An adaptive threshold wind farm ultra-short-term power prediction method, device, equipment and storage medium

Abstract

The application discloses a wind farm ultra-short-term power prediction method and device based on an adaptive threshold, an equipment and a storage medium, and relates to the technical field of wind power generation. The method comprises the following steps: obtaining a standardized time series data set according to numerical weather prediction data and measured data of a wind farm; obtaining an initial power prediction value of a wind turbine in the wind farm based on the standardized time series data set; calculating a physical rationality score based on the initial power prediction value; replacing a traditional fixed threshold with an adaptive physical threshold combined with recent operation characteristics of the wind farm to determine the physical rationality of the initial power prediction value and trigger a correction of the initial power prediction value; calculating a time series stability score after correction, adopting an adaptive time series threshold combined with meteorological change characteristics at a prediction moment to identify a rolling prediction error accumulation risk and trigger a secondary correction; and finally obtaining a total power prediction value of the wind farm based on a target power prediction value. The method can solve the problem that a prediction result is prone to violating physical laws of wind farm operation, reduce errors and improve the stability of prediction.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and in particular to an adaptive threshold method, apparatus, equipment and storage medium for ultra-short-term power prediction of wind farms. Background Technology

[0002] As an important component of clean energy, wind power has seen continuous growth in installed capacity in recent years, and its proportion in the power grid supply structure has been steadily increasing. Wind power output is characterized by randomness, fluctuation, and intermittency; significant variations in its output power can impact the frequency stability, voltage regulation, and power balance of the power grid. To ensure the safe and stable operation of the power grid, improve wind power absorption, and reduce grid dispatch costs, accurate prediction of wind power output is necessary.

[0003] Data-driven methods, with their superior nonlinear fitting capabilities, have become the mainstream technology for ultra-short-term wind power forecasting. This method trains the model using a large amount of historical operating data and meteorological data, establishing a mapping relationship between input features and output power. It can adapt to complex and ever-changing wind farm operating conditions and meet basic forecasting needs in most conventional scenarios.

[0004] However, data-driven prediction methods rely solely on statistical patterns and are not constrained by the physical laws governing wind turbine operation, easily leading to outputs that do not conform to actual operating conditions. To address this issue, existing technologies generally employ output truncation. This method can only correct values ​​exceeding obvious physical boundaries, failing to resolve the fundamental problem of predictions not conforming to physical laws. The predictions may still contain implicit inconsistencies, and forced corrections disrupt the inherent continuity of the data, introducing new errors. Existing wind power prediction methods are prone to violating the physical laws of wind farm operation, with errors accumulating continuously during rolling predictions, resulting in poor prediction accuracy. Summary of the Invention

[0005] This application provides an adaptive threshold method, apparatus, equipment, and storage medium for ultra-short-term power prediction of wind farms, which can solve the problem that the prediction results of existing methods are prone to violating the physical laws of wind farm operation, reduce errors, and improve the accuracy of prediction results.

[0006] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, this application provides an adaptive threshold method for ultra-short-term power prediction of wind farms, the method comprising: Acquire numerical weather forecast data and measured data of wind farms; A standardized time-series dataset was obtained based on numerical weather forecast data and measured data from wind farms. Based on the standardized time-series dataset, the initial power prediction value of each wind turbine in the wind farm is obtained; Based on the initial power prediction values ​​of wind turbine units and standardized time-series datasets, calculate the physical rationality score; Determine whether the physical rationality score is less than the physical threshold to obtain the first judgment result; If the first judgment result indicates that the physical rationality score is less than the physical threshold, then the initial power prediction value is corrected to obtain the corrected power prediction value of the wind turbine. Based on the corrected power prediction values ​​of wind turbine units and the standardized time series dataset, the time series stability score is calculated. Determine whether the time series stability score is less than the time series threshold to obtain a second determination result; If the second judgment result indicates that the time series stability score is less than the time series threshold, then the corrected power prediction value is corrected to obtain the target power prediction value of the wind turbine. The total power prediction value of the wind farm is obtained based on the target power prediction value of the wind turbine.

[0007] In some possible implementations, the method further includes: If the first judgment result indicates that the physical rationality score is greater than or equal to the physical threshold, then the initial power prediction value of the wind turbine is determined as the corrected power prediction value of the wind turbine.

[0008] In some possible implementations, the method further includes: If the second judgment result indicates that the time series stability score is greater than or equal to the time series threshold, then the corrected power prediction value of the wind turbine is determined as the target power prediction value of the wind turbine.

[0009] In some possible implementations, the standardized time-series dataset obtained from wind farm numerical weather forecast data and wind farm measured data includes: Spatiotemporal alignment processing is performed on numerical weather prediction data and measured data of wind farms to obtain an aligned dataset. Outliers in the aligned dataset are removed, and missing values ​​are filled in the aligned dataset after outlier removal to obtain a complete dataset. The complete dataset is then denoised, and the denoised complete dataset is standardized to obtain a standardized time series dataset.

[0010] In some possible implementations, obtaining the initial power prediction value for each wind turbine in the wind farm based on a standardized time-series dataset includes: Obtain the aerodynamic characteristic parameters of wind turbines in the wind farm; based on the standardized time-series dataset and the aerodynamic characteristic parameters of wind turbines in the wind farm, combined with operational constraints and physical mechanism constraints, calculate the initial power prediction value of each wind turbine in the wind farm.

[0011] In some possible implementations, the total power prediction value of the wind farm is obtained based on the target power prediction value of the wind turbine, including: The target power predictions of all wind turbines in the wind farm are aggregated and calculated to obtain the total power prediction of the wind farm.

[0012] In some possible implementations, the method further includes: Obtain the measured total power of the wind farm; based on the measured total power of the wind farm and the predicted total power of the wind farm, update the relevant parameters used to calculate the initial power prediction value, physical rationality score and time series stability score.

[0013] Secondly, this application provides an adaptive threshold wind farm ultra-short-term power prediction device, the device comprising: The acquisition module is used to acquire numerical weather forecast data and measured data of wind farms; and to obtain a standardized time-series dataset based on the numerical weather forecast data and measured data of wind farms. The first calculation module is used to obtain the initial power prediction value of each wind turbine in the wind farm based on the standardized time series dataset; The first judgment module is used to calculate a physical rationality score based on the initial power prediction value of the wind turbine and the standardized time series dataset; determine whether the physical rationality score is less than a physical threshold to obtain a first judgment result; if the first judgment result indicates that the physical rationality score is less than the physical threshold, then the initial power prediction value is corrected to obtain the corrected power prediction value of the wind turbine. The second judgment module is used to calculate the time series stability score based on the corrected power prediction value of the wind turbine and the standardized time series dataset; determine whether the time series stability score is less than the time series threshold, and obtain a second judgment result; if the second judgment result indicates that the time series stability score is less than the time series threshold, then the corrected power prediction value is corrected to obtain the target power prediction value of the wind turbine. The second calculation module is used to obtain the total power prediction value of the wind farm based on the target power prediction value of the wind turbine.

[0014] Thirdly, this application provides a computing device, including a memory and a processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of the first aspects.

[0015] Fourthly, this application provides a computer-readable storage medium for storing a computer program for performing the method as described in any one of the first aspects.

[0016] Fifthly, this application provides a computer program product comprising one or more computer instructions, wherein when the computer instructions are executed by a computer, the computer performs the method as described in any one of the first aspects.

[0017] As can be seen from the above technical solution, this application has at least the following beneficial effects: In this application, numerical weather prediction data and measured data of wind farms are obtained; a standardized time-series dataset is obtained based on the numerical weather prediction data and measured data of wind farms; based on the standardized time-series dataset, the initial power prediction value of each wind turbine in the wind farm is obtained; based on the initial power prediction value of the wind turbine and the standardized time-series dataset, a physical rationality score is calculated; if the physical rationality score is less than a physical threshold, the initial power prediction value is corrected to obtain the corrected power prediction value of the wind turbine; based on the corrected power prediction value of the wind turbine and the standardized time-series dataset, a time-series stability score is calculated; if the time-series stability score is less than a time-series threshold, the corrected power prediction value is further corrected to obtain the target power prediction value of the wind turbine; based on the target power prediction value of the wind turbine, the total power prediction value of the wind farm is obtained. In existing technical solutions, wind power prediction often adopts a pure data-driven model, which relies on a large amount of training data, is prone to prediction results that violate the physical laws of wind turbines, and errors are prone to accumulate and become uncontrollable during rolling prediction. The model has a large number of parameters and high inference latency, making it difficult to adapt to edge deployment. As can be seen, this application ensures that the prediction results conform to physical laws from the source by constraining physical mechanisms, performing two-level verification and dynamic correction, effectively suppressing error accumulation and improving prediction accuracy.

[0018] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0019] Figure 1 A flowchart of an adaptive threshold method for predicting ultra-short-term power in wind farms, provided as an embodiment of this application; Figure 2 A schematic diagram of an adaptive threshold wind farm ultra-short-term power prediction device provided in an embodiment of this application; Figure 3 This is a schematic diagram of a computing device provided in an embodiment of this application. Detailed Implementation

[0020] The terms "first," "second," and "third," etc., used in this application specification and accompanying drawings are used to distinguish different objects, not to limit a specific order.

[0021] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0022] Currently, in practical applications, wind power ultra-short-term power prediction largely relies on purely data-driven models. Problems such as prediction results easily violating the physical laws of the turbine and the accumulation of rolling errors are already widespread. In addition, existing technologies generally suffer from insufficient multi-source data fusion, poor spatiotemporal matching, and the tendency to lose extreme operating condition characteristics during denoising, leading to low data utilization and a significant decrease in prediction accuracy in extreme scenarios. Furthermore, high-precision models typically have a large number of parameters and high inference latency, making them difficult to adapt to deployments of low-computing-power equipment at the edge of wind farms.

[0023] In view of this, embodiments of this application provide an adaptive threshold method for ultra-short-term power prediction of wind farms. This method can be applied to a processing device, which can be a terminal or a server. Terminals include, but are not limited to, smartphones, tablets, laptops, personal digital assistants, or smart wearable devices. Servers can be cloud servers, such as central servers in a central cloud computing cluster or edge servers in an edge cloud computing cluster. Alternatively, servers can be located in a local data center. A local data center refers to a data center directly controlled by the user. Compared with existing technologies, this application avoids physically unreasonable predictions from the source and suppresses error accumulation through physical mechanism modeling, two-level verification, and dynamic correction, while improving data utilization efficiency, reducing deployment difficulty, and better suiting actual field application scenarios.

[0024] To make the technical solution of this application clearer and easier to understand, the following describes, in conjunction with the accompanying drawings, an adaptive threshold method for wind farm ultra-short-term power prediction provided by an embodiment of this application. Figure 1 As shown, this figure is a flowchart of an adaptive threshold-based ultra-short-term power prediction method for wind farms provided in an embodiment of this application. The method includes: S201. The processing equipment acquires numerical weather forecast data and measured data of the wind farm, and obtains a standardized time series dataset based on the numerical weather forecast data and measured data of the wind farm.

[0025] Numerical weather forecast data for wind farms refers to meteorological forecast data for future periods provided by meteorological departments or professional meteorological service agencies, covering the area where the wind farm is located. This includes data on wind speed, wind direction, ambient temperature, atmospheric pressure, and relative humidity. The numerical weather forecast data for wind farms is continuous forecast data for periods after the current moment. Its forecast duration and time resolution can be dynamically adapted and adjusted according to the time granularity requirements of grid dispatch, the wind farm's operation and dispatch plan, and the operational needs of ultra-short-term power forecasting. For example, meteorological forecast data at minute or 15-minute intervals after the current moment can be selected, such as 15 minutes, 1 hour, or 4 hours. Alternatively, the forecast duration can be set to any duration within the scope of ultra-short-term power forecasting, depending on the specific application scenario. This ensures that the forecast period fully covers the entire time interval required for grid dispatch and wind farm operation while avoiding interference from low-precision meteorological data caused by excessively long forecast durations, thus guaranteeing the reliability of ultra-short-term power forecast data.

[0026] Measured data from a wind farm refers to the historical operating data of each wind turbine unit within the wind farm, collected by the wind farm's SCADA (Supervisory Control and Data Acquisition) system. This data includes historical output power, rotor speed, pitch angle, nacelle yaw angle, and unit operating status code data. The measured data represents recent historical operating data up to the current moment. Its time span can be dynamically adjusted according to the actual operating characteristics of the wind farm, the degree of weather condition fluctuations, and the accuracy requirements of ultra-short-term power forecasting. For example, continuous operating data from the previous 12 or 24 hours can be selected. Alternatively, the time span can be set to any duration within the range of 6 to 48 hours, depending on the specific application scenario. This ensures that the data fully characterizes the recent operating patterns and conditions of the wind turbine units while avoiding the introduction of historical data with excessively long time spans that could reduce the accuracy of operating condition matching, thus guaranteeing the real-time performance of ultra-short-term power forecasting.

[0027] The processing equipment performs spatiotemporal alignment processing on the numerical weather forecast data and the measured data of the wind farm to obtain an aligned dataset. Outliers in the aligned dataset are removed, and missing values ​​are filled in the aligned dataset after outlier removal to obtain a complete dataset. The complete dataset is then denoised, and the denoised complete dataset is standardized to obtain a standardized time-series dataset.

[0028] The specific process is as follows: The processing equipment interpolates the numerical weather forecast data of the wind farm to the latitude and longitude coordinates of each wind turbine in the wind farm, and obtains the meteorological data sequence of the corresponding single wind turbine location. The linear interpolation method is used to unify the numerical weather forecast data of the wind farm and the measured data of the wind farm to the same time granularity, and generate an aligned dataset with perfectly matched timestamps.

[0029] The processing device identifies and marks outliers in the numerical weather forecast data of wind farms in the aligned dataset using the first anomaly detection rule; and identifies and marks outliers in the measured data of wind farms in the aligned dataset using the second anomaly detection rule, marking the data locations of all outliers as missing.

[0030] The first anomaly detection rule is a combined anomaly identification rule designed specifically for the characteristics of numerical weather prediction data. It employs a combination of the standard deviation method and the interquartile range method. For each characteristic sequence in the numerical weather prediction data, the mean, standard deviation, first quartile, third quartile, and interquartile range of that sequence are calculated simultaneously. When a data point simultaneously exceeds the range of mean plus or minus 2.5 times the standard deviation, and also exceeds the range of first quartile minus 1.2 times the interquartile range to third quartile plus 1.2 times the interquartile range, the data point is determined to be an anomaly. This first anomaly detection rule is adapted to the characteristic that numerical weather prediction data itself has a certain degree of prediction error, adopting a relatively lenient judgment standard to avoid mistakenly deleting valid meteorological trend data.

[0031] The second anomaly detection rule is a combined anomaly identification rule designed specifically for the characteristics of SCADA measured data in wind farms. It also employs a combination of the standard deviation method and the interquartile range method. For each feature sequence in the SCADA measured data, the mean, standard deviation, first quartile, third quartile, and interquartile range of that feature sequence are calculated simultaneously. When a data point exceeds the range of the mean plus or minus three times the standard deviation, or exceeds the range of the first quartile minus 1.5 times the interquartile range to the third quartile plus 1.5 times the interquartile range, that data point is determined to be an anomaly. This second anomaly detection rule is adapted to the high accuracy characteristics of SCADA measured data, employing relatively strict judgment criteria to effectively eliminate obvious anomalies caused by sensor failures, communication interruptions, etc.

[0032] Next, for all positions marked as missing, preliminary completion is performed using the median of all non-missing values ​​in the corresponding feature sequence to obtain an intermediate dataset without significant missing values. Then, the intermediate dataset is subjected to the first standardization process, and the weighted K-nearest neighbor algorithm is used to complete the positions corresponding to the original outliers. After completion, the dataset is inversely standardized back to the original scale to obtain the complete dataset.

[0033] The processing device performs denoising on each feature sequence in the complete dataset using a combination of variational mode decomposition and wavelet thresholding, retaining effective features while removing noise interference. A second standardization process is then performed on the denoised complete dataset to unify all features to the same dimension, resulting in a standardized time series dataset.

[0034] The beneficial effects of this step are as follows: it solves the problem of spatiotemporal misalignment of multi-source data through spatiotemporal alignment processing, ensuring the consistency of data in time and space dimensions; it effectively removes abnormal data while preserving data integrity to the greatest extent through categorical outlier removal and two-level missing value completion; it avoids the loss of extreme condition features while removing noise through joint denoising processing; and it improves the accuracy and stability of prediction results by eliminating the dimensional differences between different features through standardization processing.

[0035] S202. The processing equipment obtains the initial power prediction value of each wind turbine in the wind farm based on the standardized time series dataset.

[0036] Obtain the aerodynamic characteristic parameters of wind turbines in the wind farm; based on the standardized time-series dataset and the aerodynamic characteristic parameters of wind turbines in the wind farm, combined with operational constraints and physical mechanism constraints, calculate the initial power prediction value of each wind turbine in the wind farm.

[0037] The specific process is as follows: The processing equipment acquires the aerodynamic characteristic parameters of each wind turbine in the wind farm. The aerodynamic characteristic parameters of the wind turbine are inherent parameters provided by the wind turbine at the factory, including the rotor swept area, mechanical transmission efficiency, power generation efficiency, optimal pitch angle curve, cut-in wind speed, cut-out wind speed, and rated power.

[0038] The processing equipment extracts the feature values ​​corresponding to the prediction time from the standardized time series dataset, performs inverse standardization, and obtains the original dimensional operating parameters, including the wind speed at the prediction time, the ambient temperature at the prediction time, the atmospheric pressure at the prediction time, the relative humidity at the prediction time, the pitch angle at the current time, the rotor speed at the current time, the nacelle yaw angle at the current time, and the wind turbine operating status code at the current time.

[0039] The processing equipment determines the operating constraints of the wind turbine, which include the operating status constraints of the wind turbine and the operating range constraints of key parameters.

[0040] The operating status constraints of wind turbines are determined based on the operating status code of the unit, and the output rules under different operating conditions are quantified by the operating status correction coefficient. The operating range constraints for key parameters limit the allowable value ranges for pitch angle, rotor angular velocity, and nacelle yaw angle. Specifically: the allowable range for pitch angle is limited to the minimum to maximum pitch angle range of the wind turbine's pitch system mechanical design; measured pitch angle values ​​outside this range are considered abnormal sensor data and are not included in the initial power calculation. The allowable range for rotor angular velocity is limited to the range from the wind turbine's stationary angular velocity to its rated operating angular velocity; measured rotor angular velocity values ​​outside this range are considered abnormal sensor data and are not included in the initial power calculation. The allowable range for nacelle yaw angle is limited to the minimum to maximum yaw angle range of the wind turbine's yaw system mechanical design; measured nacelle yaw angle values ​​outside this range are considered abnormal sensor data and are not included in the initial power calculation.

[0041] The processing equipment then determines the physical constraints of the wind turbine, which include non-negativity constraints, upper limit constraints on output, and wind speed boundary constraints. The mathematical expressions are as follows: Nonnegativity constraint: ; Output limit constraint: ,in, Rated power of wind turbine, unit ; Wind speed boundary constraints: when hour, ;when hour, ,in, To cut into wind speed, To cut off the wind speed, the units are all... .

[0042] Based on a standardized time-series dataset and aerodynamic characteristic parameters of wind turbines in a wind farm, the processing equipment calculates the initial power prediction value for each wind turbine in the wind farm, combining operational constraints and physical mechanism constraints. The calculation formula is as follows:

[0043] in, For the predicted time The initial power prediction value, For the mechanical transmission efficiency of wind turbine units. The power generation efficiency of wind turbine units. Standard air density, For the area swept by the wind turbine, To predict time points in a standardized time series dataset The wind speed value, The wind energy utilization coefficient, To standardize the current time in the time series dataset The pitch angle value, For the predicted time The tip speed ratio, For the predicted time The deviation between the wind direction and the cabin yaw angle at the current time t. To predict time points in a standardized time series dataset Ambient temperature value, To predict time points in a standardized time series dataset Atmospheric pressure value, To predict time points in a standardized time series dataset The relative humidity value, For environmental parameter correction factors, This is the current operating status code of the wind turbine. This is the correction factor for the operating status of the wind turbine. It is set to 1 during normal operation, the value is the actual power limitation ratio during power-limited operation, and 0 during fault shutdown. For the current moment, The step size is calculated to range from 15 minutes to 4 hours.

[0044] The calculation formula is as follows:

[0045] in, This is the first environmental correction factor. The primary standard environmental parameter is... This is the second environmental correction factor. This is the second standard environmental parameter. This is the third environmental correction factor. This is the third standard environmental parameter.

[0046] The benefits of this step are as follows: it clarifies that the setting of key parameter operating range constraints is based on the inherent technical documents of the wind turbine manufacturer, ensuring the objectivity and engineering applicability of the constraints; it eliminates abnormal sensor data through parameter range pre-verification, avoiding calculation errors caused by invalid input; and it combines operational constraints and physical mechanism constraints to ensure the rationality and reliability of the initial prediction values ​​from both the input data and calculation process dimensions, further improving the accuracy of the prediction results.

[0047] In some embodiments, to further improve the prediction accuracy under complex meteorological conditions and extreme scenarios, this application may also use the initial power prediction value calculated based on the physical mechanism as a physical prior feature, fuse it with a standardized time-series dataset, and input it into an improved residual temporal convolutional network with embedded physical mechanism constraints for joint optimization calculation to obtain the final initial power prediction value. The specific process is as follows: The processing device concatenates the physical prior features at each moment with the standardized meteorological features and measured operational features at the corresponding moment to construct a sliding window time series sample set containing physical prior information. The length of the sliding window is consistent with the historical data dependency length of the ultra-short-term power prediction. The time series sample set is divided into a training set, a validation set, and a test set according to a preset ratio of 7:2:1. The training set is used for iterative updates of model parameters, the validation set is used to monitor the training process and implement early stopping to prevent overfitting, and the test set is used to evaluate the generalization ability and prediction accuracy of the model after training to ensure that the model meets the requirements of engineering applications.

[0048] The processing device constructs a physically constrained embedded improved residual temporal convolutional network as the backbone prediction model, with the improved residual module as its core unit. This improved residual module consists of a main convolutional branch and an identity mapping jump-connection branch. The main convolutional branch employs a cascaded structure of one-dimensional depthwise separable convolution, batch normalization, and gated linear unit activation to extract deep temporal correlations of input features. The identity mapping jump-connection branch directly transmits input features to the module output, avoiding the gradient vanishing and gradient exploding problems during deep network training. This application embeds wind power physical mechanism constraints, including wind speed-power characteristic curve constraints, turbine output threshold constraints, and non-negative output constraints, into the gradient update stage of the main convolutional branch of the residual module. This forces correction of feature gradients that violate these physical laws, ensuring that the backbone prediction model always follows the physical laws of wind turbine operation during feature extraction. Meanwhile, the backbone prediction model sets up a multi-scale dilated causal convolution branch to capture the short-term mutation, short-to-medium-term fluctuation and long-term trend features of wind power respectively. After the convolution block attention module adaptively weights and fuses the features at different scales, it outputs the preliminary power prediction value and 95% confidence interval.

[0049] The 95% confidence interval is used to quantify the statistical uncertainty of the prediction results, indicating that there is a 95% probability that the actual output power of the wind turbine will fall within this interval. It can provide a quantitative risk decision-making basis for grid dispatch. The narrow interval supports the formulation of power generation plans to reduce dispatch costs, while the wide interval prompts the advance reservation of peak-shaving reserve capacity to avoid grid security risks. At the same time, it can realize proactive risk warning under extreme weather conditions. Moreover, due to the embedded physical mechanism constraints, the interval will not have physical unreasonable situations such as negative lower limit or upper limit exceeding rated power, which has higher engineering credibility and can be directly adapted to the grid's 5-minute real-time dispatch requirements.

[0050] The processing equipment employs a three-layer composite loss function to collaboratively optimize the training process of the backbone prediction model. Based on the prediction results of the training set, the mean squared error loss, gradient smoothing loss, and physical constraint loss are calculated separately. The three losses are weighted and summed to obtain the total loss. The model parameters are iteratively adjusted based on the total loss. The training process is monitored in real time using a validation set. Training is stopped when the validation set loss no longer decreases for 10 consecutive rounds, resulting in a trained backbone prediction model. The prediction accuracy and physical rationality of the backbone prediction model are verified using a test set to ensure that the backbone prediction model can output prediction results that conform to physical laws under both normal and extreme operating conditions.

[0051] In some embodiments, this application also constructs a lightweight error fitting branch that shares low-level features with the backbone network. The error fitting branch shares the low-level temporal features extracted from the first three convolutional layers of the backbone network, and simultaneously extracts the historical prediction residuals and real-time meteorological features of the backbone model as additional inputs. A single-layer gated recurrent unit structure is used to achieve rapid error fitting. The training set and validation set are input into the error fitting branch for synchronous training to obtain a trained error compensation model. The trained backbone prediction model and the error compensation model together constitute a joint prediction model. In the real-time inference stage, the standardized temporal data at the current moment and the corresponding physical prior features are input into the trained backbone model to obtain a preliminary power prediction value. At the same time, the preliminary power prediction value output by the backbone model is input into the error compensation model to obtain a real-time error compensation amount. The preliminary power prediction value and the real-time error compensation amount are superimposed to calculate the final initial power prediction value after joint inference. This final initial power prediction value can be directly used as the input for calculating the physical rationality score in step S203.

[0052] In some embodiments, the processing device performs structured pruning on the trained joint prediction model, removing redundant convolutional channels whose contribution to prediction accuracy is lower than a preset contribution threshold. This achieves a preset compression ratio for the number of parameters in the joint prediction model while minimizing accuracy loss, resulting in a lightweight joint prediction model. A deployment mode combining cloud pre-training and edge incremental fine-tuning is adopted. The joint prediction model is pre-trained in the cloud based on common operational data from multiple wind farms. The lightweight joint prediction model parameters are then distributed to the wind farm edge devices via an encrypted communication channel. Incremental fine-tuning is performed on the lightweight joint prediction model based on local operational data of a preset duration from the wind farm, resulting in a final prediction model adapted to local terrain, meteorological characteristics, and turbine operating conditions. This ensures that the inference latency is less than a preset latency threshold, meeting the real-time scheduling requirements of the power grid at a preset time granularity and the deployment needs of low-computing-power devices at the wind farm edge.

[0053] S203. The processing device calculates the physical threshold based on a standardized time-series dataset.

[0054] The physical threshold is a dynamic adaptive threshold, and its setting method is as follows: The physical threshold is based on the basic physical threshold and dynamically adjusted according to the recent operating characteristics of the wind farm. The calculation formula is as follows:

[0055] in, For physical threshold, Based on the basic physical threshold, The first weighting coefficient is determined based on the historical operating data of the wind farm; This represents the standard deviation of the measured wind speed sequence over the past 24 hours. This is the second weighting coefficient. This represents the change in measured wind speed over the past hour. The third weighting coefficient, This represents the standard deviation of the measured power series over the past 24 hours. It is the fourth weighting coefficient. This represents the average forecast error over the past 24 hours. For example, It can be taken as 0.02. It can be taken as 0.03. It can be 0.000001. It can be taken as 0.0000005.

[0056] S204. The processing equipment calculates a physical rationality score based on the initial power prediction value of the wind turbine and the standardized time series dataset.

[0057] The physical rationality score is used to quantify the degree of conformity between the initial power prediction value and the historical physical laws of wind turbine operation. The calculation formula is as follows:

[0058] in, For the predicted time The physical rationality score ranges from 0 to 1. The higher the score, the more the initial prediction value conforms to the physical laws. The number of historical reference samples ranges from 24 to 96, corresponding to historical data from the past 6 to 24 hours; i is the index of the historical sample, corresponding to the historical time, and ranges from 1 to N. To standardize the wind speed value at the i-th historical moment in the time series dataset, For the standardized time series dataset, the first Measured power values ​​at a historical moment, The time decay weight for the i-th historical moment is... This refers to the rated power of the wind turbine.

[0059] The beneficial effect of this step is that wind power is directly proportional to the cube of wind speed, which is a fundamental physical law of wind power energy conversion. Under different wind speeds, the base output power of wind turbines varies significantly, and directly comparing the absolute power values ​​cannot accurately reflect the degree of conformity between the predicted value and historical operating patterns. By using the ratio of historical measured power to the cube of the wind speed at the corresponding moment as a benchmark to normalize the initial power prediction value, the influence of differences in the base power under different wind speeds can be eliminated, allowing the physical rationality score to objectively reflect the consistency between the initial prediction value and historical operating data under the same conditions. Combining time decay weights to highlight recent operating patterns quantifies the physical rationality of the initial prediction value, supporting the formulation of corrective decisions.

[0060] In some embodiments, the time decay weight The calculation is performed using an exponential decay function, and the formula is as follows: ,in The attenuation coefficient ranges from 0.01 to 0.05, and can be exemplarily set to 0.02. This ensures that historical samples closer to the current moment have a greater impact on the score, better reflecting the real-time changes in wind farm operating conditions. When using the final initial power prediction value output by the aforementioned improved residual temporal convolutional network for calculation, the 95% confidence interval of the model output can be simultaneously incorporated into the calculation benchmark for the physical rationality score. When the predicted value falls within the confidence interval, the fault tolerance threshold of the score can be appropriately increased to avoid over-correction.

[0061] Furthermore, the numerical values ​​listed in the embodiments of this application are merely illustrative and are not intended to limit the scope of protection of this application. Those skilled in the art can reasonably adjust the above numerical values ​​according to actual application scenarios, and all equivalent substitutions or adaptive modifications should fall within the scope of protection of this application.

[0062] S205. The processing equipment determines whether the physical rationality score is less than the physical threshold and obtains the first judgment result.

[0063] The beneficial effect of this step is that the operating characteristics of wind farms change with weather conditions. When wind speed fluctuates significantly, the normal fluctuation range of wind turbine output power also increases accordingly. When the historical average prediction error is large, it indicates that the model's prediction reliability under current operating conditions has decreased. Using a fixed threshold cannot adapt to these changes and is prone to misjudgments or omissions. Dynamic adaptive thresholds, by introducing parameters such as wind speed standard deviation, wind speed variation, power standard deviation, and average prediction error, can adjust the threshold size in real time according to the recent operating characteristics of the wind farm. The threshold can be appropriately lowered when wind speed fluctuations are large or model errors are large, and appropriately raised when operation is stable or model errors are small, making the correction decisions more consistent with actual operating conditions and improving the accuracy of the correction decisions.

[0064] In some embodiments, the basic physical threshold The value range is 0.7 to 0.9, and 0.85 can be used as an example. When the wind turbine is in a power-limited operation state, the physical threshold can be adjusted a second time based on the unit operation state correction coefficient, reducing the threshold by 0.05 to 0.1 to adapt to the changes in power output characteristics under power-limited conditions and avoid misjudging normal power-limited output as physically unreasonable. When a fault shutdown state of the unit is detected, the physical reasonableness score is directly determined to be 0, forcibly triggering the correction process to ensure that the prediction result is consistent with the actual operating state of the unit.

[0065] If the first judgment result indicates that the physical rationality score is less than the physical threshold, then execute S206; If the first judgment result indicates that the physical rationality score is greater than or equal to the physical threshold, then execute S212.

[0066] S206. The processing equipment corrects the initial power prediction value to obtain the corrected power prediction value of the wind turbine.

[0067] The initial power prediction value is corrected using a historical operating condition weighted correction method. The correction formula is as follows:

[0068] in, For the predicted time The corrected power prediction value, This is the initial predicted value weighting coefficient, ranging from 0.2 to 0.8. A lower physical plausibility score generally indicates a lower weighting. The smaller the value; For the first Air density at a historical moment.

[0069] The beneficial effect of this step is that when the physical rationality of the initial power prediction is insufficient, it indicates a significant deviation between the model's prediction and historical operating patterns. In this case, the initial prediction can be corrected by incorporating historical operating data under similar conditions. By converting historical measured power into equivalent power under current wind speed and air density, and then weighting the average according to time decay, a reference power value based on historical operating patterns can be obtained. Weighting the initial prediction with this reference power value allows the model's predictive ability to be preserved while introducing the constraints of historical operating patterns. The weight coefficient of the initial prediction is dynamically adjusted based on the physical rationality score; the lower the physical rationality, the smaller the weight of the initial prediction and the larger the weight of the historical reference power value, making the correction result more reliable.

[0070] S207. The processing device calculates the time series threshold based on the standardized time series dataset.

[0071] The time-series threshold is a dynamically adaptive threshold, and its setting method is as follows: The time series threshold is based on the baseline time series threshold and dynamically corrected by combining the meteorological change characteristics at the prediction time with historical rolling errors. The calculation formula is as follows:

[0072] in, For time-series thresholds, Based on time-series thresholds, This is the fifth weighting coefficient, calibrated based on historical operating data of the wind farm; To predict the change in wind speed between the current time and the previous time, It is the sixth weighting coefficient. To predict the change in wind direction between the current time and the previous time, It is the seventh weighting coefficient. This represents the average rolling forecast error over the past 24 hours. For example... It can be 0.1. You can take 0.15. It can be taken as 0.0000005.

[0073] S208. The processing equipment calculates the timing stability score based on the corrected power prediction value of the wind turbine and the standardized time series dataset.

[0074] The temporal stability score is used to quantify the degree of continuous change in the predicted power series and identify the risk of error accumulation in rolling forecasts. The calculation formula is as follows:

[0075] in, For the predicted time The time series stability score, The length of the rolling forecast window ranges from 4 to 16, corresponding to forecast data from the past 1 to 4 hours. For the first l Corrected power prediction values ​​for each historical prediction time.

[0076] The beneficial effect of this step is that although wind power fluctuates, its changes are continuous over a short period and do not involve abrupt changes. During rolling forecasting, errors accumulate gradually with each forecast step, causing discontinuous jumps in the predicted power sequence. The time-series stability score measures the degree of continuous change in the forecast sequence by calculating the relative difference between the current corrected power forecast and historical corrected power forecasts. Using a time-inverse weighting method, more recent historical forecasts have a greater impact on the score, enabling more sensitive detection of recent error accumulation. The time-series stability score can identify forecasts with a risk of error accumulation, triggering targeted error compensation operations.

[0077] S209. The processing device determines whether the timing stability score is less than the timing threshold and obtains the second judgment result.

[0078] If the second judgment result indicates that the time series stability score is less than the time series threshold, then execute S210; If the second judgment result indicates that the time stability score is greater than or equal to the time threshold, then execute S213.

[0079] The beneficial effect of this step is that when wind speed or direction changes significantly, the output power of wind turbines will exhibit normal large fluctuations, which are not due to error accumulation. Using a fixed threshold cannot distinguish between normal power fluctuations and abnormal fluctuations caused by error accumulation. Dynamic adaptive time-series thresholds, by introducing parameters such as wind speed change rate, wind direction change rate, and historical rolling error, can adjust the threshold size in real time based on meteorological characteristics and the model's historical performance. Appropriately lowering the threshold when wind speed or direction changes are large, and appropriately raising the threshold when meteorological conditions are stable or historical rolling error is small, can effectively distinguish between normal power fluctuations and abnormal error accumulation, improving the accuracy of error compensation triggering.

[0080] S210 The processing equipment corrects the modified power prediction value to obtain the target power prediction value of the wind turbine.

[0081] The historical error rolling compensation method is used to correct the revised power prediction value. The correction formula is as follows:

[0082] in, For the predicted time The target power prediction value, The total number of error compensation samples, ranging from 12 to 48, corresponds to historical error data from the past 3 to 12 hours. The compensation weight for the j-th historical error, Let $\frac{j}{j}$ be the prediction error for the $j$-th historical moment. Let J be the wind speed value at the j-th historical moment. To predict the deviation between the wind direction at a given time and the cabin yaw angle at the current time, The deviation between the measured wind direction and the measured cabin yaw angle at the j-th historical moment.

[0083] The beneficial effect of this step is that, during the rolling forecasting process, the model's prediction error has a certain continuity, and recent error characteristics can reflect current error characteristics to some extent. The historical error rolling compensation method effectively suppresses error accumulation by using historical prediction errors to compensate for the current corrected power prediction value. By multiplying the historical error by the ratio of the cube of the current wind speed to the cube of the historical wind speed, the historical error can be converted into an equivalent error under the current wind speed. Multiplying by the cosine of the wind direction deviation difference can account for the impact of wind direction deviation on the error. Using error attenuation weights ensures that more recent historical errors have a greater impact on compensation, more accurately reflecting current error characteristics. Triggering compensation only when time-series stability is insufficient avoids unnecessary compensation introducing new errors, improving the stability and accuracy of long-term forecasts.

[0084] S211. The processing equipment obtains the total power prediction value of the wind farm based on the target power prediction value of the wind turbine.

[0085] The target power predictions of all wind turbines in the wind farm are aggregated and calculated to obtain the total power prediction of the wind farm.

[0086] The specific process is as follows: The processing equipment iterates through all wind turbines in the wind farm, obtains the target power prediction value for each wind turbine at the prediction time t+τ, and sums up the target power prediction values ​​of all wind turbines to obtain the total power prediction value of the wind farm at the prediction time t+τ. The calculation formula is as follows:

[0087] Where K represents the total number of wind turbines in the wind farm. For the k-th wind turbine at the predicted time The target power prediction value, This represents the predicted total power of the wind farm at the predicted time t+τ.

[0088] The advantage of this step is that by using a method of predicting and then aggregating wind turbines to calculate the total power of the wind farm, the individual operating characteristics of each wind turbine can be fully considered, avoiding errors caused by ignoring the differences in operating conditions between turbines when directly predicting the total power. The calculation method of summing turbines one by one is simple and efficient, and can meet the real-time requirements of ultra-short-term power prediction.

[0089] S212. The processing equipment determines the initial power prediction value of the wind turbine as the corrected power prediction value of the wind turbine.

[0090] After the processing equipment determines the initial power prediction value of the wind turbine as the corrected power prediction value of the wind turbine, it executes S207.

[0091] S213. The processing equipment determines the corrected power prediction value of the wind turbine as the target power prediction value of the wind turbine.

[0092] After the processing equipment determines the corrected power prediction value of the wind turbine as the target power prediction value of the wind turbine, it executes S211.

[0093] In some possible implementations, the measured total power of the wind farm is obtained; based on the measured total power of the wind farm and the predicted total power of the wind farm, the relevant parameters used to calculate the initial power prediction, physical rationality score and time series stability score are updated.

[0094] Specifically, the processing equipment obtains the measured total power of the wind farm at a historical time t1, compares it with the predicted total power value at the corresponding time, and calculates the difference between the two as the total power prediction error.

[0095] The parameter update cycle is set to 24 hours, meaning that a parameter update operation is performed every 24 hours. The update cycles listed in the embodiments of this application are merely illustrative and are not intended to limit the scope of protection of this application. Those skilled in the art can reasonably adjust the above values ​​according to actual application scenarios, and all equivalent substitutions or adaptive modifications should fall within the scope of protection of this application.

[0096] The parameters that need to be updated include the time decay coefficient, error decay coefficient, first to fourth weighting coefficients, and fifth to seventh weighting coefficients. During the update, the parameter values ​​before the update are weighted and averaged with the optimal parameter values ​​calculated based on the total power prediction error over the past 24 hours to obtain the updated parameter values. The weights of the parameter values ​​before the update are the moving average coefficients, and the weight of the optimal parameter values ​​is 1 minus the moving average coefficient. The moving average coefficient ranges from 0.8 to 0.9, for example, 0.85. All parameters requiring updating are updated using the same weighted averaging method.

[0097] The beneficial effect of this step is that, through error feedback between the measured total power and the predicted total power, the relevant parameters of the model are updated regularly, enabling the model to adapt to long-term changes in the operating characteristics of wind farms, such as turbine performance aging, changes in surrounding terrain, and seasonal climate differences. The moving average method for updating parameters can smooth the process of parameter changes, avoid abrupt parameter changes that could lead to large fluctuations in the prediction results, and improve the long-term stability and adaptability of the prediction method.

[0098] The above text combined Figure 1 The adaptive threshold wind farm ultra-short-term power prediction method provided in the embodiments of this application has been described in detail. The apparatus and equipment provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0099] This application also provides an adaptive threshold wind farm ultra-short-term power prediction device, such as... Figure 2 As shown in the figure, this is a schematic diagram of an adaptive threshold wind farm ultra-short-term power prediction device provided in an embodiment of this application. The device includes: The acquisition module 301 is used to acquire numerical weather forecast data and measured data of wind farms; and to obtain a standardized time series dataset based on the numerical weather forecast data and measured data of wind farms. The first calculation module 302 is used to obtain the initial power prediction value of each wind turbine in the wind farm based on the standardized time series dataset; The first judgment module 303 is used to calculate a physical rationality score based on the initial power prediction value of the wind turbine and the standardized time series dataset; determine whether the physical rationality score is less than a physical threshold to obtain a first judgment result; if the first judgment result indicates that the physical rationality score is less than the physical threshold, then the initial power prediction value is corrected to obtain the corrected power prediction value of the wind turbine. The second judgment module 304 is used to calculate a time series stability score based on the corrected power prediction value of the wind turbine and the standardized time series dataset; determine whether the time series stability score is less than a time series threshold to obtain a second judgment result; if the second judgment result indicates that the time series stability score is less than the time series threshold, then the corrected power prediction value is corrected to obtain the target power prediction value of the wind turbine. The second calculation module 305 is used to obtain the total power prediction value of the wind farm based on the target power prediction value of the wind turbine.

[0100] In some possible implementations, the first judgment module 303 is further configured to determine the initial power prediction value of the wind turbine as the corrected power prediction value of the wind turbine if the first judgment result indicates that the physical rationality score is greater than or equal to the physical threshold.

[0101] In some possible implementations, the second judgment module 304 is further configured to determine the corrected power prediction value of the wind turbine as the target power prediction value of the wind turbine if the second judgment result indicates that the timing stability score is greater than or equal to the timing threshold.

[0102] In some possible implementations, the acquisition module 301 is specifically used to perform spatiotemporal alignment processing on the numerical weather forecast data and the measured data of the wind farm to obtain an aligned dataset; remove outliers from the aligned dataset; perform missing value completion processing on the aligned dataset after removing outliers to obtain a complete dataset; perform denoising processing on the complete dataset; and perform standardization processing on the denoised complete dataset to obtain a standardized time-series dataset.

[0103] In some possible implementations, the first calculation module 302 is specifically used to obtain the aerodynamic characteristic parameters of the wind turbines in the wind farm; based on the standardized time-series dataset and the aerodynamic characteristic parameters of the wind turbines in the wind farm, combined with operational constraints and physical mechanism constraints, to calculate the initial power prediction value of each wind turbine in the wind farm.

[0104] In some possible implementations, the second calculation module 305 is specifically used to aggregate the target power prediction values ​​of all wind turbine units in the wind farm to obtain the total power prediction value of the wind farm.

[0105] In some possible implementations, the device further includes: The update module is used to obtain the measured total power of the wind farm; based on the measured total power of the wind farm and the predicted total power of the wind farm, the relevant parameters used to calculate the initial power prediction value, physical rationality score and time series stability score are updated.

[0106] The adaptive threshold wind farm ultra-short-term power prediction device according to the embodiments of this application can correspond to the execution of the method described in the embodiments of this application, and the other operations and / or functions of each module / unit of the adaptive threshold wind farm ultra-short-term power prediction device are respectively for implementing Figure 1 For the sake of brevity, the corresponding processes of each method in the illustrated embodiments will not be described in detail here.

[0107] This application also provides a computing device. For example... Figure 3 As shown in the figure, this is a schematic diagram of a computing device provided in an embodiment of this application. The computing device 400 includes a bus 401, a processor 402, a communication interface 403, and a memory 404. The processor 402, the memory 404, and the communication interface 403 communicate with each other via the bus 401.

[0108] Bus 401 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0109] Processor 402 can be any one or more of the following processors: central processing unit (CPU), graphics processing unit (GPU), microprocessor (MP), or digital signal processor (DSP).

[0110] The communication interface 403 is used for communication with external devices. For example, if the computing device is a first switch, the communication interface 403 can be used for communication between the first switch and a first user terminal, or for communication between the first switch and a second switch.

[0111] Memory 404 may include volatile memory, such as random access memory (RAM). Memory 404 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0112] The memory 404 stores executable code, which the processor 402 executes to perform the aforementioned adaptive threshold wind farm ultra-short-term power prediction method.

[0113] Specifically, in achieving Figure 2 In the case of the illustrated embodiment, and Figure 2 When the modules or units of the adaptive threshold wind farm ultra-short-term power prediction device described in the embodiments are implemented in software, the following steps are performed: Figure 2 The software or program code required for the functions of each module / unit can be partially or wholly stored in memory 404. Processor 402 executes the program code corresponding to each unit stored in memory 404 and executes the aforementioned adaptive threshold wind farm ultra-short-term power prediction method.

[0114] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium capable of being stored by a computing device, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct a computing device to execute the aforementioned adaptive threshold wind farm ultra-short-term power prediction method.

[0115] This application also provides a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application are generated.

[0116] The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another website, computer, or data center via wired (e.g., coaxial cable, fiber optic) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0117] When the computer program product is executed by a computer, the computer performs any of the aforementioned adaptive threshold wind farm ultra-short-term power prediction methods. The computer program product can be a software installation package; when any of the aforementioned adaptive threshold wind farm ultra-short-term power prediction methods needs to be used, the computer program product can be downloaded and executed on the computer.

[0118] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0119] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A method for ultra-short-term power prediction of wind farms with adaptive thresholds, characterized in that, The method includes: Acquire numerical weather forecast data and measured data of wind farms; A standardized time-series dataset was obtained based on numerical weather forecast data and measured data from wind farms. Based on the standardized time-series dataset, the initial power prediction value of each wind turbine in the wind farm is obtained; Based on the initial power prediction values ​​of wind turbine units and standardized time-series datasets, calculate the physical rationality score; Determine whether the physical rationality score is less than the physical threshold to obtain the first judgment result; If the first judgment result indicates that the physical rationality score is less than the physical threshold, then the initial power prediction value is corrected to obtain the corrected power prediction value of the wind turbine. Based on the corrected power prediction values ​​of wind turbine units and the standardized time series dataset, the time series stability score is calculated. Determine whether the time series stability score is less than the time series threshold to obtain a second determination result; If the second judgment result indicates that the time series stability score is less than the time series threshold, then the corrected power prediction value is corrected to obtain the target power prediction value of the wind turbine. The total power prediction value of the wind farm is obtained based on the target power prediction value of the wind turbine.

2. The method according to claim 1, characterized in that, The method further includes: If the first judgment result indicates that the physical rationality score is greater than or equal to the physical threshold, then the initial power prediction value of the wind turbine is determined as the corrected power prediction value of the wind turbine.

3. The method according to claim 1, characterized in that, The method further includes: If the second judgment result indicates that the time series stability score is greater than or equal to the time series threshold, then the corrected power prediction value of the wind turbine is determined as the target power prediction value of the wind turbine.

4. The method according to claim 1, characterized in that, The standardized time-series dataset obtained from wind farm numerical weather prediction data and wind farm measured data includes: Spatiotemporal alignment processing is performed on numerical weather prediction data and measured data of wind farms to obtain an aligned dataset. Outliers in the aligned dataset are removed, and missing values ​​are filled in the aligned dataset after outlier removal to obtain the complete dataset. The complete dataset is denoised, and the denoised complete dataset is then standardized to obtain a standardized time series dataset.

5. The method according to claim 1, characterized in that, The initial power prediction value for each wind turbine in the wind farm, obtained based on the standardized time-series dataset, includes: Obtain the aerodynamic characteristic parameters of wind turbines in a wind farm; Based on standardized time-series datasets and aerodynamic characteristic parameters of wind turbines in wind farms, combined with operational constraints and physical mechanism constraints, the initial power prediction value of each wind turbine in the wind farm is calculated.

6. The method according to claim 1, characterized in that, The total power prediction value of the wind farm is obtained from the target power prediction value of the wind turbine, including: The target power predictions of all wind turbines in the wind farm are aggregated and calculated to obtain the total power prediction of the wind farm.

7. The method according to claim 1, characterized in that, The method further includes: Obtain the measured total power of the wind farm; Based on the measured total power of the wind farm and the predicted total power of the wind farm, the relevant parameters used to calculate the initial power prediction, physical rationality score, and time series stability score are updated.

8. A wind farm ultra-short-term power prediction device with adaptive threshold, characterized in that, The device includes: The acquisition module is used to acquire numerical weather forecast data and measured data of wind farms; and to obtain a standardized time-series dataset based on the numerical weather forecast data and measured data of wind farms. The first calculation module is used to obtain the initial power prediction value of each wind turbine in the wind farm based on the standardized time series dataset; The first judgment module is used to calculate a physical rationality score based on the initial power prediction value of the wind turbine and the standardized time series dataset; determine whether the physical rationality score is less than a physical threshold to obtain a first judgment result; if the first judgment result indicates that the physical rationality score is less than the physical threshold, then the initial power prediction value is corrected to obtain the corrected power prediction value of the wind turbine. The second judgment module is used to calculate the time series stability score based on the corrected power prediction value of the wind turbine and the standardized time series dataset; determine whether the time series stability score is less than the time series threshold, and obtain a second judgment result; if the second judgment result indicates that the time series stability score is less than the time series threshold, then the corrected power prediction value is corrected to obtain the target power prediction value of the wind turbine. The second calculation module is used to obtain the total power prediction value of the wind farm based on the target power prediction value of the wind turbine.

9. A computing device, characterized in that, Including memory and processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the method as described in any one of claims 1 to 7.

Images