Intelligent security monitoring and responding method for power generation enterprise

Abstract

The invention relates to the technical field of industrial safety management, in particular to an intelligent security monitoring and responding method for a power generation enterprise. The method comprises the following steps: acquiring environmental data of a power generation enterprise; performing enterprise abnormal event detection on the power generation enterprise environment data to obtain enterprise environment abnormal event data; performing multi-source video linkage monitoring on the enterprise environment abnormal event data to obtain multi-angle video data; and performing abnormal event analysis according to the enterprise environment abnormal event data and the multi-angle video data to obtain abnormal event risk assessment data so as to perform intelligent security monitoring and response auxiliary operation of the power generation enterprise. According to the method, the abnormal event detection and analysis capability is improved, and the safety and the operation efficiency of an enterprise are also remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial safety management, and in particular to an intelligent security monitoring and response method for a power generation enterprise. Background Art

[0002] With the rapid development of the industrial field, power generation enterprises, as an important part of energy production, have complex production environments and diverse risk factors. Traditional safety management methods mainly rely on regular inspections, single-point monitoring and manual analysis, which have the following significant shortcomings: traditional monitoring systems are only based on sensors or cameras in fixed positions, which are prone to forming monitoring blind spots and lack comprehensive coverage of environmental changes. At present, most systems rely only on a single data source (such as environmental sensor data or video data) for anomaly detection, resulting in low accuracy and reliability of detection results. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes an intelligent security monitoring and response method for a power generation enterprise to solve at least one of the above technical problems.

[0004] The present application provides a method for intelligent security monitoring and response of a power generation enterprise, comprising the following steps: Step S1: Obtain environmental data of power generation enterprises; Step S2: Perform enterprise abnormal event detection on the environmental data of the power generation enterprise to obtain enterprise environmental abnormal event data; Step S3: Perform multi-source video linkage monitoring on abnormal event data of the enterprise environment to obtain multi-angle video data; Step S4: Analyze abnormal events based on the enterprise environment abnormal event data and multi-angle video data to obtain abnormal event risk assessment data to carry out intelligent security monitoring and response auxiliary operations for power generation enterprises.

[0005] In the present invention, environmental parameters (such as temperature, humidity, smoke, vibration, etc.) are collected in real time through a sensor network, covering all key areas of power generation enterprises. Data analysis methods are used to quickly identify the location of abnormal events, providing a basis for monitoring and response. Based on abnormal event data, cameras related to the event are automatically linked to perform multi-angle real-time monitoring of the target area. The combination of multi-source video data avoids the blind spot problem of a single monitoring point and improves the comprehensiveness of monitoring coverage. Specific response suggestions or action plans (such as triggering alarms, starting fire extinguishing systems) are generated according to the risk level, significantly shortening the time from discovery to response.

[0006] Preferably, step S1 specifically includes: Step S11: deploy temperature and humidity sensors, smoke detectors, and vibration detection equipment, initialize equipment parameters, complete network connection with the central monitoring system, and obtain sensor metadata of the power generation enterprise; Step S12: collecting sensor parameters according to the sensor metadata of the power generation enterprise to obtain preliminary sensor parameter data; Step S13: Performing quality verification on the preliminary sensing parameter data to obtain verification environment data; Step S14: Preliminarily filter the verification environment data to obtain filtered environment data; Step S15: Distribute and aggregate the filtered environment data to obtain local environment data; Step S16: Perform global sensor network mapping based on local environmental data to obtain preliminary power generation enterprise environmental data; Step S17: Perform hierarchical decoding on the preliminary power generation enterprise environmental data to obtain the power generation enterprise environmental data.

[0007] In the present invention, multiple types of sensors are reasonably deployed according to the layout of enterprise equipment to ensure the comprehensiveness of environmental monitoring. Unified management and real-time monitoring of sensors are achieved through network connection, reducing the complexity of equipment management. Environmental parameters such as temperature, humidity, vibration frequency, smoke concentration, etc. are collected in real time using the deployed sensor network. Error data is removed by verification to improve data quality, laying the foundation for accurate analysis. Low-pass filtering and smoothing algorithms are used to preliminarily filter the verification data to extract effective environmental change characteristics. Distributed cache significantly reduces the data pressure of the central system and improves data transmission and processing efficiency. Local aggregation facilitates partitioned environmental analysis and independent processing. A global sensor network topology map is generated using local environmental data to perform data spatial association. The sensor monitoring range is mapped to the physical space of the enterprise through geometric projection or GIS tools. The preliminary power generation enterprise environmental data is multi-level decoded according to the enterprise environmental characteristics (such as layered by time, space, and feature dimensions) to facilitate multi-dimensional analysis.

[0008] Preferably, step S2 specifically includes: Step S21: performing feature matching according to the environmental data of the power generation enterprise and the preset abnormal event recognition model to obtain preliminary abnormal event data; Step S22: performing complex event processing on the preliminary abnormal event data to obtain complex event data; Step S23: performing abnormal event determination on the complex event data to obtain abnormal event determination data; Step S24: Perform multi-stage verification and cross-check on the abnormal event determination event to obtain enterprise environment abnormal event data.

[0009] In the present invention, potential abnormal events are quickly screened from huge environmental data through feature matching. The preliminary abnormal event data is input into a complex event processing engine (such as Apache Flink) to analyze the time, space, and causal relationship of the event. Extract composite patterns (such as the superposition of multiple environmental features or multiple events within a time window) to generate complex event data. Further narrow the scope of abnormal events based on abnormal standards and models to improve the accuracy of judgment. Multi-stage verification of abnormal event judgment data is performed, including threshold verification (such as repeated events within a time window), historical data comparison (such as the frequency of occurrence of similar events), and spatial consistency check (such as event coverage). Use a cross-check mechanism (Cross-check) to combine video data, sensor data, and log data to verify the authenticity of the event.

[0010] Preferably, step S3 specifically includes: Step S31: locking the event area according to the enterprise environment abnormal event data to obtain abnormal event area data; Step S32: activating the regional monitoring device according to the abnormal event regional data to obtain regional monitoring data; Step S33: collecting original video according to the regional monitoring data to obtain original video data; Step S34: Slicing the video according to the original video data to obtain video slice data, and performing time axis synchronization on the video slice data to obtain synchronized video segment data; Step S35: extracting video features from the synchronized video segment data to obtain video feature data, and performing abnormal picture recognition on the video feature data to obtain abnormal picture segment data; Step S36: annotate and classify the images according to the abnormal image segment data to obtain annotated video data; Step S37: construct a video network according to the monitoring location data corresponding to the regional monitoring data and the annotated video data to obtain multi-angle video data.

[0011] In the present invention, based on the abnormal event data of the enterprise environment, the abnormal event area is locked through the geographic information system (GIS) or sensor network mapping. The accuracy of the locked area is optimized by using the dynamic area modeling technology. According to the abnormal event area data, the monitoring camera in the coverage area is activated, and the parameters such as the camera angle and focal length are adjusted. The real-time video stream is extracted from the activated regional monitoring device. The storage and transmission of video data are optimized by using efficient video coding technology (such as H.265). The original video data is sliced ​​according to the time period. The time axis of the multi-source video data is corrected using the time synchronization protocol (NTP) to ensure the uniformity of the timeline. The synchronized video segment data is feature extracted to extract the target area features. The abnormal screen (such as fire, illegal intrusion, etc.) is automatically detected to generate abnormal screen segment data. The abnormal screen segment data is manually or automatically annotated, including the objects, actions and event types in the screen. The screen content is classified and the video event type is annotated. The annotated video data is associated with the geographic location data of the regional monitoring device. A multi-angle video network is constructed to display the panoramic view of the abnormal event and the screens at different angles.

[0012] Preferably, the event area locking in step S31 is specifically as follows: The video surveillance coverage area is locked according to the enterprise environment abnormal event data to obtain the first abnormal event area data; According to the enterprise environment abnormal event data, the abnormal mode propagation path area is locked to obtain the second abnormal event area data; According to the enterprise environment abnormal event data, the environment characteristic difference area is locked to obtain the third abnormal event area data; Regional fusion is performed according to the first abnormal event region data, the second abnormal event region data and the third abnormal event region data to obtain abnormal event region data.

[0013] In the present invention, based on the enterprise environment abnormal event data, the geographical distribution and coverage of video surveillance cameras are used to lock the video surveillance equipment covering the event area, and the abnormal event occurrence area is quickly locked through the coverage of the video surveillance camera to ensure the correlation between the video data and the event. The abnormal mode propagation path model is used to analyze the time and space trajectory of the event from the occurrence point to the surrounding area, lock the affected propagation area, consider the dynamic characteristics of the abnormal event propagation, and adjust the locking range in real time. Compare the difference between the abnormal event area and the baseline environmental characteristics (such as temperature, humidity, smoke concentration, etc.), lock the abnormal area with significant characteristic changes, not limited to the source of the event, but lock other affected areas through changes in environmental characteristics. The first, second, and third abnormal event area data are integrated, and regional superposition, optimization, and boundary refinement are performed to generate the final abnormal event area, integrating the advantages of multiple locking methods to ensure the comprehensiveness and accuracy of regional locking.

[0014] Preferably, the video surveillance coverage area is locked specifically as follows: Obtain basic data of monitoring equipment; Constructing a monitoring equipment spatial distribution model based on the basic data of the monitoring equipment to obtain a monitoring equipment spatial distribution model; Mapping the event location according to the abnormal event data of the enterprise environment to obtain the abnormal event location data; Perform camera mapping based on abnormal event location data and monitoring equipment spatial distribution model to obtain camera list data; Reconstruct the video scene geometry based on the camera list data to obtain a three-dimensional model of the monitoring area; Extract the visible area of ​​abnormal events based on the three-dimensional model of the monitoring area to obtain key frame data of the visible area; The monitoring area boundary is generated for the key frame data of the visible area to obtain the first abnormal event area data.

[0015] The present invention collects the basic data of all monitoring devices, including parameters such as geographic location, viewing angle, focal length, resolution, and coverage. The basic data of monitoring equipment is used to build a spatial distribution model, including the geographic coordinates, field of view, coverage, and blind spot location of the equipment. The spatial polygons of the monitoring equipment coverage area are generated using GIS tools and trigonometric calculations. The location attributes in the enterprise environment abnormal event data are geographically mapped with the monitoring equipment spatial distribution model to quickly associate the abnormal event with the specific location in the enterprise spatial layout. According to the abnormal event location data and the monitoring equipment spatial distribution model, the camera covering the location is identified. A list of cameras associated with the event is generated, including information such as camera ID, location, and coverage area. Through three-dimensional reconstruction technology, a three-dimensional scene analysis is provided for abnormal events. The visible area related to the abnormal event location is extracted from the monitoring area three-dimensional model (SA3DM), and key frame data in the visible area is generated. The visible area key frame data (VRK) is subjected to boundary analysis to generate the physical boundary of the abnormal event area.

[0016] Preferably, the abnormal mode propagation path area locking is specifically: Based on the abnormal event data of the enterprise environment, Bayesian network modeling is carried out to obtain the event dynamic propagation model; Generate a spatiotemporal node network for the event dynamic propagation model to obtain event spatiotemporal node network data; Calculate the probability of the propagation path based on the event space-time node network data to obtain the event propagation path probability data; Perform regional impact allocation on the event propagation path probability data to obtain event regional impact weight data; Multi-region propagation fusion is performed based on the event area impact weight data to obtain multi-region propagation path data; Predicting the propagation path based on the multi-region propagation path data to obtain predicted propagation path data; Reconstruct the propagation path area according to the predicted propagation path data to obtain the propagation path coverage area data; Region generation is performed based on the enterprise environment abnormal event data and the propagation path coverage area data to obtain second abnormal event area data.

[0017] In the present invention, a Bayesian network is constructed based on the abnormal event data of the enterprise environment to analyze the causal relationship and propagation law of abnormal events. The model node represents the influencing factor of the event, the edge represents the causal relationship, and the edge weight represents the probability of event propagation. Based on the event dynamic propagation model, a node network containing spatiotemporal information is constructed, the node represents the spatial position, and the edge represents the propagation path. In the event spatiotemporal node network data, the probability of event propagation path is calculated using the Markov chain. Considering the influencing factors (such as time delay and environmental resistance) on each path to adjust the propagation probability, the propagation probability of the event on different paths is accurately calculated to help identify high-risk propagation paths. According to the event propagation path probability data, the impact of path propagation is distributed to the coverage area, and a weight is assigned to each area to intuitively reflect the scope and intensity of event impact. The event area impact weight data is fused with the propagation characteristics of multiple regions to generate a global multi-region propagation path, and the multi-region propagation fusion improves the adaptability to complex propagation scenarios. Accurately predict the future propagation path of the event to help enterprises take prevention and control measures before the event expands. According to the predicted propagation path data, the coverage of the propagation path is reconstructed in combination with the regional spatial information, and the coverage area of ​​the propagation path is accurately reconstructed to reduce the boundary error in the analysis. The second abnormal event area is generated by integrating the enterprise environment abnormal event data and the propagation path coverage area data.

[0018] Preferably, the environmental characteristic difference area locking is specifically: Performing spatiotemporal mapping of environmental characteristics based on enterprise environmental abnormal event data to obtain characteristic spatiotemporal distribution map data; Generate a feature benchmark model based on the event environment feature spatiotemporal distribution map data to obtain a regional feature benchmark model; Perform real-time feature deviation analysis based on the regional feature benchmark model and feature spatiotemporal distribution map data to obtain feature deviation matrix data; Perform difference region clustering according to feature deviation matrix data to obtain cluster region data; Conduct regional similarity analysis based on clustered regional data to obtain differential regional data; Refining the difference boundary of the difference area data to obtain optimized area boundary data; Region generation is performed based on the characteristic deviation matrix data and the optimized region boundary data to obtain the third abnormal event region data.

[0019] In the present invention, based on the enterprise environmental abnormal event data, environmental features (such as temperature and humidity, vibration, and smoke concentration) are extracted and spatiotemporal distribution modeling is performed to provide an intuitive distribution of environmental features in time and space, and reveal the potential change law of the event. Based on historical environmental data and feature spatiotemporal distribution map data, a feature benchmark model is constructed, and a benchmark model is generated by region to adapt to the environmental characteristics of different locations of the enterprise. The real-time environmental feature data is compared with the regional feature benchmark model to calculate the deviation of each feature point. Based on the feature deviation matrix data, a clustering algorithm (such as DBSCAN or K-Means) is used to identify the difference area. The clustering process highlights the area with significant feature deviation and reduces the interference of invalid areas. The feature vector of the clustered area data is extracted, the similarity between the areas is calculated (such as cosine similarity or dynamic time warping), and the overlapping areas are merged through similarity analysis to improve the efficiency of regional analysis. The boundary of the difference area data is optimized, the abnormal boundary points are eliminated, the boundary curve is smoothed, and the redundant boundary points are removed to ensure the accuracy of the regional boundary. The deviation data and the boundary optimization results are combined to ensure the accuracy and reliability of the abnormal area.

[0020] Preferably, the regional integration is specifically: Perform multi-region intersection calculation according to the first abnormal event region data, the second abnormal event region data, and the third abnormal event region data to obtain intersection region data; Assign regional weights according to the intersection area data to obtain weighted regional data; Performing regional feature fusion according to weighted regional data to obtain fused regional feature data; Optimize the region boundary according to the fused region feature data to obtain first region data; Perform feature extraction based on the first abnormal event region data, the second abnormal event region data, and the third abnormal event region data to obtain first abnormal event region feature data, second abnormal event region feature data, and third abnormal event region feature data, respectively; Performing Gaussian clustering calculation on the first abnormal event regional feature data, the second abnormal event regional feature data, and the third abnormal event regional feature data to obtain regional feature Gaussian clustering data; Calculate the degree of similarity and difference based on the regional characteristic Gaussian clustering data to obtain the regional characteristic similarity and difference degree data; Perform weighted calculation on the intersection area data according to the regional feature similarities and differences to obtain the second area data; Performing regional difference distribution calculation based on the first regional data and the second regional data to obtain regional difference distribution data; Regional compensation calculation is performed on the first regional data and the second regional data according to the regional difference distribution data to obtain abnormal event regional data.

[0021] In the present invention, the geometric intersection calculation is performed on the first, second and third abnormal event area data to identify the common coverage part of the three area data. According to the source and importance of the intersection area data, a weight value is assigned to each area (such as according to the reliability or detection accuracy of the abnormal event). The environmental features of each area in the weighted area data are fused to generate comprehensive feature data. According to the fused regional feature data, the abnormal area boundary is optimized and the redundant or fuzzy boundary is eliminated. The first, second and third abnormal event area data are respectively subjected to feature extraction to extract the spatial, temporal and environmental features of each area. The extracted regional feature data are clustered using a Gaussian mixture model to output high-confidence clustering centers and regional feature distributions. Gaussian clustering identifies the commonalities and differences between regions and provides a basis for similarity and difference analysis. The similarity and difference between regions are calculated for the Gaussian clustering results to quantify the differences in regional characteristics. The intersection area data is weighted calculated using the regional feature similarity and difference degree data. According to the first and second regional data, the regional difference distribution is calculated to generate a regional risk map, provide difference information between regions, and provide data support for resource scheduling optimization. The omission and redundancy problems are solved through regional compensation to generate a complete range of abnormal areas.

[0022] Preferably, step S4 is specifically: Step S41: performing multimodal feature fusion on the enterprise environment abnormal event data and the multi-angle video data to obtain multimodal data of enterprise abnormal events; Step S42: performing Bayesian risk assessment on the multimodal data of enterprise abnormal events to obtain preliminary enterprise abnormal event risk assessment data; Step S43: performing event cascade analysis on the multimodal data of enterprise abnormal events according to the preliminary enterprise abnormal event risk assessment data to obtain event cascade graph data; Step S44: Perform potential risk assessment based on the event cascade graph data and the preset enterprise environment abnormal event assessment model to obtain secondary enterprise abnormal event risk assessment data; Step S45: Integrate the preliminary enterprise abnormal event risk assessment data and the secondary enterprise abnormal event risk assessment data to obtain abnormal event risk assessment data for performing intelligent security monitoring and response assistance operations for power generation enterprises.

[0023] In the present invention, enterprise environmental abnormal event data (such as temperature, humidity, vibration, and smoke concentration) and multi-angle video data (such as picture features captured by the camera) are fused. A Bayesian network model is used to conduct a preliminary risk assessment on the multimodal data of enterprise abnormal events to quantify the risk probability of abnormal events. Based on the preliminary enterprise abnormal event risk assessment data, causal and temporal analysis is performed on abnormal events in multimodal data to reveal the dynamic relationship between abnormal events and provide insights into complex event scenarios. Using a preset enterprise environmental abnormal event assessment model, secondary risk analysis is performed on event cascade graph data to assess the risk probability and impact range of potential abnormal events and identify secondary risks that have not yet manifested. Integrate the preliminary and secondary assessment results to form a complete risk assessment system.

[0024] The beneficial effects of the present invention are: deploying multiple types of sensors (such as temperature and humidity, vibration, smoke, etc.) to cover the core areas of power generation enterprises, realizing comprehensive collection of environmental data, forming high-quality enterprise environmental data, and ensuring the real-time, accuracy and stability of the data. A full-link real-time response from the occurrence to the detection of abnormal events is realized, shortening the event discovery time. The abnormal event detection process is deeply combined with the enterprise environmental data, and abnormal features in complex scenes (such as the superposition effect of environmental anomalies) can be identified. Dynamic monitoring of abnormal events is realized, and accurate video data with multiple dimensions and perspectives is captured, providing visual support for event analysis. Different types of abnormal area data are effectively integrated, reducing the redundant processing volume of video data and improving system efficiency. The multi-source linkage mechanism ensures the complete recording of events in complex scenes (such as secondary problems caused by equipment failure). The system can dynamically evaluate the current risks and secondary risks of abnormal events and warn the scope of disaster spread in advance. The evaluation results accurately quantify the severity and impact of the event, and provide clear action priority recommendations for enterprise decision makers. Risk assessment integrates static data and dynamic monitoring, significantly improving the comprehensiveness and depth of risk management. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings: Figure 1 A flowchart showing a method for intelligent security monitoring and response for a power generation enterprise according to an embodiment of the present invention is shown; Figure 2 A flowchart showing a method for acquiring environmental data of a power generation enterprise according to an embodiment of the present invention is provided; Figure 3 A flowchart showing a method for detecting abnormal events in an enterprise according to an embodiment of the present invention is provided; Figure 4 A flowchart showing a multi-source video linkage monitoring method according to an embodiment is shown; Figure 5 A flowchart of the steps of a method for assessing risk of abnormal events according to an embodiment is shown. DETAILED DESCRIPTION

[0026] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0027] In addition, the drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0028] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0029] See also Figures 1 to 5 , the present application provides a method for intelligent security monitoring and response of a power generation enterprise, comprising the following steps: Step S1: Obtain environmental data of power generation enterprises; Specifically, temperature and humidity sensors, smoke detectors, and vibration detection equipment are deployed in key areas within power generation enterprises (such as boiler rooms, fuel depots, and transmission lines). These devices are connected to the central monitoring system through pre-configured wireless or wired networks. The data collected by the equipment includes temperature, humidity, gas concentration, and vibration frequency, and the data is uploaded to the central monitoring system at set time intervals. The monitoring system performs real-time verification and formatting of the transmitted data, filters out invalid or abnormal data, and ensures the consistency and reliability of the collected data. Subsequently, the environmental data is divided into local data for different areas, and global data of the overall environmental characteristics of the enterprise is generated in the system.

[0030] Step S2: Perform enterprise abnormal event detection on the environmental data of the power generation enterprise to obtain enterprise environmental abnormal event data; Specifically, the central monitoring system uses the built-in abnormal event detection module to perform real-time analysis on the received environmental data. The system presets multiple abnormal detection rules, such as excessive temperature indicating equipment overload, humidity changes reflecting leakage risks, and excessive gas concentration indicating dangerous gas leakage. By matching the environmental data with the preset rules, preliminary abnormal event data is identified. Subsequently, the system performs complex event processing on the abnormal events, and further confirms the severity and diffusion trend of the events by analyzing the relationships between multiple variables in the abnormal area. The system records the confirmed abnormal event information (including event type, occurrence time, location, and impact range) to generate enterprise environmental abnormal event data.

[0031] Step S3: Perform multi-source video linkage monitoring on abnormal event data of the enterprise environment to obtain multi-angle video data; Specifically, when the monitoring system confirms an abnormal event, it immediately activates the surveillance cameras covering the area according to the location of the abnormal event. The multi-angle video data collected by the camera in real time is uploaded to the monitoring center, and the system processes the video images, including image clarity improvement, background noise elimination, and moving target detection. The images collected by multiple cameras are integrated into a complete regional monitoring video through image stitching technology. Subsequently, the system extracts key images related to the abnormal event in the video, such as clips of smoke, flames, or equipment abnormalities, and generates multi-angle video data for further abnormal event analysis.

[0032] Step S4: Analyze abnormal events based on the enterprise environment abnormal event data and multi-angle video data to obtain abnormal event risk assessment data to carry out intelligent security monitoring and response auxiliary operations for power generation enterprises.

[0033] Specifically, the monitoring center combines enterprise environmental abnormal event data with multi-angle video data for multimodal analysis. By matching the dynamic changes of environmental data and the features in the video footage, the system makes a preliminary assessment of the severity of the abnormal event. Subsequently, using historical data and preset risk assessment models, a more in-depth analysis of the potential impact scope and development trend of the current abnormal event is conducted. For example, the system can predict the path of fire spread or whether abnormal equipment will trigger other chain reactions. The system generates a complete abnormal event risk assessment report, including risk levels, recommended emergency response measures and subsequent risk points, and provides it to operators for decision-making assistance.

[0034] Preferably, step S1 specifically includes: Step S11: deploy temperature and humidity sensors, smoke detectors, and vibration detection equipment, initialize equipment parameters, complete network connection with the central monitoring system, and obtain sensor metadata of the power generation enterprise; Specifically, temperature and humidity sensors, smoke detectors, and vibration detection equipment are installed in key areas of power generation enterprises (such as boiler rooms, fuel storage areas, transmission equipment areas, etc.). Each sensor adjusts the installation location according to the specific needs of the area. For example, temperature and humidity sensors are arranged in equipment-intensive areas and storage environments, smoke detectors are arranged in flammable areas, and vibration detection equipment is installed near key mechanical equipment. The parameters of the sensor equipment (such as sampling frequency, sensitivity range, and network address) are initialized after installation. The device completes the docking of the data transmission network with the central monitoring system through a wireless network module (such as Wi-Fi or LoRa) or a wired connection. After initialization, the system records the metadata information of each sensor, including the device number, installation location, working range, and network connection status.

[0035] Step S12: collecting sensor parameters according to the sensor metadata of the power generation enterprise to obtain preliminary sensor parameter data; Specifically, environmental data is collected regularly at a set sampling frequency through deployed sensors. The data collected by each device includes: temperature and humidity sensors, which record temperature and humidity changes; smoke detectors: detect smoke concentration in the area; vibration detection equipment: monitor the vibration frequency of the equipment operation. The collected preliminary sensor parameter data is uploaded to the central monitoring system through the transmission module of the equipment, and marked and archived by timestamp and location.

[0036] Step S13: Performing quality verification on the preliminary sensing parameter data to obtain verification environment data; Specifically, after receiving the preliminary sensor parameter data, the central monitoring system verifies the data quality, including the following steps: Check whether there are missing or abnormal fields in the data packet, such as whether the temperature, humidity, and smoke concentration values ​​are empty. Determine whether each parameter value is within the normal operating range of the device, for example, whether the temperature value exceeds the device detection limit. Process obviously abnormal data (such as sudden extreme values) and replace them with the average value of the surrounding time period or the data of adjacent sensors. After completing the quality verification, output the verified environmental data.

[0037] Step S14: Preliminarily filter the verification environment data to obtain filtered environment data; Specifically, the verification environment data is initially filtered to remove redundant information and low-relevance data, such as invalid data or repeated sampling points when there is no abnormal state. The data after initial filtering retains records with strong relevance to the equipment monitoring range and time series, providing more refined input data for subsequent analysis.

[0038] Step S15: Distribute and aggregate the filtered environment data to obtain local environment data; Specifically, the initially filtered data is cached and aggregated on distributed nodes at the regional level. For example, the sensor data in each monitoring area is transmitted to the local cache server for short-term storage, and then the environmental data characteristics of the area (such as average temperature, maximum vibration value, etc.) are generated through aggregation calculation. This not only reduces the central storage pressure of the system, but also provides basic data for regional environmental analysis.

[0039] Step S16: Perform global sensor network mapping based on local environmental data to obtain preliminary power generation enterprise environmental data; Specifically, the central monitoring system globally maps the sensor network of the entire power generation enterprise through the local environmental data uploaded by the distributed nodes. The mapping process generates preliminary environmental data within the enterprise based on the geographical location of the sensors, the coverage of the monitoring area, and the current data status, including the working status of the equipment, the location of the abnormal area, and the distribution of environmental variables.

[0040] Step S17: Perform hierarchical decoding on the preliminary power generation enterprise environmental data to obtain the power generation enterprise environmental data.

[0041] Specifically, the central monitoring system decodes the global data in layers, deconstructing the complex environmental data into logical layers: real-time monitoring layer: high-frequency data reflecting the current environmental status; historical trend layer: low-frequency data describing the long-term data change trend; abnormal state layer: marking and associating abnormal event data. The environmental data after layered decoding is stored as complete environmental data of power generation enterprises to support abnormal event detection and response work.

[0042] Preferably, step S2 specifically includes: Step S21: performing feature matching according to the environmental data of the power generation enterprise and the preset abnormal event recognition model to obtain preliminary abnormal event data; Specifically, the central monitoring system performs feature matching based on the environmental data of power generation enterprises and the preset abnormal event recognition model. The abnormal event recognition model contains a variety of abnormal feature templates, for example: excessive temperature indicates equipment overload; a sharp increase in humidity reflects equipment leakage; excessive smoke concentration indicates combustion or leakage accidents. The monitoring system compares the variables such as temperature, humidity, and smoke concentration in the environmental data with these templates one by one to filter out data that meets the conditions. For example, when the temperature in a certain area continues to exceed the safety threshold and the humidity changes significantly, the system marks it as a potential abnormal event and records relevant data (such as time, location, and scope of impact). The filtered data is defined as preliminary abnormal event data.

[0043] Step S22: performing complex event processing on the preliminary abnormal event data to obtain complex event data; Specifically, for preliminary abnormal event data, the central monitoring system further analyzes the potential complexity of the abnormality. For example: check whether there are multiple abnormal features occurring at the same time, such as temperature exceeding the standard accompanied by increased smoke concentration; evaluate the time series change pattern of abnormal events, such as whether multiple consecutive abnormal readings are periodic. The monitoring system determines whether these abnormalities have the possibility of causal relationship or chain reaction by analyzing the synchronous data of multiple sensors. For example, when a high temperature event and abnormal vibration data in an adjacent area are highly consistent in time, the system merges them into a complex event. Complex event data includes event type, time series characteristics, and spatial distribution characteristics.

[0044] Step S23: performing abnormal event determination on the complex event data to obtain abnormal event determination data; Specifically, complex event data is further analyzed to determine whether it constitutes an abnormal event. For example: Combined with historical data, the severity of the current event is evaluated, such as whether it exceeds the previous range of changes. Based on the scope of the event impact, determine whether it poses a direct threat to the equipment or environment. The system generates an abnormal event determination result for each complex event, including the event type (such as fire, equipment failure), risk level (high, medium, low), and priority response recommendations. For example, when the system determines that a high temperature event accompanied by abnormal vibration causes equipment overheating and failure, it will be marked as a high-priority abnormal event.

[0045] Step S24: Perform multi-stage verification and cross-check on the abnormal event determination event to obtain enterprise environment abnormal event data.

[0046] Specifically, in order to improve the accuracy of abnormal event determination, the system performs multi-stage verification and cross-checking on the determination results: Data resampling verification: re-collect real-time data from relevant sensors to check whether abnormal features persist. Historical pattern verification: compare the current event with the historical data pattern to determine whether it conforms to the known abnormal pattern. Multi-sensor cross-check: check whether other sensor data in the same area is consistent with the determination result, for example, whether the readings of surrounding sensors support the occurrence of abnormal events. Abnormal events that pass multi-stage verification are further confirmed to generate enterprise environmental abnormal event data, including detailed event information (type, location, time and risk level).

[0047] Preferably, step S3 specifically includes: Step S31: locking the event area according to the enterprise environment abnormal event data to obtain abnormal event area data; Specifically, the monitoring system locates the area where abnormal events occur based on the enterprise environment abnormal event data. For example, if the sensor in a certain area detects abnormal high temperature and smoke, the system will determine the event area based on the area as the center and the impact range of the event. The event area data includes the area boundary, center point location and impact radius, which are used to monitor the activation of the equipment.

[0048] Step S32: activating the regional monitoring device according to the abnormal event regional data to obtain regional monitoring data; Specifically, the monitoring system searches the device database for a list of monitoring devices covering the event area. For example, the camera device in the area is activated and enters the real-time video acquisition mode. At the same time, the system checks the status of the device to ensure that the device is operating normally. The activated monitoring device begins to transmit video data to the central system to form regional monitoring data.

[0049] Step S33: collecting original video according to the regional monitoring data to obtain original video data; Specifically, the activated monitoring equipment collects video data of the event area in real time and uploads it to the central monitoring system. The original video data includes the video frame sequence and the corresponding timestamp, device number and other meta information. These data are centrally stored for processing.

[0050] Step S34: Slicing the video according to the original video data to obtain video slice data, and performing time axis synchronization on the video slice data to obtain synchronized video segment data; Specifically, the central monitoring system slices the raw video data and divides the long video stream into multiple short video clips. The slicing is based on the time interval and the time when the event occurred, ensuring that each video contains key scenes. Subsequently, the system synchronizes the time axis of the video clips collected by different cameras to ensure that all video clips are aligned within the same time range.

[0051] Step S35: extracting video features from the synchronized video segment data to obtain video feature data, and performing abnormal picture recognition on the video feature data to obtain abnormal picture segment data; Specifically, the system extracts features from the synchronized video clips and analyzes key elements in the image: identifying moving targets in the image, such as people, equipment, or smoke; extracting information such as the target's trajectory and color distribution; and detecting special visual features in the image (such as flames and smoke). Through the above feature extraction, the system generates key image information related to abnormal events.

[0052] Step S36: annotate and classify the images according to the abnormal image segment data to obtain annotated video data; Specifically, the system combines the results of video feature extraction with a deep learning algorithm to identify abnormal images. For example, when features of smoke diffusion or flame jumping are detected, the system marks these images as abnormal images. Abnormal image fragment data includes time range, abnormal type (such as smoke, flame) and location. The system labels and classifies abnormal images. For example, the images are marked as "smoke abnormality" and "high temperature equipment abnormality", and the corresponding time and space information is recorded. The labeled data is stored as labeled video data for quick retrieval, which is convenient for further analysis.

[0053] Step S37: construct a video network according to the monitoring location data corresponding to the regional monitoring data and the annotated video data to obtain multi-angle video data.

[0054] Specifically, the system integrates the annotated video data according to the geographical location of the monitoring equipment to generate a multi-angle video network. For example, through the overlapping relationship between the camera position and the picture content, the pictures from different cameras are spliced ​​to form a panoramic video of the event area. The multi-angle video data generated by the system includes time-synchronized multi-view pictures, which provides complete visual support for subsequent abnormal event analysis.

[0055] Preferably, the event area locking in step S31 is specifically as follows: The video surveillance coverage area is locked according to the enterprise environment abnormal event data to obtain the first abnormal event area data; Specifically, the system determines the first abnormal event area based on the abnormal event data of the enterprise environment and the coverage of the video surveillance equipment. For example, when an abnormal event occurs in the boiler room area, the system retrieves the monitoring equipment covering the area through the camera database, and generates a monitoring coverage area based on the camera's field of view and monitoring range. The coverage area is represented as polygonal area data, marking the monitoring range containing the abnormal event. These data are defined as the first abnormal event area data.

[0056] According to the enterprise environment abnormal event data, the abnormal mode propagation path area is locked to obtain the second abnormal event area data; Specifically, the system combines the dynamic propagation model of abnormal events to analyze the event diffusion path. For example, abnormal temperature and increased smoke are detected in a certain area. The system predicts the direction and range of smoke diffusion through historical data and propagation laws. Based on the sensor network and environmental data, the system generates the path area for event propagation. The area includes the initial event point and the potential diffusion range, which is marked as the second abnormal event area data.

[0057] According to the enterprise environment abnormal event data, the environment characteristic difference area is locked to obtain the third abnormal event area data; Specifically, the system identifies abnormal areas based on the differences in environmental characteristics inside and outside the abnormal event area. For example, near the area where the abnormal event occurred, the humidity, temperature or vibration characteristics significantly deviate from the normal range. By comparing the changes in these characteristic values, the system marks the areas that are significantly different from the normal environment, and the areas are defined as the third abnormal event area data.

[0058] Regional fusion is performed according to the first abnormal event region data, the second abnormal event region data and the third abnormal event region data to obtain abnormal event region data.

[0059] Specifically, the system fuses the first, second, and third abnormal event area data to generate the final abnormal event area. First, the system calculates the intersection of these areas to ensure that the overlapping areas are marked as high-risk areas first. Subsequently, the non-overlapping areas are ranked by importance using a weight assignment method. The fused abnormal event area data includes the main scope of the event and the affected diffusion area, which is used to monitor device activation and abnormal event analysis.

[0060] Preferably, the video surveillance coverage area is locked specifically as follows: Obtain basic data of monitoring equipment; Specifically, the system extracts basic information about monitoring devices from the monitoring device database, including device number, location coordinates, camera orientation, field of view, maximum monitoring distance, and the current working status of the device. The data records the spatial distribution and coverage of each device for subsequent regional modeling and analysis.

[0061] Constructing a monitoring equipment spatial distribution model based on the basic data of the monitoring equipment to obtain a monitoring equipment spatial distribution model; Specifically, the system constructs a spatial distribution model of monitoring equipment based on the basic data of the monitoring equipment. The monitoring range of each camera is defined as a cone model with the camera position as the vertex and extending according to the field of view and monitoring distance. Subsequently, the cones of all devices are integrated into a unified enterprise coordinate system to form a spatial distribution model of monitoring equipment covering the entire power generation enterprise, which is used to quickly query whether a certain location is within the monitoring range.

[0062] Mapping the event location according to the abnormal event data of the enterprise environment to obtain the abnormal event location data; Specifically, based on the enterprise environment abnormal event data, the system locates the specific location where the abnormal event occurred. By combining the geographic coordinates of the sensor where the event occurred or the affected area, the event location data is generated. The data describes the central location of the abnormal event and its impact range, marking the area that needs to be monitored.

[0063] Perform camera mapping based on abnormal event location data and monitoring equipment spatial distribution model to obtain camera list data; Specifically, the system generates camera list data based on the matching of event location data and the spatial distribution model of monitoring equipment. The system selects camera devices that cover the location of abnormal events; through the field of view and position relationship of the camera, further confirms the cameras with a high degree of overlap with the event area and adds them to the camera list. The camera list data includes the device numbers and properties of all devices used to monitor the event area.

[0064] Reconstruct the video scene geometry based on the camera list data to obtain a three-dimensional model of the monitoring area; Specifically, based on the camera list data, the system extracts real-time video frames from these devices and uses the camera position and orientation information to geometrically reconstruct the video scene. The geometric reconstruction process involves mapping the two-dimensional image captured by the camera into three-dimensional space to restore the spatial structure of the monitoring area. The reconstructed three-dimensional model of the monitoring area describes the terrain, buildings, and equipment layout within the event area, providing an accurate spatial reference for visual area analysis.

[0065] Extract the visible area of ​​abnormal events based on the three-dimensional model of the monitoring area to obtain key frame data of the visible area; Specifically, based on the 3D model of the monitoring area, the system extracts the visible area of ​​the abnormal event. By simulating the line of sight of each camera, the system calculates which monitoring devices can directly observe the area of ​​the abnormal event. The data of the visible area includes the relationship between the location of the event and the surrounding obstructions, and records the best viewing angle information of each camera. The system then extracts the key frames of the video from these viewing angles and marks the scenes related to the abnormal event.

[0066] The monitoring area boundary is generated for the key frame data of the visible area to obtain the first abnormal event area data.

[0067] Specifically, based on the extracted keyframe data of the visible area, the system generates the boundary of the first abnormal event area. By analyzing the visible range of different cameras, the system performs boundary fitting on the abnormal event area and generates an accurate spatial boundary. The boundary represents the monitoring range affected by the abnormal event, and combines the keyframe data to mark the abnormal features and important monitoring points in the area.

[0068] Preferably, the abnormal mode propagation path area locking is specifically: Based on the abnormal event data of the enterprise environment, Bayesian network modeling is carried out to obtain the event dynamic propagation model; Specifically, the system uses enterprise environmental abnormal event data to build a Bayesian network model that describes the dynamics of event propagation. The propagation dynamics of events are described by the relationship between nodes and edges, where nodes represent specific environmental states (such as increased temperature and increased smoke concentration) and edges represent the causal relationship between these states. Based on historical abnormal event data, the system analyzes the probability of each state affecting subsequent states. For example, a high temperature area causes abnormal equipment vibration, and abnormal vibration causes a fire. The output of the model is an event dynamic propagation model that shows how environmental anomalies expand in time and space.

[0069] Generate a spatiotemporal node network for the event dynamic propagation model to obtain event spatiotemporal node network data; Specifically, the system generates a spatiotemporal node network based on the event dynamic propagation model. Each node represents an abnormal state at a specific time and location (for example, a high temperature state at a certain point in time). The connection between nodes represents the propagation relationship between states. For example, the abnormality of a node will affect the surrounding nodes. The system analyzes the spatial location and time series characteristics of abnormal events, organizes these nodes according to time order and spatial distance, and generates event spatiotemporal node network data.

[0070] Calculate the probability of the propagation path based on the event space-time node network data to obtain the event propagation path probability data; Specifically, the system analyzes the propagation probability of each connection in the spatiotemporal node network. The historical data is used to evaluate the probability of an abnormal state propagating to neighboring nodes. For example, the probability of a high-temperature node propagating to a neighboring location depends on the range and time interval of the temperature change. By accumulating the propagation probabilities of all paths, event propagation path probability data is generated to identify the propagation direction and impact area.

[0071] Perform regional impact allocation on the event propagation path probability data to obtain event regional impact weight data; Specifically, the system distributes the impact to the enterprise area based on the probability data of the propagation path. The area is divided into multiple units (such as grids), and the system assigns an impact weight to each area unit through a linear calculation based on the location according to the probability value of the propagation path. The weight reflects the possibility that the area will be affected by the abnormal event. For example, the area unit closer to the high temperature node is assigned a higher weight.

[0072] Multi-region propagation fusion is performed based on the event area impact weight data to obtain multi-region propagation path data; Specifically, the system fuses the impact weights of each region to generate multi-region propagation path data. The fusion process combines the impact of different propagation paths on the region, superimposes the weights of all paths, and forms a unified regional impact distribution. The result is a multi-region propagation path data covering the entire enterprise, showing the main areas where abnormal events propagate.

[0073] Predicting the propagation path based on the multi-region propagation path data to obtain predicted propagation path data; Specifically, the system predicts future propagation paths based on multi-region propagation path data. The system analyzes the direction and speed of the current propagation path and uses the time-shifted average algorithm or the long-short time series convolutional network prediction algorithm to infer the range of spread of abnormal events in the future. The predicted propagation path data includes the areas that will be affected in the future and the estimated time when abnormal conditions will occur in these areas.

[0074] Reconstruct the propagation path area according to the predicted propagation path data to obtain the propagation path coverage area data; Specifically, the system uses the predicted propagation path data to reconstruct the propagation path coverage area. Based on the direction and impact range of the propagation path, the system generates a set of polygonal areas to represent the affected spatial range. Each coverage area is marked with a time node and predicted abnormal state for monitoring and response.

[0075] Region generation is performed based on the enterprise environment abnormal event data and the propagation path coverage area data to obtain second abnormal event area data.

[0076] Specifically, the system combines the initial abnormal event data and the propagation path coverage area to generate the second abnormal event area data. The system merges the initial abnormal event area with the propagation path coverage area to generate a comprehensive area, marking the key areas that need the most attention. The generated second abnormal event area data includes the center location, impact range, time distribution and propagation direction of the abnormal event.

[0077] Preferably, the environmental characteristic difference area locking is specifically: Performing spatiotemporal mapping of environmental characteristics based on enterprise environmental abnormal event data to obtain characteristic spatiotemporal distribution map data; Specifically, the system maps the environmental characteristics in time and space based on the enterprise environmental abnormal event data to generate a characteristic time and space distribution map. First, the environmental characteristic data such as temperature, humidity, smoke concentration, vibration frequency, etc. in the abnormal event area are extracted. These characteristic data are organized into a three-dimensional time and space distribution map according to the time series and spatial distribution. For example, the characteristic data at each time point corresponds to a specific grid area in the enterprise map, reflecting the change of characteristic values ​​over time and location. The characteristic time and space distribution map data fully describes the dynamic changes of environmental characteristics in the abnormal event area.

[0078] Generate a feature benchmark model based on the event environment feature spatiotemporal distribution map data to obtain a regional feature benchmark model; Specifically, the system generates a regional feature benchmark model by analyzing the feature spatiotemporal distribution map data. Based on historical data, the normal feature value range of each region is calculated, for example, the average value and fluctuation range of normal temperature, humidity and smoke concentration in a specific region. The generated feature benchmark model provides reference data under normal conditions for each grid area for deviation analysis.

[0079] Perform real-time feature deviation analysis based on the regional feature benchmark model and feature spatiotemporal distribution map data to obtain feature deviation matrix data; Specifically, the system compares the current feature data with the feature benchmark model and analyzes the real-time deviation. For each grid area, the deviation between the current feature value and the benchmark value is calculated, such as whether the temperature is higher than the normal range, whether the humidity has dropped abnormally, etc. The deviation analysis results are expressed in the form of a feature deviation matrix, where each matrix cell corresponds to a region and its feature deviation data, describing the environmental changes in the abnormal event area.

[0080] Perform difference region clustering according to feature deviation matrix data to obtain cluster region data; Specifically, based on the feature deviation matrix data, the system performs cluster analysis on the regions and identifies regions with similar feature changes. The system groups grids with similar feature deviations into a cluster region. For example, the region with both high temperature and high humidity is clustered into one unit. The data of each cluster region includes a list of the grids to which it belongs and a general description of the feature deviation. The output result is the cluster region data.

[0081] Conduct regional similarity analysis based on clustered regional data to obtain differential regional data; Specifically, the system performs similarity analysis on different clustered areas and identifies areas with significant differences. The system compares the characteristic deviations of each clustered area to identify which areas have significant differences in environmental characteristics, such as a certain area with significantly lower humidity than other areas. The output difference area data includes the locations of these areas with significant differences and their main characteristic deviations.

[0082] Refining the difference boundary of the difference area data to obtain optimized area boundary data; Specifically, the system refines the boundary of the difference region based on the difference region data. First, the preliminary boundary of the difference region is extracted, for example, a boundary polygon is generated based on the spatial distribution of feature deviations in the grid. Then, the spatial boundary of the difference region is optimized by smoothing the boundary or filling the gaps in the boundary to make it more consistent with the actual situation. The output optimized region boundary data describes the precise range of the difference region.

[0083] Region generation is performed based on the characteristic deviation matrix data and the optimized region boundary data to obtain the third abnormal event region data.

[0084] Specifically, the system combines the characteristic deviation matrix data and the optimized regional boundary data to generate the third abnormal event region data. The difference area is further integrated and annotated to form a complete abnormal event region, describing the spatial distribution and characteristic changes of the event. The third abnormal event region data includes the overall outline of the difference area, the internal characteristic distribution, and the deviation from the environmental baseline.

[0085] Preferably, the regional integration is specifically: Perform multi-region intersection calculation according to the first abnormal event region data, the second abnormal event region data, and the third abnormal event region data to obtain intersection region data; Specifically, the system calculates the intersection of multiple regions based on the first abnormal event region data, the second abnormal event region data, and the third abnormal event region data. The spatial overlapping parts of the three regions are identified, such as grids or location points included in multiple regions. The overlapping parts are defined as intersection region data. The intersection region reflects the common parts of all abnormal event regions and is the core impact area of ​​the event.

[0086] Assign regional weights according to the intersection area data to obtain weighted regional data; Specifically, based on the intersection area data, the system assigns weights to the area units. The weight reflects the importance of the area. For example, the importance of each unit in the intersection area can be determined by the degree of coverage of the three abnormal event areas. The system assigns a value to each unit. For example, a unit covered by three areas at the same time gets a higher weight, while a unit covered by only one area has a lower weight. After the weighted area data is output, it provides support for subsequent regional feature analysis.

[0087] Performing regional feature fusion according to weighted regional data to obtain fused regional feature data; Specifically, the system fuses the feature data in the region based on the weighted regional data. The system performs weighted averaging on the feature values ​​(such as temperature, humidity, smoke concentration, etc.) in the intersection region to ensure that the calculation result of each regional feature comprehensively considers the contribution of multiple regions to the feature. The fused feature data not only contains the common features of the abnormal region, but also retains the weight differences in the event description of different regions. The output fused regional feature data can be used to accurately describe the environmental characteristics of the comprehensive abnormal region.

[0088] Optimize the region boundary according to the fused region feature data to obtain first region data; Specifically, the system optimizes the boundaries of the fused region feature data. First, the boundary points with significant anomalies in the fused features are extracted and the initial boundaries are marked. Then, the boundaries are optimized by smoothing the boundary lines, completing the missing parts of the boundaries, or removing boundary noise points, so that they more accurately reflect the spatial distribution of the events. The first region data is generated, which represents the optimized fused region.

[0089] Perform feature extraction based on the first abnormal event region data, the second abnormal event region data, and the third abnormal event region data to obtain first abnormal event region feature data, second abnormal event region feature data, and third abnormal event region feature data, respectively; Specifically, the system extracts feature data from the first, second, and third abnormal event regions, respectively. The feature data of each region includes environmental variables such as the average temperature, humidity, vibration amplitude, and smoke concentration of the region. After extraction, feature data of the first abnormal event region, feature data of the second abnormal event region, and feature data of the third abnormal event region are generated, respectively, for further analysis.

[0090] Performing Gaussian clustering calculation on the first abnormal event regional feature data, the second abnormal event regional feature data, and the third abnormal event regional feature data to obtain regional feature Gaussian clustering data; Specifically, the system performs Gaussian clustering calculations on the feature data of the three regions to identify the distribution patterns of the features. By analyzing the distribution patterns of the feature data, the system clusters the grids with similar features in the region into one category. The generated regional feature Gaussian clustering data describes the similarities or differences between the features of different regions.

[0091] Calculate the degree of similarity and difference based on the regional characteristic Gaussian clustering data to obtain the regional characteristic similarity and difference degree data; Specifically, based on Gaussian clustering data, the system calculates the degree of similarity and difference of regional features. By comparing the center values ​​and distribution ranges of different regional clusters, the system identifies feature similarities and differences. The system outputs regional feature similarity and difference data to measure the differences in features between different abnormal regions.

[0092] Perform weighted calculation on the intersection area data according to the regional feature similarities and differences to obtain the second area data; Specifically, the system re-weights the intersection area data based on the regional feature similarities and differences. Areas with greater feature similarities and differences are given higher weights, indicating that these areas play a more important role in the impact of the event. The system outputs the second area data, which describes the re-weighted regional features.

[0093] Performing regional difference distribution calculation based on the first regional data and the second regional data to obtain regional difference distribution data; Specifically, the system calculates the difference distribution of the first region data and the second region data. By comparing the spatial distribution and feature distribution of the two regions, the system identifies which locations have significant differences in the two regions. The regional difference distribution data includes the difference value and the spatial location of the difference, providing a basis for regional compensation.

[0094] Regional compensation calculation is performed on the first regional data and the second regional data according to the regional difference distribution data to obtain abnormal event regional data.

[0095] Specifically, the system performs compensation calculations on the first region data and the second region data based on the regional difference distribution data to generate abnormal event region data. For regions with large differences, the system adjusts the feature values ​​or weights so that these regions can better reflect the true situation of the event. The generated abnormal event region data contains the core area, boundary information, and comprehensive feature description of the event, which can fully characterize the impact scope and nature of the abnormal event.

[0096] Preferably, step S4 is specifically: Step S41: performing multimodal feature fusion on the enterprise environment abnormal event data and the multi-angle video data to obtain multimodal data of enterprise abnormal events; Specifically, the system integrates the enterprise environmental abnormal event data with multi-angle video data to generate multimodal data of enterprise abnormal events. The system extracts environmental data related to abnormal events (such as temperature, humidity, and smoke concentration) and video data from multiple monitoring devices (such as key frames and moving target information). The environmental data and video data are aligned according to timestamps and spatial positions to ensure that the two types of data can jointly describe abnormal events in the same time and area. The system combines environmental features (such as temperature increase) with video features (such as smoke diffusion and flame appearance) to form a multimodal feature vector that describes the comprehensive characteristics of abnormal events. The output multimodal data contains a detailed description of time, space, environmental status, and video features.

[0097] Step S42: performing Bayesian risk assessment on the multimodal data of enterprise abnormal events to obtain preliminary enterprise abnormal event risk assessment data; Specifically, the system performs Bayesian risk assessment on the multimodal data of enterprise abnormal events and generates preliminary enterprise abnormal event risk assessment data. The characteristics of multimodal data are used as input for risk assessment, including environmental anomalies (such as high temperature, smoke) and visual anomalies (such as flames, personnel activities). By analyzing historical data and current features, the risks caused by abnormal events are assessed. For example, smoke diffusion and increased equipment vibration indicate fire risks. According to the risk assessment results, abnormal events are divided into three risk levels: low, medium, and high. The output preliminary risk assessment data includes the location, time, feature details and risk level of the event.

[0098] Step S43: performing event cascade analysis on the multimodal data of enterprise abnormal events according to the preliminary enterprise abnormal event risk assessment data to obtain event cascade graph data; Specifically, based on the preliminary risk assessment data, the system performs event cascade analysis on multimodal data and generates event cascade graph data. The system treats each abnormal event as a node, and the node attributes include event time, location, risk level, and feature information. The association between events is judged based on the time series and spatial distribution. For example, abnormal high temperatures and smoke in adjacent areas indicate that the events have a chain reaction. The connection between events is represented as a graph, in which the weight of each edge reflects the strength of the association between the two events. The output event cascade graph data shows the association structure between abnormal events, providing a basis for potential risk assessment.

[0099] Step S44: Perform potential risk assessment based on the event cascade graph data and the preset enterprise environment abnormal event assessment model to obtain secondary enterprise abnormal event risk assessment data; Specifically, the system evaluates potential risks and generates secondary enterprise abnormal event risk assessment data based on the event cascade graph data and the preset enterprise environment abnormal event assessment model (constructed by deep learning algorithm through historical data, or mathematical model construction through regression calculation of historical data). The system analyzes the propagation direction and scope of abnormal events based on the event connection relationship in the cascade graph. For example, high temperature extends to adjacent areas and causes equipment failure. Calculate the joint impact of multiple events, such as smoke diffusion and vibration abnormalities jointly causing a fire. Calculate the potential risk level for each event based on the propagation and cumulative effects of the event. The output secondary risk assessment data includes the location, time and potential risk level of the secondary event.

[0100] Step S45: Integrate the preliminary enterprise abnormal event risk assessment data and the secondary enterprise abnormal event risk assessment data to obtain abnormal event risk assessment data for performing intelligent security monitoring and response assistance operations for power generation enterprises.

[0101] Specifically, the system integrates the preliminary enterprise abnormal event risk assessment data and the secondary enterprise abnormal event risk assessment data to generate complete abnormal event risk assessment data. The system combines the direct risk of the preliminary risk assessment and the potential risk of the secondary assessment to form a comprehensive risk level. For each event, the system marks its direct risk, potential risk and comprehensive risk level to form a comprehensive risk description. The generated risk assessment data contains a comprehensive description of the event (location, time, characteristics, risk level), which is used to guide the intelligent security monitoring and response of power generation enterprises.

[0102] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0103] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An intelligent security monitoring and response method for a power generation enterprise, characterized in that: The following steps are involved: Step S1: Obtain environmental data of power generation enterprises; Step S2: Perform enterprise abnormal event detection on the environmental data of the power generation enterprise to obtain enterprise environmental abnormal event data; Step S3: Perform multi-source video linkage monitoring on abnormal event data of the enterprise environment to obtain multi-angle video data; Step S4: Analyze abnormal events based on abnormal event data of the enterprise environment and multi-angle video data to obtain abnormal event risk assessment data to perform intelligent security monitoring and response auxiliary operations for power generation enterprises; The event area locking in step S31 is specifically as follows: The video surveillance coverage area is locked according to the enterprise environment abnormal event data to obtain the first abnormal event area data; According to the enterprise environment abnormal event data, the abnormal mode propagation path area is locked to obtain the second abnormal event area data; According to the enterprise environmental abnormal event data, the environmental characteristic difference area is locked to obtain the third abnormal event area data; Performing regional fusion according to the first abnormal event region data, the second abnormal event region data, and the third abnormal event region data to obtain abnormal event region data; The specific locking of the abnormal mode propagation path area is as follows: Based on the abnormal event data of the enterprise environment, Bayesian network modeling is carried out to obtain the event dynamic propagation model; Generate a spatiotemporal node network for the event dynamic propagation model to obtain event spatiotemporal node network data; Calculate the probability of the propagation path based on the event space-time node network data to obtain the event propagation path probability data; Perform regional impact allocation on the event propagation path probability data to obtain event regional impact weight data; Multi-region propagation fusion is performed based on the event area impact weight data to obtain multi-region propagation path data; Predicting the propagation path based on the multi-region propagation path data to obtain predicted propagation path data; Reconstruct the propagation path area according to the predicted propagation path data to obtain the propagation path coverage area data; Region generation is performed based on the enterprise environment abnormal event data and the propagation path coverage area data to obtain second abnormal event area data.

2. The method according to claim 1, characterized in that: Step S1 is specifically as follows: Step S11: deploy temperature and humidity sensors, smoke detectors, and vibration detection equipment, initialize equipment parameters, complete network connection with the central monitoring system, and obtain sensor metadata of the power generation enterprise; Step S12: collecting sensor parameters according to the sensor metadata of the power generation enterprise to obtain preliminary sensor parameter data; Step S13: Performing quality verification on the preliminary sensing parameter data to obtain verification environment data; Step S14: Preliminarily filter the verification environment data to obtain filtered environment data; Step S15: Distribute and aggregate the filtered environment data to obtain local environment data; Step S16: Perform global sensor network mapping based on local environmental data to obtain preliminary power generation enterprise environmental data; Step S17: Perform hierarchical decoding on the preliminary power generation enterprise environmental data to obtain the power generation enterprise environmental data.

3. The method according to claim 2, characterized in that Step S2 is specifically as follows: Step S21: performing feature matching according to the environmental data of the power generation enterprise and the preset abnormal event recognition model to obtain preliminary abnormal event data; Step S22: performing complex event processing on the preliminary abnormal event data to obtain complex event data; Step S23: performing abnormal event determination on the complex event data to obtain abnormal event determination data; Step S24: Perform multi-stage verification and cross-check on the abnormal event determination event to obtain enterprise environment abnormal event data.

4. The method according to claim 1, characterized in that Step S3 is specifically as follows: Step S31: locking the event area according to the enterprise environment abnormal event data to obtain abnormal event area data; Step S32: activating the regional monitoring device according to the abnormal event regional data to obtain regional monitoring data; Step S33: collecting original video according to the regional monitoring data to obtain original video data; Step S34: Slicing the video according to the original video data to obtain video slice data, and performing time axis synchronization on the video slice data to obtain synchronized video segment data; Step S35: extracting video features from the synchronized video segment data to obtain video feature data, and performing abnormal picture recognition on the video feature data to obtain abnormal picture segment data; Step S36: annotate and classify the images according to the abnormal image segment data to obtain annotated video data; Step S37: construct a video network according to the monitoring location data corresponding to the regional monitoring data and the annotated video data to obtain multi-angle video data.

5. The method according to claim 1, characterized in that The specific areas of video surveillance coverage are as follows: Obtain basic data of monitoring equipment; Constructing a monitoring equipment spatial distribution model based on the basic data of the monitoring equipment to obtain a monitoring equipment spatial distribution model; Mapping the event location according to the abnormal event data of the enterprise environment to obtain the abnormal event location data; Perform camera mapping based on abnormal event location data and monitoring equipment spatial distribution model to obtain camera list data; Reconstruct the video scene geometry based on the camera list data to obtain a three-dimensional model of the monitoring area; Extract the visible area of ​​abnormal events based on the three-dimensional model of the monitoring area to obtain key frame data of the visible area; The monitoring area boundary is generated for the key frame data of the visible area to obtain the first abnormal event area data.

6. The method according to claim 1, characterized in that The specific areas of environmental feature difference locking are: Performing spatiotemporal mapping of environmental characteristics based on enterprise environmental abnormal event data to obtain characteristic spatiotemporal distribution map data; Generate a feature benchmark model based on the event environment feature spatiotemporal distribution map data to obtain a regional feature benchmark model; Perform real-time feature deviation analysis based on the regional feature benchmark model and feature spatiotemporal distribution map data to obtain feature deviation matrix data; Perform difference region clustering according to feature deviation matrix data to obtain cluster region data; Conduct regional similarity analysis based on clustered regional data to obtain differential regional data; Refining the difference boundary of the difference area data to obtain optimized area boundary data; Region generation is performed based on the characteristic deviation matrix data and the optimized region boundary data to obtain the third abnormal event region data.

7. The method according to claim 1, characterized in that The specific regional integration is as follows: Perform multi-region intersection calculation according to the first abnormal event region data, the second abnormal event region data, and the third abnormal event region data to obtain intersection region data; Assign regional weights according to the intersection area data to obtain weighted regional data; Performing regional feature fusion according to weighted regional data to obtain fused regional feature data; Optimize the region boundary according to the fused region feature data to obtain first region data; Perform feature extraction based on the first abnormal event region data, the second abnormal event region data, and the third abnormal event region data to obtain first abnormal event region feature data, second abnormal event region feature data, and third abnormal event region feature data, respectively; Performing Gaussian clustering calculation on the first abnormal event regional feature data, the second abnormal event regional feature data, and the third abnormal event regional feature data to obtain regional feature Gaussian clustering data; Calculate the degree of similarity and difference based on the regional characteristic Gaussian clustering data to obtain the regional characteristic similarity and difference degree data; Perform weighted calculation on the intersection area data according to the regional feature similarities and differences to obtain the second area data; Performing regional difference distribution calculation based on the first regional data and the second regional data to obtain regional difference distribution data; Regional compensation calculation is performed on the first regional data and the second regional data according to the regional difference distribution data to obtain abnormal event regional data.

8. The method according to claim 1, characterized in that Step S4 is specifically as follows: Step S41: performing multimodal feature fusion on the enterprise environment abnormal event data and the multi-angle video data to obtain multimodal data of enterprise abnormal events; Step S42: performing Bayesian risk assessment on the multimodal data of enterprise abnormal events to obtain preliminary enterprise abnormal event risk assessment data; Step S43: performing event cascade analysis on the multimodal data of enterprise abnormal events according to the preliminary enterprise abnormal event risk assessment data to obtain event cascade graph data; Step S44: Perform potential risk assessment based on the event cascade graph data and a preset enterprise environment abnormal event assessment model to obtain secondary enterprise abnormal event risk assessment data; Step S45: Integrate the preliminary enterprise abnormal event risk assessment data and the secondary enterprise abnormal event risk assessment data to obtain abnormal event risk assessment data for performing intelligent security monitoring and response assistance operations for power generation enterprises.

Images