CN-121980466-A - Icing fan power prediction method and system considering data quality control

CN121980466ACN 121980466 ACN121980466 ACN 121980466ACN-121980466-A

Abstract

A method for predicting the power of an icing fan based on data quality control includes such steps as obtaining the meteorological data and power data of the fan in the icing scene of a region to be tested in a set time range, building a multi-dimensional quality control system of measured meteorological data, detecting abnormal data, making abnormal labels, inputting the meteorological data with abnormal labels, training the real-time abnormal detection model of meteorological data, real-time correcting the abnormal data marked as suspicious or missing based on space-time characteristic analysis, sample enhancement of the meteorological correction data and wind power data based on the generation countermeasure network, training the power prediction model of the icing fan, and inputting the enhanced meteorological correction data and wind power data to the trained power prediction model to obtain the final power prediction result of the icing fan based on data quality control. The wind power prediction method and the wind power prediction device can improve the accuracy of wind power prediction and the capability of the power grid dispatching side to wind power fluctuation.

Inventors

LIU YANG
YU YIXIAO
XIE BOYU
ZHANG SHAOFENG
GUO YINA
TENG WEIJUN
SUN XIN
Gu Qingfa
YANG MING
LI MENGLIN

Assignees

国网河南省电力公司电力科学研究院
山东大学

Dates

Publication Date: 20260505
Application Date: 20260225

Claims (10)

1. The method for predicting the power of the icing fan by considering the data quality control is characterized by comprising the following steps of: the method comprises the steps of obtaining historical meteorological data and corresponding power data of a fan in an icing scene of a region to be detected in a set icing period, constructing a measured meteorological data multidimensional quality control system, detecting the meteorological data, and obtaining an abnormal label corresponding to each meteorological data; Constructing a real-time meteorological data anomaly detection model, and training the real-time meteorological data anomaly detection model by using the anomaly data with the anomaly tags; reading the real-time acquired meteorological data by using the trained real-time meteorological data anomaly detection model to perform instant anomaly judgment to obtain a meteorological sequence with anomaly marks of a target site; extracting space-time input features in a weather sequence with an abnormal mark of the target site based on a space-time characteristic analysis method; based on the weather sequence with the abnormal mark of the target site and the space-time input characteristics, training the data real-time correction model to obtain weather correction data; And carrying out real-time prediction by using the trained icing fan power prediction model to obtain an icing fan power prediction result.
2. The method for predicting the power of the ice-coating fan taking data quality control into consideration as set forth in claim 1, wherein the method comprises the following steps of: The actually measured meteorological data multidimensional quality control system comprises data completeness test, value domain rationality test, logic relevance test, time sequence fluctuation test and space coordination test; The meteorological data distribution abnormal label which does not pass through the data completeness test is marked as 2, the meteorological data distribution abnormal label which does not pass through the value range rationality test, the logic relevance test, the time sequence fluctuation test or the space coordination test is marked as 1, and the abnormal labels corresponding to the rest data labels are marked as 0.
3. The method for predicting the power of the ice-coating fan taking data quality control into consideration as claimed in claim 2, wherein the method comprises the following steps of: judging whether abnormal data are iced or not according to the ice-covered shutdown criterion of the wind turbine generator to cause shutdown, if so, changing the abnormal label of the abnormal data into 0; the shutdown criteria include the following conditions: The temperature is smaller than a set temperature threshold value, the relative humidity is larger than a set humidity threshold value, and the wind speed is in a set wind speed interval; when all conditions are satisfied, the judgment result is yes.
4. The method for predicting the power of the ice-coating fan taking data quality control into consideration as set forth in claim 1, wherein the method comprises the following steps of: the space-time characteristic analysis method is used for extracting space-time input characteristics in a meteorological sequence with an abnormal mark of the target site, and comprises the following specific steps: In the time characteristic analysis, calculating an improved Pearson correlation coefficient between a historical time observation value of a certain meteorological element of a target site and a current time value observation value of the same meteorological element; In the space characteristic analysis, calculating an improved Pearson correlation coefficient between the contemporaneous weather sequence observed value of a certain adjacent site of the target site and the contemporaneous weather sequence observed value of the target site; And determining the weather sequence with the abnormal marks of the target site, which meets the timing delay condition and the corresponding improved Pearson correlation coefficient meets the coefficient threshold condition, as the space-time input characteristic.
5. The method for predicting the power of the ice-coating fan taking into consideration data quality control as set forth in claim 4, wherein: The specific calculation method for improving the Pearson correlation coefficient comprises the following steps: Fitting a multiple linear regression model of X and Y; data of the ith data point And (3) with Subtracting the corresponding multiple linear regression models respectively to obtain a pure residual sequence And Calculating the clean residual sequence And Corresponding sample mean And An improved Pearson correlation coefficient is calculated.
6. The method for predicting the power of the ice-coating fan taking data quality control into consideration as set forth in claim 1, wherein the method comprises the following steps of: the data real-time correction model is built based on the hybrid neural network model; training the data real-time correction model based on the weather sequence with the abnormal mark and the space-time input characteristics of the target site to obtain weather correction data, wherein the specific steps comprise: The mixed neural network model is a long-short-term memory network-deep neural network mixed model, the meteorological sequence with the abnormal mark of the target site is input into the long-term memory network, and the space-time input characteristic is input into the deep neural network; fusing the outputs of the long-term memory network and the deep neural network by using an implicit layer to obtain weather correction data; The data real-time correction model is trained using a composite loss function that is a weighted sum of weighted mean square error, focal-MSE loss for outliers, and a time-space consistency constraint.
7. The method for predicting the power of the ice-coating fan taking data quality control into consideration as set forth in claim 1, wherein the method comprises the following steps of: the icing fan power prediction model is constructed by using LightGBM algorithm; Constructing a weighted sum of square loss, a physical consistency constraint term and a model complexity regularization term as a loss function; the calculation method of the physical consistency constraint item comprises the following steps: calculating absolute values of differences between the icing fan power prediction results of all samples in the icing fan power prediction model and corresponding fan reference power curve reference values, summing all the absolute values and calculating an average value, and multiplying the average value by a super parameter lambda to obtain a calculation result of the physical consistency constraint term.
8. An icing fan power prediction system taking into account data quality control using the method of any of claims 1-7, comprising an anomalous weather data detection module, a real-time anomalous weather data detection module, a weather correction data acquisition module and a power prediction module, characterized in that: The abnormal meteorological data detection module is used for acquiring historical meteorological data and corresponding power data of a fan in an icing scene in a set icing period of a region to be detected; constructing a measured meteorological data multidimensional quality control system, and detecting the meteorological data to obtain an abnormal label corresponding to each meteorological data; the real-time abnormal meteorological data detection module is used for constructing a real-time meteorological data abnormal detection model, training the real-time meteorological data abnormal detection model by using the abnormal data with the abnormal label, reading the real-time acquired meteorological data by using the trained real-time meteorological data abnormal detection model, and performing instant abnormal judgment to obtain a meteorological sequence with an abnormal mark of a target site; The system comprises a weather correction data acquisition module, a data real-time correction model and a weather correction data acquisition module, wherein the space-time input characteristics in a weather sequence with an abnormal mark of the target site are extracted based on a space-time characteristic analysis method; and the power prediction module is used for training an icing fan power prediction model based on the power data and the weather modification data, and performing real-time prediction by using the trained icing fan power prediction model to obtain an icing fan power prediction result.
9. A terminal comprises a processor and a storage medium, and is characterized in that: The storage medium is used for storing instructions; the processor being operative according to the instructions to perform the steps of the method according to any one of claims 1-7.
10. Computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any of claims 1-7.

Description

Icing fan power prediction method and system considering data quality control Technical Field The invention belongs to the technical field of wind power prediction, and particularly relates to an icing fan power prediction method and system considering data quality control. Background As a clean and renewable energy resource, wind energy plays a key role in promoting economic growth, ensuring energy safety, and slowing down climate change. The method promotes the orderly development of the wind power industry and has profound significance for optimizing the national energy structure and realizing low-carbon transformation. Developing accurate wind power prediction is an important approach for coping with wind energy randomness and volatility challenges, and is important for enhancing power grid regulation and control capability, maintaining stable operation of a system and promoting efficient utilization of clean energy. However, with the increase of global climate complexity, extreme meteorological events such as freezing rain, cold tide and the like frequently occur, and the fan blade is easy to be covered with ice, so that abnormal fluctuation of the output characteristics of the power station is caused. Especially in high latitude or high altitude areas, the phenomenon of blade icing is more common in low temperature environment, which not only directly influences the power output of a unit, but also threatens the stable operation of a wind power plant. In view of the above, most of traditional wind power prediction models are constructed by depending on numerical weather forecast and historical operation data, and special effects caused by extreme meteorological conditions such as icing and the like cannot be fully considered. On the area scale, because of the complex space-time coupling relation between wind power plant groups, the existing prediction model such as the patent document with publication number of CN117638926A cannot effectively integrate space-time correlation between stations, and is difficult to accurately reflect the integral fluctuation of cluster output. In addition, the reliability of the scheme of the patent document with publication number CN119442918B is established on the basis of numerical weather forecast and icing physical simulation, and a link for quality control and correction of the original input data is not included. Therefore, the prior art ignores the problem that the poor quality of the measured meteorological data further reduces the reliability of the input data, which limits the generalization capability of the model. Together, these problems cause the traditional model to have a large prediction bias in extreme weather, which constitutes a potential risk for the safety and reliability of the power system. Disclosure of Invention In order to solve the defects in the prior art, the invention provides the method and the system for predicting the power of the icing fan by considering the data quality control. The invention adopts the following technical scheme. An icing fan power prediction method considering data quality control comprises the following steps: the method comprises the steps of obtaining historical meteorological data and corresponding power data of a fan in an icing scene of a region to be detected in a set icing period, constructing a measured meteorological data multidimensional quality control system, detecting the meteorological data, and obtaining an abnormal label corresponding to each meteorological data; Constructing a real-time meteorological data anomaly detection model, and training the real-time meteorological data anomaly detection model by using the anomaly data with the anomaly tags; reading the real-time acquired meteorological data by using the trained real-time meteorological data anomaly detection model to perform instant anomaly judgment to obtain a meteorological sequence with anomaly marks of a target site; extracting space-time input features in a weather sequence with an abnormal mark of the target site based on a space-time characteristic analysis method; based on the weather sequence with the abnormal mark of the target site and the space-time input characteristics, training the data real-time correction model to obtain weather correction data; And carrying out real-time prediction by using the trained icing fan power prediction model to obtain an icing fan power prediction result. Further preferably, the actually measured meteorological data multidimensional quality control system comprises a data completeness check, a value range rationality check, a logic relevance check, a time sequence fluctuation check and a space coordination check; The meteorological data distribution abnormal label which does not pass through the data completeness test is marked as 2, the meteorological data distribution abnormal label which does not pass through the value range rationality test, the logic relevance test, the time sequence fluctua