Power station meteorological disaster classification and early warning method and system based on multi-source data

Abstract

The application discloses a power station meteorological disaster classification and early warning method and system based on multi-source data, relates to the technical field of meteorological disaster early warning, and comprises the following steps: obtaining meteorological big data of a target power station area, and synchronously collecting group perception data; identifying the group perception data to obtain micro-scale disaster auxiliary information, cross-verification with the meteorological big data to form a preliminary disaster event set; constructing a multi-dimensional disaster factor set containing at least two meteorological factors based on the two; performing multi-disaster compound analysis to generate a compound disaster risk index, conducting disaster grading early warning according to the index, and realizing accurate early warning of power station meteorological disasters. The application solves the technical problems that traditional power station meteorological disaster early warning relies on single official meteorological big data, cannot accurately identify micro-scale disasters, and can only predict single disasters, leading to lagging early warning and data fragmentation, achieves accurate identification of micro-scale disasters and multi-disaster compound early warning, and meets the technical effects of accurate meteorological disaster early warning and targeted protection for power stations.

Description

Technical Field

[0001] This invention relates to the field of meteorological disaster early warning technology, and in particular to a method and system for classifying and issuing early warnings of meteorological disasters in power plants based on multi-source data. Background Technology

[0002] The safe and stable operation of power plants is crucial for energy supply, and meteorological disasters are a key factor threatening power plant safety. Accurate meteorological disaster early warning is a core prerequisite for power plant disaster prevention and mitigation. Currently, power plant meteorological disaster early warning largely relies on official meteorological big data, achieving warnings through monitoring single meteorological factors. While these methods play a role in macro-level disaster early warning, as power plants demand higher timeliness and accuracy in disaster early warning, the application of traditional technologies reveals significant limitations. Traditional methods struggle to identify localized micro-scale disasters and can only predict single disasters, leading to delayed warnings and incomplete data, failing to meet the needs of power plants for accurate meteorological disaster early warning and targeted protection. Summary of the Invention

[0003] This application provides a method and system for classifying and warning of meteorological disasters in power plants based on multi-source data. It is used to solve the technical problems of traditional power plant meteorological disaster warning relying on a single official meteorological big data, which cannot accurately identify micro-scale disasters and can only predict single disasters, resulting in delayed warnings and one-sided data.

[0004] The first aspect of this application provides a method for classifying and issuing early warnings of meteorological disasters at power plants based on multi-source data. The method includes: acquiring meteorological big data of a target power plant area and simultaneously collecting crowd perception data related to the target power plant area, wherein the crowd perception data includes social media data, news video data, and other crowd perception data; identifying disaster events from the crowd perception data to obtain micro-scale disaster auxiliary information, and cross-validating the micro-scale disaster auxiliary information with the meteorological big data to form a preliminary disaster event set; constructing a multi-dimensional disaster factor set based on the meteorological big data and the preliminary disaster event set, the multi-dimensional disaster factor set containing at least two meteorological factors; performing multi-hazard composite analysis based on the multi-dimensional disaster factor set to generate a composite disaster risk index, and issuing a disaster classification early warning based on the composite disaster risk index.

[0005] A second aspect of this application provides a power plant meteorological disaster classification and early warning system based on multi-source data. The system includes: a mass perception data acquisition module, used to acquire meteorological big data of a target power plant area and simultaneously acquire mass perception data related to the target power plant area, wherein the mass perception data includes social media data, news video data, and other mass perception data; a preliminary disaster event set acquisition module, used to identify disaster events from the mass perception data, acquire micro-scale disaster auxiliary information, and cross-validate the micro-scale disaster auxiliary information with the meteorological big data to form a preliminary disaster event set; a multi-dimensional disaster factor set construction module, used to construct a multi-dimensional disaster factor set based on the meteorological big data and the preliminary disaster event set, wherein the multi-dimensional disaster factor set contains at least two meteorological factors; and a disaster classification and early warning execution module, used to perform multi-hazard composite analysis based on the multi-dimensional disaster factor set, generate a composite disaster risk index, and perform disaster classification and early warning based on the composite disaster risk index.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] This application acquires meteorological big data of the target power plant area and simultaneously collects group perception data including social media and news video data. AI is used to extract text / visual features and geographic information, and spatial clustering is used to generate microscale disaster auxiliary information. This information is then cross-validated with meteorological big data to form a preliminary disaster event set. Based on these two sets, a multi-dimensional disaster factor set containing basic and derived factors is constructed. The intensity of factor chain effects is calculated using a hierarchical influence relationship model, and a composite disaster risk index is generated by overlaying characteristic coefficients of the power plant area. Combined with spatial location features, a graded early warning signal with geofencing is generated, achieving precise classification and early warning of meteorological disasters at the power plant. This results in accurate identification of microscale disasters and multi-hazard composite early warning, meeting the technical requirements for precise early warning and targeted protection against meteorological disasters at the power plant. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the power plant meteorological disaster classification and early warning method based on multi-source data provided in this application embodiment.

[0010] Figure 2 This is a schematic diagram of the structure of a power plant meteorological disaster classification and early warning system based on multi-source data provided in an embodiment of this application.

[0011] Figure labeling: Group perception data acquisition module 1, preliminary disaster event set acquisition module 2, multi-dimensional disaster factor set construction module 3, disaster classification early warning execution module 4. Detailed Implementation

[0012] This application provides a method and system for classifying and warning of meteorological disasters in power plants based on multi-source data. It is used to solve the technical problems of traditional power plant meteorological disaster warning relying on a single official meteorological big data, which cannot accurately identify micro-scale disasters and can only predict single disasters, resulting in delayed warnings and one-sided data.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] Example 1, as Figure 1 As shown, a power plant meteorological disaster classification and early warning method based on multi-source data is described, wherein the method includes:

[0016] Step A100: Obtain meteorological big data of the target power station area and simultaneously collect group perception data related to the target power station area, wherein the group perception data includes social media data, news video data and other group perception data.

[0017] Specifically, when conducting meteorological disaster classification and early warning for power plants based on multi-source data, the first step is to identify the specific geographical area of ​​the target power plant and use this as the boundary for data collection. Based on this area, real-time meteorological big data for the target power plant area is acquired by connecting to the monitoring network of official meteorological departments, the meteorological monitoring terminals deployed independently by the power plant, and the data transmission link of meteorological satellites. This meteorological big data includes core meteorological indicators such as wind speed, humidity, temperature, and rainfall. The data collection frequency is set according to the early warning requirements; for example, temperature and humidity data are updated every 30 minutes, wind speed data is recorded every 15 minutes, and rainfall data is statistically analyzed hourly. This standardized meteorological data can intuitively reflect the macro-level meteorological changes in the target area, providing authoritative basic data support for subsequent disaster analysis.

[0018] Next, crowd perception data related to the target power plant area was collected simultaneously. For social media data, by calling the open application programming interfaces (APIs) of mainstream social media platforms, text information in user-posted content with geographical locations tagged as the target power plant area and its surroundings was filtered out, and the geocoding associated with these texts was extracted to determine the specific latitude and longitude of the disaster descriptions. For news video data, video content published by local news websites and short news video platforms covering the target power plant area was selectively crawled. At the same time, the location information at the time of video shooting was extracted from the metadata information of the videos or the accompanying text descriptions of news reports to clarify the specific geographical locations corresponding to these visual features. In addition, other crowd perception data included disaster information submitted by residents of communities around the power plant through online reporting platforms built by local emergency management departments, and abnormal weather notices issued by surrounding enterprises near the factory area, further enriching the source dimensions of crowd perception data.

[0019] By acquiring meteorological big data of the target power plant area and simultaneously collecting social media data, news video data, and other group perception data, a multi-dimensional data support system was built. This system compensates for the limitations of traditional single data sources in acquiring micro-scale disaster information and lays a data foundation for the accurate identification and early warning of meteorological disasters at the power plant in the future.

[0020] Step A200: Identify disaster events from the group perception data, obtain microscale disaster auxiliary information, and cross-validate the microscale disaster auxiliary information with the meteorological big data to form a preliminary disaster event set.

[0021] Optionally, firstly, text features containing disaster description semantics and associated geocoding are extracted from social media data; visual features and shooting location information that conform to the preset disaster pattern are identified from news video data; then, spatial clustering is performed based on geocoding and shooting location information, and combined with text features and visual features, to generate microscale disaster auxiliary information with disaster type labels. The specific steps are explained in detail in A210-A230.

[0022] Next, a spatiotemporal alignment matrix is ​​established by comparing the disaster type labels with the corresponding spatial location monitoring indicators in the meteorological big data. Then, the validity of the disaster events is verified based on the matrix, and all valid disaster events are aggregated according to spatial location to generate a preliminary disaster event set containing the disaster type-spatial distribution mapping relationship. The specific steps are explained in detail in A240-A250.

[0023] Step A300: Based on the meteorological big data and the preliminary disaster event set, construct a multi-dimensional disaster factor set, which contains at least two meteorological factors.

[0024] In one embodiment of this application, a target disaster type combination is determined based on the disaster type-spatial distribution mapping relationship of the preliminary disaster event set. Based on this combination, basic meteorological factors are extracted from meteorological big data and derived disaster factors are extracted from the preliminary disaster event set. Then, a mapping relationship matrix between the two is established to obtain a multi-dimensional disaster factor set. The specific steps are described in detail in A310-A330.

[0025] Step A400: Perform multi-hazard composite analysis based on the multi-dimensional disaster factor set to generate a composite disaster risk index, and conduct disaster classification and early warning based on the composite disaster risk index.

[0026] Specifically, firstly, a hierarchical influence relationship model of each factor in a multi-dimensional disaster factor set is established, then the chain effect intensity between factors is calculated based on this, and finally, the intensity is superimposed with the power station area characteristic coefficient to output a composite disaster risk index. The specific steps are explained in detail in A410-A430.

[0027] Next, the corresponding warning level is triggered based on the continuous numerical range of the composite disaster risk index. Then, the warning level is superimposed with the spatial location features in the preliminary disaster event set to generate a graded warning signal with geofencing. The specific steps are explained in detail in A440-A450.

[0028] Furthermore, step A200 in the method provided in this application embodiment includes:

[0029] A210: Extract textual features containing disaster description semantics and their associated geocoding from the social media data.

[0030] A220: Identify visual features and their shooting location information that conform to a preset disaster pattern from the news video data.

[0031] A230: Based on the geocoding and shooting location information, spatial clustering is performed, and combined with the text features and visual features, microscale disaster auxiliary information with disaster type labels is generated.

[0032] Specifically, the first step is to process social media data. Natural language processing (NLP) techniques are used to perform semantic analysis and feature extraction on the social media data. The process involves first filtering text content posted by users within the target power plant area and its surrounding geographical range on social media. The filtered text undergoes data preprocessing to remove irrelevant symbols, redundant expressions, and non-target language content. Then, word segmentation technology is used to break down continuous text into independent word units. Subsequently, a pre-built disaster domain keyword library (covering disaster-related words and expressions such as tornadoes, hail, strong winds, and rainstorms) is used for keyword matching to initially filter out text that may contain disaster descriptions. Next, a pre-trained language model is used, fine-tuned in conjunction with disaster-specific corpora, to perform deep semantic understanding on the initially filtered text. This determines whether the text truly contains descriptive semantics of the occurrence and development of a disaster, excluding text that mentions a disaster but does not actually occur. Finally, from the semantically verified text, the core disaster descriptive semantic text features are extracted and associated with the text's inherent geographical location information or parsed from specific locations mentioned in the text to obtain geocoding, thus completing the semantic analysis and feature extraction of the social media data.

[0033] Subsequently, the location service interface of the social media platform is invoked or the specific location information mentioned in the text is parsed. Combined with Geographic Information System (GIS) technology, the geocoding associated with this text content—that is, the precise latitude and longitude coordinates—is obtained to determine the spatial location corresponding to the disaster description. For example, when a user posts on social media that a tornado passed 500 meters east of a power station, the textual features of the disaster description "tornado" can be extracted, and the user's location latitude and longitude at the time of posting can be obtained as the associated geocoding.

[0034] Next, the news video data was processed. First, the relevant news videos of the target power station area were analyzed frame by frame. A disaster visual recognition model was pre-trained. When building the disaster visual recognition model, the vortex shape of tornadoes, the dense falling characteristics of hail, and the water accumulation of rainstorms were sorted out. Visual feature parameters such as shape, motion trajectory, and color distribution corresponding to each disaster were extracted to establish a disaster visual feature library. Then, the target detection architecture based on YOLO convolutional neural network was selected as the basic framework of the model to build a network structure that includes feature extraction, feature matching, and disaster classification.

[0035] During the training phase, a large number of video samples of different disaster types were collected, covering disaster footage under different lighting, weather, and shooting angles. Keyframe extraction, disaster area labeling, and type label assignment were performed on the samples. The labeled samples were divided into training and validation sets according to the proportion. The model was iteratively trained using the training set, and the model's learning rate, convolution kernel size, and other parameters were adjusted in real time using the validation set until the model's recognition accuracy for the preset disaster patterns on the validation set reached the preset standard. The model input consisted of video frames parsed frame by frame from news videos related to the target power plant area, and the output consisted of the recognition results of the video frames, including whether there were visual features that matched the preset disaster patterns. If so, the corresponding disaster type and the position coordinates of the visual feature in the video frame were output simultaneously, providing data support for subsequent extraction of shooting location information and generation of micro-scale disaster auxiliary information.

[0036] Then, the system identifies whether visual features in the video frames match a preset disaster pattern. Once relevant visual features are identified, such as a large number of white, granular falling objects matching a hail pattern or a rotating column of air matching a tornado pattern, the corresponding video segments are recorded. Simultaneously, the system extracts the video's shooting location information from the news video's metadata (such as GPS location records of the shooting equipment) or the accompanying text descriptions of the news report to determine the specific spatial range of the visual features.

[0037] Next, after extracting the features and location information from social media data and news video data, they need to be integrated. A spatial clustering algorithm is used to group data with similar spatial distances into the same cluster based on the acquired geocoding and shooting location information. For example, social media text data and news video data with a latitude and longitude deviation within 100 meters are grouped together, thus achieving spatial correlation between data from different sources. Subsequently, textual and visual features within the same cluster are cross-validated and comprehensively judged. If the text features mention hail, and visual features of hail falling are identified in videos from the same spatial location, then the disaster type in that area can be determined to be a hail disaster, and the spatial area corresponding to that cluster group is labeled with a hail disaster type. Through the above process, micro-scale disaster auxiliary information with disaster type labels is finally generated. This information contains both a clear disaster type and a precise spatial location.

[0038] By employing natural language processing technology to extract semantic text features and associated geocoding of disaster descriptions from social media data, using computer vision technology to identify the visual features of preset disaster patterns and shooting location information from news video data, and integrating data with spatial clustering algorithms and combining feature annotations to label disaster types, the system achieves the effect of accurately obtaining microscale disaster auxiliary information with clear disaster types and spatial locations.

[0039] Furthermore, step A200 in the method provided in this application embodiment includes:

[0040] A240: Compare the disaster type labels with the corresponding spatial location monitoring indicators in the meteorological big data to establish a spatiotemporal alignment matrix between meteorological data and mass perception data.

[0041] A250: Based on the spatiotemporal alignment matrix, verify the validity of disaster events, and aggregate all valid disaster events according to their spatial location to generate a preliminary disaster event set containing the disaster type-spatial distribution mapping relationship.

[0042] Optionally, when cross-validating microscale disaster auxiliary information with meteorological big data, the corresponding meteorological monitoring indicators are first determined based on the disaster type labels in the microscale disaster auxiliary information. If the disaster type label is strong wind, the corresponding meteorological big data monitoring indicators include average wind force level and gust level; if the label is heavy rain, the corresponding indicator is the rainfall amount within a specific time period.

[0043] Subsequently, using Geographic Information System (GIS) technology, the spatial locations of the micro-scale disaster auxiliary information annotations were matched with monitoring stations within the same latitude and longitude range in the meteorological big data, ensuring a precise spatial correspondence. Simultaneously, the recording periods of the two types of data were synchronized according to timestamps to guarantee data consistency in the temporal dimension. By performing spatial matching and temporal synchronization, disaster type labels were compared one by one with the corresponding spatial locations and meteorological monitoring indicator values ​​for the same time periods, establishing a spatiotemporal alignment matrix between meteorological data and collective perception data. The matrix clearly presents the correspondence between disaster type labels and meteorological monitoring indicators at different spatial locations and time points.

[0044] Next, the validity of disaster events is verified based on the spatiotemporal alignment matrix. When the disaster report density in the group perception data exceeds the first threshold and the meteorological data anomaly degree of the corresponding spatiotemporal node exceeds the second threshold, the disaster event is confirmed to be valid. The group perception noise data that fails the verification is removed to form a preliminary set of spatiotemporally calibrated disaster events. The specific steps are explained in detail in A251-A252.

[0045] Next, all verified valid disaster events were collected, and a spatial aggregation method was adopted: First, spatial division criteria were determined, and a distance threshold based on the target power station was set, such as a radius of 5 kilometers, or a fixed grid was divided according to latitude and longitude, or the range was divided according to the actual geographical area; then, the spatial location information such as latitude and longitude of all verified valid disaster events was extracted; then, according to the division criteria, valid disaster events whose spatial locations fall within the same range were grouped into the same area group, for example, valid hail disaster events within 5 kilometers of the power station were aggregated into one group, and valid rainstorm disaster events in the downstream valley area were aggregated into another group. During the aggregation process, the disaster type and specific spatial range of each group of events were recorded, and finally, a preliminary disaster event set containing the disaster type-spatial distribution mapping relationship was generated. This set can intuitively reflect the distribution of different disaster types in the target power station area.

[0046] By establishing a spatiotemporal alignment matrix by comparing disaster type labels with meteorological monitoring indicators of corresponding spatial locations, and after validity verification, valid disaster events are aggregated according to spatial location, thus achieving the effect of generating a preliminary disaster event set containing the mapping relationship between disaster type and spatial distribution.

[0047] Furthermore, step A250 in the method provided in this application embodiment includes:

[0048] A251: When the disaster report density in the group perception data exceeds the first threshold and the meteorological data anomaly degree of the corresponding spatiotemporal node exceeds the second threshold, the disaster event is confirmed to be valid and verified by meteorological data.

[0049] A252: Eliminate the group-perceived noise data that has not passed meteorological data verification to form a preliminary set of disaster events with spatiotemporal calibration.

[0050] Specifically, before verifying the validity of disaster events based on the spatiotemporal alignment matrix, it is necessary to determine the first and second thresholds required for verification. The first threshold corresponds to the disaster report density in the crowd perception data. Its setting should refer to the crowd feedback patterns of historical micro-scale disaster events in the target power plant area. For example, when local hail or short-term strong winds occurred in the area in the past 5 years, the average number of valid disaster reports per square kilometer per unit space and per hour per unit time in the crowd perception data can be statistically analyzed. The average value can be slightly increased to finally determine the first threshold, ensuring that it can cover the reporting density range when a real disaster occurs, while avoiding misjudgments triggered by a small number of scattered reports.

[0051] The second threshold corresponds to the anomaly of meteorological data and needs to be determined based on long-term meteorological monitoring data of the target power station area. First, calculate the normal fluctuation range of core meteorological indicators such as wind speed, rainfall, and humidity, and then count the proportion of these indicators that exceed the normal range when historical disasters occur. For example, the anomaly of wind speed during the outer influence of historical typhoons often reaches more than 40%. Based on this, the second threshold can be set to 35% to accurately capture significant anomalies related to disasters in meteorological data.

[0052] After determining the first and second thresholds, the disaster report density in the mass sensing data is calculated. First, based on the spatial location information in the spatiotemporal alignment matrix, the target power plant area is divided into several equal-area spatial units, such as 1 square kilometer per unit. Simultaneously, time windows are divided according to the time dimension, such as 1 hour per window, ensuring that each combination of spatial unit and time window corresponds to a specific spatiotemporal node. Then, the number of reports containing the disaster type label corresponding to that node in the mass sensing data within each spatiotemporal node is counted. The number of reports is then divided by the area of ​​the spatial unit to obtain the disaster report density for that spatiotemporal node. This density is compared with the first threshold to determine whether it exceeds the first threshold.

[0053] Next, the meteorological data anomaly degree of the corresponding spatiotemporal node is calculated. Meteorological big data monitoring indicators corresponding to that spatiotemporal node are extracted from the spatiotemporal alignment matrix. For example, if the disaster type label is short-term strong wind, the average wind speed monitoring value within the time window of that node is extracted; if the label is heavy rain, the cumulative rainfall monitoring value is extracted. Then, the extracted monitoring value is compared with the normal range of the meteorological indicator, and the anomaly degree is calculated as (monitoring value - upper limit of normal range) / upper limit of normal range × 100%. If the monitoring value is lower than the lower limit of the normal range, a similar calculation is performed. For example, for the spatiotemporal node corresponding to the aforementioned short-term strong wind, the average wind speed monitoring value is 12 m / s, and the upper limit of the normal range for wind speed in this area is 8 m / s. The calculated anomaly degree is (12-8) / 8 × 100% = 50%, which is compared with the second threshold of 35%, and it is determined to exceed the second threshold.

[0054] Then, when the disaster report density of a certain spatiotemporal node exceeds the first threshold and the meteorological data anomaly exceeds the second threshold, the disaster event corresponding to that node is confirmed as valid and verified by meteorological data. If only one of the threshold conditions is met, for example, if the report density of a certain spatiotemporal node exceeds the first threshold but the meteorological data anomaly does not exceed the second threshold, or if the report density does not exceed the first threshold but the meteorological data anomaly exceeds the second threshold, then the disaster event is determined to be invalid.

[0055] Subsequently, all group perception data that failed meteorological data verification were classified as noise data and removed. This noise data may include user-incorrectly sent non-disaster information, duplicate submissions of the same report, etc. Finally, the verified valid disaster events were labeled according to their spatiotemporal nodes in the spatiotemporal alignment matrix, clarifying the occurrence time, specific spatial location, and disaster type of each valid event, forming a preliminary spatiotemporally labeled disaster event set.

[0056] By combining historical data to determine dual thresholds, calculating the disaster report density and meteorological data anomaly of spatiotemporal nodes, and jointly judging the validity of disaster events with dual thresholds and removing noisy data, the effect of forming a preliminary set of spatiotemporally calibrated disaster events was achieved.

[0057] Furthermore, step A252 in the method provided in this application embodiment includes:

[0058] A252-1: Label each event in the preliminary disaster event set with a disaster type tag, and associate each disaster type with the list of affected equipment based on the power plant equipment vulnerability database.

[0059] Specifically, when processing the initial disaster event set, the first step is to label each event in the set with a disaster type. Those skilled in the art have pre-established a standardized meteorological disaster type system that covers common micro-scale and complex disaster types found in power plants, such as strong winds, rainstorms, hailstorms, tornadoes, and landslides, with each disaster type corresponding to a clear characteristic description.

[0060] Next, the key information of each event in the preliminary disaster event set, including meteorological monitoring data in the spatiotemporal alignment matrix and disaster description features in the group perception data, is compared one by one with the features in the standardized disaster type system. After determining the disaster type of each event, it is labeled with the corresponding disaster type tag. For example, if an event in the preliminary disaster event set records solid precipitation with a diameter of about 8 mm in an area 3 km south of the target power station at a certain time, and the group report mentions hail hitting the roof, and the meteorological data and disaster description match the feature of hail disaster in the standardized system as solid precipitation with a diameter ≥ 5 mm, then the event is labeled with a hail disaster tag.

[0061] After completing the disaster type labeling, a pre-built power plant equipment vulnerability database is invoked. This database, established by those skilled in the art based on long-term power plant operation and maintenance data, equipment factory parameters, and historical disaster damage records, covers basic information on various core equipment of the power plant, such as equipment name, installation location, operating parameter thresholds, and the correlation between different disaster types and equipment impacts. It clarifies which equipment is prone to failure, damage, or performance degradation when a certain type of disaster occurs. For example, the database records the correlation between transmission lines and strong wind disasters: when the wind speed exceeds 10 m / s, transmission lines are prone to conductor galloping; the correlation between cable trenches and rainstorm disasters: when the rainfall exceeds 40 mm / hour, cable trenches are prone to water accumulation leading to cable insulation damage; and the correlation between outdoor monitoring equipment and hail disasters: when the hailstone diameter exceeds 6 mm, the monitoring equipment lens is prone to breakage.

[0062] Subsequently, the association between disaster types and the list of affected equipment was established. For each event tagged with a disaster type, a matching query was performed in the power plant equipment vulnerability database using a data retrieval algorithm: using the disaster type tag of the event as the search keyword, all equipment information related to that disaster type was extracted from the database, including equipment name, equipment number, installation location, and possible failure types. This extracted equipment information was organized according to equipment function categories to form a list of affected equipment corresponding to the disaster event, and the list was bound to the event itself for storage, ensuring that each disaster event could be associated with specific affected equipment. For example, for an event tagged with a strong wind disaster, by searching the power plant equipment vulnerability database, equipment associated with strong wind disasters, such as transmission lines, outdoor disconnect switches, and cooling tower fans, was extracted and organized to form a list of affected equipment: transmission lines, numbered L-01 to L-05, connecting the power plant's main transformer and the step-up substation; outdoor disconnect switches, numbered QS-12 to QS-15, installed in the western area of ​​the power plant; cooling tower fans, numbered F-03 to F-06, were also bound to this strong wind disaster event.

[0063] By establishing a standardized disaster type system to label events, calling up a pre-built power plant equipment vulnerability database, and searching for related equipment using disaster type as a keyword to form a list, the goal of ensuring that each disaster event in the initial disaster event set corresponds to a clear list of affected equipment was achieved.

[0064] Furthermore, step A300 in the method provided in this application embodiment includes:

[0065] A310: Determine the target disaster type combination based on the disaster type-spatial distribution mapping relationship in the preliminary disaster event set.

[0066] A320: Based on the target disaster type combination, extract relevant basic meteorological factors from the meteorological big data, and extract derived disaster factors from the preliminary disaster event set.

[0067] A330: Establish the mapping relationship matrix between the basic meteorological factors and the derived disaster factors to obtain the multi-dimensional disaster factor set.

[0068] Specifically, when constructing a multi-dimensional disaster factor set, the process begins with a preliminary disaster event set. This set already contains a disaster type-spatial distribution mapping relationship, which clearly records the specific spatial distribution of different disaster types within the target power plant area. Next, this mapping relationship is analyzed in depth, statistically analyzing the co-occurrence frequency and temporal correlation of different disaster types within the same spatial unit, such as a 1-square-kilometer grid divided according to the power plant area, or adjacent spatial units. Disaster types that co-occur frequently and exhibit causal or superimposed effects are grouped together to determine the target disaster type combination.

[0069] Next, after determining the target disaster type combination, disaster factors are extracted dimensionally. For basic meteorological factors, relevant data are screened and extracted from meteorological big data based on the core impact meteorological indicators of each disaster type in the target disaster type combination. If the target disaster type combination includes heavy rain, rainfall data for the corresponding spatial area needs to be extracted, such as 1-hour rainfall and 24-hour rainfall; if it includes short-term strong winds, average wind speed and gust wind speed data need to be extracted; if it includes low-temperature related disasters, temperature and humidity data are extracted. These basic meteorological factors directly reflect the meteorological conditions for the occurrence of disasters and provide the basic data support for subsequent analysis.

[0070] For the extraction of derivative disaster factors, the preliminary disaster event set is used as the core. Combining the characteristics of disaster types and the actual impact on the power station, indirect disaster impact indicators related to basic meteorological factors are extracted from this set. For example, for the combination of rainstorm and landslide, the probability of landslide when this combination of disasters occurs is extracted from the preliminary disaster event set, calculated by statistically analyzing the ratio of the number of landslides under similar historical meteorological conditions to the total number of events; for the combination of low temperature and rainfall, the risk of equipment icing is extracted, referring to the records of equipment icing events under low temperature and rainfall weather in the preliminary disaster event set, and extrapolating it based on current temperature and humidity data; for thunderstorm-related combinations, the frequency of lightning is extracted, that is, the number of lightning strikes per unit time when thunderstorm disaster events occur in the preliminary disaster event set.

[0071] Next, after extracting the basic meteorological factors and derived disaster factors, a mapping matrix between the two needs to be established. Combining statistical analysis methods with historical disaster data, the influence and correlation patterns of changes in basic meteorological factors on derived disaster factors are analyzed, and these patterns are quantified and entered into the matrix. For example, historical data statistics show that for every 10 mm increase in rainfall, the probability of landslides increases by an average of 18%; for every 3 m / s increase in gust wind speed, the risk of damage to outdoor equipment (derived factor) increases by 12%; and for every temperature below 0°C and relative humidity above 80%, the risk of equipment icing increases by 25%. These correlations and quantified values ​​are organized according to the structure of basic meteorological factor - derived disaster factor - correlation strength to form a mapping matrix. This matrix clearly presents the correspondence and influence strength between each basic meteorological factor and its corresponding derived disaster factor. Finally, by integrating the basic meteorological factors, derived disaster factors, and their mapping matrix, a multi-dimensional disaster factor set containing at least two meteorological factors is obtained.

[0072] By analyzing the disaster type-spatial distribution mapping relationship of the preliminary disaster event set, the target disaster type combination is determined. Based on the combination, basic meteorological factors are extracted from meteorological big data and derived disaster factors are extracted from the preliminary disaster event set. Then, a mapping relationship matrix between the two is established, which achieves the effect of constructing a multi-dimensional disaster factor set containing at least two meteorological factors.

[0073] Furthermore, step A400 in the method provided in this application embodiment includes:

[0074] A410: Establish a hierarchical influence relationship model for each factor in the multidimensional disaster factor set.

[0075] A420: Calculate the intensity of the chain effect between factors based on the hierarchical influence relationship model.

[0076] A430: The composite disaster risk index is output by superimposing the intensity of the chain effect and the regional characteristic coefficient of the power station.

[0077] In one embodiment, when performing multi-hazard composite analysis based on a multi-dimensional disaster factor set to generate a composite disaster risk index, the process begins by establishing a hierarchical influence relationship model for each factor in the multi-dimensional disaster factor set. The construction process of this model is detailed in steps A411-A414, involving the calculation of the intensity of the chain effect between factors. Key parameters are first extracted from the hierarchical influence relationship model, including the correspondence between each primary factor and secondary factor, the preset transmission attenuation coefficient, and the quantitative values ​​of different factors, such as the real-time monitoring value of wind speed (primary factor) and the estimated value of landslide probability (secondary factor).

[0078] Subsequently, the direct cascading effect strength of a single primary factor on its associated secondary factors is calculated. For example, if the primary factor is wind speed, with a real-time monitoring value of 12 m / s, and the corresponding secondary factor is transmission line galloping risk, the conduction attenuation coefficient between the two in the hierarchical influence relationship model is 0.3. Using the formula: Single cascading effect strength = Primary factor quantification value × Conduction attenuation coefficient, the direct cascading effect strength of this group of factors can be calculated as 12 × 0.3 = 3.6. If multiple primary factors jointly affect the same secondary factor, such as rainfall and soil moisture content jointly affecting the probability of landslides, then the cascading effect strength of each primary factor on the secondary factor needs to be calculated separately, and then a weighted sum is obtained. The weights are determined based on the contribution of each primary factor to the secondary factor in historical data, such as rainfall with a weight of 0.6 and soil moisture content with a weight of 0.4, thus obtaining the comprehensive cascading effect strength of the secondary factor. Finally, the intensity of the chain reaction between all primary and secondary factors is summarized to form the total chain reaction intensity under multiple disasters. For example, in the target power station area, the chain reaction intensity of wind speed-line galloping is 3.6, the chain reaction intensity of rainfall-landslide is 4.2, and the chain reaction intensity of low temperature-equipment icing is 2.8. The total chain reaction intensity is 3.6+4.2+2.8=10.6.

[0079] Next, the regional characteristic coefficient of the power station is introduced and superimposed. The regional characteristic coefficient of the power station is a correction parameter determined based on the unique attributes of the area where the target power station is located. Its value needs to comprehensively consider factors such as regional topography, equipment layout characteristics, and the degree of impact of historical disasters. Specifically, geographical data of the power station area is first collected, such as terrain slope, altitude, and surrounding vegetation coverage; equipment operation and maintenance data, such as the service life and protection level of key equipment; and historical disaster records, such as the number of typhoons, rainstorms, and other disasters that have occurred in the area in the past 10 years and the resulting equipment damage rate. Then, the analytic hierarchy process (AHP) is used to quantify and score these factors. For example, if the power station is located in a mountainous area with a terrain slope >25°, which is prone to landslides, the score is 0.3; if the average service life of key equipment exceeds 15 years, the protection capability is reduced, and the score is 0.2; if there have been more than 3 rainstorm disasters in the past 5 years and the equipment damage rate reaches 15%, the historical risk is high, and the score is 0.2; then the regional characteristic coefficient of the power station is 1 + 0.3 + 0.2 + 0.2 = 1.7, where the base value is set to 1, and the scores of each influencing factor are positively superimposed. If the power station is located on a plain, with a terrain slope of <5° (score 0), equipment service life of less than 5 years (score 0), and historical disaster damage rate of <3% (score 0), then the regional characteristic coefficient is 1.

[0080] During the superposition process, a multiplicative model of total cascading effect intensity × power plant regional characteristic coefficient is adopted, or a weight allocation is introduced according to actual needs, such as a weight of 0.7 for cascading effect intensity and a weight of 0.3 for regional characteristic coefficient. The formula is: Composite Disaster Risk Index = Total Cascading Effect Intensity × 0.7 + Total Cascading Effect Intensity × Regional Characteristic Coefficient × 0.3, to ensure that the index can reflect the cascading effects of multiple disasters and also fit the actual regional risk of the power plant. Taking a total cascading effect intensity of 10.6 and a regional characteristic coefficient of 1.7 as an example, the composite disaster risk index calculated by the multiplicative model is 10.6 × 1.7 = 18.02, while the composite disaster risk index calculated by the weighted model is 10.6 × 0.7 + 10.6 × 1.7 × 0.3 = 7.42 + 5.406 = 12.826. The specific model selection can be determined according to the early warning accuracy requirements of the power plant.

[0081] Finally, the results of the superposition operation are standardized to map the values ​​to the range of 0-100, which facilitates subsequent classification. The final output is a composite disaster risk index, which can be directly used in the subsequent disaster classification and early warning process.

[0082] By calculating the intensity of the total chain effect between factors based on the hierarchical influence relationship model, determining the regional characteristic coefficient by combining the regional attributes of the power station, and superimposing the two through a quantitative model, the result is a composite disaster risk index that can accurately reflect the chain effect of multiple disasters and the risk characteristics of the power station area.

[0083] Furthermore, step A410 in the method provided in this application embodiment includes:

[0084] A411: The factors that have a direct physical effect on power plant equipment are concentrated in the multi-dimensional disaster factors and are regarded as first-level factors.

[0085] A412: Factors that indirectly influence the primary factors are considered as secondary factors.

[0086] A413: Establish a nonlinear mapping relationship between primary and secondary factors, quantify the intensity of chain effects through the transmission attenuation coefficient, and construct a hierarchical influence relationship model.

[0087] A414: Wherein, the conduction attenuation coefficient is dynamically adjusted according to the correlation strength between factors in historical disaster events.

[0088] Optionally, when constructing a hierarchical impact model, firstly, primary factors are identified from a multi-dimensional disaster factor set. Specifically, this involves defining the criteria for direct physical effects by considering the physical structural characteristics and operating principles of the power plant equipment. The criteria define a factor as one that can directly affect the material, structure, or operating state of the equipment without requiring other intermediate disasters or influences. Those skilled in the art systematically examine each factor in the multi-dimensional disaster factor set and designate those meeting this criterion as primary factors. For example, wind speed, a factor in the multi-dimensional disaster factor set, can directly exert a blowing force on the power plant's transmission lines, causing conductor galloping and tower tilting, which constitutes a direct physical effect. Rainfall can directly erode the foundations of outdoor equipment and soak cable trenches, also constituting a direct physical effect. Therefore, wind speed, rainfall, temperature (low temperatures directly cause equipment components to freeze and crack), and hail impact force are identified as primary factors.

[0089] Next, we further screen secondary factors. The core characteristic of secondary factors is that they need to indirectly affect primary factors. Therefore, based on the effects of primary factors, we analyze them in conjunction with the chain reaction pattern of disasters. Using a causal relationship mining algorithm, we extract factors from multi-dimensional disaster factors that have a path of primary factor → intermediate influence → secondary factor → equipment impact. For example, the primary factor of rainfall leads to increased surface soil moisture content, which in turn triggers landslides. Landslides impact the foundations of power station transmission towers. In this case, the landslide factor needs to indirectly affect the equipment through rainfall; therefore, the landslide probability is classified as a secondary factor. Similarly, the primary factor of temperature below 0℃ causes water vapor in the air to condense into ice on the equipment surface. Equipment icing reduces insulation performance and can cause short circuits. The equipment icing risk factor needs to indirectly affect the equipment through the primary factor of low temperature; therefore, the equipment icing risk is classified as a secondary factor.

[0090] Subsequently, a nonlinear mapping relationship between primary and secondary factors is established. Since the influence of primary factors on secondary factors is not a simple linear relationship—for example, when rainfall is low, the probability of landslides increases slowly, but when rainfall exceeds a certain threshold, the probability of landslides increases exponentially—a backpropagation (BP) neural network model from a nonlinear regression model can be used to construct the mapping relationship. Taking the primary factor of rainfall and the secondary factor of landslide probability as an example, historical disaster data for the past 10 years in the target power station area are first collected, including information such as the number of landslides and their scale corresponding to different rainfall intervals. Rainfall data is used as the model input, and landslide probability as the output. The model is iteratively trained using a gradient descent algorithm to fit a nonlinear curve of rainfall and landslide probability. When rainfall is below 50 mm, the landslide probability remains below 5%; when rainfall is between 50 and 100 mm, the probability rises to 15%-30%; and when rainfall exceeds 100 mm, the probability jumps to over 60%, thus forming a nonlinear mapping relationship between the two.

[0091] Based on the established mapping relationship, the intensity of the chain reaction is quantified using a conduction attenuation coefficient. The initial value of the conduction attenuation coefficient needs to be calculated based on the correlation strength between primary and secondary factors in historical disaster events. For example, in statistical historical data, for every 5 m / s increase in wind speed (primary factor), the average increase in the probability of tree collapse (secondary factor) is considered. If, in the past 5 major wind disasters, the probability of tree collapse increased by an average of 20% for every 5 m / s increase in wind speed, then the initial conduction attenuation coefficient is set to 0.2. Simultaneously, considering that factors such as vegetation coverage, terrain slope, and equipment aging around the power station will change over time, the conduction attenuation coefficient is dynamically adjusted: for each new disaster event, the actual correlation data between primary and secondary factors in that event is included in the historical database, the correlation strength is recalculated, and the conduction attenuation coefficient is updated. For example, if the vegetation coverage in a certain area decreases from 60% to 30%, and in a newly occurring major wind disaster, the probability of tree collapse increases to 35% for every 5 m / s increase in wind speed, then the conduction attenuation coefficient is adjusted from 0.2 to 0.35. By using the aforementioned nonlinear mapping relationship and dynamically adjusted conduction attenuation coefficient, a hierarchical influence relationship model is finally constructed.

[0092] By screening primary factors based on the physical effects of equipment, determining secondary factors according to the chain reaction path, establishing mapping relationships using nonlinear models, and dynamically adjusting the conduction attenuation coefficient based on historical data, the hierarchical influence relationship model with clear correlation between primary and secondary factors and quantifiable chain effect intensity was achieved.

[0093] Furthermore, step A400 in the method provided in this application embodiment includes:

[0094] A440: Trigger the corresponding early warning level based on the continuous numerical range of the composite disaster risk index.

[0095] A450: The warning level is superimposed with the spatial location features in the preliminary disaster event set to generate a graded warning signal with geofencing.

[0096] In one embodiment, the continuous numerical range of the composite disaster risk index and the corresponding warning level for each range are first defined. The range division is based on the power plant's historical disaster data, equipment safety tolerance capabilities, and industry warning standards. Those skilled in the art first analyze meteorological disaster events such as typhoons, rainstorms, and hail that have occurred at the target power plant over the past 5-10 years, extract the composite disaster risk index for each disaster, and simultaneously statistically analyze the actual impact data such as equipment damage rate and production interruption duration under this index. Then, combined with the maximum risk tolerance threshold of the power plant's core equipment (such as transmission lines, main transformers, and outdoor switchgear), for example, if the highest composite disaster risk index that a certain type of main transformer can withstand is 65, the index is divided into continuous ranges with clear risk gradients, and a corresponding warning level is matched for each range.

[0097] Specifically, an index of 0-25 is set as Level 1 warning, low risk; historical data shows that disasters in this range only cause minor abnormal noises in individual equipment, requiring no shutdown. 26-55 is set as Level 2 warning, medium risk, which may lead to decreased insulation performance of outdoor equipment, requiring increased inspection. 56-85 is set as Level 3 warning, high risk, easily causing equipment tripping or partial shutdown. 86 and above are set as Level 4 warning, extremely high risk, potentially causing widespread equipment damage and prolonged power outages. Once the current composite disaster risk index is obtained through multi-hazard composite analysis, matching this index with a preset range will trigger the corresponding warning level. For example, if the calculated index is 48, falling within the 26-55 range, a Level 2 warning will be directly triggered.

[0098] Subsequently, the triggered warning level needs to be overlaid with the spatial location characteristics in the preliminary disaster event set. The preliminary disaster event set records the specific spatial location information of each valid disaster event, including the latitude and longitude range of the disaster impact and the functional zones of the power plant involved, such as the power plant production area, oil storage area, transmission line corridor, and surrounding ancillary facilities area. The precise overlay of the two events is achieved using Geographic Information System (GIS) technology: First, the spatial location data from the initial disaster event set is imported into the GIS. Using the system's built-in coordinate matching function, the impact boundary of each disaster event is precisely marked on the electronic map, along with the corresponding area name and latitude / longitude. Second, the triggered warning levels are associated with these spatial areas. Based on the composite disaster risk index of different areas, their respective warning levels are determined. If the composite disaster risk index of a spatial area is 62, triggering a Level 3 warning, the boundary of the production area is outlined in red in the GIS, forming a geographic fence. Inside the fence, a Level 3 warning is marked – currently affected by combined heavy rain and strong winds; production equipment needs to be shut down in advance for protection. If the composite disaster risk index of another area is 22, triggering a Level 1 warning, the boundary of the office area is marked with a blue fence, indicating a Level 1 warning – low risk; normal office operations are unaffected. Through the overlay process described above, a tiered warning signal with clear geographic fences is finally generated. This signal can intuitively display the differences in disaster risk between different areas of the power plant.

[0099] By combining historical data of the power plant with equipment thresholds to divide the continuous intervals of the composite disaster risk index to trigger corresponding early warning levels, and relying on the geographic information system to overlay the early warning level with the spatial location characteristics of the preliminary disaster event set to generate a geofence, the effect of providing precise zoning and clearly defined geofenced graded early warning signals for different areas of the power plant is achieved.

[0100] In summary, the power plant meteorological disaster classification and early warning method based on multi-source data provided in this application has the following technical effects:

[0101] This application acquires meteorological big data of the target power plant area and simultaneously collects group perception data, including social media data, news video data, and other group perception data. It identifies disaster events from the group perception data to obtain micro-scale disaster auxiliary information, cross-validates this information with meteorological big data to form a preliminary disaster event set. Based on the meteorological big data and the preliminary disaster event set, it constructs a multi-dimensional disaster factor set containing at least two meteorological factors. Based on the multi-dimensional disaster factor set, it performs multi-hazard composite analysis to generate a composite disaster risk index. Then, it combines the composite disaster risk index with the spatial location characteristics of the preliminary disaster event set to generate a graded early warning signal with geofencing, thereby achieving accurate classification and early warning of meteorological disasters at the power plant. This makes the early warning results more timely and targeted, achieving accurate identification of micro-scale disasters and multi-hazard composite early warning, meeting the technical requirements for accurate early warning and targeted protection against meteorological disasters at the power plant.

[0102] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a power plant meteorological disaster classification and early warning system based on multi-source data. The system includes:

[0103] The crowd perception data acquisition module 1 is used to acquire meteorological big data of the target power station area and simultaneously acquire crowd perception data related to the target power station area. The crowd perception data includes social media data, news video data and other crowd perception data.

[0104] The preliminary disaster event set acquisition module 2 is used to identify disaster events from the group perception data, obtain microscale disaster auxiliary information, and cross-validate the microscale disaster auxiliary information with the meteorological big data to form a preliminary disaster event set.

[0105] Multi-dimensional disaster factor set construction module 3, which constructs a multi-dimensional disaster factor set based on the meteorological big data and the preliminary disaster event set, the multi-dimensional disaster factor set containing at least two meteorological factors.

[0106] The disaster classification and early warning execution module 4 is used to perform multi-hazard composite analysis based on the multi-dimensional disaster factor set, generate a composite disaster risk index, and perform disaster classification and early warning based on the composite disaster risk index.

[0107] Furthermore, the preliminary disaster event set acquisition module 2 is used to perform the following steps:

[0108] Text features containing disaster description semantics and their associated geocodings are extracted from the social media data; visual features conforming to a preset disaster pattern and their shooting location information are identified from the news video data; spatial clustering is performed based on the geocodings and shooting location information, and microscale disaster auxiliary information with disaster type labels is generated by combining the text features and visual features.

[0109] Furthermore, the preliminary disaster event set acquisition module 2 is used to perform the following steps:

[0110] The disaster type labels are compared with the corresponding spatial location monitoring indicators in meteorological big data to establish a spatiotemporal alignment matrix between meteorological data and mass perception data. Based on the spatiotemporal alignment matrix, the validity of disaster events is verified, and all valid disaster events are aggregated according to spatial location to generate a preliminary disaster event set containing the disaster type-spatial distribution mapping relationship.

[0111] Furthermore, the preliminary disaster event set acquisition module 2 is used to perform the following steps:

[0112] When the disaster report density in the group perception data exceeds the first threshold and the meteorological data anomaly degree of the corresponding spatiotemporal node exceeds the second threshold, the disaster event is confirmed to be valid and verified by meteorological data; group perception noise data that fails to pass meteorological data verification is removed to form a preliminary set of spatiotemporally calibrated disaster events.

[0113] Furthermore, the preliminary disaster event set acquisition module 2 is used to perform the following steps:

[0114] Each event in the preliminary disaster event set is labeled with a disaster type tag, and each disaster type is associated with a list of affected equipment based on the power plant equipment vulnerability database.

[0115] Furthermore, the multi-dimensional disaster factor set construction module 3 is used to perform the following steps:

[0116] Based on the disaster type-spatial distribution mapping relationship in the preliminary disaster event set, a target disaster type combination is determined; based on the target disaster type combination, relevant basic meteorological factors are extracted from the meteorological big data, and derived disaster factors are extracted from the preliminary disaster event set; a mapping relationship matrix between the basic meteorological factors and the derived disaster factors is established to obtain the multi-dimensional disaster factor set.

[0117] Furthermore, the disaster classification and early warning execution module 4 is used to perform the following steps:

[0118] A hierarchical influence relationship model of each factor in the multi-dimensional disaster factor set is established; the intensity of the chain effect between factors is calculated based on the hierarchical influence relationship model; the chain effect intensity is superimposed with the power plant area characteristic coefficient to output the composite disaster risk index.

[0119] Furthermore, the disaster classification and early warning execution module 4 is used to perform the following steps:

[0120] The factors that have a direct physical effect on power plant equipment are concentrated into primary factors; the factors that indirectly affect the power plant equipment through the primary factors are designated as secondary factors; a nonlinear mapping relationship between the primary and secondary factors is established, and the intensity of the chain effect is quantified by the transmission attenuation coefficient to construct a hierarchical influence relationship model; wherein, the transmission attenuation coefficient is dynamically adjusted according to the correlation strength between factors in historical disaster events.

[0121] Furthermore, the disaster classification and early warning execution module 4 is used to perform the following steps:

[0122] Based on the continuous numerical range of the composite disaster risk index, the corresponding early warning level is triggered; the early warning level is superimposed with the spatial location features in the preliminary disaster event set to generate a graded early warning signal with geofencing.

[0123] The power plant meteorological disaster classification and early warning system based on multi-source data provided in this embodiment of the invention can execute the power plant meteorological disaster classification and early warning method based on multi-source data provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0124] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0125] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for classifying and issuing early warning of meteorological disasters in power plants based on multi-source data, characterized in that: The method includes: Meteorological big data of the target power station area is acquired, and crowd perception data related to the target power station area is collected simultaneously. The crowd perception data includes social media data, news video data and other crowd perception data. Disaster event identification is performed on the group perception data to obtain micro-scale disaster auxiliary information, and the micro-scale disaster auxiliary information is cross-validated with the meteorological big data to form a preliminary disaster event set. Based on the meteorological big data and preliminary disaster event set, a multi-dimensional disaster factor set is constructed, which includes at least two meteorological factors; Based on the multi-dimensional disaster factor set, a multi-hazard composite analysis is performed to generate a composite disaster risk index, and disaster classification and early warning are carried out based on the composite disaster risk index. Disaster event identification is performed on the aforementioned group perception data to obtain micro-scale disaster auxiliary information, including: Extract textual features containing disaster description semantics and their associated geocoding from the social media data; Identify visual features and their shooting location information that match a preset disaster pattern from the news video data; Spatial clustering is performed based on the geocoding and shooting location information, and combined with the text features and visual features, microscale disaster auxiliary information with disaster type labels is generated; The microscale disaster auxiliary information is cross-validated with the meteorological big data to form a preliminary disaster event set, including: The disaster type labels are compared with the monitoring indicators of the corresponding spatial locations in the meteorological big data to establish a spatiotemporal alignment matrix between meteorological data and mass perception data. Based on the spatiotemporal alignment matrix, the validity of disaster events is verified, and all valid disaster events are aggregated according to spatial location to generate a preliminary disaster event set containing the disaster type-spatial distribution mapping relationship; Based on the aforementioned meteorological big data and preliminary disaster event set, a multi-dimensional disaster factor set is constructed. This multi-dimensional disaster factor set contains at least two meteorological factors, including: Based on the disaster type-spatial distribution mapping relationship in the preliminary disaster event set, determine the target disaster type combination; Based on the target disaster type combination, relevant basic meteorological factors are extracted from the meteorological big data, and derived disaster factors are extracted from the preliminary disaster event set; Establish a mapping matrix between the basic meteorological factors and the derived disaster factors to obtain the multi-dimensional disaster factor set; Based on the aforementioned multi-dimensional disaster factor set, a multi-hazard composite analysis is performed to generate a composite disaster risk index, including: Establish a hierarchical influence relationship model for each factor in the multidimensional disaster factor set; Calculate the intensity of the chain effect between factors based on the hierarchical influence relationship model; The combined disaster risk index is output by superimposing the intensity of the chain effect and the regional characteristic coefficient of the power station.

2. The power plant meteorological disaster classification and early warning method based on multi-source data as described in claim 1, characterized in that, Based on the aforementioned spatiotemporal alignment matrix, the validity of the disaster event is verified, including: When the disaster report density in the group perception data exceeds the first threshold and the meteorological data anomaly degree of the corresponding spatiotemporal node exceeds the second threshold, the disaster event is confirmed to be valid and verified by meteorological data. After removing the group-perceived noise data that failed to pass meteorological data verification, a preliminary set of disaster events with spatiotemporal calibration is formed.

3. The power plant meteorological disaster classification and early warning method based on multi-source data as described in claim 2, characterized in that, The method further includes: Each event in the preliminary disaster event set is labeled with a disaster type tag, and each disaster type is associated with a list of affected equipment based on the power plant equipment vulnerability database.

4. The power plant meteorological disaster classification and early warning method based on multi-source data as described in claim 1, characterized in that, The construction of the hierarchical influence relationship model includes: The factors that have a direct physical effect on power plant equipment are concentrated into the multi-dimensional disaster factors and are regarded as primary factors. Factors that indirectly influence the primary factors are considered as secondary factors. Establish a nonlinear mapping relationship between primary and secondary factors, quantify the intensity of chain effects through the transmission attenuation coefficient, and construct a hierarchical influence relationship model. The conduction attenuation coefficient is dynamically adjusted based on the correlation strength between factors in historical disaster events.

5. The power plant meteorological disaster classification and early warning method based on multi-source data as described in claim 1, characterized in that, Disaster classification and early warning are carried out based on the aforementioned composite disaster risk index, including: Based on the continuous numerical range of the composite disaster risk index, the corresponding early warning level is triggered. The warning level is superimposed with the spatial location features in the preliminary disaster event set to generate a graded warning signal with geofencing.

6. A power plant meteorological disaster classification and early warning system based on multi-source data, characterized in that, The system is used to implement the power plant meteorological disaster classification and early warning method based on multi-source data as described in any one of claims 1-5, the system comprising: The crowd perception data acquisition module is used to acquire meteorological big data of the target power station area and simultaneously acquire crowd perception data related to the target power station area. The crowd perception data includes social media data, news video data and other crowd perception data. The preliminary disaster event set acquisition module is used to identify disaster events from the group perception data, acquire micro-scale disaster auxiliary information, and cross-validate the micro-scale disaster auxiliary information with the meteorological big data to form a preliminary disaster event set. A multi-dimensional disaster factor set construction module constructs a multi-dimensional disaster factor set based on the meteorological big data and the preliminary disaster event set, wherein the multi-dimensional disaster factor set contains at least two meteorological factors; The disaster classification and early warning execution module is used to perform multi-hazard composite analysis based on the multi-dimensional disaster factor set, generate a composite disaster risk index, and perform disaster classification and early warning based on the composite disaster risk index.

Images