A wind farm blade icing disaster forecasting and early warning method based on multi-source meteorological fusion

Abstract

The application discloses a wind farm blade icing disaster forecasting and early warning method based on multi-source meteorological fusion, and particularly relates to the field of wind farm operation safety and meteorological information processing, and comprises the following steps: collecting wind speed change information, humidity change information, temperature change information and blade adhesion change information of multiple edge nodes, performing arrangement processing on the collected information according to node positions, performing cutting processing in time sequence, identifying wind speed mutation segments, humidity accumulation segments, temperature drop segments and adhesion growth segments and dividing the segments into characteristic change segments; the wind speed change, humidity change, temperature change and blade adhesion change of each node are arranged into the characteristic change segments which can reflect the icing mechanism evolution sequence, and the spatio-temporal dislocation and mechanism conflict are eliminated in the cross-node comparison, section corresponding and feature zoning construction layer-by-layer fusion process, so that the regional icing situation can be predicted and warned.

Description

Technical Field

[0001] This invention relates to the field of wind farm operation safety and meteorological information processing technology, and more specifically, to a method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion. Background Technology

[0002] In the current wind farm blade icing monitoring system, multi-source meteorological and equipment status data are collected independently by multiple distributed edge nodes. Due to differences in micro-topography, turbulent structure, and wind field redistribution, these nodes actually perceive icing processes with very different mechanisms: windward slopes are mostly characterized by high-speed inertial impaction-type icing, while valley bottoms are prone to rime accumulation dominated by moisture retention, and leeward slopes often experience freeze-thaw cycle-type icing. As a result, the observation sequences formed have significant differences in temporal rhythm, mechanistic attributes, and spatial representativeness, and are far from simple spatial sampling of the same disaster process.

[0003] However, existing forecasting and early warning methods rely on unstable edge data transmission, directly using these heterogeneous, time-stamped, and inconsistent multi-source observations as fusion inputs, and constructing a unified regional icing situation field through interpolation, normalization, or weighting. This fusion framework, which substitutes numerical consistency for mechanistic consistency, will mix rapidly growing icing signals with slowly accumulating icing signals in cloud-based situation prediction, and may even cause incorrect compensation due to transmission delays at some key nodes, thus reconstructing the real icing evolution into a unified, illusory situation.

[0004] Therefore, under the actual technical conditions of the coexistence of multi-source meteorological fusion and unstable edge data transmission, existing methods mistakenly treat spatiotemporally asynchronous and mechanistically heterogeneous local icing observations as different fragments of the same situation, resulting in a systematic deviation between the regional icing disaster field generated by cloud-based situation prediction and the actual physical process. This makes it difficult to meet the reliability requirements of refined icing disaster early warning for wind farms, which has become an urgent technical problem to be solved. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a wind farm blade icing disaster forecasting and early warning method based on multi-source meteorological fusion. Under the conditions of unstable edge data transmission and heterogeneous multi-source observation mechanisms, the wind speed changes, humidity changes, temperature changes, and blade adhesion changes of each node are organized into characteristic change segments that can reflect the evolution sequence of icing mechanism. In the layer-by-layer fusion process of cross-node comparison, segment correspondence, and characteristic partitioning construction, the spatiotemporal misalignment and mechanism conflict are eliminated, thereby achieving accurate prediction and stable early warning of regional icing situation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion, comprising:

[0007] S1. By collecting wind speed change information, humidity change information, temperature change information and blade attachment change information from multiple edge nodes, the collected information is sorted according to node position and segmented according to time sequence to identify wind speed change segments, humidity accumulation segments, temperature decrease segments and attachment amount increase segments and divide them into characteristic change segments.

[0008] S2. By analyzing the characteristic change segments of each node, the succession relationship between the wind speed change segment and the adhesion growth segment, the humidity accumulation segment and the adhesion growth segment, and the temperature decrease segment and the adhesion growth segment are obtained. Based on the analysis results, the characteristic succession sequence of each node is obtained.

[0009] S3. By performing comparison processing on the feature continuation sequences of multiple nodes according to the node positions, and performing corresponding processing according to the order of feature change segments, the segment correspondence data between nodes is obtained from the comparison processing and the corresponding processing.

[0010] S4. By performing grouping processing based on the segment correspondence data between nodes, nodes with the same characteristic continuation sequence arrangement pattern are grouped into the same group, and the nodes in each group are reorganized according to the order of characteristic change segments to obtain a feature partition set.

[0011] S5. By performing combination processing on the feature partition set according to the spatial adjacency relationship between partitions, and by performing combination processing according to the trend relationship of feature change segments within partitions, the regional icing disaster forecast and early warning results are obtained from the combination processing.

[0012] In a preferred embodiment, S1 further includes reading wind speed change information, humidity change information, temperature change information and blade attachment change information of multiple edge nodes, obtaining a set of change information belonging to the same node, and sorting and outputting all change information belonging to the same node in sequence according to the node position and the time order of each change information recording itself as the change information sequence corresponding to the node, and excluding change information that does not belong to the same node and not entering subsequent processing.

[0013] The recording time of each change information in the change information sequence is obtained, and time-by-time segmentation is performed. In the time-by-time segmentation, the change information with continuous recording time is grouped into the same time segment according to whether the recording time of adjacent change information is continuous, and the change information with discontinuous recording time is output as an independent time segment. All time segments are output together as a time segment set.

[0014] For the wind speed change information, humidity change information, temperature change information, and blade adhesion change information in each time segment of the time segment set, wind speed change analysis processing, humidity change analysis processing, temperature change analysis processing, and adhesion change analysis processing are performed respectively.

[0015] In a preferred embodiment, S1 further includes comparing the magnitude of change based on the order of recording the change information within the same time period in each analysis process, and determining whether the time period meets the conditions of sudden wind speed change, humidity accumulation, temperature decrease, and adhesion increase.

[0016] After the analysis is completed, the time segments that meet the wind speed change condition are output as wind speed change segments, the time segments that meet the humidity accumulation condition are output as humidity accumulation segments, the time segments that meet the temperature decrease condition are output as temperature decrease segments, the time segments that meet the adhesion growth condition are output as adhesion growth segments, and the time segments that do not meet any of the above conditions are output as unclassified time segments, and the unclassified time segments are excluded.

[0017] The start and end times of each segment are obtained by reading segments of sudden wind speed changes, cumulative humidity, temperature decrease, and adhesion increase. Segment-by-segment combination processing is performed, and multiple segments belonging to the same node with consecutive start and end times are sequentially combined and output as feature change segments of the corresponding node. Segments with discontinuous start and end times are output as independent feature change segments.

[0018] In a preferred embodiment, S2 further includes reading the feature change segments of each node, obtaining the segment type and start and end time, and arranging the feature change segments belonging to the same node in the order of their start and end times to output a time-ordered segment sequence, and recording the feature change segments with repeated start and end times or abnormal start and end time intervals during the sorting process as sorting deviation information.

[0019] Obtain two adjacent feature change segments in a time-ordered segment sequence and perform step-by-step comparison processing: when the previous segment is a wind speed change segment and the next segment is an adhesion increase segment, the previous segment is a humidity accumulation segment and the next segment is an adhesion increase segment, or the previous segment is a temperature decrease segment and the next segment is an adhesion increase segment, output the two segments as a consecutive segment pair.

[0020] When the preceding and following feature change segments do not belong to any of the above exclusive combination relationships, the two segments are output as a non-contiguous segment pair, and the difference between their start and end times is recorded as comparison deviation information.

[0021] In a preferred embodiment, S2 further includes reading contiguous fragment pairs, obtaining all contiguous fragment pairs belonging to the same node, and performing fragment chain construction processing.

[0022] In the fragment chain construction process, when processing the first contiguous fragment pair, a fragment chain is constructed using the first contiguous fragment pair. When processing subsequent contiguous fragment pairs, if the start time of the subsequent contiguous fragment pair is consecutively connected to the end time of the last contiguous fragment pair in the fragment chain, then the subsequent contiguous fragment pair is added to the fragment chain; otherwise, a new fragment chain is constructed using the subsequent contiguous fragment pair.

[0023] During the construction of the fragment chain, at least one of the following situations, namely time jump, order reversal and fragment missing, that occurs within the fragment chain, is recorded as fragment chain deviation information. After the construction is completed, all fragment chains are output as a fragment chain set.

[0024] The start and end time order of the time-ordered segment sequence is adjusted according to the sorting deviation information, the combination relationship is checked according to the comparison deviation information, and the discontinuous segments in the segment chain set are corrected according to the segment chain deviation information.

[0025] When a conflict occurs between start and end times, the start and end times of the characteristic change segment are re-divided, and the corrected time-ordered segment sequence and the corrected continuation segment pair are output.

[0026] By reading the corrected contiguous fragment pairs, the start and end time order of each contiguous fragment pair is obtained, and sequence induction processing is performed. In the sequence induction processing, contiguous fragment pairs within the same fragment chain are recorded sequentially according to the continuity of start and end times and merged and output as the characteristic contiguous sequence of the corresponding node. When a node does not have a contiguous fragment pair, the characteristic contiguous sequence of that node is set to a sequence that does not contain a contiguous fragment pair.

[0027] In a preferred embodiment, S3 further includes obtaining the node positions corresponding to each feature continuation sequence and performing node-by-node sorting processing on each feature continuation sequence.

[0028] In the node-by-node sorting process, the continuation segments within each feature continuation sequence are formed into a time sorting chain according to the order of their start and end times, and the continuation sequences between each feature continuation sequence are formed into a position sorting chain according to the order of their node positions. The time sorting chain and the position sorting chain are output together as the sorting benchmark set.

[0029] By reading the sorting benchmark set, the successive fragment pairs in the time sorting chain and the successive fragment pairs corresponding to adjacent nodes in the position sorting chain are obtained. Time comparison windows are established one by one according to the start and end times of the successive fragment pairs in the time sorting chain. In each time comparison window, the fragment type and start and end time differences of the successive fragment pairs of adjacent nodes are compared.

[0030] Record cases where the difference between start and end times does not meet the continuity condition as corresponding deviation information, and output cases where the difference between start and end times meets the continuity condition and the segment types are consistent as cross-node corresponding segment pairs.

[0031] By reading cross-node corresponding fragment pairs, performing group-by-group aggregation processing, cross-node corresponding fragment pairs that belong to the same fragment type and have a sequential relationship in the time sorting chain of multiple nodes are aggregated into cross-node corresponding chains in turn, and cross-node corresponding fragment pairs that do not have a sequential relationship are formed into independent corresponding chains, and the cross-node corresponding chains and independent corresponding chains are output as a set of corresponding chains.

[0032] By reading the corresponding chain set and performing segment construction processing, the corresponding deviation information is fed back and corrected according to the continuity of each corresponding chain in the node position and the order of the time sequence chain. The cross-node segment combination is formed in sequence according to the corrected corresponding chain, and the cross-node segment combination is output as the segment correspondence data between nodes.

[0033] In a preferred embodiment, S4 further includes obtaining the node position and segment combination corresponding to each node in the segment corresponding data and performing node-by-node indexing processing. In the node-by-node indexing processing, a node position index chain is constructed in the order of node positions, and a segment fragment index chain is constructed in the order of feature change fragments recorded in the segment combination. The node position index chain and the segment fragment index chain are output together as a segment index set.

[0034] By reading the adjacent nodes of the node position index chain in the segment index set, and the corresponding feature change segment sequence of the adjacent nodes in the segment fragment index chain, and performing cross-filtering processing;

[0035] In the cross-screening process, the sequence of consecutive feature change segments in the feature change segment sequence corresponding to adjacent nodes is compared to form the segment correspondence. The segment correspondence with the consecutive relationship is output as the node segment correspondence chain, and the segment correspondence with the consecutive relationship is output as the node segment deviation information. All node segment correspondence chains are output as the segment correspondence chain set.

[0036] In a preferred embodiment, S4 further includes obtaining all feature change segments in the fragment corresponding chain set by reading the fragment corresponding chain set, establishing a fragment sequence chain based on the order in which the feature change segments appear in the fragment corresponding chain set, and comparing the start and end times of adjacent feature change segments in the fragment sequence chain with the continuity relationship of the corresponding node positions in the node position index chain.

[0037] The feature change segments with continuous relationships are aggregated to form grouped reconstruction chains, and the feature change segments with non-continuous relationships are formed into independent grouped reconstruction chains. All grouped reconstruction chains are output as the initial reconstruction chain set.

[0038] Based on the start and end time differences recorded in the node segment deviation information, the start and end time order of the corresponding feature change segments in the segment sequence chain is corrected. The corrected start and end times are reused to form the grouped reconstruction chain, and the corrected grouped reconstruction chain is merged with the initial reconstruction chain set to output the updated reconstruction chain set.

[0039] Based on the continuity relationship of each group reconstruction chain in the updated reconstruction chain set at the node position and the order of feature change segments, the group reconstruction chains with valid continuity relationships are aggregated in sequence to form the first type of feature partition, and the group reconstruction chains with invalid continuity relationships are set as the second type of feature partition, and the first type of feature partition and the second type of feature partition are output together as the feature partition set.

[0040] In a preferred embodiment, S5 further includes obtaining the node positions corresponding to each feature partition in the feature partition set, and connecting the feature partitions with consecutive node positions in sequence to form a spatial continuous chain according to the order of the node positions, recording the feature partitions with discontinuous node positions as spatial interval segments, and outputting the spatial continuous chain and spatial interval segments together as a spatial relationship set.

[0041] By reading the feature partition set, all feature change segments within each feature partition are obtained, and a trend change sequence is formed according to the registration order of the feature change segments. Feature change segments with continuous trend direction are connected in sequence to form a trend continuous chain, and feature change segments with discontinuous trend direction are recorded as trend interval segments. The trend continuous chain and trend interval segments are output together as a trend relationship set.

[0042] Read the spatial relationship set and the trend relationship set, and obtain the adjacent feature partitions in the spatial continuous chain, as well as the feature change segment sequence corresponding to the adjacent feature partitions in the trend relationship set. Compare the node position continuous relationship and the trend direction continuous relationship one by one. Record the comparison results that satisfy both continuous relationships as the spatial trend correspondence chain, and record the comparison results that do not satisfy both relationships as the spatial trend deviation information. Output all spatial trend correspondence chains as the cross correspondence set.

[0043] When the spatial trend deviation information records a discontinuous trend direction, the corresponding feature change segments in the trend relationship set are adjusted in sequence; when the spatial trend deviation information records a discontinuous node position, the corresponding feature partitioning in the spatial relationship set is adjusted in sequence; the corrected spatial relationship set and the trend relationship set are recombined and output as an updated structure set.

[0044] The algorithm obtains the node position order of each feature partition in the updated structure set and the trend change relationship of the corresponding feature change segments. It then combines the feature partitions with continuous node positions and consistent trend directions to form a continuous integration chain. The feature partitions with node position intervals and trend direction intervals are recorded as independent integration segments. Finally, the continuous integration chain and the independent integration segments are output together as the regional icing disaster forecast and early warning results.

[0045] The technical effects and advantages of this invention are as follows:

[0046] This method first separates the physical mechanism of characteristic change segments at the edge, and then uses cross-node sorting, corresponding chain construction and multi-level deviation feedback correction in the cloud to prevent spatiotemporally asynchronous and mechanism-heterogeneous local observations from being misjudged as the same situation segment. This fundamentally eliminates the systematic bias of cloud situation prediction and achieves refined and reliable regional icing early warning.

[0047] By identifying sudden changes in wind speed, accumulated humidity, temperature drop, and increased adhesion in time segments, constructing consecutive segment pairs according to the continuation conditions, and using segment chain deviation information for correction, the single-node feature continuation sequence can truly express the local icing trigger chain, avoiding segment misalignment interference caused by edge transmission delay.

[0048] By cross-comparing time sorting chains, position sorting chains, cross-node corresponding segment pairs and corresponding chain sets, and then using deviation information feedback to correct segment combinations, the characteristic change segments between different nodes are re-established in space to restore the continuous relationship, thereby repairing the spatial break caused by unstable edge data.

[0049] By performing node indexing, fragment chain extraction, group reconstruction chain generation and correction on the corresponding data of the segments, the resulting feature partition set has both the continuity of node positions and the consistency of mechanism evolution, ensuring that each partition is a spatial-temporal substructure of the real icing evolution process, rather than a simple numerical clustering result.

[0050] By cross-comparing spatial and trend continuous chains, a dual correction of spatial and trend relationships is achieved. Then, a continuous integration chain is generated from the updated structure set, so that the regional icing situation maintains both spatial continuity and consistency of mechanism and trend. Attached Figure Description

[0051] Figure 1 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Refer to the instruction manual appendix Figure 1 An embodiment of the present invention provides a method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion, comprising:

[0054] S1. By collecting wind speed change information, humidity change information, temperature change information and blade attachment change information from multiple edge nodes, the collected information is sorted according to node position and segmented according to time sequence to identify wind speed change segments, humidity accumulation segments, temperature decrease segments and attachment amount increase segments and divide them into characteristic change segments.

[0055] S2. By analyzing the characteristic change segments of each node, the succession relationship between the wind speed change segment and the adhesion growth segment, the humidity accumulation segment and the adhesion growth segment, and the temperature decrease segment and the adhesion growth segment are obtained. Based on the analysis results, the characteristic succession sequence of each node is obtained.

[0056] S3. By performing comparison processing on the feature continuation sequences of multiple nodes according to the node positions, and performing corresponding processing according to the order of feature change segments, the segment correspondence data between nodes is obtained from the comparison processing and the corresponding processing.

[0057] S4. By performing grouping processing based on the segment correspondence data between nodes, nodes with the same characteristic continuation sequence arrangement pattern are grouped into the same group, and the nodes in each group are reorganized according to the order of characteristic change segments to obtain a feature partition set.

[0058] S5. By performing combination processing on the feature partition set according to the spatial adjacency relationship between partitions, and by performing combination processing according to the trend relationship of feature change segments within partitions, the regional icing disaster forecast and early warning results are obtained from the combination processing.

[0059] In S1, the wind speed change information, humidity change information, temperature change information and blade attachment change information of multiple edge nodes are also read to obtain the change information set belonging to the same node. All change information belonging to the same node is sorted and output as the change information sequence corresponding to the node according to the node position and the time order of each change information. Change information that does not belong to the same node is excluded and does not enter the subsequent processing.

[0060] The recording time of each change information in the change information sequence is obtained, and time-by-time segmentation is performed. In the time-by-time segmentation, the change information with continuous recording time is grouped into the same time segment according to whether the recording time of adjacent change information is continuous, and the change information with discontinuous recording time is output as an independent time segment. All time segments are output together as a time segment set.

[0061] For the wind speed change information, humidity change information, temperature change information, and blade adhesion change information in each time segment of the time segment set, wind speed change analysis processing, humidity change analysis processing, temperature change analysis processing, and adhesion change analysis processing are performed respectively.

[0062] In S1, it also includes comparing the magnitude of change based on the order of the recorded time of change information within the same time period in each analysis and processing, and determining whether the time period meets the conditions of sudden wind speed change, humidity accumulation, temperature decrease, and adhesion increase.

[0063] After the analysis is completed, the time segments that meet the wind speed change condition are output as wind speed change segments, the time segments that meet the humidity accumulation condition are output as humidity accumulation segments, the time segments that meet the temperature decrease condition are output as temperature decrease segments, the time segments that meet the adhesion growth condition are output as adhesion growth segments, and the time segments that do not meet any of the above conditions are output as unclassified time segments. Unclassified time segments are excluded so that they do not enter any segment combination processing.

[0064] The start and end times of each segment are obtained by reading segments of sudden wind speed changes, cumulative humidity, temperature decrease, and adhesion increase. Segment-by-segment combination processing is performed, and multiple segments belonging to the same node with consecutive start and end times are sequentially combined and output as feature change segments of the corresponding node. Segments with discontinuous start and end times are output as independent feature change segments.

[0065] In S2, it also includes reading the feature change segments of each node, obtaining the segment type and start and end time, and arranging the feature change segments belonging to the same node in the order of their start and end times to output a time-ordered segment sequence. Feature change segments with repeated start and end times or abnormal start and end time intervals during the sorting process are recorded as sorting deviation information.

[0066] Obtain two adjacent feature change segments in a time-ordered segment sequence and perform step-by-step comparison processing: when the previous segment is a wind speed change segment and the next segment is an adhesion increase segment, the previous segment is a humidity accumulation segment and the next segment is an adhesion increase segment, or the previous segment is a temperature decrease segment and the next segment is an adhesion increase segment, output the two segments as a consecutive segment pair.

[0067] When the preceding and following feature change segments do not belong to any of the above exclusive combination relationships, the two segments are output as a non-contiguous segment pair, and the difference between their start and end times is recorded as comparison deviation information.

[0068] S2 also includes reading contiguous fragment pairs, obtaining all contiguous fragment pairs belonging to the same node, and performing fragment chain construction processing;

[0069] In the fragment chain construction process, when processing the first contiguous fragment pair, a fragment chain is constructed using the first contiguous fragment pair. When processing subsequent contiguous fragment pairs, if the start time of the subsequent contiguous fragment pair is consecutively connected to the end time of the last contiguous fragment pair in the fragment chain, then the subsequent contiguous fragment pair is added to the fragment chain; otherwise, a new fragment chain is constructed using the subsequent contiguous fragment pair.

[0070] During the construction of the fragment chain, at least one of the following situations, namely time jump, order reversal and fragment missing, that occurs within the fragment chain, is recorded as fragment chain deviation information. After the construction is completed, all fragment chains are output as a fragment chain set.

[0071] The start and end time order of the time-ordered segment sequence is adjusted according to the sorting deviation information, the combination relationship is checked according to the comparison deviation information, and the discontinuous segments in the segment chain set are corrected according to the segment chain deviation information.

[0072] When a conflict occurs between start and end times, the start and end times of the characteristic change segment are re-divided, and the corrected time-ordered segment sequence and the corrected continuation segment pair are output.

[0073] By reading the corrected contiguous fragment pairs, the start and end time order of each contiguous fragment pair is obtained, and sequence induction processing is performed. In the sequence induction processing, contiguous fragment pairs within the same fragment chain are recorded sequentially according to the continuity of start and end times and merged and output as the characteristic contiguous sequence of the corresponding node. When a node does not have a contiguous fragment pair, the characteristic contiguous sequence of that node is set to a sequence that does not contain a contiguous fragment pair.

[0074] In S3, it is also included to obtain the node positions corresponding to each feature continuation sequence and to perform node-by-node sorting processing on each feature continuation sequence.

[0075] In the node-by-node sorting process, the continuation segments within each feature continuation sequence are formed into a time sorting chain according to the order of their start and end times, and the continuation sequences between each feature continuation sequence are formed into a position sorting chain according to the order of their node positions. The time sorting chain and the position sorting chain are output together as the sorting benchmark set.

[0076] By reading the sorting benchmark set, the successive fragment pairs in the time sorting chain and the successive fragment pairs corresponding to adjacent nodes in the position sorting chain are obtained. Time comparison windows are established one by one according to the start and end times of the successive fragment pairs in the time sorting chain. In each time comparison window, the fragment type and start and end time differences of the successive fragment pairs of adjacent nodes are compared.

[0077] Record cases where the difference between start and end times does not meet the continuity condition as corresponding deviation information, and output cases where the difference between start and end times meets the continuity condition and the segment types are consistent as cross-node corresponding segment pairs.

[0078] By reading cross-node corresponding fragment pairs, performing group-by-group aggregation processing, cross-node corresponding fragment pairs that belong to the same fragment type and have a sequential relationship in the time sorting chain of multiple nodes are aggregated into cross-node corresponding chains in turn, and cross-node corresponding fragment pairs that do not have a sequential relationship are formed into independent corresponding chains, and the cross-node corresponding chains and independent corresponding chains are output as a set of corresponding chains.

[0079] By reading the corresponding chain set and performing segment construction processing, the corresponding deviation information is fed back and corrected according to the continuity of each corresponding chain in the node position and the order of the time sequence chain. The cross-node segment combination is formed in sequence according to the corrected corresponding chain, and the cross-node segment combination is output as the segment correspondence data between nodes.

[0080] In S4, it also includes obtaining the node position and segment combination corresponding to each node in the segment data and performing node-by-node indexing. In the node-by-node indexing, a node position index chain is constructed in the order of node positions, and a segment fragment index chain is constructed in the order of feature change fragments recorded in the segment combination. The node position index chain and the segment fragment index chain are output together as a segment index set.

[0081] By reading the adjacent nodes of the node position index chain in the segment index set, and the corresponding feature change segment sequence of the adjacent nodes in the segment fragment index chain, and performing cross-filtering processing;

[0082] In the cross-screening process, the sequence of consecutive feature change segments in the feature change segment sequence corresponding to adjacent nodes is compared to form the segment correspondence. The segment correspondence with the consecutive relationship is output as the node segment correspondence chain, and the segment correspondence with the consecutive relationship is output as the node segment deviation information. All node segment correspondence chains are output as the segment correspondence chain set.

[0083] In S4, it also includes obtaining all feature change segments in the fragment corresponding chain set by reading the fragment corresponding chain set, establishing a fragment sequence chain based on the order in which the feature change segments appear in the fragment corresponding chain set, and comparing the start and end times of adjacent feature change segments in the fragment sequence chain with the continuity relationship of the corresponding node positions in the node position index chain.

[0084] The feature change segments with continuous relationships are aggregated to form grouped reconstruction chains, and the feature change segments with non-continuous relationships are formed into independent grouped reconstruction chains. All grouped reconstruction chains are output as the initial reconstruction chain set.

[0085] Based on the start and end time differences recorded in the node segment deviation information, the start and end time order of the corresponding feature change segments in the segment sequence chain is corrected. The corrected start and end times are reused to form the grouped reconstruction chain, and the corrected grouped reconstruction chain is merged with the initial reconstruction chain set to output the updated reconstruction chain set.

[0086] Based on the continuity relationship of each group reconstruction chain in the updated reconstruction chain set at the node position and the order of feature change segments, the group reconstruction chains with valid continuity relationships are aggregated in sequence to form the first type of feature partition, and the group reconstruction chains with invalid continuity relationships are set as the second type of feature partition, and the first type of feature partition and the second type of feature partition are output together as the feature partition set.

[0087] In S5, it also includes obtaining the node positions corresponding to each feature partition in the feature partition set, and connecting the feature partitions with consecutive node positions to form a spatial continuous chain according to the order of the node positions, recording the feature partitions with discontinuous node positions as spatial interval segments, and outputting the spatial continuous chain and spatial interval segments together as a spatial relationship set.

[0088] By reading the feature partition set, all feature change segments within each feature partition are obtained, and a trend change sequence is formed according to the registration order of the feature change segments. Feature change segments with continuous trend direction are connected in sequence to form a trend continuous chain, and feature change segments with discontinuous trend direction are recorded as trend interval segments. The trend continuous chain and trend interval segments are output together as a trend relationship set.

[0089] Read the spatial relationship set and the trend relationship set, and obtain the adjacent feature partitions in the spatial continuous chain, as well as the feature change segment sequence corresponding to the adjacent feature partitions in the trend relationship set. Compare the node position continuous relationship and the trend direction continuous relationship one by one. Record the comparison results that satisfy both continuous relationships as the spatial trend correspondence chain, and record the comparison results that do not satisfy both relationships as the spatial trend deviation information. Output all spatial trend correspondence chains as the cross correspondence set.

[0090] When the spatial trend deviation information records a discontinuous trend direction, the corresponding feature change segments in the trend relationship set are adjusted in sequence; when the spatial trend deviation information records a discontinuous node position, the corresponding feature partitioning in the spatial relationship set is adjusted in sequence; the corrected spatial relationship set and the trend relationship set are recombined and output as an updated structure set.

[0091] The algorithm obtains the node position order of each feature partition in the updated structure set and the trend change relationship of the corresponding feature change segments. It then combines the feature partitions with continuous node positions and consistent trend directions to form a continuous integration chain. The feature partitions with node position intervals and trend direction intervals are recorded as independent integration segments. Finally, the continuous integration chain and the independent integration segments are output together as the regional icing disaster forecast and early warning results.

[0092] This solution stems from a renewed understanding of the characteristics of wind farm field observations. Changes in wind speed, humidity, temperature, and blade adhesion are independently collected by multiple edge nodes. These edge nodes are inherently heterogeneous before data generation due to the influence of micro-topography and local wind field differences. Furthermore, because edge data transmission is unstable, issues such as timestamp misalignment and delayed data uploads often occur. If the cloud directly treats multi-source data as simple fragments of the same process when performing situation prediction, it will not only lose the true icing mechanism but also splice the independent icing processes of different nodes into an incorrect continuous situation. Therefore, this solution no longer uses the traditional numerical calibration method in S1. Instead, it organizes the change information collected by each node according to the node location and time sequence to form continuous or independent time segments. Then, based on the change amplitude of each change information within the time segment, it identifies wind speed change segments, humidity accumulation segments, temperature decrease segments, and blade adhesion increase segments to ensure that the subsequent processing is of characteristic change segments with clear physical meaning, rather than loose records affected by edge data transmission.

[0093] When extracting the icing mechanism within a single node, this scheme does not directly splice the segments in the order of occurrence in S2. Instead, it compares the succession relationship between characteristic change segments. Only when a segment of sudden wind speed change, humidity accumulation, or temperature decrease is immediately followed by a segment of increased adhesion amount does it form a succession segment pair. Through the construction of segment chains and feedback correction of various deviation information, the characteristic succession sequence becomes the mechanism chain expression of the real icing triggering mechanism at that node.

[0094] After obtaining the mechanism of a single node, the solution further addresses the issue of how to align cross-node data in space in S3 and S4. By constructing time sorting chains, position sorting chains, cross-node corresponding segment pairs, corresponding chain sets, and segment combinations within the region, and combining them with a feedback correction mechanism, a feature partition set that can simultaneously maintain spatial continuity and mechanism consistency is finally obtained. The existence of the feature partition set is essentially to ensure that the system can maintain spatial situational consistency when facing edge data transmission delays and data loss, rather than causing a jump in regional situational prediction due to data breakpoints.

[0095] In the final S5, this scheme does not directly use the feature partition set as the regional icing result. Instead, it forms a spatial continuous chain based on spatial adjacency and a trend continuous chain based on the trend changes of feature change segments within the partition. Then, through cross-comparison between spatial and trend relationships, feedback correction of deviation information, and secondary integration of the updated structure, a continuous integrated chain and independent integrated segments within the region are formed. In this process, the spatial continuity of feature partitions is used to repair the time discontinuity caused by edge data transmission, the trend continuity is used to maintain the temporal evolution consistency of the icing mechanism, and the spatial trend corresponding chain serves as the basic unit for situation prediction. This ensures that the regional icing disaster forecast and early warning results can reflect the real physical mechanism and are compatible with the observation gaps caused by unstable edge data transmission, thereby achieving stable regional icing situation prediction.

[0096] 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 forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion, characterized in that, include: S1. By collecting wind speed change information, humidity change information, temperature change information and blade attachment change information from multiple edge nodes, the collected information is sorted according to node position and segmented according to time sequence to identify wind speed change segments, humidity accumulation segments, temperature decrease segments and attachment amount increase segments and divide them into characteristic change segments. S2. By analyzing the characteristic change segments of each node, the succession relationship between the wind speed change segment and the adhesion growth segment, the humidity accumulation segment and the adhesion growth segment, and the temperature decrease segment and the adhesion growth segment are obtained. Based on the analysis results, the characteristic succession sequence of each node is obtained. S3. By performing comparison processing on the feature continuation sequences of multiple nodes according to the node positions, and performing corresponding processing according to the order of feature change segments, the segment correspondence data between nodes is obtained from the comparison processing and the corresponding processing. S4. By performing grouping processing based on the segment correspondence data between nodes, nodes with the same characteristic continuation sequence arrangement pattern are grouped into the same group, and the nodes in each group are reorganized according to the order of characteristic change segments to obtain a feature partition set. S5. By performing combination processing on the feature partition set according to the spatial adjacency relationship between partitions, and performing combination processing according to the trend relationship of feature change segments within partitions, the regional icing disaster forecast and early warning results are obtained from the combination processing. In S5, it also includes obtaining the node positions corresponding to each feature partition in the feature partition set, and connecting the feature partitions with consecutive node positions to form a spatial continuous chain according to the order of the node positions, recording the feature partitions with discontinuous node positions as spatial interval segments, and outputting the spatial continuous chain and spatial interval segments together as a spatial relationship set. By reading the feature partition set, all feature change segments within each feature partition are obtained, and a trend change sequence is formed according to the registration order of the feature change segments. Feature change segments with continuous trend direction are connected in sequence to form a trend continuous chain, and feature change segments with discontinuous trend direction are recorded as trend interval segments. The trend continuous chain and trend interval segments are output together as a trend relationship set. Read the spatial relationship set and the trend relationship set, and obtain the adjacent feature partitions in the spatial continuous chain, as well as the feature change segment sequence corresponding to the adjacent feature partitions in the trend relationship set. Compare the node position continuous relationship and the trend direction continuous relationship one by one. Record the comparison results that satisfy both continuous relationships as the spatial trend correspondence chain, and record the comparison results that do not satisfy both relationships as the spatial trend deviation information. Output all spatial trend correspondence chains as the cross correspondence set. When the spatial trend deviation information records a discontinuous trend direction, adjustments are made sequentially to the corresponding feature change segments in the trend relationship set. When the location of the spatial trend deviation information recording node is discontinuous, the corresponding feature partition order in the spatial relationship set is adjusted; the corrected spatial relationship set and the trend relationship set are then recombined and output as an updated structure set. The algorithm obtains the node position order of each feature partition in the updated structure set and the trend change relationship of the corresponding feature change segments. It then combines the feature partitions with continuous node positions and consistent trend directions to form a continuous integration chain. The feature partitions with node position intervals and trend direction intervals are recorded as independent integration segments. Finally, the continuous integration chain and the independent integration segments are output together as the regional icing disaster forecast and early warning results.

2. The method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion according to claim 1, characterized in that: In S1, the wind speed change information, humidity change information, temperature change information and blade attachment change information of multiple edge nodes are also read to obtain the change information set belonging to the same node. All change information belonging to the same node is sorted and output as the change information sequence corresponding to the node according to the node position and the time order of each change information. Change information that does not belong to the same node is excluded and does not enter the subsequent processing. The recording time of each change information in the change information sequence is obtained, and time-by-time segmentation is performed. In the time-by-time segmentation, the change information with continuous recording time is grouped into the same time segment according to whether the recording time of adjacent change information is continuous, and the change information with discontinuous recording time is output as an independent time segment. All time segments are output together as a time segment set. For the wind speed change information, humidity change information, temperature change information, and blade adhesion change information in each time segment of the time segment set, wind speed change analysis processing, humidity change analysis processing, temperature change analysis processing, and adhesion change analysis processing are performed respectively.

3. The method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion according to claim 2, characterized in that: In S1, it also includes comparing the magnitude of change based on the order of the recorded time of change information within the same time period in each analysis and processing, and determining whether the time period meets the conditions of sudden wind speed change, humidity accumulation, temperature decrease, and adhesion increase. After the analysis is completed, the time segments that meet the wind speed change condition are output as wind speed change segments, the time segments that meet the humidity accumulation condition are output as humidity accumulation segments, the time segments that meet the temperature decrease condition are output as temperature decrease segments, the time segments that meet the adhesion growth condition are output as adhesion growth segments, and the time segments that do not meet any of the above conditions are output as unclassified time segments, and the unclassified time segments are excluded. The start and end times of each segment are obtained by reading segments of sudden wind speed changes, cumulative humidity, temperature decrease, and adhesion increase. Segment-by-segment combination processing is performed, and multiple segments belonging to the same node with consecutive start and end times are sequentially combined and output as feature change segments of the corresponding node. Segments with discontinuous start and end times are output as independent feature change segments.

4. The method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion according to claim 3, characterized in that: In S2, it also includes reading the feature change segments of each node, obtaining the segment type and start and end time, and arranging the feature change segments belonging to the same node in the order of their start and end times to output a time-ordered segment sequence. Feature change segments with repeated start and end times or abnormal start and end time intervals during the sorting process are recorded as sorting deviation information. Obtain two adjacent feature change segments in a time-ordered segment sequence and perform step-by-step comparison processing: when the previous segment is a wind speed change segment and the next segment is an adhesion increase segment, the previous segment is a humidity accumulation segment and the next segment is an adhesion increase segment, or the previous segment is a temperature decrease segment and the next segment is an adhesion increase segment, output the two segments as a consecutive segment pair. When the preceding and following feature change segments do not belong to any of the above exclusive combination relationships, the two segments are output as a non-contiguous segment pair, and the difference between their start and end times is recorded as comparison deviation information.

5. The wind farm blade icing disaster forecasting and early warning method based on multi-source meteorological fusion according to claim 4, characterized in that: S2 also includes reading contiguous fragment pairs, obtaining all contiguous fragment pairs belonging to the same node, and performing fragment chain construction processing; In the fragment chain construction process, when processing the first contiguous fragment pair, a fragment chain is constructed using the first contiguous fragment pair. When processing subsequent contiguous fragment pairs, if the start time of the subsequent contiguous fragment pair is consecutively connected to the end time of the last contiguous fragment pair in the fragment chain, then the subsequent contiguous fragment pair is added to the fragment chain; otherwise, a new fragment chain is constructed using the subsequent contiguous fragment pair. During the construction of the fragment chain, at least one of the following situations, namely time jump, order reversal and fragment missing, that occurs within the fragment chain, is recorded as fragment chain deviation information. After the construction is completed, all fragment chains are output as a fragment chain set. The start and end time order of the time-ordered segment sequence is adjusted according to the sorting deviation information, the combination relationship is checked according to the comparison deviation information, and the discontinuous segments in the segment chain set are corrected according to the segment chain deviation information. When a conflict occurs between start and end times, the start and end times of the characteristic change segment are re-divided, and the corrected time-ordered segment sequence and the corrected continuation segment pair are output. By reading the corrected contiguous fragment pairs, the start and end time order of each contiguous fragment pair is obtained, and sequence induction processing is performed. In the sequence induction processing, contiguous fragment pairs within the same fragment chain are recorded sequentially according to the continuity of start and end times and merged and output as the characteristic contiguous sequence of the corresponding node. When a node does not have a contiguous fragment pair, the characteristic contiguous sequence of that node is set to a sequence that does not contain a contiguous fragment pair.

6. The method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion according to claim 5, characterized in that: In S3, it is also included to obtain the node positions corresponding to each feature continuation sequence and to perform node-by-node sorting processing on each feature continuation sequence. In the node-by-node sorting process, the continuation segments within each feature continuation sequence are formed into a time sorting chain according to the order of their start and end times, and the continuation sequences between each feature continuation sequence are formed into a position sorting chain according to the order of their node positions. The time sorting chain and the position sorting chain are output together as the sorting benchmark set. By reading the sorting benchmark set, the successive fragment pairs in the time sorting chain and the successive fragment pairs corresponding to adjacent nodes in the position sorting chain are obtained. Time comparison windows are established one by one according to the start and end times of the successive fragment pairs in the time sorting chain. In each time comparison window, the fragment type and start and end time differences of the successive fragment pairs of adjacent nodes are compared. Record cases where the difference between start and end times does not meet the continuity condition as corresponding deviation information, and output cases where the difference between start and end times meets the continuity condition and the segment types are consistent as cross-node corresponding segment pairs. By reading cross-node corresponding fragment pairs, performing group-by-group aggregation processing, cross-node corresponding fragment pairs that belong to the same fragment type and have a sequential relationship in the time sorting chain of multiple nodes are aggregated into cross-node corresponding chains in turn, and cross-node corresponding fragment pairs that do not have a sequential relationship are formed into independent corresponding chains, and the cross-node corresponding chains and independent corresponding chains are output as a set of corresponding chains. By reading the corresponding chain set and performing segment construction processing, the corresponding deviation information is fed back and corrected according to the continuity of each corresponding chain in the node position and the order of the time sequence chain. The cross-node segment combination is formed in sequence according to the corrected corresponding chain, and the cross-node segment combination is output as the segment correspondence data between nodes.

7. The method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion according to claim 6, characterized in that: In S4, it also includes obtaining the node position and segment combination corresponding to each node in the segment data and performing node-by-node indexing. In the node-by-node indexing, a node position index chain is constructed in the order of node positions, and a segment fragment index chain is constructed in the order of feature change fragments recorded in the segment combination. The node position index chain and the segment fragment index chain are output together as a segment index set. By reading the adjacent nodes of the node position index chain in the segment index set, and the corresponding feature change segment sequence of the adjacent nodes in the segment fragment index chain, and performing cross-filtering processing; In the cross-screening process, the sequence of consecutive feature change segments in the feature change segment sequence corresponding to adjacent nodes is compared to form the segment correspondence. The segment correspondence with the consecutive relationship is output as the node segment correspondence chain, and the segment correspondence with the consecutive relationship is output as the node segment deviation information. All node segment correspondence chains are output as the segment correspondence chain set.

8. The method for forecasting and early warning of wind farm blade icing disasters based on multi-source meteorological fusion according to claim 7, characterized in that: In S4, it also includes obtaining all feature change segments in the fragment corresponding chain set by reading the fragment corresponding chain set, establishing a fragment sequence chain based on the order in which the feature change segments appear in the fragment corresponding chain set, and comparing the start and end times of adjacent feature change segments in the fragment sequence chain with the continuity relationship of the corresponding node positions in the node position index chain. The feature change segments with continuous relationships are aggregated to form grouped reconstruction chains, and the feature change segments with non-continuous relationships are formed into independent grouped reconstruction chains. All grouped reconstruction chains are output as the initial reconstruction chain set. Based on the start and end time differences recorded in the node segment deviation information, the start and end time order of the corresponding feature change segments in the segment sequence chain is corrected. The corrected start and end times are reused to form the grouped reconstruction chain, and the corrected grouped reconstruction chain is merged with the initial reconstruction chain set to output the updated reconstruction chain set. Based on the continuity relationship of each group reconstruction chain in the updated reconstruction chain set at the node position and the order of feature change segments, the group reconstruction chains with valid continuity relationships are aggregated in sequence to form the first type of feature partition, and the group reconstruction chains with invalid continuity relationships are set as the second type of feature partition, and the first type of feature partition and the second type of feature partition are output together as the feature partition set.

Images