirely on manual review and handling, resulting in high labor costs and delayed risk response, making it difficult to meet the needs of refined security management in large-scale scenarios.
[0003] While existing patents have alleviated the problem of false alarms in single-source monitoring to some extent, they still suffer from issues such as idle computing power on edge devices and a huge computing load on the cloud; the weighting factors cannot adapt to dynamic environmental scenarios such as equipment aging, sudden changes in lighting, rain, snow, and seasonal noise, resulting in large quantification errors in reliable metrics and a tendency to generate false alarms; and the overall process is an open-loop, one-way process: only disposal instructions are issued, without on-site equipment execution feedback, secondary risk verification, or a model self-updating mechanism, so the system cannot autonomously optimize with real security events, and the long-term operation cannot continuously reduce false alarms and missed alarms. Summary of the Invention
[0004] The purpose of this invention is to provide a security monitoring method and system based on big data, which significantly reduces the false alarm and missed alarm rates of security systems and realizes intelligent security closed-loop management with refined multi-scenario and sustainable self-learning capabilities.
[0005] To achieve the above objectives, the present invention provides a security monitoring method based on big data, comprising the following steps: S1. Receive edge data and preprocess it; S2. Perform multi-dimensional temporal alignment and data normalization on the processed edge data; S3. Based on the results of S2, construct a three-dimensional dynamic adaptive confidence calculation model and calculate the independent confidence and confidence interval of confidence fluctuation. S4. Construct a credibility adjustment instruction list using the data obtained from S2 and S3; S5. Adjust the independent credibility level by level based on the credibility adjustment instruction list to obtain the final comprehensive credibility. S6. Judge the final comprehensive credibility obtained in S5 to obtain the refined hierarchical verification result; S7. Based on the refined hierarchical verification results, retrieve the pre-built four-dimensional intelligent security strategy library of the system to generate linkage response instructions; S8. Complete the monitoring process and perform iterative optimization.
[0006] Preferably, the process of S1 is as follows: S11. Each edge node receives real-time data streams from multiple heterogeneous sensors in parallel and performs three-layer structured feature extraction of features from the base layer, anomaly layer, and environmental interference layer. S12. Calculate the comprehensive noise correction coefficient based on the characteristics of the environmental interference layer, and perform graded attenuation compensation on the original instantaneous signal-to-noise ratio using the comprehensive noise correction coefficient, outputting a high-precision corrected instantaneous signal-to-noise ratio. ; S13, Configure low and high thresholds, through... The process of completing the initial credibility prediction for edge local data is as follows: like If the data is below the low threshold, it is determined that the data is pure environmental noise and equipment background noise, and is discarded directly. like If the data is between the low and high thresholds, it is labeled as weakly anomalous data, marked with suspicious interference labels, and stored in a lightweight manner with feature parameters uploaded. like If the value exceeds the high threshold, the data is considered valid. The features at this point are marked as valid high-value abnormal features, and the data is uploaded in its entirety.
[0007] Preferably, the process of S2 is as follows: S21. Receive the multi-source heterogeneous data uploaded by S1, including three-layer feature data, corrected instantaneous signal-to-noise ratio, suspected interference tags, and the status of multiple types of heterogeneous sensors; S22. Filter the uploaded data for legality: remove data with lost packets, abnormal data exceeding the range, data with disordered timing, and blank or invalid data to ensure that the data entering the analysis link is complete and valid; S23. Using the unified system timestamp as a benchmark, perform time-series alignment and padding for all nodes and all types of feature data to ensure that all feature data at the same judgment time belong to the same time window and eliminate judgment deviations caused by time misalignment. S24. Perform a range normalization operation on the multimodal features that have completed time-series alignment and padding, and map different physical quantities to the standard range of 0~1. S25. Retrieve environmental parameters and equipment deployment attributes of the current monitoring area, automatically establish a scene classification tag library, and accurately classify three types of scenes: indoor closed scenes, outdoor open scenes, and equipment operation and maintenance scenes.
[0008] Preferably, the process of S3 is as follows: S31. Read the data obtained in S2 to acquire three types of core operating condition parameters, including time parameters. Equipment parameters and scene parameters ; S32. Dynamically generate weighting factors based on the three types of core parameters in S31. Alarm accuracy rate compared to historical data ; S33. Read the system constant maximum signal-to-noise ratio and normalize it to obtain... ; S34. Calculate the independent confidence level of each feature index. The process is as follows: ; S35. Perform standard deviation statistics on the feature confidence values sampled multiple times within the current time window to generate confidence intervals for confidence fluctuations.
[0009] Preferably, the process of S4 is as follows: S41. Retrieve the pre-built multimodal collaborative rule library and prioritize the selection of valid rules that are suitable for the current scenario based on the scenario adaptation rules: In enclosed indoor scenarios: Automatically tighten anomaly detection thresholds and weaken rules that mitigate large-scale environmental interference; Outdoor open scenarios: Relax the tolerance for noise and light interference, and strengthen the multi-feature linkage rules; Equipment maintenance scenarios: Prioritize matching rules derived from equipment vibration and electrical anomalies; S42, Synchronous parallel matching supports five original rule categories: rule of contradiction, rule of interference cancellation, rule of scenario adaptation and rule of derivative, to perform multi-dimensional intelligent correction of credibility; S43. Set arbitration rules for rule conflict priority, with the following priorities: Derivative rules > Scene adaptation rules > Interference cancellation rules > Supporting rules / Contradictory rules; S44. Sort, deduplicate, and standardize all effective rule modification instructions, and output a list of credibility adjustment instructions. Each instruction includes: target feature number, adjustment type, adjustment range, confidence interval constraint identifier, and effective priority.
[0010] Preferably, the process of S5 is as follows: The instructions obtained in S51 and S44 are rearranged according to the conflict priority arbitration rules set in S43; S52. Intelligent arbitration for conflict scenarios where the same feature indicator is acted upon by multiple instructions simultaneously: Higher-priority instructions override conflicting lower-priority instructions. When a reverse adjustment occurs in a peer instruction, the risk attribute of the current scenario is used to determine whether to adopt the instruction with increased trustworthiness for suspected high-risk scenarios and the instruction with decreased trustworthiness for scenarios with environmental interference. Redundant instructions that are duplicated, invalid, or have reverse cancellation are removed to generate a unique and ordered sequence of final execution instructions; S53. Under the constraint of the confidence interval for confidence fluctuation obtained in S3, execute the sorted confidence adjustment instructions one by one: If the adjusted confidence level exceeds the upper limit of the confidence interval, it will be automatically truncated and locked to the maximum value of the interval. If the adjusted confidence level is lower than the lower limit of the confidence interval, it will be automatically clamped and locked to the minimum value of the interval. S54. After completing all instruction calibrations, output the final comprehensive confidence level of each feature with scene label, confidence interval marker, and conflict arbitration and boundary calibration.
[0011] Preferably, the process of S6 is as follows: S61. Receive the final comprehensive credibility of each feature obtained in S5, and call the adaptive credibility threshold according to the current scenario; S62. Read the current risk type and scenario attributes, and dynamically and adaptively adjust the feature overlap matching threshold: If intrusion, fire, and equipment damage are classified as high-risk, the matching threshold is lowered to 60%, while the ability to detect rare risks is improved. If personnel lingering and minor noise are classified as low-risk, the matching threshold is raised to 80%, while suppressing false alarms due to minor fluctuations. Meanwhile, the overall threshold for indoor scenes has been increased by 10%, while the overall threshold for complex outdoor scenes has been decreased by 10%. S63. Introducing a three-dimensional weighting factor—feature overlap weight, event importance weight, and time-series decay weight—to obtain the precise weighted causal association strength, the process is as follows: S631. The percentage of samples in a statistical historical period that are marked as real security incidents; S632. Multiply the result of S631 by the feature overlap coefficient between the current feature combination and the historical segment; S633, then multiply by the event importance weight corresponding to the event severity level; S634. Finally, multiply by the temporal decay weight based on time distance to obtain the precise weighted causal association strength; S64. By accurately weighting the causal relationship strength, the confidence interval of the credibility fluctuation, and the inherent risk level of the current scenario, determine which category the current standardized verification result belongs to among the five levels: extremely high risk, high risk, medium risk, low risk, and no risk, and also include the risk credibility fluctuation level.
[0012] Preferably, the process of S7 is as follows: S71. Collect risk level, security level of incident area, scene type label and risk credibility fluctuation level; S72. Based on the pre-built four-dimensional intelligent security strategy library, perform precise retrieval and matching: Based on the risk level, a basic emergency response plan will be implemented. The severity of measures will be adjusted according to the regional security level, with the intensity of measures automatically upgraded in core areas. The device linkage logic is adapted according to the scenario type: indoors, it focuses on access control and sound and light alarms; outdoor, it focuses on personnel dispatching and trajectory tracking; and device scenarios focus on equipment inspection and shutdown protection. The level of conservatism in handling cases is adjusted based on the level of credibility fluctuations; if the fluctuations are large, a review mechanism is added to avoid mishandling. S73. Conduct risk assessment and event classification, as follows: S731. Accurately determine the specific type of risk event; S732. Further subdivide the risk levels and match them with corresponding security and emergency response standards; S733. Generate the handling priority for this event. When multiple events occur concurrently, prioritize handling high-risk events. S74. Generate a multi-level differentiated linkage instruction set based on the results of S73, and output it according to the execution object. Platform alert levels: tiered pop-up alerts in the background, automatic archiving of event logs, and locking and retaining abnormal videos; Mobile dispatch hierarchy: Security personnel's APP accurately pushes location, event type, and handling tasks; On-site equipment linkage hierarchy: hierarchical sound and light alarm, one-way / two-way locking of area access control, and pan-tilt-zoom (PTZ) monitoring for focusing and tracking; Specialized emergency response levels: High-risk scenarios trigger fire standby, area isolation, and equipment power-off protection; Timing control: All linkage commands are triggered sequentially according to a preset timing sequence to avoid system congestion caused by simultaneous activation of devices; S75. All linkage instructions, scheduling plans, handling priorities, and review requirements are uniformly encapsulated into structured handling instruction messages, carrying the unique event number, scene tag, risk level, and execution time limit, and sent to edge nodes and security terminal devices.
[0013] Preferably, the process of S8 is as follows: S81: Real-time acquisition of edge device execution status, sensor secondary detection data, security personnel handling feedback, device linkage response latency and on-site handling screen data, generating multi-dimensional execution feedback data streams and transmitting them back to the cloud in real time; In response to abnormal situations, the system automatically generates fault alarms and triggers backup response plans. S82. Combining the secondary perception data from the execution feedback and the on-site handling results, recalculate the comprehensive credibility and causal correlation strength of each feature in the current scenario, dynamically adjust the original risk level, correct the risk judgment deviation, and output the final accurate risk assessment result. S83. Standardize and archive all data of this complete event and store it in the historical feature fragment database; S84. Initiate multi-dimensional autonomous iterative updates, automatically triggering incremental updates of the rule base, fine-tuning of security strategy base parameters, adaptive optimization of scene thresholds, and iterative calibration of dynamic weight parameters; S85. If the same feature combination in the same scenario is misjudged / missed 3 times in a row, the cause of the misjudgment will be automatically analyzed, and corresponding interference cancellation rules and scenario adaptation rules will be generated in a targeted manner. The dynamic weight parameters and matching thresholds in the scenario will be optimized simultaneously to avoid similar misjudgment problems from the bottom layer of the algorithm. S86. Migrate the optimized scene rules, weight parameters, and judgment thresholds to similar scenes to achieve rapid iteration of the entire system.
[0014] The present invention also provides a security monitoring system based on big data, which adopts a security monitoring method based on big data, including an edge perception preprocessing module for edge data acquisition and a cloud-based collaborative analysis and judgment module for security monitoring; The edge perception preprocessing module includes a multi-sensor parallel acquisition unit, a hierarchical feature extraction unit, a dynamic noise correction unit, an edge local confidence initial screening and filtering unit, and an edge execution feedback acquisition unit. The cloud-based collaborative analysis and resolution module includes a data temporal alignment and normalization unit, a dynamic adaptive independent credibility calculation unit, a spatiotemporal correlation multimodal collaborative reasoning unit, an interval constraint credibility adjustment unit, a fuzzy matching saliency verification unit, a multi-level strategy mapping instruction generation unit, a closed-loop feedback correction and iteration unit, and a scenario adaptive management unit.
[0015] Therefore, the security monitoring method and system based on big data described above, as used in this invention, has the following advantages: 1. Bandwidth and latency optimization, edge local filtering of invalid interference data, reduction of cloud data transmission volume, and shortening of overall security event response latency, suitable for large-scale sensor deployment in parks, factories, and large communities; 2. The system can be autonomously and sustainably optimized. The closed-loop feedback and iteration mechanism eliminates the need for regular manual maintenance of rules and policy libraries, significantly reducing the manpower costs of security operation and maintenance. 3. Through a three-dimensional mapping strategy library of five-level risk classification + scenario-risk-verification results, security, access control, sound and light and fire protection equipment can be scheduled on demand to avoid ineffective waste of resources; 4. The security monitoring system is compatible with various security scenarios such as indoor buildings, outdoor parks, underground utility tunnels, and construction sites. Fuzzy matching thresholds and dynamic weight parameters can be adaptively configured.
[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0017] Figure 1 This is an overall flowchart of a security monitoring method and system based on big data according to the present invention; Figure 2 This is a process diagram illustrating the iterative optimization of a security monitoring method and system based on big data according to the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Specific model specifications need to be selected and determined according to the actual specifications of the device, etc. The specific selection calculation method adopts existing technology in the art, and therefore will not be described in detail.
[0019] Example 1; like Figure 1 and Figure 2 As shown, this invention provides a security monitoring method based on big data, comprising the following steps: S1. Receive edge data and preprocess it; S11. Each edge node receives real-time data streams from multiple heterogeneous sensors in parallel and performs three-layer structured feature extraction of features from the base layer, anomaly layer, and environmental interference layer. S12. Calculate the comprehensive noise correction coefficient based on the characteristics of the environmental interference layer, and perform graded attenuation compensation on the original instantaneous signal-to-noise ratio using the comprehensive noise correction coefficient, outputting a high-precision corrected instantaneous signal-to-noise ratio. ; S13, Configure low and high thresholds, through... The process of completing the initial credibility prediction for edge local data is as follows: like If the data is below the low threshold, it is determined that the data is pure environmental noise and equipment background noise, and is discarded directly. like If the data is between the low and high thresholds, it is labeled as weakly anomalous data, marked with suspicious interference labels, and stored in a lightweight manner with feature parameters uploaded. like If the value exceeds the high threshold, the data is considered valid. The features at this point are marked as valid high-value abnormal features, and the data is uploaded in its entirety.
[0020] S2. Perform multi-dimensional temporal alignment and data normalization on the processed edge data; S21. Receive the multi-source heterogeneous data uploaded by S1, including three-layer feature data, corrected instantaneous signal-to-noise ratio, suspected interference tags, and the status of multiple types of heterogeneous sensors; S22. Filter the uploaded data for legality: remove data with lost packets, abnormal data exceeding the range, data with disordered timing, and blank or invalid data to ensure that the data entering the analysis link is complete and valid; S23. Using the unified system timestamp as a benchmark, perform time-series alignment and padding for all nodes and all types of feature data to ensure that all feature data at the same judgment time belong to the same time window and eliminate judgment deviations caused by time misalignment. S24. Perform a range normalization operation on the multimodal features that have completed time-series alignment and padding, and map different physical quantities to the standard range of 0~1. S25. Retrieve environmental parameters and equipment deployment attributes of the current monitoring area, automatically establish a scene classification tag library, and accurately classify three types of scenes: indoor closed scenes, outdoor open scenes, and equipment operation and maintenance scenes.
[0021] S3. Based on the results of S2, construct a three-dimensional dynamic adaptive confidence calculation model and calculate the independent confidence and confidence interval of confidence fluctuation. S31. Read the data obtained in S2 to acquire three types of core operating condition parameters, including time parameters. Equipment parameters and scene parameters ; S32. Dynamically generate weighting factors based on the three types of core parameters in S31. Alarm accuracy rate compared to historical data ; S33. Read the system constant maximum signal-to-noise ratio and normalize it to obtain... ; S34. Calculate the independent confidence level of each feature index. The process is as follows: ; S35. Perform standard deviation statistics on the feature confidence values sampled multiple times within the current time window to generate confidence intervals for confidence fluctuations.
[0022] S4. Construct a credibility adjustment instruction list using the data obtained from S2 and S3; S41. Retrieve the pre-built multimodal collaborative rule library and prioritize the selection of valid rules that are suitable for the current scenario based on the scenario adaptation rules: In enclosed indoor scenarios: Automatically tighten anomaly detection thresholds and weaken rules that mitigate large-scale environmental interference; Outdoor open scenarios: Relax the tolerance for noise and light interference, and strengthen the multi-feature linkage rules; Equipment maintenance scenarios: Prioritize matching rules derived from equipment vibration and electrical anomalies; S42, Synchronous parallel matching supports five original rule categories: rule of contradiction, rule of interference cancellation, rule of scenario adaptation and rule of derivative, to perform multi-dimensional intelligent correction of credibility; S43. Set arbitration rules for rule conflict priority, with the following priorities: Derivative rules > Scene adaptation rules > Interference cancellation rules > Supporting rules / Contradictory rules; S44. Sort, deduplicate, and standardize all effective rule modification instructions, and output a list of credibility adjustment instructions. Each instruction includes: target feature number, adjustment type, adjustment range, confidence interval constraint identifier, and effective priority.
[0023] S5. Adjust the independent credibility level by level based on the credibility adjustment instruction list to obtain the final comprehensive credibility. The instructions obtained in S51 and S44 are rearranged according to the conflict priority arbitration rules set in S43; S52. Intelligent arbitration for conflict scenarios where the same feature indicator is acted upon by multiple instructions simultaneously: Higher-priority instructions override conflicting lower-priority instructions. When a reverse adjustment occurs in a peer instruction, the risk attribute of the current scenario is used to determine whether to adopt the instruction with increased trustworthiness for suspected high-risk scenarios and the instruction with decreased trustworthiness for scenarios with environmental interference. Redundant instructions that are duplicated, invalid, or have reverse cancellation are removed to generate a unique and ordered sequence of final execution instructions; S53. Under the constraint of the confidence interval for confidence fluctuation obtained in S3, execute the sorted confidence adjustment instructions one by one: If the adjusted confidence level exceeds the upper limit of the confidence interval, it will be automatically truncated and locked to the maximum value of the interval. If the adjusted confidence level is lower than the lower limit of the confidence interval, it will be automatically clamped and locked to the minimum value of the interval. S54. After completing all instruction calibrations, output the final comprehensive confidence level of each feature with scene label, confidence interval marker, and conflict arbitration and boundary calibration.
[0024] S6. Judge the final comprehensive credibility obtained in S5 to obtain the refined hierarchical verification result; S61. Receive the final comprehensive credibility of each feature obtained in S5, and call the adaptive credibility threshold according to the current scenario; S62. Read the current risk type and scenario attributes, and dynamically and adaptively adjust the feature overlap matching threshold: If intrusion, fire, and equipment damage are classified as high-risk, the matching threshold is lowered to 60%, while the ability to detect rare risks is improved. If personnel lingering and minor noise are classified as low-risk, the matching threshold is raised to 80%, while suppressing false alarms due to minor fluctuations. Meanwhile, the overall threshold for indoor scenes has been increased by 10%, while the overall threshold for complex outdoor scenes has been decreased by 10%. S63. Introducing a three-dimensional weighting factor—feature overlap weight, event importance weight, and time-series decay weight—to obtain the precise weighted causal association strength, the process is as follows: S631. The percentage of samples in a statistical historical period that are marked as real security incidents; S632. Multiply the result of S631 by the feature overlap coefficient between the current feature combination and the historical segment; S633, then multiply by the event importance weight corresponding to the event severity level; S634. Finally, multiply by the temporal decay weight based on time distance to obtain the precise weighted causal association strength; S64. By accurately weighting the causal relationship strength, the confidence interval of the credibility fluctuation, and the inherent risk level of the current scenario, determine which category the current standardized verification result belongs to among the five levels: extremely high risk, high risk, medium risk, low risk, and no risk, and also include the risk credibility fluctuation level.
[0025] S7. Based on the refined hierarchical verification results, retrieve the pre-built four-dimensional intelligent security strategy library of the system to generate linkage response instructions; S71. Collect risk level, security level of incident area, scene type label and risk credibility fluctuation level; S72. Based on the pre-built four-dimensional intelligent security strategy library, perform precise retrieval and matching: Based on the risk level, a basic emergency response plan will be implemented. The severity of measures will be adjusted according to the regional security level, with the intensity of measures automatically upgraded in core areas. The device linkage logic is adapted according to the scenario type: indoors, it focuses on access control and sound and light alarms; outdoor, it focuses on personnel dispatching and trajectory tracking; and device scenarios focus on equipment inspection and shutdown protection. The level of conservatism in handling cases is adjusted based on the level of credibility fluctuations; if the fluctuations are large, a review mechanism is added to avoid mishandling. S73. Conduct risk assessment and event classification, as follows: S731. Accurately determine the specific type of risk event; S732. Further subdivide the risk levels and match them with corresponding security and emergency response standards; S733. Generate the handling priority for this event. When multiple events occur concurrently, prioritize handling high-risk events. S74. Generate a multi-level differentiated linkage instruction set based on the results of S73, and output it according to the execution object. Platform alert levels: tiered pop-up alerts in the background, automatic archiving of event logs, and locking and retaining abnormal videos; Mobile dispatch hierarchy: Security personnel's APP accurately pushes location, event type, and handling tasks; On-site equipment linkage hierarchy: hierarchical sound and light alarm, one-way / two-way locking of area access control, and pan-tilt-zoom (PTZ) monitoring for focusing and tracking; Specialized emergency response levels: High-risk scenarios trigger fire standby, area isolation, and equipment power-off protection; Timing control: All linkage commands are triggered sequentially according to a preset timing sequence to avoid system congestion caused by simultaneous activation of devices; S75. All linkage instructions, scheduling plans, handling priorities, and review requirements are uniformly encapsulated into structured handling instruction messages, carrying the unique event number, scene tag, risk level, and execution time limit, and sent to edge nodes and security terminal devices.
[0026] S8. Complete the monitoring process and perform iterative optimization.
[0027] S81: Real-time acquisition of edge device execution status, sensor secondary detection data, security personnel handling feedback, device linkage response latency and on-site handling screen data, generating multi-dimensional execution feedback data streams and transmitting them back to the cloud in real time; In response to abnormal situations, the system automatically generates fault alarms and triggers backup response plans. S82. Combining the secondary perception data from the execution feedback and the on-site handling results, recalculate the comprehensive credibility and causal correlation strength of each feature in the current scenario, dynamically adjust the original risk level, correct the risk judgment deviation, and output the final accurate risk assessment result. S83. Standardize and archive all data of this complete event and store it in the historical feature fragment database; S84. Initiate multi-dimensional autonomous iterative updates, automatically triggering incremental updates of the rule base, fine-tuning of security strategy base parameters, adaptive optimization of scene thresholds, and iterative calibration of dynamic weight parameters; S85. If the same feature combination in the same scenario is misjudged / missed 3 times in a row, the cause of the misjudgment will be automatically analyzed, and corresponding interference cancellation rules and scenario adaptation rules will be generated in a targeted manner. The dynamic weight parameters and matching thresholds in the scenario will be optimized simultaneously to avoid similar misjudgment problems from the bottom layer of the algorithm. S86. Migrate the optimized scene rules, weight parameters, and judgment thresholds to similar scenes to achieve rapid iteration of the entire system.
[0028] The present invention also provides a security monitoring system based on big data, which adopts a security monitoring method based on big data, including an edge perception preprocessing module for edge data acquisition and a cloud-based collaborative analysis and judgment module for security monitoring; The edge perception preprocessing module includes a multi-sensor parallel acquisition unit, a hierarchical feature extraction unit, a dynamic noise correction unit, an edge local confidence initial screening and filtering unit, and an edge execution feedback acquisition unit. The cloud-based collaborative analysis and resolution module includes a data temporal alignment and normalization unit, a dynamic adaptive independent credibility calculation unit, a spatiotemporal correlation multimodal collaborative reasoning unit, an interval constraint credibility adjustment unit, a fuzzy matching saliency verification unit, a multi-level strategy mapping instruction generation unit, a closed-loop feedback correction and iteration unit, and a scenario adaptive management unit.
[0029] The scene adaptive management unit is used to identify the monitoring scene type in real time, adapt to various algorithm thresholds, and migrate and optimize model parameters to achieve differentiated intelligent judgment of multiple scenes. The closed-loop feedback correction and iteration unit establishes a two-way data interaction channel with the fuzzy matching saliency verification unit, the spatiotemporal correlation multimodal collaborative reasoning unit, and the scene adaptive management unit, respectively, to dynamically correct risk judgment, incrementally update the rule base and strategy base, and iteratively optimize scene adaptation parameters.
[0030] Example 2; This method was applied to a security scenario in a certain park, and the process is as follows: Edge preprocessing; Each camera, sound sensor, ground vibration sensor, and infrared smoke detector gateway in the park serves as an edge node, extracting three layers of features locally. During periods of strong sunlight in summer, when light intensity fluctuates significantly, the noise correction coefficient is automatically calculated to attenuate the original SNR. Noise data with a confidence level below 0.3 is filtered at the edge, and only high-value features such as crowd gathering, abnormal noise, and sudden vibration changes are uploaded to the cloud.
[0031] Cloud-based trustworthiness calculation; For devices that have been in use for 3 years (aging state e=3) and are exposed to strong sunlight at noon (t), the dynamic weight α(t,e) is reduced from the default 0.5 to 0.2 to decrease the real-time SNR weight and improve the accuracy of historical alarms. Weights, and synchronously output the fluctuation range of the confidence level within 0.05.
[0032] Spacetime rule reasoning; A sudden increase in population density and continuous abnormal noises occurring simultaneously in the same area triggers a support rule, increasing the credibility of both features; severe vibrations without any population features trigger a contradictory rule, reducing the credibility of the vibrations; excessive fluctuations in strong midday light trigger an interference cancellation rule, weakening the credibility of abnormal population density.
[0033] Fuzzy matching significance verification; The current feature combination is "crowd gathering + continuous abnormal noise". There are no completely identical segments in the historical database, but there are similar cluster alarm segments with an overlap of 75%. The calculated proportion of real events is 82%, and the causal relationship strength is 0.615, which exceeds the significance threshold of 0.5, and is judged as high risk.
[0034] Tiered handling; Matching the 3D strategy library, for high-risk outdoor intrusion scenarios in the park, the following instructions are generated: platform red pop-up alarm, on-site audio and visual alarm, dispatching 2 security personnel to the incident area, and locking the access control of surrounding entrances and exits.
[0035] Closed-loop iteration; Security personnel confirm illegal gatherings and intrusions on-site and send the feedback data back to the cloud; the complete event is stored in the historical database, and the rule database is automatically updated incrementally to improve the benchmark value of the causal correlation strength of similar feature combinations.
[0036] Therefore, the present invention employs a security monitoring method and system based on big data, which has the following advantages: 1. Bandwidth and latency optimization, edge local filtering of invalid interference data, reduction of cloud data transmission volume, and shortening of overall security event response latency, suitable for large-scale sensor deployment in parks, factories, and large communities; 2. The system can be autonomously and sustainably optimized. The closed-loop feedback and iteration mechanism eliminates the need for regular manual maintenance of rules and policy libraries, significantly reducing the manpower costs of security operation and maintenance. 3. Through a three-dimensional mapping strategy library of five-level risk classification + scenario-risk-verification results, security, access control, sound and light and fire protection equipment can be scheduled on demand to avoid ineffective waste of resources; 4. The security monitoring system is compatible with various security scenarios such as indoor buildings, outdoor parks, underground utility tunnels, and construction sites. Fuzzy matching thresholds and dynamic weight parameters can be adaptively configured.
[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A big data based security monitoring method, characterized in that, Includes the following steps: S1. Receive edge data and preprocess it, as follows: S11. Each edge node receives real-time data streams from multiple heterogeneous sensors in parallel and performs three-layer structured feature extraction of features from the base layer, anomaly layer, and environmental interference layer. S12, based on the feature of the environmental interference layer, a comprehensive noise correction coefficient is calculated, and the original instantaneous signal-to-noise ratio is compensated by graded attenuation through the comprehensive noise correction coefficient, and a high-precision corrected instantaneous signal-to-noise ratio is output ; S13, Configure low and high thresholds, through... The process of completing the initial credibility prediction for edge local data is as follows: like If the data is below the low threshold, it is determined that the data is pure environmental noise and equipment background noise, and is discarded directly. like If the data is between the low and high thresholds, it is labeled as weakly anomalous data, marked with suspicious interference labels, and stored in a lightweight manner with feature parameters uploaded. like If the value is greater than the high threshold, the data is considered valid data, and the features at this time are marked as valid high-value abnormal features. The data should be uploaded in its entirety. S2. Perform multi-dimensional temporal alignment and data normalization on the processed edge data; S3. Based on the results of S2, construct a three-dimensional dynamic adaptive confidence calculation model, and calculate the independent confidence and confidence interval of confidence fluctuation. The process is as follows: S31. Read the data obtained in S2 to acquire three types of core operating condition parameters, including time parameters. Equipment parameters and scene parameters ; S32. Dynamically generate weighting factors based on the three types of core parameters in S31. Alarm accuracy rate compared to historical data ; S33. Read the system constant maximum signal-to-noise ratio and normalize it to obtain... ; S34. Calculate the independent confidence level of each feature index. The process is as follows: ; S35. Perform standard deviation statistics on the confidence values of features sampled multiple times within the current time window to generate confidence intervals for confidence fluctuations. S4. Construct a confidence adjustment instruction list using the data obtained from S2 and S3, as follows: S41. Retrieve the pre-built multimodal collaborative rule library and prioritize the selection of valid rules that are suitable for the current scenario based on the scenario adaptation rules: In enclosed indoor scenarios: Automatically tighten anomaly detection thresholds and weaken rules that mitigate large-scale environmental interference; Outdoor open scenarios: Relax the tolerance for noise and light interference, and strengthen the multi-feature linkage rules; Equipment maintenance scenarios: Prioritize matching derived rules related to equipment vibration and electrical anomalies; S42, Synchronous parallel matching supports five original rule categories: rule of contradiction, rule of interference cancellation, rule of scenario adaptation and rule of derivative, to perform multi-dimensional intelligent correction of credibility; S43. Set arbitration rules for rule conflict priority, with the following priorities: Derivative rules > Scene adaptation rules > Interference cancellation rules > Supporting rules / Contradictory rules; S44. Sort, deduplicate, and standardize all effective rule modification instructions, and output a list of credibility adjustment instructions. Each instruction includes: target feature number, adjustment type, adjustment range, confidence interval constraint identifier, and effective priority; S5. Based on the credibility adjustment instruction list, the independent credibility is corrected level by level to obtain the final comprehensive credibility. The process is as follows: The instructions obtained in S51 and S44 are rearranged according to the conflict priority arbitration rules set in S43; S52. Intelligent arbitration for conflict scenarios where the same feature indicator is acted upon by multiple instructions simultaneously: Higher-priority instructions override conflicting lower-priority instructions. When a reverse adjustment occurs in a peer instruction, the risk attribute of the current scenario is used to determine whether to adopt the instruction with increased trustworthiness for suspected high-risk scenarios and the instruction with decreased trustworthiness for scenarios with environmental interference. Redundant instructions that are duplicated, invalid, or have reverse cancellation are removed to generate a unique and ordered sequence of final execution instructions; S53. Under the constraint of the confidence interval for confidence fluctuation obtained in S3, execute the sorted confidence adjustment instructions one by one: If the adjusted confidence level exceeds the upper limit of the confidence interval, it will be automatically truncated and locked to the maximum value of the interval. If the adjusted confidence level is lower than the lower limit of the confidence interval, it will be automatically clamped and locked to the minimum value of the interval. S54. After completing all instruction calibrations, output the final comprehensive confidence level of each feature with scene label, confidence interval marker, and conflict arbitration and boundary calibration. S6. Judge the final comprehensive credibility obtained in S5 to obtain the refined hierarchical verification result; S7. Based on the refined hierarchical verification results, retrieve the pre-built four-dimensional intelligent security strategy library of the system to generate linkage response instructions; S8. Complete the monitoring process and perform iterative optimization.
2. The security monitoring method based on big data according to claim 1, characterized in that, The process of S2 is as follows: S21. Receive the multi-source heterogeneous data uploaded by S1, including three-layer feature data, corrected instantaneous signal-to-noise ratio, suspected interference tags, and the status of multiple types of heterogeneous sensors; S22. Filter the uploaded data for legality: remove data with lost packets, abnormal data exceeding the range, data with disordered timing, and blank or invalid data to ensure that the data entering the analysis link is complete and valid; S23. Using the unified system timestamp as a benchmark, perform time-series alignment and padding for all nodes and all types of feature data to ensure that all feature data at the same judgment time belong to the same time window and eliminate judgment deviations caused by time misalignment. S24. Perform a range normalization operation on the multimodal features that have completed time-series alignment and padding, and map different physical quantities to the standard range of 0~1. S25. Retrieve environmental parameters and equipment deployment attributes of the current monitoring area, automatically establish a scene classification tag library, and accurately classify three types of scenes: indoor closed scenes, outdoor open scenes, and equipment operation and maintenance scenes.
3. The security monitoring method based on big data according to claim 2, characterized in that, The process of S6 is as follows: S61. Receive the final comprehensive credibility of each feature obtained in S5, and call the adaptive credibility threshold according to the current scenario; S62. Read the current risk type and scenario attributes, and dynamically and adaptively adjust the feature overlap matching threshold: If intrusion, fire, and equipment damage are classified as high-risk, the matching threshold is lowered to 60%, while the ability to detect rare risks is improved. If personnel lingering and minor noise are classified as low-risk, the matching threshold is raised to 80%, while suppressing false alarms due to minor fluctuations. Meanwhile, the overall threshold for indoor scenes has been increased by 10%, while the overall threshold for complex outdoor scenes has been decreased by 10%. S63. Introducing a three-dimensional weighting factor—feature overlap weight, event importance weight, and time-series decay weight—to obtain the precise weighted causal association strength, the process is as follows: S631. The percentage of samples in a statistical historical period that are marked as real security incidents; S632. Multiply the result of S631 by the feature overlap coefficient between the current feature combination and the historical segment; S633, then multiply by the event importance weight corresponding to the event severity level; S634. Finally, multiply by the temporal decay weight based on time distance to obtain the precise weighted causal association strength; S64. By accurately weighting the causal relationship strength, the confidence interval of the credibility fluctuation, and the inherent risk level of the current scenario, determine which category the current standardized verification result belongs to among the five levels: extremely high risk, high risk, medium risk, low risk, and no risk, and also include the risk credibility fluctuation level.
4. The security monitoring method based on big data according to claim 3, characterized in that, The process for S7 is as follows: S71. Collect risk level, security level of incident area, scene type label and risk credibility fluctuation level; S72. Based on the pre-built four-dimensional intelligent security strategy library, perform precise retrieval and matching: Based on the risk level, a basic emergency response plan will be implemented. The severity of measures will be adjusted according to the regional security level, with the intensity of measures automatically upgraded in core areas. The device linkage logic is adapted according to the scenario type: indoors, it focuses on access control and sound and light alarms; outdoor, it focuses on personnel dispatching and trajectory tracking; and device scenarios focus on equipment inspection and shutdown protection. The level of conservatism in handling cases is adjusted based on the level of credibility fluctuations; if the fluctuations are large, a review mechanism is added to avoid mishandling. S73. Conduct risk assessment and event classification, as follows: S731. Accurately determine the specific type of risk event; S732. Further subdivide the risk levels and match them with corresponding security and emergency response standards; S733. Generate the handling priority for this event. When multiple events occur concurrently, prioritize handling high-risk events. S74. Generate a multi-level differentiated linkage instruction set based on the results of S73, and output it according to the execution object. Platform alert levels: tiered pop-up alerts in the background, automatic archiving of event logs, and locking and retaining abnormal videos; Mobile dispatch hierarchy: Security personnel's APP accurately pushes location, event type, and handling tasks; On-site equipment linkage hierarchy: hierarchical sound and light alarm, one-way / two-way locking of area access control, and pan-tilt-zoom (PTZ) monitoring for focusing and tracking; Specialized emergency response levels: High-risk scenarios trigger fire standby, area isolation, and equipment power-off protection; Timing control: All linkage commands are triggered sequentially according to a preset timing sequence to avoid system congestion caused by simultaneous activation of devices; S75. All linkage instructions, scheduling plans, handling priorities, and review requirements are uniformly encapsulated into structured handling instruction messages, carrying the unique event number, scene tag, risk level, and execution time limit, and sent to edge nodes and security terminal devices.
5. A security monitoring method based on big data according to claim 4, characterized in that, The process of S8 is as follows: S81: Real-time acquisition of edge device execution status, sensor secondary detection data, security personnel handling feedback, equipment linkage response latency and on-site handling screen data, generating multi-dimensional execution feedback data streams and transmitting them back to the cloud in real time; In response to abnormal situations, the system automatically generates fault alarms and triggers backup response plans. S82. Combining the secondary perception data from the execution feedback and the on-site handling results, recalculate the comprehensive credibility and causal correlation strength of each feature in the current scenario, dynamically adjust the original risk level, correct the risk judgment deviation, and output the final accurate risk assessment result. S83. Standardize and archive all data of this complete event and store it in the historical feature fragment database; S84. Initiate multi-dimensional autonomous iterative updates, automatically triggering incremental updates of the rule base, fine-tuning of security strategy base parameters, adaptive optimization of scene thresholds, and iterative calibration of dynamic weight parameters; S85. If the same feature combination in the same scenario is misjudged / missed 3 times in a row, the cause of the misjudgment will be automatically analyzed, and corresponding interference cancellation rules and scenario adaptation rules will be generated in a targeted manner. The dynamic weight parameters and matching thresholds in the scenario will be optimized simultaneously to avoid similar misjudgment problems from the bottom layer of the algorithm. S86. Migrate the optimized scene rules, weight parameters, and judgment thresholds to similar scenes to achieve rapid iteration of the entire system.
6. A security monitoring system based on big data, employing the security monitoring method based on big data as described in any one of claims 1-5, characterized in that: This includes an edge perception preprocessing module for edge data acquisition and a cloud-based collaborative analysis and processing module for security monitoring; The edge perception preprocessing module includes a multi-sensor parallel acquisition unit, a hierarchical feature extraction unit, a dynamic noise correction unit, an edge local confidence initial screening and filtering unit, and an edge execution feedback acquisition unit. The cloud-based collaborative analysis and resolution module includes a data temporal alignment and normalization unit, a dynamic adaptive independent credibility calculation unit, a spatiotemporal correlation multimodal collaborative reasoning unit, an interval constraint credibility adjustment unit, a fuzzy matching saliency verification unit, a multi-level strategy mapping instruction generation unit, a closed-loop feedback correction and iteration unit, and a scenario adaptive management unit.
Citation Information
Patent Citations
Security alarm information data interaction system and method
CN118486152A
Intelligent Internet of Things public security management and control system and method based on multi-source data fusion
CN120263824A
Electrical complete equipment state monitoring and early warning system based on multi-parameter intelligent sensing
CN121431995A
