An artificial intelligence-based wind power prediction data processing method

Abstract

This application discloses an artificial intelligence-based wind power prediction data processing method, relating to the field of new energy power system operation and control technology. The method includes: acquiring the predicted wind power of a target wind farm at the current moment; determining the cluster type label; loading a classification reinforcement learning correction model corresponding to the cluster type; calculating correction coefficients through the classification reinforcement learning correction model; and correcting the predicted wind power based on the correction coefficients to obtain the corrected wind power. This application, by introducing cluster type labels and loading corresponding classification reinforcement learning correction models, adaptively corrects wind power prediction based on the characteristics of different wind farm clusters. This effectively solves the problem that traditional unified models cannot cope with extreme weather and special operating conditions, avoiding correction failure or reverse correction caused by misjudging large deviations as noise, and significantly improving the accuracy and robustness of wind power prediction.

Description

Technical Field

[0001] This application relates to the field of new energy power system operation and control technology, and in particular to a wind power prediction data processing method based on artificial intelligence. Background Technology

[0002] Wind energy is inherently highly random, volatile, and intermittent. Large-scale wind power grid connection poses significant challenges to power system power balance, peak shaving, frequency regulation, and safe and stable operation. Currently, wind power prediction technology based on Numerical Weather Prediction (NWP) and deep learning algorithms is generally used. However, in practical engineering applications, a deviation between predicted and measured power is unavoidable. This deviation stems not only from the meteorological forecasting errors inherent in NWP itself but also from the uncertainties in the physical response of wind turbines in complex environments. If an effective error correction mechanism is not established and the raw prediction data is directly used for scheduling, inaccurate predictions will lead to increased wind curtailment, grid performance penalties, and even grid frequency oscillations. Therefore, targeted error correction based on the raw predictions is necessary to improve the reliability of wind power prediction.

[0003] To address the aforementioned prediction bias issues, relevant technologies primarily improve accuracy through data post-processing and combined model optimization. In statistics and machine learning, related technologies widely employ error correction models, utilizing historical error sequences to train models such as Support Vector Machines, Long Short-Term Memory Networks (LSTM), or XGBoost, attempting to uncover the distribution patterns of errors and compensate for future predictions. In signal processing, some technologies employ Empirical Mode Decomposition (EMD) or Variational Mode Decomposition (VMD) techniques to decompose wind power sequences into components of different frequencies, smoothing the prediction curve by eliminating high-frequency noise components, thereby reducing the impact of random errors. Furthermore, some solutions attempt to introduce real-time anemometer data or radar data to perform short-term corrections to the NWP input field, aiming to improve prediction quality from the source of the input.

[0004] However, these related technologies still have limitations in handling "maximum deviations" or "long-tail events" caused by extreme weather (such as cold waves, icing, typhoons, and extreme gusts) or special operating conditions (such as power grid curtailment and equipment failure). First, most existing error correction models are trained based on the assumption of a normal distribution or the goal of minimizing the mean square error, tending to fit stationary data that occurs with a high probability. For numerical abrupt changes caused by extreme weather, the models often treat these abrupt changes as noise and smooth them out, leading to correction failure or even reverse correction at critical moments. Second, these technologies rely too heavily on meteorological data and lack a perception mechanism for non-meteorological factors such as power grid curtailment, equipment failure, and maintenance shutdowns, making it difficult to identify sudden power drops caused by special operating conditions.

[0005] Therefore, current conventional models suffer from severe lag when faced with such large deviations. They also typically use uniform model parameters to handle all time periods, lacking an adaptive switching mechanism for different causes of deviations. As a result, they cannot accurately eliminate large deviations when facing complex and ever-changing real-world operating environments, and thus cannot meet the engineering requirements for high reliability. Summary of the Invention

[0006] This application provides a wind power prediction data processing method based on artificial intelligence to overcome the shortcomings of the aforementioned related technologies. The technical solution is as follows: In a first aspect, this application provides a wind power prediction data processing method based on artificial intelligence, including: Obtain the predicted wind power of the target wind farm at the current moment; The cluster type label is determined based on the operating data of the target wind farm during a preset time period prior to the current moment; The classification reinforcement learning correction model corresponding to the cluster type is loaded according to the cluster type label, and the correction coefficient is calculated by the classification reinforcement learning correction model based on the predicted wind power and the operating data within the preset time period. The predicted wind power is corrected based on the correction coefficient, and the corrected wind power at the current moment is obtained. The classification reinforcement learning correction model is trained based on a pre-built sample set, which is constructed based on historical operation data of multiple wind farms in historical periods.

[0007] In one alternative embodiment of the first aspect, the process of constructing the sample set includes: Historical operating data of multiple wind farms in different historical periods are acquired; wherein, the historical operating data includes historical measured power, historical predicted power, numerical weather forecast data, and station operating status data; The corresponding maximum deviation data were obtained by filtering the historical operation data of each wind farm for each period within the historical period. Sample data are constructed based on the maximum deviation data of each wind farm. The sample inputs of the sample data are historical predicted power, numerical weather forecast data and station operation status data. The sample label is the historical measured power corresponding to the sample input. Identify the dominant driving factors for each wind farm based on the maximum deviation data; The cluster type label corresponding to each wind farm is determined based on the location information and dominant driving factors of each wind farm, and the cluster type to which each wind farm belongs is determined based on the cluster type label. A corresponding sample set is constructed based on the collection of sample data for each cluster type.

[0008] In one alternative embodiment of the first aspect, the step of obtaining corresponding maximum deviation data based on the historical operating data of each wind farm for each period within the historical period includes: The wind power generation capacity at the corresponding time is determined based on the station's operational status data. Calculate the power deviation between the historical measured power and the historical predicted power at each time point in each time period for each wind farm; Calculate the ratio of the power deviation to the wind power unit's operating capacity to obtain the relative degree of deviation; If the relative deviation value is greater than the preset deviation threshold, the historical operating data within the corresponding time period is determined to be extremely deviation data.

[0009] In one alternative of the first aspect, the numerical weather forecast data includes forecast temperature data and forecast wind speed data, and the station operation status data includes wind farm humidity data and wind farm icing status data. The process of identifying the dominant driving factors based on the maximum deviation data of each wind farm includes: The forecast temperature data, forecast wind speed data, wind farm humidity data, and wind farm icing status data in the maximum deviation data are respectively used as influencing factor nodes, and the power deviation is used as the target node to generate a node set. Generate directed edges from each influencing factor node to the target node in the node set, perform correlation tests on each directed edge, retain the directed edges that pass the correlation test, and construct a causal graph network based on every two nodes with directed edges. Calculate the causal factor strength corresponding to each directed edge in the causal graph network; For each directed edge, if the causal factor strength is greater than a preset strength threshold, then the influence factor node of the directed edge is determined as the dominant driving factor.

[0010] In one alternative embodiment of the first aspect, determining the cluster type label corresponding to each wind farm based on its location information and dominant driving factor includes: The location information is determined based on the station operation status data of each wind farm, and the latitude of the wind farm is extracted. The mapping table is traversed according to each dominant driving factor and dimension of the wind farm. The cluster type label corresponding to each wind farm is determined according to the pre-built cluster type label and the mapping relationship between the dimension and the dominant driving factor in the mapping table.

[0011] In one alternative embodiment of the first aspect, the training process of the classification reinforcement learning correction model includes the following steps: Input the sample set of each cluster type into the classification reinforcement learning correction model of the same cluster type for training; Obtain the power correction result obtained by the classification reinforcement learning correction model based on the sample input; Construct a reward function based on the difference between the power correction result and the sample label corresponding to the sample input; The model parameters of the classification reinforcement learning correction model are adjusted according to the value of the reward function; After determining that the model has converged, output the trained classification reinforcement learning correction model.

[0012] In one alternative embodiment of the first aspect, the reward function is constructed based on the difference between the power correction result and the sample label corresponding to the sample input, and the weight coefficient of each term in the reward function is determined according to the cluster type. The formula for the reward function is expressed as follows: ; ; in, For the reward function, These are the weighting coefficients, and the values ​​of the two weighting coefficients are determined based on the cluster type. For power correction results, The historical measured power corresponding to the sample label. For the installed capacity of the station, This is the corrected power ramp-up value. The historical predicted power in the sample input, The power ramp-up threshold.

[0013] Secondly, this application also provides a wind power prediction data processing device based on artificial intelligence, comprising: The data acquisition unit is used to acquire the predicted wind power of the target wind farm at the current moment; The data processing unit is used to determine the cluster type label based on the operating data of the target wind farm during a preset period before the current time; The wind power correction unit is used to load a classification reinforcement learning correction model corresponding to the cluster type according to the cluster type label, and calculate the correction coefficient through the classification reinforcement learning correction model based on the predicted wind power and the operating data in the preset time period. The wind power correction unit is also used to correct the predicted wind power based on the correction coefficient, and process it to obtain the corrected wind power at the current moment. The classification reinforcement learning correction model is trained based on a pre-built sample set, which is constructed based on historical operation data of multiple wind farms in historical periods.

[0014] Thirdly, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method provided by the first aspect of this application or any implementation thereof.

[0015] Fourthly, this application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided by the first aspect of this application or any implementation thereof.

[0016] This application introduces cluster type labels and loads corresponding classification reinforcement learning correction models to adaptively correct for the characteristics of different wind farm clusters. This effectively solves the problem that traditional unified models cannot cope with extreme weather and special operating conditions, avoids the failure of correction or reverse correction caused by misjudging large deviations as noise, and significantly improves the accuracy and robustness of wind power prediction.

[0017] By constructing a sample set based on maximum deviation data and identifying the dominant driving factors, the influence of various factors on power deviation was accurately mined, and the maximum deviation samples were accurately identified. Based on location information and dominant driving factors, cluster classification was performed and models were trained separately, establishing differentiated reward functions and action spaces. This overcame the shortcomings of existing technologies, such as lack of mechanistic explanation and inability to handle non-meteorological factors. It ensured that the model could significantly reduce prediction errors in complex scenarios after convergence, while taking into account the physical ramping constraints of the power system and improving the safety and economy of new energy grid connection. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a wind power prediction data processing method based on artificial intelligence, provided in an embodiment of this application. Figure 2 This is one of the schematic diagrams illustrating the effect of the artificial intelligence-based wind power prediction data processing method provided in the embodiments of this application; Figure 3 This is the second schematic diagram illustrating the effect of the artificial intelligence-based wind power prediction data processing method provided in the embodiments of this application; Figure 4This is a schematic diagram of the structure of a wind power prediction data processing device based on artificial intelligence provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to such process, method, product, or apparatus.

[0022] It should be noted that the terms "first" and "second" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects. It is understood that "first" and "second" can be interchanged in a specific order or sequence where permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in an order other than those described or illustrated herein.

[0023] The present application will now be described in detail with reference to specific embodiments.

[0024] Next, combine Figure 1 This application introduces an artificial intelligence-based wind power prediction data processing method provided by its embodiments. For details, please refer to... Figure 1 , Figure 1 This illustration shows a flowchart of a wind power prediction data processing method based on artificial intelligence, provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps: S101, Obtain the predicted wind power of the target wind farm at the current moment; S102, determine the cluster type label based on the operating data of the target wind farm during a preset time period before the current moment; S103, Load the classification reinforcement learning correction model corresponding to the cluster type according to the cluster type label; S104, the correction coefficient is calculated by the classification reinforcement learning correction model based on the predicted wind power and the operating data within the preset time period; S105, Based on the correction coefficient, correct the predicted wind power and process it to obtain the corrected wind power at the current moment.

[0025] It should be noted that during the operation of a wind farm, the predicted wind power of the wind farm in a given period can generally be calculated using neural network models or mathematical models. However, the predicted wind power obtained by related technologies often deviates from the actual measured power of the wind farm in the corresponding period. The embodiments of this application can correct the predicted wind power calculated by general models to avoid the occurrence of large deviations.

[0026] Specifically, in S101, any wind farm can be selected as the research object, which is the target wind farm whose wind power prediction needs to be corrected. Any neural network model or mathematical model can be used to calculate the wind power prediction at the current moment based on the operating data of the target wind farm in a preset period before the current moment. This application embodiment does not limit this.

[0027] In some embodiments, the classification reinforcement learning correction model provided in this application can construct an agent according to the label of each cluster type to achieve adaptive processing for each cluster type. This can effectively solve the problem of single models not being suitable for cross-regional applications and significantly reduce prediction errors in extreme scenarios. The process of constructing the sample set for training includes the following steps: S201, acquire historical operating data of multiple wind farms in historical periods; wherein, the historical operating data includes historical measured power, historical predicted power, numerical weather forecast data, and station operating status data.

[0028] Understandably, multiple wind farms located in different regions within a large geographical area can be selected for data collection, making it easier to cover different types of clusters.

[0029] Understandably, historical operating data of wind farms in a historical period refers to known operating data recorded within the corresponding period.

[0030] For example, taking historical period a and the immediately following historical period b as examples, the historical measured power and historical predicted power of this application embodiment are explained. A sample set is constructed based on the historical operating data of historical period b. The historical measured power can be understood as the actual operating power of the wind farm at each moment in historical period b. The historical predicted power can be understood as the prediction result of historical period b obtained based on the historical operating data of historical period a.

[0031] Following S201, each type of data in the historical operational data can be preprocessed separately. A preset time resolution can be used to align the time scale of each data type, for example, a 15-minute time resolution. Max-min normalization can then be performed on each data type, using the following formula: ; in, The original data before normalization. and These are the minimum and maximum values ​​of the original data before normalization. This represents the data after normalization.

[0032] Subsequent steps can be performed based on the normalized data.

[0033] S202, based on the historical operating data of each wind farm in each period of the historical period, the corresponding maximum deviation data are obtained by filtering.

[0034] S202 specifically includes the following steps: S2021, Determine the wind power generation capacity at the corresponding time based on the station operation status data; S022, calculate the power deviation between the historical measured power and the historical predicted power for each wind farm at each time point within each time period. ; S2023, Calculate the power deviation The ratio of the wind power unit's operating capacity to the relative deviation is obtained using the formula: ; in, The degree of relative deviation, The historical measured power corresponding to the sample label. The historical predicted power in the sample input, The wind power generation capacity is t, where t represents each moment in the time period and T represents the total number of moments in the time period. For example, the time period can be set to 3 months and the time resolution can be selected as 15 minutes.

[0035] S2024, compare the relative deviation degree with the preset deviation threshold.

[0036] If the relative deviation value is greater than the preset deviation threshold, then execute S2025 to determine the historical operating data within the corresponding time period as extremely large deviation data.

[0037] Otherwise, execute S2026 to determine that the corresponding data is normal data and there is no significant deviation.

[0038] By following the steps above, we can identify data with extremely large deviations in all historical data and exclude ordinary data, thereby reducing the amount of data and improving the training efficiency of the model. In addition, constructing a sample set based solely on data with extremely large deviations is beneficial to improving the accuracy of the model in correcting data with extremely large deviations.

[0039] S203, Sample data is constructed based on the maximum deviation data of each wind farm. The sample inputs of the sample data are historical predicted power, numerical weather forecast data and station operation status data. The sample label is the historical measured power corresponding to the sample input. S204 identifies the dominant driving factors for each wind farm based on the maximum deviation data.

[0040] Specifically, the numerical weather forecast data includes forecast temperature data and forecast wind speed data, and the wind farm operation status data includes wind farm humidity data and wind farm icing status data. S204 specifically includes the following steps: S2041 uses the forecasted temperature data, forecasted wind speed data, wind farm humidity data, and wind farm icing status data from the maximum deviation data as influencing factor nodes X, and the power deviation as the target node Y, to generate a node set. ; in, To forecast wind speed data, To forecast temperature data, This indicates humidity data for wind farms. This represents data on the icing status of a wind farm. This indicates power deviation.

[0041] S2042, Generate directed edges from each influencing factor node in the node set to the target node. ; S2043, for each directed edge Perform correlation tests separately, including: For each directed edge Perform a higher-order partial correlation test (Fisher-Z Test). If X and Y are independent under a given condition set S (i.e., ...), ... If the condition is met, then execute S2044 to remove the edge between X and Y.

[0042] Otherwise, execute S2045 to retain the directed edges that pass the correlation test.

[0043] S2046, a causal graph network is constructed based on every two nodes with directed edges; S2047, Calculate the causal factor strength corresponding to each directed edge in the causal graph network, using the formula: ; The above formula quantifies the influence of influencing factors on power deviation, that is, it calculates the causal factor strength of factor X's influence on power deviation. X can be any of them, It is a dry budget, which means that the state of variables is "forced" to be set by humans (in a simulation experiment) to eliminate the interference of other environmental factors; The mathematical expectation (i.e., average value) of the power deviation.

[0044] This explanation uses the influencing factors of wind farm icing status data as an example. This represents the mathematical expectation of the power deviation under the condition of forced icing. This represents the expected power deviation under the condition of forced non-icing. The formula, through subtraction, calculates the system results under the two intervention settings of "forced icing" versus "forced non-icing". The average difference.

[0045] S2048, for each directed edge, compare the causal factor strength with the preset strength threshold.

[0046] If the causal factor strength meets the preset strength threshold, then execute S2049 to determine the influence factor node of the directed edge as the dominant driving factor.

[0047] Otherwise, execute S20410 to determine that the influence factor node of the directed edge is not the dominant driving factor.

[0048] Understandably, the factors affecting power deviation are not limited to those mentioned above. These four types of meteorological / environmental factors are also affected by non-meteorological factors such as man-made power rationing, power dispatching instructions, and intrinsic mechanism factors (such as equipment failure, sudden changes in physical laws, system instability, etc.). The same steps as in S204 can be used to identify whether these non-meteorological factors are the dominant driving factors. This application does not limit this.

[0049] S205, determine the cluster type label corresponding to each wind farm based on the location information and dominant driving factor of each wind farm, and determine the cluster type to which each wind farm belongs based on the cluster type label.

[0050] Specifically, the location information can be determined based on the operating status data of each wind farm, and the latitude of the wind farm can be extracted. The mapping table is traversed according to each dominant driving factor and dimension of the wind farm. The cluster type label corresponding to each wind farm is determined according to the pre-built cluster type label and the mapping relationship between the dimension and the dominant driving factor in the mapping table.

[0051] For example, for wind farms with latitude > 45.7°N and whose dominant driving factor is wind farm icing status data, they are identified as belonging to high-altitude icing clusters and can be labeled as cluster A; For wind farms located between 44.0° and 45.7°N latitude, with forecasted wind speed data and intrinsic mechanism factors as the dominant driving factors, they are identified as plain high-wave action type clusters and can be labeled as cluster B. For latitudes <44.0°N, the dominant driving factor is Wind farms with basic errors in both types of data are identified as belonging to low-risk, stable clusters and can be labeled as cluster C.

[0052] S206, construct the corresponding sample set based on the set of sample data for each cluster type.

[0053] In some embodiments, the training process of the classification reinforcement learning correction model in S103 may include the following steps: S301, input the sample set of each cluster type into the classification reinforcement learning correction model of the same cluster type for training.

[0054] It should be noted that reinforcement learning agents corresponding to the classification reinforcement learning correction model can be pre-constructed according to the cluster type. The agent uses prediction power, causal factor strength, historical error trend characteristics, and climbing rate as the state space, and the power correction coefficient as the action space, specifically represented as follows: ; in, Indicates the input state of the intelligent agent; The strength parameter is calculated by weighting the causal factor strength based on the strength of each dominant driving factor. The weight of each dominant driving factor can be set according to the value of the causal factor strength. This indicates the historical error trend characteristics of the historical predicted power relative to the historical measured power within a preset step size k before the current time. Specifically, the power deviation between the historical predicted power and the historical measured power within a preset step size k before the current time can be extracted, a power deviation time series can be constructed, and the result can be obtained by performing a weighted moving average on the series. The power ramp rate is indicated.

[0055] Specifically, All of these can be obtained through sample input processing, thereby adjusting the sample input to an input state that is easy for the model to process.

[0056] S302, Obtain the power correction result obtained by the classification reinforcement learning correction model based on the sample input processing, including: Through action space The power generation correction factor is used to adjust the historical predicted power in the sample input. : ; in, As a correction factor, its value range generally falls within... Inside.

[0057] S303, construct a reward function based on the difference between the power correction result and the sample label corresponding to the sample input, and determine the weight coefficient of each term in the reward function according to the cluster type, applying the formula: ; ; in, For the reward function, These are the weighting coefficients, and the values ​​of the two weighting coefficients are determined based on the cluster type. For power correction results, The historical measured power corresponding to the sample label. For the installed capacity of the station, This is the corrected power ramp-up value. The historical predicted power in the sample input, The power ramp-up threshold.

[0058] Specifically, different reward functions can be set according to different cluster types, thereby enabling differentiated training for different clusters to obtain classification reinforcement learning correction models adapted to different clusters. For example, for cluster A: a reward function can be set... Reduce physical constraint penalties, allowing the model to make significant downward adjustments when icing signals are detected to match actual downtime conditions; for cluster B: [the following settings can be configured] This significantly increases the ramp rate penalty, forcing the model to output a smooth curve and suppressing non-physical "spiking" fluctuations; for cluster C: [the following settings can be configured] The focus is on optimizing the RMSE (Real Estate Management and Security) indicators for the entire station.

[0059] S304, Adjust the model parameters of the classification reinforcement learning correction model according to the value of the reward function.

[0060] Understandably, during model training and refinement, the reward function... The corrected power With actual power The deviation between the values ​​is quantified as a penalty value, thereby guiding the policy network of the classification reinforcement learning correction model in this application embodiment to continuously adjust its parameters during offline training to maximize the cumulative reward. This enables the model to output the optimal corrected wind power output for prediction fluctuations caused by various extreme factors such as icing during the real-time correction phase. S305, after determining that the model has converged, output the trained classification reinforcement learning correction model.

[0061] By setting the number of iterations or determining that all samples in the sample set have participated in training, the model convergence can be confirmed. This allows for the generation of classification reinforcement learning correction models corresponding to each cluster type, which can then be used to process wind power prediction data for wind farms falling into the corresponding cluster type.

[0062] Based on the understanding of the above training process, specifically in S102, the operating data can be understood as the data related to the wind turbines generated by the target wind farm during the operation of the corresponding time period, as well as the collected environmental data. Specifically, it can include the measured power, predicted power, numerical weather forecast data (such as predicted temperature and wind speed) and the station operation status data (such as the operating humidity, icing degree, and capacity of the wind turbines) of the target wind farm at each moment in the corresponding time period.

[0063] In some embodiments, the step of determining the cluster type label in S102 may include: The location information of the target wind farm is determined based on the operating data of the target wind farm during a preset period before the current time, and the dominant driving factors of the target wind farm are analyzed and determined. The dominant driving factor of the target wind farm can be easily obtained by traversing the mapping relationship between the location information, the type of dominant driving factor and the cluster type label recorded in the pre-constructed mapping relationship table. For details, please refer to the description of the relevant embodiment in S103, which will not be repeated here.

[0064] In some embodiments, in S103, a classification reinforcement learning correction model for processing the predicted wind power of the target wind farm can be determined according to the cluster type label, and then the correction coefficient of the predicted wind power can be obtained based on the input data processing through the classification reinforcement learning correction model with the same cluster type label.

[0065] Specifically, the operational data of the target wind farm and the predicted wind power at the current moment can be processed into the form of input data for the classification reinforcement learning correction model, specifically in the form of the agent's input state given in S301, which will not be elaborated here.

[0066] Specifically, in this embodiment of the application, the predicted wind power of the target wind farm at each moment can be corrected to obtain the corrected wind power at each moment, thereby generating a power curve showing the change of the corrected wind power over time.

[0067] In some specific embodiments, historical operating data from 51 wind farms distributed in different regions over one year were selected for verification, including wind farms in clusters A and B.

[0068] The baseline model and the traditional BP neural network were selected as the models for predicting wind power.

[0069] The maximum deviation data and ordinary data are extracted from the historical operation data respectively. The maximum deviation sample is constructed based on the maximum deviation data, and the ordinary sample is constructed based on the ordinary data. The set of the maximum deviation sample and the ordinary sample is denoted as the global sample, that is, all samples. The root mean square error can be selected as the evaluation index.

[0070] The experimental results for the two clusters are shown in Table 1 and Table 2 below: Table 1 Comparison of Correction Effects for Cluster A

[0071] Figure 2 The example illustrates a comparison of the wind power output obtained by the baseline model and the classification reinforcement learning correction model (ours) provided in this application on samples from cluster A, in conjunction with Table 1 and... Figure 3 As shown, the classification reinforcement learning correction model provided in this application significantly reduces the root mean square error (RMSE) of the extremely biased samples from 45.2MW to 28.6MW, a reduction of 36.7%.

[0072] Table 2 Comparison of Correction Effects in Cluster B

[0073] Figure 3 The illustration shows the curves of the corrected wind power obtained by the classification reinforcement learning correction model provided in the embodiments of this application, the predicted wind power obtained by the baseline model, and the measured power changing with time step, combined with... Figure 3 As shown in Table 2, in cluster B, the classification reinforcement learning correction model provided in this application embodiment effectively suppresses spurious fluctuations in the predicted power curve, reducing the volatility by approximately 18.5%, and achieving a capture rate of up to 82% for mechanistic drops (compared to only 15% for the baseline model); where the capture rate is used to measure the model's ability to detect or predict specific events (here, "mechanistic drops").

[0074] In summary, the method proposed in this application can accurately identify the dominant driving factors of each wind farm, allowing the model to more accurately correct the predicted power based on the actual dominant driving factors. Furthermore, by introducing different weight parameters for the reward function for wind farms in different clusters, the model provided in this application can achieve significant correction effects in wind farms with different physical characteristics. The above data verifies the effectiveness of the "multi-cluster differentiated modeling" in this application.

[0075] The following are apparatus embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the method embodiments of this application.

[0076] Please see below. Figure 4 The diagram below illustrates the structure of an apparatus provided in an exemplary embodiment of this application. The apparatus includes: The data acquisition unit is used to acquire the predicted wind power of the target wind farm at the current moment; The data processing unit is used to determine the cluster type label based on the operating data of the target wind farm during a preset period before the current time; The wind power correction unit is used to load a classification reinforcement learning correction model corresponding to the cluster type according to the cluster type label, and calculate the correction coefficient through the classification reinforcement learning correction model based on the predicted wind power and the operating data in the preset time period. The wind power correction unit is also used to correct the predicted wind power based on the correction coefficient, and process it to obtain the corrected wind power at the current moment. The classification reinforcement learning correction model is trained based on a pre-built sample set, which is constructed based on historical operation data of multiple wind farms in historical periods.

[0077] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the functional modules described above when executing the methods of the above embodiments. In practical applications, the functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the implementation process can be found in the method embodiments, which will not be repeated here.

[0078] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described above.

[0079] Please see Figure 5This is a structural block diagram of an electronic device provided in an embodiment of this application.

[0080] like Figure 5 As shown, the electronic device includes a processor and a memory.

[0081] In this embodiment, the processor is the control center of the computer system, and can be a processor of a physical machine or a processor of a virtual machine. The processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor can be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array).

[0082] A processor can also include a main processor and a coprocessor. The main processor is used to process data in the wake-up state and is also called the CPU (Central Processing Unit). The coprocessor is a low-power processor used to process data in the standby state.

[0083] The memory may include one or more computer-readable storage media, which may be non-transitory. The memory may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments of this application, the non-transitory computer-readable storage media in the memory are used to store at least one instruction, which is executed by a processor to implement the methods in the embodiments of this application.

[0084] In some embodiments, the electronic device further includes a peripheral device interface and at least one peripheral device. The processor, memory, and peripheral device interface are connected via a bus or signal line. Each peripheral device is connected to the peripheral device interface via a bus, signal line, or circuit board. Specifically, the peripheral device includes: a display screen, a camera, and audio circuitry. The peripheral device interface can be used to connect at least one I / O (Input / Output) related peripheral device to the processor and memory.

[0085] In some embodiments of this application, the processor, memory, and peripheral device interfaces are integrated on the same chip or circuit board; in other embodiments of this application, any one or two of the processor, memory, and peripheral device interfaces can be implemented on separate chips or circuit boards. This application does not specifically limit the implementation in this regard.

[0086] The electronic device structural block diagrams shown in the embodiments of this application do not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0087] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the methods in any of the foregoing embodiments. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0088] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A wind power prediction data processing method based on artificial intelligence, characterized in that, include: Obtain the predicted wind power of the target wind farm at the current moment; The cluster type label is determined based on the operating data of the target wind farm during a preset time period prior to the current moment; The classification reinforcement learning correction model corresponding to the cluster type is loaded according to the cluster type label, and the correction coefficient is calculated by the classification reinforcement learning correction model based on the predicted wind power and the operating data within the preset time period. The predicted wind power is corrected based on the correction coefficient, and the corrected wind power at the current moment is obtained. The classification reinforcement learning correction model is trained based on a pre-built sample set, which is constructed based on historical operation data of multiple wind farms in historical periods.

2. The wind power prediction data processing method based on artificial intelligence according to claim 1, characterized in that, The process of constructing the sample set includes: Historical operating data of multiple wind farms in different historical periods are acquired; wherein, the historical operating data includes historical measured power, historical predicted power, numerical weather forecast data, and station operating status data; The corresponding maximum deviation data were obtained by filtering the historical operation data of each wind farm for each period within the historical period. Sample data are constructed based on the maximum deviation data of each wind farm. The sample inputs of the sample data are historical predicted power, numerical weather forecast data and station operation status data. The sample label is the historical measured power corresponding to the sample input. Identify the dominant driving factors for each wind farm based on the maximum deviation data; The cluster type label corresponding to each wind farm is determined based on the location information and dominant driving factors of each wind farm, and the cluster type to which each wind farm belongs is determined based on the cluster type label. A corresponding sample set is constructed based on the collection of sample data for each cluster type.

3. The wind power prediction data processing method based on artificial intelligence according to claim 2, characterized in that, The maximum deviation data obtained by filtering the historical operating data of each wind farm for each period within the historical period includes: The wind power generation capacity at the corresponding time is determined based on the station's operational status data. Calculate the power deviation between the historical measured power and the historical predicted power at each time point in each time period for each wind farm; Calculate the ratio of the power deviation to the wind power unit's operating capacity to obtain the relative degree of deviation; If the relative deviation value is greater than the preset deviation threshold, the historical operating data within the corresponding time period is determined to be extremely deviation data.

4. The wind power prediction data processing method based on artificial intelligence according to claim 3, characterized in that, The numerical weather forecast data includes forecast temperature data and forecast wind speed data, and the station operation status data includes wind farm humidity data and wind farm icing status data. The process of identifying the dominant driving factors based on the maximum deviation data of each wind farm includes: The forecast temperature data, forecast wind speed data, wind farm humidity data, and wind farm icing status data in the maximum deviation data are respectively used as influencing factor nodes, and the power deviation is used as the target node to generate a node set. Generate directed edges from each influencing factor node to the target node in the node set, perform correlation tests on each directed edge, retain the directed edges that pass the correlation test, and construct a causal graph network based on every two nodes with directed edges. Calculate the causal factor strength corresponding to each directed edge in the causal graph network; For each directed edge, if the causal factor strength is greater than a preset strength threshold, then the influence factor node of the directed edge is determined as the dominant driving factor.

5. The wind power prediction data processing method based on artificial intelligence according to claim 3, characterized in that, The step of determining the cluster type label corresponding to each wind farm based on its location information and dominant driving factors includes: The location information is determined based on the station operation status data of each wind farm, and the latitude of the wind farm is extracted. The mapping table is traversed according to each dominant driving factor and dimension of the wind farm. The cluster type label corresponding to each wind farm is determined according to the pre-built cluster type label and the mapping relationship between the dimension and the dominant driving factor in the mapping table.

6. A wind power prediction data processing method based on artificial intelligence according to any one of claims 2-5, characterized in that, The training process of the classification reinforcement learning correction model includes the following steps: Input the sample set of each cluster type into the classification reinforcement learning correction model of the same cluster type for training; Obtain the power correction result obtained by the classification reinforcement learning correction model based on the sample input; Construct a reward function based on the difference between the power correction result and the sample label corresponding to the sample input; The model parameters of the classification reinforcement learning correction model are adjusted according to the value of the reward function; After determining that the model has converged, output the trained classification reinforcement learning correction model.

7. The wind power prediction data processing method based on artificial intelligence according to claim 6, characterized in that, The reward function is constructed based on the difference between the power correction result and the sample label corresponding to the sample input, and the weight coefficient of each term in the reward function is determined according to the cluster type. The formula of the reward function is expressed as follows: ； ； in, For the reward function, These are the weighting coefficients, and the values ​​of the two weighting coefficients are determined based on the cluster type. For power correction results, The historical measured power corresponding to the sample label. For the installed capacity of the station, This is the corrected power ramp-up value. The historical predicted power in the sample input, The power ramp-up threshold.

8. A wind power prediction data processing device based on artificial intelligence, characterized in that, include: The data acquisition unit is used to acquire the predicted wind power of the target wind farm at the current moment; The data processing unit is used to determine the cluster type label based on the operating data of the target wind farm during a preset period before the current time; The wind power correction unit is used to load a classification reinforcement learning correction model corresponding to the cluster type according to the cluster type label, and calculate the correction coefficient through the classification reinforcement learning correction model based on the predicted wind power and the operating data in the preset time period. The wind power correction unit is also used to correct the predicted wind power based on the correction coefficient, and process it to obtain the corrected wind power at the current moment. The classification reinforcement learning correction model is trained based on a pre-built sample set, which is constructed based on historical operation data of multiple wind farms in historical periods.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images