A new energy station safe operation supervision method based on multi-dimensional data

Abstract

This invention discloses a method for safety operation and maintenance supervision of new energy power plants based on multidimensional data, specifically in the field of safety operation and maintenance supervision. It addresses the problems of strong concealment of biological activity, difficulty in identifying risk propagation paths, insufficient reliability of early warnings, and low efficiency in operation and maintenance resource scheduling in existing new energy power plants. By acquiring multi-source fragmented observation data of historical fault events, it extracts traces of biological activity, remnants, and equipment exposure features to construct weakly correlated feature vectors. Furthermore, it establishes a probabilistic relationship network by combining equipment spatial topology and power plant exposure paths, outputting the posterior distribution of equipment biological disturbance risks and the equipment chain of exposure propagation. Further, it constructs a dynamic and reliable mapping by combining historical biological activity early warning records, and generates operation and maintenance protection task initiation time windows and resource scheduling schemes based on a joint supervision cost model. This achieves dynamic supervision and collaborative resource scheduling of biological disturbance risks in new energy power plants, improving the efficiency of power plant safety operation and maintenance supervision.

Description

Technical Field

[0001] This invention relates to the field of safety operation and maintenance supervision technology, and more specifically, to a method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data. Background Technology

[0002] With the accelerating pace of large-scale construction of new energy power plants, photovoltaic power stations, wind farms, and energy storage stations are typically located in mountainous, desert, coastal, and vegetated areas. These stations contain a large number of devices spread over a wide area, and many cable trays, combiner boxes, inverters, transformer substations, and power transmission and distribution facilities operate in semi-open environments, making them susceptible to the activities of rodents, birds, and small wild animals. In actual operation, biological activity often spreads along exposed paths such as drainage ditches, cable channels, fence edges, and equipment maintenance access routes within the station, easily leading to problems such as cable gnawing, insulation damage, equipment short circuits, abnormal heat dissipation, and communication interruptions.

[0003] Due to the complex layout of equipment in new energy power stations, biological activities are characterized by nighttime concealment, regional migration, and intermittency. It is usually difficult to obtain complete biological activity process data directly on site. The operation and maintenance side relies more on scattered fault records for manual troubleshooting, which leads to a lag in judging the cause of the fault and difficulty in timely identifying the scope of risk propagation.

[0004] Meanwhile, the intensity of biological activity in different regions fluctuates dynamically with seasonal changes, climate change, and changes in site environmental conditions. Most existing operation and maintenance management methods adopt fixed-cycle inspections or simple alarm response methods, lacking the ability to jointly analyze the risk of biological activity transmission, the reliability of early warnings, and the relationship between equipment exposure. It is difficult to dynamically adjust the inspection sequence, personnel allocation, and material scheduling plan according to actual risk changes, which can easily lead to problems such as delayed response in high-risk areas, repeated investment of inspection resources, and low efficiency in scheduling protection tasks, thereby affecting the safe and stable operation of new energy power plants. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a new energy power station safety operation and maintenance supervision method based on multi-dimensional data to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for safety operation and maintenance supervision of new energy power plants based on multidimensional data includes the following steps: S1. Obtain multi-source fragmented observation data of historical fault events in new energy power stations, extract traces of biological activity, residues and equipment exposure features according to equipment location, and form a weakly correlated feature vector; S2. Using faults caused by biological activities as hidden nodes and weakly correlated feature vectors as explicit nodes, a probabilistic relationship network is constructed by combining equipment spatial topology and site exposure paths. The fault correlation strength between equipment is statistically analyzed and the conditional probability parameters are estimated. The posterior distribution of biological disturbance risk for each equipment and the exposure propagation equipment chain are output. S3. Statistically analyze the historical advance amount and corresponding actual situation of historical biological activity warning records, extract reliable change trends according to seasonal time period and station area, and fit to form a dynamic reliable mapping with reliability decreasing with advance amount. S4. Receive the latest biological activity early warning record and corresponding lead time, retrieve the dynamic reliable mapping and the current risk posterior distribution of each device, calculate the regulatory priority of each device in the exposure and transmission device chain, and generate a joint regulatory sequence. S5. Input the joint supervision sequence into the joint supervision cost model to solve for the operation and maintenance protection task start time window and resource scheduling scheme that minimizes the comprehensive supervision cost; S6. If a fault event occurs at the site within the preset period after the operation and maintenance protection task is executed, fragmented observations will be added and incorporated into the weakly correlated feature vector and the probabilistic relationship network parameters will be updated.

[0008] As a further aspect of the present invention, in S1, the formation of the weakly correlated feature vector specifically includes: Target detection is performed on operation and maintenance inspection images of historical fault events, and the coverage area, attachment location and distribution pattern of biological residues are extracted as residue association features; The vegetation coverage around the equipment, the distance from the water source, and the historical temperature and humidity during the same period were used as the equipment exposure characteristics. The infrared pyroelectric trigger pulse sequence with a preset time width is extracted centered on the time of the fault occurrence. The variation coefficient of the pulse interval and the trigger density are calculated to form biological activity trace features. Based on the matching of biological activity trace features, it is determined whether the historical fault events are caused by biological activities. The features associated with the remains, the equipment exposure features, and the traces of biological activity are concatenated into vectors to form a weakly correlated feature vector for the failure event.

[0009] As a further aspect of the present invention, in S2, the output of the posterior distribution of biological disturbance risk and the exposure propagation device chain for each device specifically includes: Faults caused by biological activity are defined as hidden nodes, and weakly associated feature vectors are defined as explicit nodes. Connection edges are constructed and the strength of associated edges are determined based on the spatial topological adjacency relationship between equipment, the exposed path connectivity relationship of the station, and the co-occurrence frequency of historical biological activities across equipment in the same period. Based on the historical fault handling records, the conditional probability relationship between hidden nodes and visible nodes is statistically analyzed, and the conditional probability parameters are estimated iteratively using the expectation-maximization algorithm to obtain the biological activity fault association probability corresponding to various weak correlation features. The current weakly correlated feature vectors of each device are input into the probabilistic relation network for parameter estimation. The biological activity risk propagation path is extracted by combining the correlation edge strength, the exposure propagation device chain is generated, and the posterior probability distribution of each device's failure caused by biological activity is output.

[0010] As a further aspect of the present invention, in step S3, fitting and forming a dynamic reliability mapping in which reliability decays with lead time specifically includes: Extract historical release records of biological activity early warning records, and associate the lead time of each early warning record with biological activity traces and remains in the target area within the corresponding time window to form early warning-real-time correlation data pairs; The associated warning-real-time data is hierarchically grouped by seasonal time period code and station area code. The seasonal time period code is taken from the quarter identifier to which the warning was issued, and the station area code is taken from the spatial grid index corresponding to the warning coverage area. Based on real-time correlation data, we determine whether the biological activity warning is accurate. Within each group, we calculate the ratio of forecasts that are accurate to the actual situation under different lead times. Using the lead time as the independent variable and the accuracy ratio as the dependent variable, we fit the reliability decay curve for each group. The confidence decay curves of all groups are stored by indexing seasonal time period codes and station area codes to form a dynamic confidence mapping. The mapping inputs are lead time, seasonal time period codes and station area codes, and the output is the early warning confidence value under the corresponding conditions.

[0011] As a further aspect of the present invention, in step S4, generating the joint regulatory sequence specifically includes: The latest biological activity warning received is analyzed, and the lead time, coverage spatial grid index, and seasonal time period code of the warning release time are extracted as query keys. The corresponding warning confidence value is retrieved from the dynamic confidence mapping. Using the coverage space grid index as the range, extract the devices and their posterior probability values ​​that fall within the corresponding range from the risk posterior distribution of each device, and screen the on-chain devices that fall within the corresponding range in the exposure propagation device chain as devices to be regulated; For each device in the set of monitored devices, the posterior probability value is multiplied by the early warning confidence value to obtain a monitoring priority score, and the devices are arranged in descending order of scores to generate a joint monitoring sequence.

[0012] As a further aspect of the present invention, in S5, the joint regulatory cost model is specifically a comprehensive regulatory cost function that includes delay cost, resource cost, and path cost.

[0013] As a further aspect of the present invention, in step S5, solving for the operation and maintenance protection task startup time window and resource scheduling scheme that minimizes the overall monitoring cost specifically includes: The delay cost is the sum of the regulatory priority scores of each device to be regulated in the joint regulatory sequence and the product of the time elapsed between the task start time and the warning reception time; Resource costs are the sum of the products of the manpower and hours required to perform the task and the material consumption and the corresponding unit price. The path cost is the product of the total actual movement path length formed by the inspection personnel in completing the inspection tasks of each piece of equipment to be monitored in the joint supervision sequence and the movement cost per unit distance. With the goal of minimizing the comprehensive supervision cost function, and with the constraints of the inspection completion time limit of the equipment to be supervised, the upper limit of available manpower hours, and the upper limit of the total amount of schedulable materials as the boundary conditions, the search is conducted to find the task start time window, inspection sequence, and manpower and material allocation scheme that satisfy all constraints and minimize the comprehensive supervision cost function. The output is the operation and maintenance protection task start time window and resource scheduling scheme.

[0014] As a further aspect of the present invention, in step S6, updating the probabilistic relationship network parameters specifically includes: If a fault occurs in the equipment at the site within a preset period after the task is executed, the corresponding multi-source fragmented observation data is obtained, weak correlation feature vectors are extracted as new samples, the new samples are merged into the existing weak correlation feature vector set, and the conditional probability parameters between hidden nodes and explicit nodes in the probability relationship network are incrementally corrected.

[0015] The technical effects and advantages of the new energy power station safety operation and maintenance supervision method based on multi-dimensional data of the present invention are as follows: This invention constructs a probabilistic relationship network based on the exposure path of a new energy power station by analyzing the correlation between traces of biological activity, characteristics of residual materials, and equipment exposure features in historical failure events. This enables dynamic inference and propagation path identification of equipment failure risks caused by biological activity, improving the accuracy of biological disturbance risk identification in complex power station environments. Furthermore, by establishing a dynamic reliability mapping where reliability changes with the lead time of the warning, differentiated assessment of the reliability of biological activity warnings is achieved for different seasons and different power station areas, avoiding the problem of inaccurate warning responses under fixed rules. Based on this, the risk posterior distribution, exposure propagation equipment chain, and warning reliability are jointly monitored and ranked. Combined with inspection time limits, manpower hours, and material scheduling constraints, comprehensive regulatory cost optimization is performed, improving resource utilization efficiency and task response efficiency in operation and maintenance protection tasks, reducing redundant inspections, path redundancy, and response delays in high-risk areas, thereby enhancing the safety operation and maintenance supervision capabilities of new energy power stations. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a new energy power station safety operation and maintenance supervision method based on multi-dimensional data according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1 Figure 1 This invention presents a method for safety operation and maintenance supervision of new energy power stations based on multidimensional data, which includes the following steps: S1. Obtain multi-source fragmented observation data of historical fault events in new energy power stations, extract traces of biological activity, residues and equipment exposure features according to equipment location, and form a weakly correlated feature vector; S2. Using faults caused by biological activities as hidden nodes and weakly correlated feature vectors as explicit nodes, a probabilistic relationship network is constructed by combining equipment spatial topology and site exposure paths. The fault correlation strength between equipment is statistically analyzed and the conditional probability parameters are estimated. The posterior distribution of biological disturbance risk for each equipment and the exposure propagation equipment chain are output. S3. Statistically analyze the historical advance amount and corresponding actual situation of historical biological activity warning records, extract reliable change trends according to seasonal time period and station area, and fit to form a dynamic reliable mapping with reliability decreasing with advance amount. S4. Receive the latest biological activity early warning record and corresponding lead time, retrieve the dynamic reliable mapping and the current risk posterior distribution of each device, calculate the regulatory priority of each device in the exposure and transmission device chain, and generate a joint regulatory sequence. S5. Input the joint supervision sequence into the joint supervision cost model to solve for the operation and maintenance protection task start time window and resource scheduling scheme that minimizes the comprehensive supervision cost; S6. If a fault event occurs at the site within the preset period after the operation and maintenance protection task is executed, fragmented observations will be added and incorporated into the weakly correlated feature vector and the probabilistic relationship network parameters will be updated.

[0019] In S1, a weakly correlated feature vector is formed.

[0020] When performing target detection on maintenance and inspection images of historical fault events, visible light and infrared images generated during historical inspections of the site are first retrieved and organized according to the equipment number, shooting time, and shooting location of the faulty equipment. For the same fault event, inspection images within 24 hours before and after the fault are prioritized for analysis. For images showing low light at night, backlighting, or obstruction by rain or snow, image quality screening is first performed to remove images with large areas of blur, overexposure, or target areas obscured by more than a set proportion. This set proportion is determined based on the statistical results of historical valid inspection images; in this embodiment, the obscured area of ​​the main equipment outline is no more than 30% of the overall equipment outline area. Subsequently, the main equipment area in the retained images is located, and the corresponding target detection model is called according to the equipment type. Different equipment structure templates are used for detection of transformer substations, combiner boxes, and inverters to avoid errors in identifying the location of remaining objects due to differences in equipment structures. During target detection, feathers, rodent droppings, chewed debris, and nest remains in the images were identified and statistically analyzed according to the surface area of ​​the equipment. The equipment's heat dissipation holes, cable interfaces, bottom support areas, and top cover edges were divided into different attachment areas. For the identified residues, their coverage area on the equipment surface was calculated, and distribution morphology features were extracted based on the degree of aggregation of residues in each area. When the same type of residue appeared consecutively in the cable interface area and the bottom area of ​​the equipment, it was determined that the equipment had a continuous biological activity path. Subsequently, the site's geographic information records were retrieved to obtain vegetation coverage information within a 50-meter radius of the equipment. This information, combined with the actual distance between the equipment and drainage ditches, pools, or site water storage facilities, formed the equipment exposure characteristics. Vegetation coverage was obtained by statistically analyzing the proportion of green pixels in UAV inspection images, and the distance to water sources was calculated using coordinates from the site's GIS deployment. For historical temperature and humidity data, environmental monitoring records from the site for the fifteen days before and after the month of the fault were retrieved. Temperature and humidity variation curves were generated based on hourly averages, and the duration of high humidity periods at night was recorded. If the nighttime air humidity continuously exceeded 85% for more than four hours, it was marked as a high-humidity exposure environment. Finally, the characteristics of the residual materials and the equipment exposure characteristics were uniformly converted into standardized numerical descriptions to form static environmental correlation data for the corresponding fault event.

[0021] When extracting an infrared pyroelectric trigger pulse sequence with a preset time width centered on the time of the fault occurrence, trigger logs of infrared pyroelectric sensors around the faulty equipment are retrieved for the corresponding time periods before and after the fault, and the complete trigger sequence is restored in chronological order. In this embodiment, the preset time width is determined based on the statistical results of historical biological activity duration, preferably from six hours before the fault to two hours after the fault as the analysis window to ensure coverage of the complete process of biological activity approaching, staying, and leaving the equipment. Subsequently, the time intervals of continuous pulses in the trigger log are statistically analyzed. When the time interval between adjacent pulses is consistently less than a set time threshold, it is determined that there is continuous activity behavior. The time threshold is determined based on the statistical data of normal environmental disturbances at the site at night. The interval variation coefficient is further calculated for the pulse interval sequence. When the interval change shows a continuous low fluctuation state, it is determined that the active target has stable staying characteristics; when the interval change fluctuates frequently and is accompanied by high-density continuous triggering, it is determined that there is rapid movement or multi-target activity. For the trigger density, it is statistically analyzed according to the number of effective triggers per unit time, and compared with the average trigger level of historical normal periods near the faulty equipment. When the current trigger density exceeds twice the historical average, it is marked as an abnormal biological activity state. Subsequently, the obtained pulse interval variation coefficient and trigger density are combined to form biological activity trace features, and matching rules are established based on historically confirmed biological activity fault samples. Specifically, the biological activity trace features in historical fault samples are classified and stored according to three behavioral patterns: rodent activity, bird loitering, and small animal climbing. Rodent activity typically corresponds to continuous low-interval triggering features at night, bird loitering typically corresponds to intermittent triggering features during the day, and small animal climbing corresponds to short-term high-frequency triggering features. For the current fault event, its biological activity trace features are compared with the historical behavioral pattern database class by class. When the matching degree between the current feature and a certain historical behavioral pattern exceeds a set matching ratio, the fault event is determined to be a biological activity-induced fault. In this embodiment, the matching ratio is determined based on the statistics of historically confirmed samples, preferably 70% as the matching judgment threshold. After the fault type is determined, the residue association features, equipment exposure features, and biological activity trace features are concatenated into vectors in a unified order, with the residue association features located at the beginning of the vector, the equipment exposure features in the middle of the vector, and the biological activity trace features at the end of the vector, to form a structurally unified weakly associated feature vector.

[0022] In S2, the posterior distribution of biological disturbance risk for each device and the exposure propagation device chain are output.

[0023] After defining faults caused by biological activity as hidden nodes and the generated weakly correlated feature vectors as explicit nodes, the spatial topological relationships between equipment within the site are restored based on the site equipment layout diagram. For equipment continuously connected within the same electrical area via cable trays, cable trenches, fence edges, or maintenance passages, spatial topological adjacency is determined. Further, exposed path connectivity is extracted. Specifically, the site's internal cable tray routing diagram, drainage ditch distribution diagram, fence structure diagram, and equipment maintenance passage layout diagram are retrieved. Locations that can form continuous biological movement paths are defined as exposed path nodes, and exposed path connectivity is established based on whether there are continuous, unobstructed passage areas between exposed path nodes. For example, when two junction boxes are continuously connected via the same cable tray, and there is no enclosed isolation structure in the middle of the cable tray, an exposed path connectivity relationship is determined between the two junction boxes; when equipment is only spatially close but isolated by an enclosed fence, no exposed path connectivity relationship is established. Subsequently, the frequency of cross-device co-occurrence of historical biological activities within the same time period was statistically analyzed. Using a 30-minute time window, historical infrared pyroelectric trigger records and inspection trace records were traversed. When adjacent devices both exhibited abnormal triggering or residue records within the same time window, a cross-device co-occurrence event was counted. For device pairs with frequent continuous co-occurrence, the strength of the corresponding association edge was increased. The determination of the association edge strength did not employ a simple reciprocal distance method, but rather a comprehensive assessment of the actual movement path length between devices, the continuity of exposed paths, and historical co-occurrence frequency. The actual movement path length was obtained statistically based on the actual path lengths of cable trays, cable trenches, and maintenance passages between devices. The continuity of exposed paths was graded based on the number of path interruptions, the number of enclosed isolation structures, and the degree of nighttime lighting obstruction. In this embodiment, uninterrupted and unobstructed paths were defined as highly connected paths, paths with only a single isolation point were defined as moderately connected paths, and paths with multiple isolation structures were defined as low-connectivity paths. The historical co-occurrence frequency was normalized based on the cumulative co-occurrence count in historical records. After completing the construction of the associated edges, the device nodes, hidden nodes, and visible nodes are mapped together into the probabilistic relationship network. The device nodes form a propagation relationship through the associated edges, and the visible nodes and hidden nodes form a risk inference relationship through the conditional probability relationship, so as to construct a probabilistic relationship network structure that conforms to the propagation characteristics of biological activities in new energy power plants.

[0024] When analyzing the conditional probability relationship between hidden and visible nodes based on historical fault handling records, historically completed fault work orders are retrieved, and fault events confirmed by on-site maintenance personnel to be caused by biological activity are selected. For each fault record, the corresponding weakly correlated feature vector is read, and the frequency of different biological activity trace features, residue association features, and equipment exposure features in biological activity faults is statistically analyzed. For example, when rodent excrement is concentrated at the bottom of the equipment and accompanied by continuous high-frequency pyroelectric triggering at night, the corresponding feature combination is marked as an associated feature. Subsequently, the expectation-maximization algorithm is used to iteratively estimate the conditional probability parameters between hidden and visible nodes, where the input for visible nodes is the weakly correlated feature vector, and the output for hidden nodes is whether the fault is caused by biological activity. In specific implementation, the conditional probability parameters are initialized based on historically confirmed fault samples, and the probability of the hidden node state for unconfirmed fault samples is estimated based on the current parameters. The estimation result is used as the hidden state distribution. Then, the conditional probability relationship between visible and hidden nodes is re-analyzed, the parameter values ​​are updated, and the hidden state estimation is performed again. The expectation-maximization algorithm employs a combination of fixed iteration rounds and convergence criteria. Iteration stops when the parameter change is less than 5% of the historical statistical mean for two consecutive rounds. If the convergence condition is not met, the next round of parameter estimation continues, with a maximum of twenty iterations. After completing the iterations, the probability of biological activity triggering fault associations corresponding to different weakly correlated features is obtained.

[0025] After inputting the current weakly correlated feature vectors of each device into the probabilistic relationship network for parameter estimation, the posterior probability distribution of biologically-induced failures for each device is calculated based on the associated features of the remnants, traces of biological activity, and exposure features of the device. Device nodes with posterior probabilities continuously higher than a set risk threshold are defined as high-risk starting nodes. The risk threshold is determined statistically based on historically confirmed biological activity failure samples, preferably using the mean posterior probability of historical failure samples. Subsequently, starting from the high-risk starting node, adjacent device nodes are searched level by level along associated edges with strength higher than the propagation threshold. The propagation threshold is determined statistically based on historical cross-device biological activity diffusion records, selecting the median of the historical effective propagation edge strength distribution as the propagation threshold. During the search, not only are adjacent devices judged to meet the edge strength condition, but also the existence of continuous exposure paths between devices and whether the current time period belongs to a biological activity period. When the above conditions are met, adjacent devices are added to the propagation path, and the search continues; when the associated edge strength is lower than the propagation threshold, the exposure path is interrupted, or the posterior probability of the device is continuously lower than the risk threshold, the current direction of expansion is stopped. When multiple branches appear within the same propagation path, the main propagation path is retained according to the strength of the associated edges from high to low, and secondary propagation paths are recorded as auxiliary propagation branches. Finally, the continuous sequence of device nodes obtained from the search is defined as an exposed propagation device chain; for example, "combiner box—cable tray interface—inverter—distribution cabinet" constitutes a complete propagation device chain. For each propagation device chain, its chain length, propagation direction, and the number of high-risk devices are further recorded, and the posterior probability distribution of biological activity faults corresponding to each device is output.

[0026] In S3, a dynamic reliability mapping is fitted to form a reliability that decreases with lead time.

[0027] When extracting historical release records of biological activity early warnings, the historical early warning logs of new energy power plants are retrieved, and the early warning release time, coverage area, duration, and source recorded in the logs are uniformly organized. In this embodiment, biological activity early warning records include joint early warning information formed by perimeter infrared monitoring equipment, nighttime video analysis equipment, and regional pyroelectric arrays. For early warning events triggered by multiple monitoring sources within the same time period, they are merged according to the principle of temporal proximity. When the release time interval between multiple early warning records does not exceed ten minutes and the coverage areas overlap continuously, they are merged into the same early warning event to avoid duplicate statistics that could lead to a shift in reliability. Subsequently, the lead time for each early warning record is calculated, using the time interval from the early warning release time to the first appearance of valid biological activity traces in the corresponding area as the lead time. The confirmation of valid biological activity traces does not solely rely on a single pyroelectric trigger, but simultaneously meets three conditions: abnormal enhancement of the infrared pyroelectric pulse, the presence of biological residues in the inspection image, and the appearance of continuous activity trajectories around the equipment in the corresponding area. Cases with only a single abnormal trigger but no subsequent residue records are not included in the statistics of valid biological activity events. Subsequently, each early warning record is associated with biological activity traces and remains in the target area within the corresponding time window, forming an early warning-real-time correlation data pair. In practice, starting from the early warning issuance time, inspection images, pyroelectric triggering records, and on-site maintenance records within the corresponding lead time are retrieved within the target area, and matched according to spatial grid positions. A valid spatial correlation is determined when the spatial deviation between the early warning coverage area and the actual biological activity area does not exceed one station grid unit. The station area grid is formed based on equipment deployment density, with each spatial grid covering adjacent equipment groups and corresponding cable tray passage areas. After completing the early warning-real-time correlation, the correlation data is hierarchically grouped according to seasonal time period codes and station area codes. The seasonal time period code is determined based on the quarter to which the early warning was issued; for example, March to May is defined as the first quarter activity period, and June to August as the second quarter activity period. The station area code is determined based on the spatial grid index corresponding to the early warning coverage area. For station grids with mountain obstructions, proximity to water sources, or dense vegetation, separate independent area codes are used to avoid distortion of reliable trends due to mixed statistics of different environmental conditions. Ultimately, a warning-real-time correlation dataset was formed, which is managed hierarchically based on time characteristics and regional exposure characteristics.

[0028] When determining whether a biological activity warning has been predicted based on real-time correlation data, the real-time correlation data corresponding to each warning record is read to determine whether there are actually persistent traces of biological activity in the target area within the warning lead time coverage period. The prediction hit determination is performed using a joint matching method. A warning record is considered a prediction hit when the target area simultaneously meets the following conditions: First, the pyroelectric pulse trigger density is higher than the historical normal value for the same period; second, there are clear records of attached debris in the inspection images; third, there are activity traces around the equipment in two or more consecutive inspection cycles. Cases with only short-term, single-time abnormal triggers or isolated debris records are not considered prediction hits to avoid the influence of occasional events such as wind-blown debris or minor environmental disturbances on the statistical results. Subsequently, the prediction hit ratio is calculated for different lead times within each group. In practice, the lead time is discretely divided into fixed time intervals, for example, multiple lead time segments are formed with a 30-minute statistical interval. The total number of warning records and the number of hit records in each segment are counted, and the corresponding hit ratio is calculated. When the number of valid samples within a certain lead time interval is lower than the set sample lower limit, a separate credibility statistical result is not generated. Instead, it is merged into the statistics of adjacent lead time intervals to avoid abnormal credibility fluctuations caused by small samples. The sample lower limit is determined based on the number of historical warning records at the airfield, selecting no fewer than fifty groups of valid warning records. Subsequently, the credibility change trend under each group is fitted with the lead time as the independent variable and the hit rate as the dependent variable. During the fitting process, a single linear change method is not adopted; instead, piecewise change curves are formed based on the actual statistical distribution of different regions and seasons. For example, in high-temperature and high-humidity areas in summer, the warning credibility will remain relatively stable within a short lead time range, while in low-temperature areas in winter, the credibility will decrease significantly with the increase of the lead time. For statistical intervals with local abnormal fluctuations, a smoothing correction method for adjacent time intervals is used to avoid abrupt changes in the credibility curve caused by a single abnormal event. Finally, credibility decay curves corresponding to different seasonal periods and different airfield regions are generated.

[0029] When storing the confidence decay curves formed by all groups according to seasonal time period codes and station area codes, the confidence curves corresponding to each group are uniformly numbered, and a mapping relationship between the area index and the time index is established. The seasonal time period code serves as the primary index, and the station area code serves as the secondary index. Each season and the same area corresponds to a set of independent confidence decay curves. For situations where multiple different exposure environments exist within the same station area, such as densely populated cable tray areas and fence edge areas, sub-area indexes are further divided based on differences in historical biological activity frequencies to ensure that the confidence mapping reflects local environmental differences. Subsequently, the confidence values ​​corresponding to each lead time segment in the confidence decay curve are stored discretely, with the confidence values ​​directly generated from historical hit ratio statistics. For areas with no new warning records for a long period, the most recent valid statistical result is retained as the current confidence mapping; when the number of consecutively added valid warning records in an area reaches a set update quantity, the confidence curve fitting for the corresponding area is re-executed. In this embodiment, the update quantity is determined based on the historical warning frequency of the station, and an area curve update is performed after twenty consecutively added valid warning records. This ultimately forms a dynamic reliability mapping, where the input includes lead time, seasonal time period code, and station area code, and the output is the warning reliability value under the corresponding conditions. For example, when the input is "second quarter of summer, dense bridge area, lead time of two hours", the warning reliability result under the corresponding statistical conditions is directly output.

[0030] In step S4, a joint regulatory sequence is generated.

[0031] When parsing the latest received biological activity warning, the release time, coverage area, and duration of the warning are read from the warning record, and the corresponding lead time is calculated based on the warning release time. Specifically, the current warning release time is compared with the statistical results of the first occurrence time of historical biological activity in the corresponding area, and the time difference between the two is used as the current warning lead time. Subsequently, the spatial grid index corresponding to the warning coverage area is read, and the corresponding seasonal time period code is determined according to the month to which the warning release time belongs. In this embodiment, the site spatial grid is formed according to the equipment deployment density and the cable tray connection area. Each grid covers the corresponding equipment group, cable tray, and adjacent maintenance passage area to ensure that the spatial grid can reflect the actual movement range of biological activity. For warning records that cover multiple spatial grids simultaneously, the equipment distribution in each spatial grid is statistically analyzed, and the primary and secondary coverage grids are determined according to the proportion of the warning coverage area. Subsequently, the corresponding warning confidence value is retrieved from the dynamic confidence mapping using the lead time, spatial grid index, and seasonal time period code as query keys. During the retrieval process, spatial grid indexes consistent with the current warning coverage area are prioritized for matching. When the current grid lacks sufficient historical records, statistical results from adjacent areas with cable tray connectivity or continuous fence pathways to the current area are sequentially retrieved to supplement the results, thus avoiding credibility distortion due to insufficient historical samples in local areas. In this embodiment, when the number of valid historical warning records in the target area is less than twenty, adjacent exposure path connected areas are automatically merged to form a credibility statistical result. After completing the credibility retrieval, the devices located within the corresponding area and their corresponding posterior probability values ​​are extracted from the risk posterior distribution of each device, using the current warning coverage spatial grid as the scope. Subsequently, the exposure propagation device chain is further read, and devices on the chain whose propagation paths pass through the current coverage spatial grid are screened. In specific implementation, the judgment is not simply based on the physical coordinates of the devices, but also simultaneously on whether the devices are located in the same exposure propagation path. For example, when the current warning area is located in the middle section of the cable tray, in addition to extracting devices within the current area, devices on the chain with continuous exposure path relationships to the current area in the upstream and downstream paths of the cable tray are also extracted simultaneously to reflect the propagation characteristics of biological activity moving continuously along the cable tray, cable trench, and fence. Devices that are only spatially adjacent but lack exposed path connectivity are excluded from the set of devices to be monitored, thus preventing a large number of invalid devices from entering the monitoring sequence. Further path continuity checks are performed on the exposed propagation device chain. When there are closed gaps, equipment modification breaks, or cable tray blockage records between adjacent devices in the chain, link expansion in the corresponding direction is stopped, and only device nodes in the continuous exposed path are retained. Finally, devices that meet the spatial coverage conditions and are located in the continuous exposed propagation path are identified as devices to be monitored, and the path position order of each device in the exposed propagation device chain, the corresponding posterior probability value, and the spatial grid information to which it belongs are recorded.

[0032] When calculating the regulatory priority score for each device in the set of monitored devices, the posterior probability value and warning confidence value of the corresponding device are read. The posterior probability value characterizes the likelihood of the device's current failure caused by biological activity, while the warning confidence value characterizes the historical effectiveness of the current warning under the corresponding seasonal and regional conditions. Subsequently, both types of data undergo unified dimensional processing, specifically converting both the posterior probability value and the warning confidence value into standardized values ​​between zero and one. When the device's posterior probability value is lower than the lower limit of the historical risk distribution, it is uniformly corrected to the historical minimum effective risk value to prevent devices with extremely low probability from completely losing their regulatory ranking ability. When the warning confidence value is higher than the historical statistical upper limit, it is truncated according to the historical maximum confidence value to prevent abnormal statistical values ​​from causing an imbalance in regulatory priority. Then, the standardized posterior probability value and the warning confidence value are multiplied to calculate the regulatory priority score. For devices located at the beginning of the exposure propagation device chain, the distinction between the beginning and end is 50%, and further path position correction is performed. The positional order of the device in the corresponding propagation path is statistically analyzed. When the device is within the first third of the propagation chain, a path propagation correction coefficient is added to the regulatory priority score. In this embodiment, the path propagation correction coefficient is determined based on historical propagation chain diffusion statistics, selecting a correction range between 1.1 and 1.3; no additional correction is applied when the device is located at the tail end of the propagation chain. After correction, the regulatory priority scores of all devices to be monitored are sorted according to their numerical values, and devices preceding the propagation chain are retained first when scores are the same. Furthermore, for multiple devices with similar scores within the same cable tray area, the corresponding devices are grouped into a continuous monitoring task group to avoid repeated path travel during inspection. Finally, a joint monitoring sequence is formed according to the regulatory priority scores from high to low, where devices at the beginning of the sequence are given priority to enter the operation and maintenance protection task scheduling scope, while devices at the tail end of the sequence are monitored later based on current resource conditions. For each device in the joint monitoring sequence, the corresponding early warning confidence value, posterior probability value, propagation chain position, and spatial grid information are recorded simultaneously.

[0033] In S5, the optimal time window and resource scheduling scheme for starting maintenance and protection tasks with the lowest overall regulatory cost are determined.

[0034] When constructing the joint regulatory cost model, the system reads the regulatory priority score, spatial grid location, exposure propagation device chain position, and corresponding early warning reception time for each device to be regulated in the joint regulatory sequence. Simultaneously, it reads the current on-duty location of inspection personnel, the inventory status of protective materials, and the road access status within the site. Subsequently, the delay cost, resource cost, and path cost in the comprehensive regulatory cost function are calculated. The delay cost characterizes the degree of regulatory risk accumulation caused by task initiation delays for the devices to be regulated. In practice, for each device to be regulated in the joint regulatory sequence, the time interval between the task initiation time and the corresponding early warning reception time is calculated, and this time interval is combined with the device's regulatory priority score to form a delay cost. In this embodiment, risk is not directly judged based on absolute time length, but rather on differentiated delay accumulation based on differences in device regulatory priorities. For example, for devices located at the beginning of the exposure propagation device chain with a high regulatory priority score, the delay cost increases significantly faster than for ordinary devices as the task initiation delay continues to increase; for devices located at the end of the propagation chain with a low score, the delay cost increases relatively slowly. To avoid overall scheduling imbalance caused by a single abnormal score, a range correction process is applied to the regulatory priority score. When a score continuously exceeds the historical risk distribution upper limit, it is truncated according to the historical maximum effective score. Resource cost is used to characterize the manpower and material consumption generated during the execution of regulatory tasks. Specifically, the inspection man-hour requirements corresponding to each regulatory task are statistically analyzed, including equipment disassembly and inspection time, protective device installation time, and on-site verification time, and the actual available man-hours are determined in conjunction with the current duty personnel's shift records. Subsequently, the consumption of materials such as protective netting, avoidance devices, insulation covering materials, and cable tray sealing materials required for task execution is statistically analyzed, and the corresponding unit price is read from the site maintenance material procurement records. In this embodiment, the manpower man-hour unit price is determined based on the site's historical maintenance labor cost statistics, and the material unit price is determined based on the most recent centralized procurement settlement record to avoid scheduling instability caused by temporary market price fluctuations. Path cost is used to characterize the movement costs generated during the inspection process. Specifically, based on the spatial grid location of equipment in the joint supervision sequence and the internal road traffic relationships of the site, the actual movement path length formed by inspection personnel when sequentially completing corresponding supervision tasks is calculated, and the path cost is formed by combining the unit distance movement cost. The actual movement path is not directly calculated as the straight-line distance between equipment, but rather based on the actual inspection routes within the site, including cable tray passages, maintenance roads, and fenced detour areas. For areas that are inaccessible due to site construction closures, nighttime blockades, or equipment maintenance obstructing traffic, detour routes are automatically replanned, and the corresponding path length is recalculated. In this embodiment, the unit distance movement cost is determined jointly based on historical inspection vehicle fuel consumption records and manual movement time statistics.When multiple devices located in the same exposure and propagation equipment chain exist consecutively along the inspection path, priority is given to maintaining continuous inspection of the path to reduce the accumulation of path costs caused by repeated detours. Ultimately, the delay costs, resource costs, and path costs are summed together to form a comprehensive regulatory cost function value.

[0035] When scheduling maintenance and protection tasks with the goal of minimizing the comprehensive regulatory cost function, the inspection completion time limit for each device to be regulated in the joint regulatory sequence is read. This inspection completion time limit is determined based on the device's regulatory priority score, its position in the exposure and propagation device chain, and the current warning reliability. Subsequently, the total available inspection time limit for the current site is calculated, and the remaining available time is calculated based on the current shift schedule, the duration of tasks already performed by personnel, and the nighttime continuous operation restriction rules. Personnel who have continuously performed high-intensity inspection tasks for more than the prescribed duration are no longer included in the current scheduling scope. In this embodiment, the nighttime continuous inspection duration is determined based on the site's historical safety operation system, preferably taking six consecutive hours as the upper limit for a single high-intensity inspection. Furthermore, the total amount of dispatchable materials in the current site is calculated, including the inventory of avoidance equipment, protective materials, and mobile lighting equipment, excluding inventory already locked for other regulatory tasks. After completing the constraint statistics, multiple candidate scheduling schemes are formed according to different task start time windows, inspection sequences, and manpower and material allocation methods. For each candidate scheme, the corresponding comprehensive regulatory cost function value is calculated, and each item is verified to meet the constraints of inspection completion time limit, manpower hour limit, and total material quantity. If a candidate scheme has equipment exceeding the inspection completion time limit, manpower hour exceeding the limit, or material allocation exceeding the limit, the corresponding scheme is directly eliminated. For candidate schemes that meet all constraints, the comprehensive regulatory cost function values ​​are further compared, and the scheduling scheme with lower delay cost and better path continuity is prioritized. For example, when the comprehensive costs of two schemes are close, the scheme that can continuously complete the inspection of the front-end equipment of the same exposure propagation equipment chain is selected first to reduce frequent cross-regional movement during the inspection process. Finally, the candidate scheme with the smallest comprehensive regulatory cost function value is determined as the target operation and maintenance protection task scheme, and the corresponding task start time window, inspection sequence, and manpower and material allocation results are output. Among them, the task start time window specifically records the time range within which each regulatory task is allowed to start execution, the inspection sequence records the order of equipment inspection, and the manpower and material allocation results record the task area and corresponding material collection information of each inspection personnel for actual operation and maintenance supervision execution at the site.

[0036] In step S6, the parameters of the probability relationship network are updated.

[0037] If a fault event still occurs in the equipment within the site within a preset period after the completion of the operation and maintenance protection task, multi-source fragmented observation data from the corresponding fault event period will be retrieved. The preset period is determined based on the statistical results of the historical biological activity duration at the site, selecting 24 to 72 hours after the completion of the operation and maintenance task as the fault tracking period to ensure coverage of the complete process of biological activity re-spreading or re-entering the equipment area within the site. Subsequently, according to the spatial grid location of the faulty equipment, inspection images, infrared pyroelectric pulse records, on-site maintenance records, cable tray channel status records, and environmental temperature and humidity change records before and after the fault are extracted and aligned according to a unified timeline. For inspection images, image sequences from six hours before the fault to two hours after the fault are extracted first, and the surface condition changes of the corresponding equipment are restored according to the equipment number and shooting area. For pyroelectric pulse records, continuous trigger sequences within the fault area are restored, and the changes in trigger frequency before and after the fault are statistically analyzed. Furthermore, on-site verification is performed on newly added fault events to confirm whether the fault is caused by biological activity. In practice, on-site maintenance personnel manually confirm the issue based on residues on the equipment surface, the location of insulation damage, cable chewing marks, and activity traces inside the cable tray. This confirmation is then combined with anomalies in pyroelectric pulses within the corresponding time period to form the final confirmation result. Fault events with only occasional mechanical damage or construction disturbance are not included in the new sample statistics. A new biological activity fault event is only identified when all three conditions are met: the presence of biological activity traces, the location of equipment damage matching historical biological activity areas, and the existence of abnormal pyroelectric triggering records within the corresponding time period. After confirmation, the residue association features, equipment exposure features, and biological activity trace features corresponding to the new fault event are re-extracted according to existing weak correlation feature vector construction rules. Residue association features include the residue attachment area, coverage area, and distribution continuity; equipment exposure features include vegetation coverage around the faulty equipment, distance from drainage ditches and water sources, and environmental temperature and humidity within the corresponding time period; biological activity trace features include pyroelectric pulse triggering density, pulse interval changes, and continuous activity duration.

[0038] After feature extraction, new weakly correlated feature vectors are formed according to a unified feature arrangement order, and the new samples are merged into the existing weakly correlated feature vector set. For new samples with excessive repetition with historical samples, sample merging is performed to avoid the same fault event from repeatedly entering the statistical set. In this embodiment, when the matching ratio between new samples and historical samples in terms of the distribution location of remnants, pyroelectric pulse characteristics, and equipment exposure environment all exceed 90%, they are marked as duplicate samples, and only the frequency of occurrence of the corresponding samples is updated, without adding new independent sample records. Subsequently, the conditional probability parameters between hidden nodes and visible nodes in the probability relationship network are incrementally corrected. In specific implementation, the frequency changes of various weakly correlated features in biological activity faults after the entry of new samples are statistically analyzed, and the corresponding conditional probability values ​​are recalculated in combination with the confirmation results corresponding to the new fault events. For example, when the number of times cable tray edge remnants and continuous high-frequency pyroelectric pulses at night co-occur in new fault events increases, the conditional probability correlation between the corresponding visible node and the hidden node of the biological activity fault is increased by a preset scale; when a certain type of historical feature does not reappear in the new samples, the corresponding conditional probability value is decreased. During incremental correction, not all historical parameters are retrained; instead, only the explicit node parameters involved in the new samples are locally updated. If multiple consecutive new samples originate from the same exposed propagation device chain region, the strength of the associated edges within the corresponding propagation chain region is simultaneously increased to reflect the actual situation of biological activity continuously spreading along the fixed bridge path. Furthermore, to avoid significant fluctuations in probability parameters due to a single abnormal failure, a single parameter change upper limit is set for each round of incremental correction. In this embodiment, the magnitude of a single parameter correction is determined based on historical parameter fluctuation statistics, selecting that a single update does not exceed 15% of the historical average parameter value. When the cumulative number of new samples reaches the set update scale, a complete probability network parameter verification is performed again, rereading the recently confirmed fault events and checking the consistency of the posterior probability output results corresponding to the current conditional probability parameters. If it is found that a certain type of equipment continuously exhibits low posterior probabilities but frequent actual faults, the conditional probability weight of the corresponding equipment exposure feature is further increased; if it is found that a certain type of region has a long-term high posterior probability but has not actually experienced a fault, the strength of the corresponding associated edges is reduced. The dynamic incremental correction of the probability relationship network parameters is finally completed, so that the subsequent posterior distribution of biological activity risk and the exposure propagation equipment chain can continuously reflect the current changes in the actual biological activity risk of the new energy power station.

[0039] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0040] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0041] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0042] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0043] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0044] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0045] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0046] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0047] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data, characterized in that, Includes the following steps: S1. Obtain multi-source fragmented observation data of historical fault events in new energy power stations, extract traces of biological activity, residues and equipment exposure features according to equipment location, and form a weakly correlated feature vector; S2. Using faults caused by biological activities as hidden nodes and weakly correlated feature vectors as explicit nodes, a probabilistic relationship network is constructed by combining equipment spatial topology and site exposure paths. The fault correlation strength between equipment is statistically analyzed and the conditional probability parameters are estimated. The posterior distribution of biological disturbance risk for each equipment and the exposure propagation equipment chain are output. S3. Statistically analyze the historical advance amount and corresponding actual situation of historical biological activity warning records, extract reliable change trends according to seasonal time period and station area, and fit to form a dynamic reliable mapping with reliability decreasing with advance amount. S4. Receive the latest biological activity early warning record and corresponding lead time, retrieve the dynamic reliable mapping and the current risk posterior distribution of each device, calculate the regulatory priority of each device in the exposure and transmission device chain, and generate a joint regulatory sequence. S5. Input the joint supervision sequence into the joint supervision cost model to solve for the operation and maintenance protection task start time window and resource scheduling scheme that minimizes the comprehensive supervision cost; S6. If a fault event occurs at the site within the preset period after the operation and maintenance protection task is executed, fragmented observations will be added and incorporated into the weakly correlated feature vector and the probabilistic relationship network parameters will be updated.

2. The method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data according to claim 1, characterized in that, In S1, the weakly correlated feature vector specifically includes: Target detection is performed on operation and maintenance inspection images of historical fault events, and the coverage area, attachment location and distribution pattern of biological residues are extracted as residue association features; The vegetation coverage around the equipment, the distance from the water source, and the historical temperature and humidity during the same period were used as the equipment exposure characteristics. The infrared pyroelectric trigger pulse sequence with a preset time width is extracted centered on the time of the fault occurrence. The variation coefficient of the pulse interval and the trigger density are calculated to form biological activity trace features. Based on the matching of biological activity trace features, it is determined whether the historical fault events are caused by biological activities. The features associated with the remains, the equipment exposure features, and the traces of biological activity are concatenated into vectors to form a weakly correlated feature vector for the failure event.

3. The method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data according to claim 1, characterized in that, In S2, the output of the posterior distribution of biological disturbance risk for each device and the exposure transmission device chain specifically includes: Faults caused by biological activity are defined as hidden nodes, and weakly associated feature vectors are defined as explicit nodes. Connection edges are constructed and the strength of associated edges are determined based on the spatial topological adjacency relationship between equipment, the exposed path connectivity relationship of the station, and the co-occurrence frequency of historical biological activities across equipment in the same period. Based on the historical fault handling records, the conditional probability relationship between hidden nodes and visible nodes is statistically analyzed, and the conditional probability parameters are estimated iteratively using the expectation-maximization algorithm to obtain the biological activity fault association probability corresponding to various weak correlation features. The current weakly correlated feature vectors of each device are input into the probabilistic relation network for parameter estimation. The biological activity risk propagation path is extracted by combining the correlation edge strength, the exposure propagation device chain is generated, and the posterior probability distribution of each device's failure caused by biological activity is output.

4. The method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data according to claim 1, characterized in that, In step S3, fitting a dynamic reliability mapping whose reliability decreases with lead time specifically includes: Extract historical release records of biological activity early warning records, and associate the lead time of each early warning record with biological activity traces and remains in the target area within the corresponding time window to form early warning-real-time correlation data pairs; The associated warning-real-time data is hierarchically grouped by seasonal time period code and station area code. The seasonal time period code is taken from the quarter identifier to which the warning was issued, and the station area code is taken from the spatial grid index corresponding to the warning coverage area. Based on real-time correlation data, we determine whether the biological activity warning is accurate. Within each group, we calculate the ratio of forecasts that are accurate to the actual situation under different lead times. Using the lead time as the independent variable and the accuracy ratio as the dependent variable, we fit the reliability decay curve for each group. The confidence decay curves of all groups are stored by indexing seasonal time period codes and station area codes to form a dynamic confidence mapping. The mapping inputs are lead time, seasonal time period codes and station area codes, and the output is the early warning confidence value under the corresponding conditions.

5. The method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data according to claim 1, characterized in that, In step S4, generating the joint regulatory sequence specifically includes: The latest biological activity warning received is analyzed, and the lead time, coverage spatial grid index, and seasonal time period code of the warning release time are extracted as query keys. The corresponding warning confidence value is retrieved from the dynamic confidence mapping. Using the coverage space grid index as the range, extract the devices and their posterior probability values ​​that fall within the corresponding range from the risk posterior distribution of each device, and screen the on-chain devices that fall within the corresponding range in the exposure propagation device chain as devices to be regulated; For each device in the set of monitored devices, the posterior probability value is multiplied by the early warning confidence value to obtain a monitoring priority score, and the devices are arranged in descending order of scores to generate a joint monitoring sequence.

6. The method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data according to claim 1, characterized in that, In S5, the joint regulatory cost model is specifically a comprehensive regulatory cost function that includes delay costs, resource costs, and path costs.

7. The method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data according to claim 1, characterized in that, In S5, the specific steps for finding the operation and maintenance protection task start time window and resource scheduling scheme that minimizes the overall monitoring cost include: The delay cost is the sum of the regulatory priority scores of each device to be regulated in the joint regulatory sequence and the product of the time elapsed between the task start time and the warning reception time; Resource costs are the sum of the products of the manpower and hours required to perform the task and the material consumption and the corresponding unit price. The path cost is the product of the total actual movement path length formed by the inspection personnel in completing the inspection tasks of each piece of equipment to be monitored in the joint supervision sequence and the movement cost per unit distance. With the goal of minimizing the comprehensive supervision cost function, and with the constraints of the inspection completion time limit of the equipment to be supervised, the upper limit of available manpower hours, and the upper limit of the total amount of schedulable materials as the boundary conditions, the search is conducted to find the task start time window, inspection sequence, and manpower and material allocation scheme that satisfy all constraints and minimize the comprehensive supervision cost function. The output is the operation and maintenance protection task start time window and resource scheduling scheme.

8. The method for safety operation and maintenance supervision of new energy power stations based on multi-dimensional data according to claim 1, characterized in that, In step S6, updating the probabilistic relationship network parameters specifically includes: If a fault occurs in the equipment at the site within a preset period after the task is executed, the corresponding multi-source fragmented observation data is obtained, weak correlation feature vectors are extracted as new samples, the new samples are merged into the existing weak correlation feature vector set, and the conditional probability parameters between hidden nodes and explicit nodes in the probability relationship network are incrementally corrected.

Images