Wind power output power prediction method based on wide-area space-time relevance mining

Abstract

The invention discloses a wind power output power prediction method based on wide-area space-time relevance mining, and relates to the technical field of wind power, and the method comprises the steps: connecting a target wind turbine unit, and obtaining historical output power data and historical meteorological data; determining a first failure cycle and a first time distribution map; determining a second failure cycle, a second time distribution map, a third failure cycle and a third time distribution map; performing wide-area time axis relevance mining, and determining a time sequence relevance queue conforming to stack operation characteristics; determining a spatial association queue conforming to stack operation characteristics; and based on the historical output power data and the historical meteorological data, in combination with the time sequence association queue and the space association queue, performing wind power output power prediction to obtain a power prediction result. The technical problems that in the prior art, wind power output power prediction precision is low, and spatio-temporal data relevance cannot be fully utilized are solved, and the technical effect of improving wind power output power prediction precision is achieved.

Description

Technical Field

[0001] The present invention relates to the field of wind power technology, and in particular to a wind power output power prediction method based on wide-area spatiotemporal correlation mining. Background Art

[0002] With the continuous growth of global energy demand and the rapid development of renewable energy, wind energy, as an important green energy, has gradually been widely used around the world. However, with the continuous increase in wind power installed capacity and the expansion of wind farms, wind power output power prediction faces increasing challenges. Traditional wind power output power prediction methods usually rely on meteorological data and simple time series analysis, which have the problems of low prediction accuracy and inability to fully consider the complex relationship between spatiotemporal data. Especially when the weather changes drastically or the synergy of different units in the wind farm is complex, the existing prediction methods are less effective and it is difficult to meet the needs of high-precision and rapid response for power grid dispatching and wind farm operation management. Summary of the invention

[0003] The present application provides a wind power output power prediction method based on wide-area spatiotemporal correlation mining, which is used to solve the technical problems of low wind power output power prediction accuracy and inability to fully utilize spatiotemporal data correlation in the prior art.

[0004] In view of the above problems, the present application provides a wind power output prediction method based on wide-area spatiotemporal correlation mining.

[0005] The present application provides a wind power output power prediction method based on wide-area spatiotemporal correlation mining, the method comprising: Connect the target wind turbine to obtain historical output power data and historical meteorological data, wherein the timestamp of the historical output power data corresponds to the timestamp of the historical meteorological data one by one; determine the first failure period and the first time distribution map based on the target wind turbine; determine the second failure period and the second time distribution map, the third failure period and the third time distribution map through the transformer unit and the converter unit corresponding to the target wind turbine; conduct wide-area time axis correlation mining based on the first failure period and the first time distribution map, the second failure period and the second time distribution map, the third failure period and the third time distribution map, and determine the time series correlation queue that meets the stack operation characteristics; determine the space correlation queue that meets the stack operation characteristics based on the wind farm where the target wind turbine is located and the corresponding transformer unit and converter unit; conduct wind power output power forecasting based on the historical output power data and the historical meteorological data, combined with the time series correlation queue and the space correlation queue, and obtain the power forecasting result.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The present application connects to a target wind turbine, obtains historical output power data and historical meteorological data, and the timestamps of the historical output power data correspond one-to-one to the timestamps of the historical meteorological data; based on the target wind turbine, determines the first failure period and the first time distribution map; through the transformer unit and the converter unit corresponding to the target wind turbine, determines the second failure period and the second time distribution map, the third failure period and the third time distribution map; based on the first failure period and the first time distribution map, the second failure period and the second time distribution map, the third failure period and the third time distribution map, performs wide-area time axis correlation mining to determine a time series correlation queue that meets the stack operation characteristics; based on the wind farm where the target wind turbine is located and the corresponding transformer unit and converter unit, determines a space correlation queue that meets the stack operation characteristics; based on the historical output power data and the historical meteorological data, combined with the time series correlation queue and the space correlation queue, performs wind power output power prediction to obtain power prediction results. The present invention solves the technical problems of low wind power output power prediction accuracy and inability to fully utilize the correlation between time and space data in the prior art. By connecting the target wind turbine group, historical output power and meteorological data are obtained, and the failure cycle and time distribution map of the wind turbine group are analyzed based on these data, and the time series and space correlation queues are combined to perform wide-area time axis correlation mining. The time series and space correlation queues of the stack operation characteristics are used to combine historical data to predict wind power output, thereby achieving the technical effect of improving the accuracy of predicting wind power output. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic flow chart of a method for predicting wind power output based on wide-area spatiotemporal correlation mining provided in an embodiment of the present application; Figure 2 A schematic diagram of a flow chart of determining a spatial correlation queue that meets stack operation characteristics in a wind power output power prediction method based on wide-area spatiotemporal correlation mining provided in an embodiment of the present application. DETAILED DESCRIPTION

[0009] The present application provides a wind power output power prediction method based on wide-area spatiotemporal correlation mining, which is used to solve the technical problems of low wind power output power prediction accuracy and inability to fully utilize spatiotemporal data correlation in the prior art. By connecting the target wind turbine, historical output power and meteorological data are obtained, and the failure cycle and time distribution map of the wind turbine are analyzed based on these data. The wide-area time axis correlation mining is carried out in combination with the time series and space correlation queues of the stack operation characteristics. The wind power output power is predicted in combination with historical data, so as to achieve the technical effect of improving the accuracy of predicted wind power output.

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0011] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. 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 that are not explicitly listed or inherent to these processes, methods, products or devices.

[0012] Examples, such as Figure 1 As shown, the present application provides a wind power output prediction method based on wide-area spatiotemporal correlation mining, the method comprising: Step S100: connecting to a target wind turbine to obtain historical output power data and historical meteorological data, wherein the timestamp of the historical output power data corresponds to the timestamp of the historical meteorological data.

[0013] In an embodiment of the present application, a connection is first established with the control system of the target wind turbine through a communication protocol (such as Modbus, OPC, etc.). Through this connection, the historical output power data and historical meteorological data stored in the historical database of the unit are obtained from the data acquisition system of the wind turbine, such as the SCADA system. During the data acquisition process, it is ensured that the timestamp of the historical output power data corresponds to the timestamp of the historical meteorological data, that is, each output power value of the wind turbine corresponds to a meteorological data with the same timestamp. Specifically, when querying historical data, the data stored in the historical database of the target wind turbine is extracted according to the time series through the database query interface (such as SQL), ensuring that the wind turbine output power and meteorological data are aligned in time.

[0014] Furthermore, in the method provided in the embodiment of the application, the timestamp of the historical output power data corresponds one-to-one with the timestamp of the historical meteorological data, and further includes: A sliding alignment window is determined by aligning the timestamp of the historical output power data and the timestamp of the historical meteorological data; a first data stream processing framework corresponding to the historical output power data is established based on the sliding alignment window and in combination with the timestamp of the historical meteorological data; a second data stream processing framework corresponding to the historical meteorological data is established based on the sliding alignment window and in combination with the timestamp of the historical meteorological data; and the original data in the storage unit of the target wind turbine is cleaned through the first data stream processing framework and the second data stream processing framework.

[0015] In an embodiment of the present application, the sliding alignment window is first determined by aligning the timestamps of the historical output power data and the historical meteorological data. Specifically, the timestamps of the historical data are first compared to ensure that the output power data and the meteorological data can be aligned in time. If there is a difference in the timestamps of the two, a sliding match is performed by setting a fixed time interval, such as 1 hour, so that the output power data and the meteorological data can be processed in the same time period. In this way, a sliding alignment window is finally obtained to ensure that the data can be processed synchronously in the same time dimension.

[0016] Then, based on the determined sliding alignment window and combined with the timestamp of the historical output power data, the first data stream processing framework is established. This framework is used to process data streams related to the output power of wind turbines and supports real-time or batch processing. In this framework, the power data stream is first divided into time periods according to the set sliding window by referring to the timestamp. Then, methods such as data cleaning, missing value filling, and outlier detection are used to ensure the accuracy and completeness of the power data. Through this process, a processed output power data stream is finally obtained, which provides clear and effective data input for subsequent analysis.

[0017] Based on the same sliding alignment window, the second data stream processing framework is established in combination with the timestamps of historical meteorological data. Similar to the first framework, this framework processes time series data related to meteorological conditions. Through this framework, meteorological data is preprocessed by cleaning, standardization, and normalization to ensure the quality of meteorological data. In this step, the data processing framework automatically removes invalid data, fills in missing values, and performs smoothing to obtain a processed meteorological data stream, ensuring that the data is accurate and suitable for subsequent analysis models.

[0018] Finally, the original data in the target wind turbine storage unit is cleaned through the first data stream processing framework and the second data stream processing framework. The data cleaning process includes denoising the original output power data and meteorological data, filling missing values, removing outliers, etc., to ensure the consistency and reliability of the data quality. After data cleaning, the cleaned historical output power data and the corresponding cleaned historical meteorological data are finally obtained.

[0019] Step S200: Based on the target wind turbine, determine a first failure period and a first time distribution map.

[0020] In the embodiment of the present application, the historical operation data and fault records of the target wind turbine are first extracted from the monitoring system of the wind turbine, such as the SCADA system. These data include key operating parameters such as the output power, speed, temperature, vibration, etc. of the wind turbine, as well as the specific time and type of the wind turbine failure. To ensure the accuracy of the data, the historical operation data of the wind turbine is regularly obtained from the control system through a standard data communication protocol, such as Modbus or OPC protocol, and the data is correctly aligned according to the timestamp.

[0021] Next, the failure event statistics method is used to analyze the collected fault records. By counting the specific time of each fan failure and calculating the time interval between failures, the periodic pattern of fan failures can be identified. For example, by analyzing the frequency of failures in different time periods, time periods with short failure intervals can be identified, which may be a precursor to the performance degradation of the fan. For periodic failures, the average occurrence cycle of failure events and their association with the operating status of the fan can also be calculated. Specifically, the failure cycle of the fan is determined by calculating the time difference between each failure (i.e., the failure interval) and calculating the mean and variance of the failure interval. This cycle reflects the health status of the fan equipment in normal operation and possible failure modes.

[0022] Once the failure cycle is determined, the next step is to use time series analysis to generate the first time distribution map. In this process, the failure time points of the wind turbines are combined with the corresponding operating data to analyze the distribution pattern of the failure events in the time dimension. First, a timeline is constructed to arrange all the failure occurrence times in chronological order. Then, through the frequency analysis of the failure occurrence, the time distribution map of the failure events is drawn. The time distribution map shows the concentration of failure events in different time periods, revealing the periodic changes of wind turbine failures within a year, a quarter, or even a month. For example, if the wind turbine fails more frequently in certain months or seasonal weather conditions, these periodic changes can be clearly seen from the time distribution map.

[0023] Through frequency distribution analysis and time interval calculation, the high-incidence period of fan failure in different time periods can be accurately determined, thereby constructing the first failure cycle and first time distribution map of the fan.

[0024] Step S300: determining a second failure period and a second time distribution spectrum, a third failure period and a third time distribution spectrum through the voltage conversion unit and the current conversion unit corresponding to the target wind turbine.

[0025] In the embodiment of the present application, the operating data of the transformer unit and the converter unit of the target wind turbine are first obtained. The transformer unit is responsible for adjusting the voltage of the wind turbine generator set and converting the electricity generated by the generator set into a voltage level suitable for transmission to the power grid; while the converter unit is used to convert the AC power generated by the wind turbine into DC power or adjust the frequency of the AC power to meet the needs of the power grid. By connecting to the monitoring system of the wind turbine (such as the SCADA system), the historical operating status, load conditions, fault records and other related data of the transformer unit and the converter unit are obtained, which provide a basis for the subsequent failure cycle analysis.

[0026] Next, the historical fault data of the transformer unit and converter unit are analyzed by using the fault event statistics method. Each time a transformer unit or converter unit fails, the time and type of the fault are recorded. By counting these fault events, the failure cycle of the transformer unit and converter unit is analyzed. For example, the time interval between each fault is counted, and its average time interval and variance are calculated to determine the failure cycle of the transformer unit and converter unit. If the time interval between faults is short, it may indicate that the equipment has a high failure rate under certain specific conditions, which can help determine the periodic characteristics of the failure mode.

[0027] Once the failure cycle of the transformer unit and the converter unit is determined, the next step is to generate the second time distribution map and the third time distribution map by combining the historical fault records and the operating data through the time series analysis method. First, the time points of the transformer unit and the converter unit failure are arranged in chronological order to construct the time axis. Then, based on the time interval of the fault event, the frequency of the fault event in different time periods is analyzed, and the second time distribution map and the third time distribution map are drawn to reveal the distribution law of the fault frequency of the transformer unit and the converter unit on different time scales. Through these maps, the failure tendency of the transformer unit and the converter unit in a specific time period (such as under certain workloads or specific weather conditions) can be identified, thereby providing a basis for the operation and maintenance of the wind turbine and fault warning.

[0028] Through the above steps, the second failure period and the third failure period of the voltage conversion unit and the current conversion unit, as well as the corresponding second time distribution spectrum and the third time distribution spectrum are finally determined.

[0029] Step S400: Based on the first failure cycle and the first time distribution map, the second failure cycle and the second time distribution map, and the third failure cycle and the third time distribution map, wide-area time axis correlation mining is performed to determine a timing correlation queue that meets the stack operation characteristics.

[0030] In the embodiment of the present application, based on the first failure cycle and the first time distribution map, the second failure cycle and the second time distribution map, and the third failure cycle and the third time distribution map, the respective key time node sets and associated features are first extracted. These key time nodes reflect the key moments when the equipment fails in different time periods, and the associated features reveal the time relationship between different equipment failures. Through these key node sets and associated features, potential associated indicators are analyzed, which help identify the potential connections between different equipment failures.

[0031] Next, based on the extracted potential correlation indicators, wide-area timeline correlation mining is performed. This process aims to comprehensively analyze the fault events of different devices according to the timeline and identify the possible causal relationships and timing patterns between them. Through this analysis, the timing correlation queue that meets the characteristics of stack operations is finally determined.

[0032] Furthermore, in the method provided in the embodiment of the application, based on the first failure cycle and the first time distribution map, the second failure cycle and the second time distribution map, and the third failure cycle and the third time distribution map, wide-area time axis correlation mining is performed to determine the timing correlation queue that meets the stack operation characteristics, and further includes: The first key time node set and the first correlation feature are extracted through the first failure cycle and the first time distribution map, the second failure cycle and the second time distribution map; the second key time node set and the second correlation feature are extracted through the first failure cycle and the first time distribution map, the third failure cycle and the third time distribution map; based on the first key time node set and the first correlation feature, the second key time node set and the second correlation feature, potential correlation indicators are determined; wide-area time axis correlation mining is performed through the potential correlation indicators to determine a timing correlation queue that meets the stack operation characteristics.

[0033] In the embodiment of the present application, the first key time node set and the first correlation feature are first extracted through the first failure cycle and the first time distribution map, the second failure cycle and the second time distribution map. Specifically, firstly, by analyzing the fault distribution of the target wind turbine as a whole (the first failure cycle and the first time distribution map) and its transformer unit (the second failure cycle and the second time distribution map), the sliding window method is used to perform frequency statistics of fault events on the time axis, and the time clustering algorithm (such as the density-based clustering algorithm DBSCAN) is combined to identify the high-incidence interval of fault events. These high-incidence intervals form the first key time node set, and each time node represents a significant high-incidence period of faults. Then, based on the obtained first key time node set, the time correlation between the overall operating parameters of the wind turbine (such as output power, speed, etc.) and the operating state of the transformer unit (such as voltage, current load, etc.) is extracted, and correlation analysis, such as the Pearson correlation coefficient, is used to quantify the relationship between the operating parameters and the fault. By calculating the correlation between the wind turbine operating parameters and the time when the fault occurs, the parameters closely related to the occurrence of the fault are identified. For example, when certain parameters (such as high load, temperature, etc.) have a significant correlation with the frequency of fault occurrence or other fault characteristics, these parameters can be used as potential first associated features. Next, the set similarity threshold is used to filter out operating parameters with high correlation, such as the relationship between high load and transformer unit voltage fluctuation, or the correlation between certain temperature ranges and fan fault occurrence. These highly correlated parameters constitute the first associated feature set.

[0034] Next, the same method is used to extract the second key time node set and the second correlation feature from the first failure cycle and the first time distribution map and the third failure cycle and the third time distribution map. Specifically, the relationship between the failure cycle of the target wind turbine as a whole (the first failure cycle and the first time distribution map) and the failure cycle of the converter unit (the third failure cycle and the third time distribution map) is first analyzed, the sliding window method is used to count the frequency of fault events, and the time clustering algorithm (such as the density-based clustering algorithm DBSCAN) is combined to identify the high-incidence period of faults of the converter unit. These high-incidence periods form the second key time node set, and each time node represents a high-incidence interval of converter unit faults. After obtaining the second key time node set, a time correlation analysis is performed to extract the relationship between the overall operating parameters of the target wind turbine (such as output power, wind speed, speed, etc.) and the operating status of the converter unit (such as voltage, current load, etc.). The correlation analysis (such as the Pearson correlation coefficient) is used to calculate the correlation between the converter unit fault moment and the operating parameters, and identify the operating parameters that have a significant impact on the converter unit failure. For example, if certain parameters (such as wind speed and output power) are highly correlated with the fault frequency or fault characteristics of the converter unit, these parameters can be used as the second correlation features. Through the pre-set similarity threshold, operating parameters with high correlation are screened out, such as the relationship between wind speed and converter unit fault or the relationship between power fluctuation and converter unit current load, and these highly correlated parameters are used as the second correlation features.

[0035] Then, based on the first key time node set and the first correlation feature, the second key time node set and the second correlation feature, the potential correlation index is determined. Specifically, the failure events of the wind turbine and its transformer unit and converter unit are first introduced. Then, based on these data, the fault distribution characteristics in different time periods are analyzed to determine the high-incidence and low-incidence periods. Finally, by comparing the high-incidence and low-incidence periods of different units, the potential correlation index is determined.

[0036] Finally, the wide-area time axis correlation mining is carried out through potential correlation indicators to identify fault modes with time correlation. In this process, the sliding window technology is first applied to divide the entire time axis into several small time periods. According to the aforementioned correlation indicators, the similarity between failure events in these time periods is measured. By sorting and correlating the events in each time period, the failure events with high correlation in a specific time period are identified. These events appear as adjacent or close timing patterns on the time axis. Therefore, through the stack operation characteristics (i.e., the first-in, last-out timing processing method), these events are organized in the order of their occurrence, thus forming a timing correlation queue.

[0037] Furthermore, in the method provided in the embodiment of the application, determining the potential correlation index based on the first key time node set and the first correlation feature, the second key time node set and the second correlation feature, further includes: Failure events are introduced, and the failure events include fault nodes of the target wind turbine and the corresponding transformer units and converter units; based on the first key time node set and the first correlation feature, the distribution characteristics and change trends of the failure events in different time periods are analyzed to determine the first high-incidence period and the first low-incidence period; based on the second key time node set and the second correlation feature, the distribution characteristics and change trends of the failure events in different time periods are analyzed to determine the second high-incidence period and the second low-incidence period; based on the first high-incidence period and the second high-incidence period, and the first low-incidence period and the second low-incidence period, potential correlation indicators are determined.

[0038] In the embodiment of the present application, failure events are first introduced, and failure events include fault nodes of the target wind turbine and its transformer unit and current converter unit. These failure events are derived from the monitoring system of the wind turbine (such as the SCADA system) and include the time point of the fault, the fault type, and the possible duration.

[0039] Next, based on the first key time node set and the first correlation feature, the distribution characteristics and change trends of failure events in different time periods are analyzed. Specifically, the frequency of failure events is first counted using the sliding window method, and time periods are divided, such as by hour, day, and month. In each time period, the number of fault events of wind turbines and transformer units is counted, and then the fault frequency of each time period is obtained. Combined with time clustering methods, such as the density-based clustering algorithm DBSCAN, high-incidence fault periods are further identified, and these periods are the first high-incidence periods. At the same time, based on similar statistical methods, the first low-incidence period, that is, the time period with fewer faults, is identified.

[0040] Next, similar methods are used to analyze the distribution characteristics and change trends of failure events in different time periods based on the second key time node set and the second correlation feature, and then identify the second high-incidence period and the second low-incidence period related to the converter unit. These analyses also use the sliding window method to count the fault frequency, and combine the time clustering algorithm to divide the time interval of the event to determine the high-incidence and low-incidence periods of faults.

[0041] Finally, potential correlation indicators are determined based on comparative analysis between the first high-incidence period and the second high-incidence period, and between the first low-incidence period and the second low-incidence period. Specifically, an overlap analysis is first performed to identify the commonalities and differences between the two periods. By calculating the frequency distribution of the overlapping parts of the time periods, the first correlation feature set (i.e., features related to the high-incidence period) and the second correlation feature set (i.e., features related to the low-incidence period) are extracted. Then, these two feature sets are compared and optimized, and the potential correlation indicators are finally determined by screening and eliminating redundant or non-significantly correlated features.

[0042] Furthermore, in the method provided in the embodiment of the application, determining the potential correlation index based on the first high-incidence period and the second high-incidence period, and the first low-incidence period and the second low-incidence period, further includes: An overlapping analysis is performed on the first high-incidence period and the second high-incidence period to extract a first associated feature set; an overlapping analysis is performed on the first low-incidence period and the second low-incidence period to extract a second associated feature set; based on the first associated feature set and the second associated feature set, elimination optimization is performed to determine the potential associated indicator.

[0043] In an embodiment of the present application, first, a time window sliding method is used for the overlapping analysis of the first high-incidence period and the second high-incidence period. This method divides the time period into multiple sub-intervals by setting a sliding window on the time axis, and counts the frequency of fault events occurring in each window. Through this method, detailed fault distribution information of the high-incidence period is obtained. Then, by performing an overlapping analysis on the time windows of the two time periods, the intersection of the two time periods on the time axis is calculated, and their overlapping time periods are extracted. In this way, the common fault area between the first high-incidence period and the second high-incidence period can be clearly identified. This overlapping part extracts the first associated feature set, that is, the fault types, abnormal modes or operating characteristics that appear in common in these two time periods, representing the potential association relationship between the unit and the transformer unit during the period when faults frequently occur.

[0044] Then, for the overlap analysis of the first low-incidence period and the second low-incidence period, the time window sliding method is also used. During the low-incidence period, fault events rarely occur, and the focus of the overlap analysis is to find the commonalities of these rare fault modes. The specific steps are the same as the high-incidence period. First, the time window of the fault event is divided, and then the intersection of the low-incidence period is calculated. There may be some rare but important fault features in these intersection periods, such as failure modes under certain specific operating conditions or environmental conditions. Through this method, the second set of associated features is extracted, that is, the fault features or conditions that frequently occur during the low-incidence period.

[0045] Next, based on the first set of associated features and the second set of associated features, feature screening and optimization methods are used to eliminate and optimize potential associated indicators. Specifically, first, correlation analysis, such as the Pearson correlation coefficient, is used to evaluate the relationship between different features, and those information features with significant correlation or high commonality are screened out. In this process, irrelevant or weakly correlated features are screened out using thresholds pre-set by technical experts, and those features that are closely related to the occurrence of faults are retained. Through this process, potential associated indicators with high correlation are finally obtained.

[0046] Step S500: Based on the wind farm where the target wind turbine is located and the corresponding voltage conversion unit and current conversion unit, a spatial association queue that meets the stack operation characteristics is determined.

[0047] In the embodiment of the present application, the spatial layout of the wind turbines in the power station is first analyzed to determine the first spatial feature and the spatial distribution map. Then, the transformer unit and the current transformer unit corresponding to the wind turbine are analyzed to determine the second spatial feature and the distribution map, and the third spatial feature and the distribution map, respectively. Finally, these spatial features and distribution maps are combined to perform wide-area spatial axis correlation mining to determine the spatial correlation queue that meets the stack operation characteristics.

[0048] Further, such as Figure 2 As shown, in the method provided in the embodiment of the application, based on the wind farm where the target wind turbine is located and the corresponding transformer unit and current converter unit, a spatial association queue that meets the stack operation characteristics is determined, and further includes: Based on the wind farm where the target wind turbine is located, a first spatial feature and a first spatial distribution map are determined; through the transformer unit and the current converter unit corresponding to the target wind turbine, a second spatial feature and a second spatial distribution map, a third spatial feature and a third spatial distribution map are determined; based on the first spatial feature and the first spatial distribution map, the second spatial feature and the second spatial distribution map, the third spatial feature and the third spatial distribution map, wide-area spatial axis correlation mining is performed to determine a spatial correlation queue that meets the stack operation characteristics.

[0049] In the embodiment of the present application, based on the wind farm where the target wind turbine is located, the first spatial feature and the first spatial distribution map are first determined. This step is performed by using a geographic information system (GIS) to perform a spatial analysis of the geographical location information and the surrounding environment of the wind turbine. Specifically, GIS is used to process the location of the wind turbine, climate conditions (such as wind speed, temperature, humidity, etc.), terrain data, and power grid layout to form a first spatial distribution map that shows the spatial relationship between the wind turbine and the surrounding environment.

[0050] Next, the second spatial features and the second spatial distribution map are determined for the transformer units and the current converter units corresponding to the target wind turbines. For this purpose, a spatial clustering method (such as the DBSCAN clustering algorithm) is used to analyze the geographical location information of the transformer units and the current converter units. Through this method, the spatial layout of the transformer units and the current converter units, as well as the spatial distance and relative distribution between them, are identified, and then the second spatial distribution map is constructed to reveal the spatial relationship between these units and the wind turbines.

[0051] In further analysis, the third space characteristics and the third space distribution map are determined through the power flow relationship of the target wind turbine and its supporting facilities. To this end, the power flow analysis method (such as power flow analysis) is used to combine the power connection path and electrical characteristics (such as voltage, current load, etc.) between the wind turbine and the transformer unit and the converter unit to generate the third space distribution map, which reflects the power flow and interdependence between the wind turbine and related equipment in the power grid.

[0052] Finally, based on the first spatial feature and the first spatial distribution map, the second spatial feature and the second spatial distribution map, the third spatial feature and the third spatial distribution map, the spatial correlation queue that meets the stack operation characteristics is determined through wide-area spatial axis correlation mining. Specifically, spatial data analysis technology is used in combination with stack operation characteristics (i.e., one end is fixed and the other end is operated) to perform correlation mining on the key nodes in the above-mentioned spatial distribution maps, identify the spatiotemporal dependency between wind turbines, transformer units, and converter units, thereby forming a spatial correlation queue that meets the stack operation characteristics, and updating the queue order according to the spatial position and power flow direction.

[0053] Step S600: Based on the historical output power data and historical meteorological data, combined with the time-series correlation queue and the space-series correlation queue, the wind power output power is predicted to obtain a power prediction result.

[0054] In an embodiment of the present application, based on historical output power data and historical meteorological data, an initialization cluster is first configured, which includes a static cluster division and a dynamic cluster division. The static cluster division is used to reflect the long-term operating characteristics of the wind turbine, mainly based on the trend changes in historical data; the dynamic cluster division is adjusted according to real-time changes to capture short-term data fluctuations and changes. Subsequently, based on these two parts of the cluster, appropriate time scale constraints are set to ensure that the prediction model is effectively calculated within a reasonable time range. Finally, under the premise of these time scale constraints, the time-series correlation queue and the space-series correlation queue are combined to calculate the wind power output power through the prediction model to obtain the final power prediction result.

[0055] Furthermore, in the method provided in the embodiment of the application, based on the historical output power data and the historical meteorological data, combined with the time series correlation queue and the space correlation queue, the wind power output power is predicted to obtain the power prediction result, and the method also includes: Based on the historical output power data and historical meteorological data, an initialization cluster is configured, and the initialization cluster includes a static cluster division and a dynamic cluster division; according to the static cluster division and the dynamic cluster division in the initialization cluster, a time scale constraint is formulated; under the limitation of the time scale constraint, a wind power output power forecast is performed to obtain a power forecast result.

[0056] In an embodiment of the present application, based on historical output power data and historical meteorological data, firstly, an initialization cluster is configured by analyzing the data, and the cluster includes a static cluster division and a dynamic cluster division. The static cluster division clusters the long-term historical data through a clustering algorithm (such as K-means or DBSCAN algorithm) to extract the long-term operating characteristics of the wind farm, such as the basic correlation trend between wind speed and power output, the long-term impact of meteorological conditions on output power, etc. The dynamic cluster division captures the rapidly changing characteristics in the meteorological data by clustering short-term real-time data, such as sudden changes in wind speed and the dynamic impact of temperature and humidity fluctuations in a short period of time on the power output of the wind turbine. Through this process, the configuration of the initialization cluster is completed.

[0057] Next, based on the static cluster division and dynamic cluster division in the initialized cluster, a time scale constraint is formulated. Specifically, starting from the static cluster division, short-term iterative guidance is performed in combination with the time-related queue and the space-related queue, and the first guidance indicator is set. The time length corresponding to the short-term iterative guidance is a certain period of time. Next, according to the dynamic cluster division, ultra-short-term iterative guidance is performed in combination with the time-related queue and the space-related queue, and the second guidance indicator is set. The time length corresponding to the ultra-short-term iterative guidance is shorter. Finally, the time scale constraint is formulated by combining the first guidance indicator and the second guidance indicator.

[0058] After that, the wind power output power is predicted under the constraints of time scale. At this time, the wind power output power prediction model is used for prediction. In order to train the wind power output power prediction model, historical output power data and historical meteorological data are used as input data. The training process uses the long short-term memory network (LSTM). During the training process, the LSTM model takes historical data (such as meteorological factors such as wind speed, temperature, humidity, and wind power output power at the corresponding time point) as input to learn the temporal relationship between these inputs and outputs. The training process includes data preparation, model training, and the role of time scale constraints. The input data includes historical meteorological data (such as wind speed, temperature, etc.) and corresponding historical output power data. These data are organized into a time series format and are standardized or normalized to ensure that different data dimensions have consistent scales. The LSTM model is trained through the back propagation algorithm and the gradient descent method to learn the relationship between meteorological factors and power output in historical data. During the training process, LSTM effectively captures long-term trend information and short-term volatility characteristics through its built-in memory mechanism. The time scale constraint helps the LSTM model decide how to process input data in different time ranges based on the results of static and dynamic cluster divisions, and optimizes the model's power prediction capabilities for different time periods. After training, the LSTM model can predict wind power output in a specific time period in the future based on real-time meteorological data and historical power data.

[0059] After training is completed, the wind power output prediction model will make predictions based on real-time meteorological data and power data, and generate wind power output prediction results for future time periods based on set time scale constraints (such as short-term or ultra-short-term).

[0060] Furthermore, in the method provided in the embodiment of the application, according to the static cluster division and the dynamic cluster division in the initialization cluster, the time scale constraint is formulated, and further includes: According to the static cluster division in the initialization cluster, combined with the time-series associated queue and the space-series associated queue, short-term iterative guidance is performed, and the first guidance indicator is set, wherein the time length corresponding to the short-term iterative guidance is According to the dynamic cluster division in the initialization cluster, combined with the timing association queue and the space association queue, ultra-short-term iterative guidance is performed, and the second guidance indicator is set, wherein the time length corresponding to the ultra-short-term iterative guidance is ,and ; Formulate time scale constraints through the first guiding indicator and the second guiding indicator.

[0061] In the embodiment of the present application, according to the static cluster division in the initialization cluster, firstly, the time-series associated queue and the space-series associated queue are combined to perform short-term iterative guidance, and the first guidance indicator is set, wherein the time length corresponding to the short-term iterative guidance is . Specifically, the static cluster division extracts long-term stable meteorological characteristics by clustering long-term historical data (such as meteorological data such as wind speed, temperature, humidity, and power output data), and captures the time dependency between wind power output power and meteorological data through time series correlation queues. For example, long-term trends in historical power data are analyzed through sliding window technology. The spatial correlation queue analyzes the spatial synergy between different wind turbines through the spatial autoregressive model (SAR) to identify the power output correlation of adjacent wind turbines under the same meteorological conditions. Combining these methods, through short-term iterative guidance, key time nodes related to short-term meteorological changes are identified, and the time length is set. (such as a few hours or a day), providing a stable time scale constraint for subsequent predictions.

[0062] Then, according to the dynamic cluster division in the initialization cluster, combined with the time-series correlation queue and the space-series correlation queue, ultra-short-term iterative guidance is performed, and the second guidance indicator is set. The time length corresponding to the ultra-short-term iterative guidance is ,and The dynamic cluster division captures the immediate relationship between meteorological factors and power output by analyzing short-term data (such as hourly or minute-level data), and uses the ARIMA model in combination with the time series correlation queue to model the fluctuation relationship between wind power output and meteorological conditions in the short term. The spatial correlation queue helps capture the spatial variation between wind turbines at this stage and identify their coordinated response under rapidly changing meteorological conditions. In this way, ultra-short-term time scales The time (e.g. minutes or hours) is determined and set to ensure that the model can quickly respond to the impact of short-term meteorological changes on power output.

[0063] The time scale constraint is finally formulated by combining the first and second guiding indicators. The first guiding indicator is used to set short-term forecasts with a longer time scale (such as a few hours to a day), while the second guiding indicator helps set ultra-short-term forecasts with a shorter time scale (such as a few minutes to a few hours).

[0064] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects: The present application connects to a target wind turbine, obtains historical output power data and historical meteorological data, and the timestamps of the historical output power data correspond one-to-one to the timestamps of the historical meteorological data; based on the target wind turbine, determines the first failure period and the first time distribution map; through the transformer unit and the converter unit corresponding to the target wind turbine, determines the second failure period and the second time distribution map, the third failure period and the third time distribution map; based on the first failure period and the first time distribution map, the second failure period and the second time distribution map, the third failure period and the third time distribution map, performs wide-area time axis correlation mining to determine a time series correlation queue that meets the stack operation characteristics; based on the wind farm where the target wind turbine is located and the corresponding transformer unit and converter unit, determines a space correlation queue that meets the stack operation characteristics; based on the historical output power data and the historical meteorological data, combined with the time series correlation queue and the space correlation queue, performs wind power output power prediction to obtain power prediction results. The present invention solves the technical problems of low wind power output power prediction accuracy and inability to fully utilize the correlation between time and space data in the prior art. By connecting the target wind turbine group, historical output power and meteorological data are obtained, and the failure cycle and time distribution map of the wind turbine group are analyzed based on these data, and the time series and space correlation queues are combined to perform wide-area time axis correlation mining. The time series and space correlation queues of the stack operation characteristics are used to combine historical data to predict wind power output, thereby achieving the technical effect of improving the accuracy of predicting wind power output.

[0065] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0066] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0067] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A wind power output prediction method based on wide-area spatiotemporal correlation mining, characterized in that: The method comprises: Connecting to a target wind turbine, acquiring historical output power data and historical meteorological data, wherein the timestamp of the historical output power data corresponds to the timestamp of the historical meteorological data; Based on the target wind turbine, determining a first failure period and a first time distribution spectrum; Determine the second failure period and the second time distribution spectrum, the third failure period and the third time distribution spectrum through the voltage conversion unit and the current conversion unit corresponding to the target wind turbine; Based on the first failure cycle and the first time distribution map, the second failure cycle and the second time distribution map, and the third failure cycle and the third time distribution map, wide-area time axis correlation mining is performed to determine a timing correlation queue that meets the stack operation characteristics; Based on the wind farm where the target wind turbine is located and the corresponding voltage conversion unit and current conversion unit, determine a spatial association queue that meets the stack operation characteristics; Based on the historical output power data and historical meteorological data, combined with the time series correlation queue and the space series correlation queue, wind power output power prediction is performed to obtain a power prediction result.

2. The method for predicting wind power output based on wide-area spatiotemporal correlation mining according to claim 1, characterized in that: Based on the wind farm where the target wind turbine is located and the corresponding voltage conversion unit and current conversion unit, determining a spatial association queue that meets the stack operation characteristics, the method includes: Based on the wind farm where the target wind turbine is located, determining a first spatial feature and a first spatial distribution map; Determine the second spatial characteristics and the second spatial distribution map, the third spatial characteristics and the third spatial distribution map through the voltage conversion unit and the current conversion unit corresponding to the target wind turbine; Based on the first spatial feature and the first spatial distribution map, the second spatial feature and the second spatial distribution map, and the third spatial feature and the third spatial distribution map, wide-area spatial axis correlation mining is performed to determine a spatial correlation queue that meets the stack operation characteristics.

3. The method for predicting wind power output based on wide-area spatiotemporal correlation mining according to claim 1, characterized in that: The timestamps of the historical output power data correspond one-to-one to the timestamps of the historical meteorological data, and the method comprises: Determine a sliding alignment window by aligning the timestamp of the historical output power data and the timestamp of the historical meteorological data; Based on the sliding alignment window and in combination with the timestamp of the historical output power data, a first data stream processing framework corresponding to the historical output power data is established; Based on the sliding alignment window and in combination with the timestamp of the historical meteorological data, a second data stream processing framework corresponding to the historical meteorological data is established; The original data in the storage unit of the target wind turbine group is cleaned by the first data stream processing framework and the second data stream processing framework.

4. The method for predicting wind power output based on wide-area spatiotemporal correlation mining according to claim 2, characterized in that: Based on the first failure cycle and the first time distribution map, the second failure cycle and the second time distribution map, and the third failure cycle and the third time distribution map, wide-area time axis correlation mining is performed to determine a timing correlation queue that meets the stack operation characteristics, the method comprising: Extracting a first key time node set and a first correlation feature through the first failure cycle and the first time distribution map, the second failure cycle and the second time distribution map; Extracting a second key time node set and a second correlation feature through the first failure cycle and the first time distribution map, the third failure cycle and the third time distribution map; Determining a potential correlation indicator based on the first key time node set and the first correlation feature, the second key time node set and the second correlation feature; Wide-area time axis correlation mining is performed through the potential correlation indicators to determine a timing correlation queue that meets the stack operation characteristics.

5. The method for predicting wind power output based on wide-area spatiotemporal correlation mining according to claim 4, characterized in that: Based on the first key time node set and the first correlation feature, the second key time node set and the second correlation feature, determining a potential correlation indicator, the method includes: Introducing a failure event, wherein the failure event includes a fault node of a target wind turbine and a corresponding transformer unit and a current transformer unit; Based on the first key time node set and the first correlation feature, analyzing the distribution characteristics and change trends of failure events in different time periods, and determining a first high-incidence period and a first low-incidence period; Based on the first key time node set and the first correlation feature, analyzing the distribution characteristics and change trends of failure events in different time periods, and determining the second high-incidence period and the second low-incidence period; Based on the first high-incidence time period and the second high-incidence time period, and the first low-incidence time period and the second low-incidence time period, a potential correlation indicator is determined.

6. The method for predicting wind power output based on wide-area spatiotemporal correlation mining according to claim 5, characterized in that: Based on the first high-incidence period and the second high-incidence period, and the first low-incidence period and the second low-incidence period, determining a potential correlation index, the method includes: Performing overlapping analysis on the first high incidence period and the second high incidence period to extract a first correlation feature set; Performing overlapping analysis on the first low incidence period and the second low incidence period to extract a second correlation feature set; Based on the first association feature set and the second association feature set, elimination optimization is performed to determine the potential association index.

7. The method for predicting wind power output based on wide-area spatiotemporal correlation mining according to claim 6, characterized in that: Based on the historical output power data and the historical meteorological data, combined with the time series correlation queue and the space series correlation queue, the wind power output power is predicted to obtain the power prediction result, and the method includes: Based on the historical output power data and the historical meteorological data, configuring an initialization cluster, wherein the initialization cluster includes a static cluster division and a dynamic cluster division; Formulate a time scale constraint according to the static cluster division and the dynamic cluster division in the initialized cluster; Under the limitation of the time scale constraint, the wind power output power is predicted to obtain the power prediction result.

8. The method for predicting wind power output based on wide-area spatiotemporal correlation mining according to claim 7, characterized in that: According to the static cluster division and the dynamic cluster division in the initialization cluster, a time scale constraint is formulated, and the method comprises: According to the static cluster division in the initialization cluster, combined with the time-series associated queue and the space-series associated queue, short-term iterative guidance is performed, and the first guidance indicator is set, wherein the time length corresponding to the short-term iterative guidance is ; According to the dynamic cluster division in the initialization cluster, combined with the time-series correlation queue and the space-series correlation queue, ultra-short-term iterative guidance is performed, and the second guidance indicator is set, wherein the time length corresponding to the ultra-short-term iterative guidance is ,and ; A time scale constraint is formulated through the first guiding indicator and the second guiding indicator.

Images