Cold wave prediction method and device based on fan shutdown record, equipment and medium

Abstract

This invention belongs to the field of wind power generation forecasting, and discloses a method, device, equipment, and medium for cold wave forecasting based on wind turbine shutdown records. The method includes: acquiring numerical weather forecast characteristics of the area to be predicted for future periods; inputting the numerical weather forecast characteristics of the future periods into a pre-trained spatiotemporal prediction model to obtain preliminary prediction results of gridded cold wave labels for the area to be predicted for future periods; and smoothing and correcting the preliminary prediction results of the gridded cold wave labels for the future periods to obtain gridded cold wave prediction labels for the area to be predicted for future periods. This invention constructs a regional-scale cold wave impact label system based on historical wind turbine shutdown data, establishes a mapping relationship between meteorological elements and the state of cold wave impact, and breaks through the limitations of traditional static meteorological threshold definitions for cold waves. It can provide quantifiable evidence of cold wave trend evolution for wind power forecasting tasks and guide the model to switch modes when a cold wave event occurs.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation prediction, and specifically relates to a method, device, equipment and medium for predicting cold waves based on wind turbine shutdown records. Background Technology

[0002] With the accelerated construction of new power systems, the installed capacity and grid connection ratio of new energy power generation such as wind power and photovoltaics continue to rise. However, the large-scale integration of high-proportion renewable energy sources has significantly increased the sensitivity of the power system to meteorological conditions, and the coupling degree between system operation stability and meteorological factors is deepening. In recent years, against the backdrop of global warming, extreme weather events have occurred frequently, especially sudden and widespread extreme weather processes such as cold waves, which can easily trigger problems such as a sharp drop in renewable energy output and wind turbine icing, posing a serious threat to the safe and stable operation of the power system.

[0003] Currently, early warnings for cold waves rely heavily on traditional weather forecasts, which are significantly out of sync with the actual operating conditions of wind farms. Existing meteorological standards mainly define cold wave events based on the magnitude of temperature drop and the minimum temperature threshold. However, this definition fails to establish an explicit correlation with the drastic power fluctuations and abnormal shutdown patterns of wind farms. The start and end times of cold waves as defined by meteorology often deviate significantly from the actual periods of sudden power drops and large-scale unit shutdowns at wind farms. This makes it difficult for prediction models built based on meteorological cold wave labels to accurately capture the true response characteristics of wind farms.

[0004] Significant differences exist in topography, spatial distribution of meteorological elements, and layout of new energy units between different regions. The massive amount of data, high-dimensional spatiotemporal characteristics, complex spatial dependencies, and dynamic evolutionary relationships will also significantly increase the learning cost and training difficulty of traditional sequence models.

[0005] Current cold wave event assessments are mostly based on static threshold rules for meteorological elements such as temperature and temperature drop, focusing on a meteorological perspective. This leads to a structural deviation between cold wave event predictions and actual wind power fluctuations, making it difficult to accurately depict the actual impact of cold waves on wind power output. Summary of the Invention

[0006] The purpose of this invention is to provide a method, device, equipment, and medium for cold wave prediction based on wind turbine shutdown records, to solve the technical problem that existing cold wave event determination methods struggle to accurately characterize the actual impact of cold waves on wind power output. This invention constructs a regional-scale cold wave impact labeling system based on historical wind turbine shutdown data, establishing a mapping relationship between meteorological elements and cold wave impact states, achieving reliable identification of future cold wave impact scenarios across all grid points. This method overcomes the limitations of traditional static meteorological threshold definitions for cold waves, deeply coupling cold wave event characterization with the actual operating state response of wind farms. It can provide quantifiable evidence of cold wave trend evolution for wind power forecasting (WPF) tasks, guiding models to switch modes when cold wave events occur.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a cold wave prediction method based on wind turbine shutdown records, comprising: Obtain numerical weather forecast characteristics of the area to be predicted for future periods; The numerical weather forecast features of the future time period are input into the pre-trained spatiotemporal prediction model to obtain the preliminary prediction results of the gridded cold wave labels of the area to be predicted in the future time period. The preliminary prediction results of the gridded cold wave labels for the future time period are smoothed and corrected to obtain the gridded cold wave prediction labels for the area to be predicted for the future time period. The pre-trained spatiotemporal prediction model is obtained by training based on the historical gridded cold wave labels of the area to be predicted and the historical gridded numerical weather forecast features of the corresponding time period; the historical gridded cold wave labels are determined based on the ratio of the station's shutdown capacity to its installed capacity.

[0008] A further improvement of the present invention is that the historical gridded cold wave label is obtained by: dividing the area to be predicted into grids to obtain multiple grids; determining the cold wave impact level label corresponding to each grid based on the shutdown records of the stations in the area to be predicted, thereby obtaining the historical gridded cold wave label.

[0009] A further improvement of the present invention is as follows: based on the shutdown records of the stations in the area to be predicted, the cold wave impact level label corresponding to each grid is determined, including: based on the location of each station and the preset spatial impact radius, the spatial impact area corresponding to each station is determined, and the grids in the spatial impact area are used as the impact grids of the corresponding stations; based on the ratio of shutdown capacity to installed capacity in the spatial impact area corresponding to each station, the cold wave impact level label corresponding to the spatial impact area is determined, and the cold wave impact level label is assigned to the grids in the spatial impact area.

[0010] A further improvement of the present invention is that: the cold wave impact level label includes multiple labels, and the levels of the multiple cold wave impact level labels are set sequentially from low to high according to the ratio of shutdown capacity to installed capacity.

[0011] A further improvement of the present invention is that: both the historical gridded numerical weather forecast features and the numerical weather forecast features for future periods include at least one meteorological element among wind speed component, temperature, dew point temperature and precipitation.

[0012] A further improvement of the present invention is that the pre-trained spatiotemporal prediction model is a convolutional long short-term memory network model; the convolutional long short-term memory network model includes an input gate, a forget gate, candidate unit states, cell states, an output gate, and hidden states.

[0013] A further improvement of the present invention is that: the preliminary prediction results of the gridded cold wave labels for the future time period are smoothed and corrected, including: the preliminary prediction results of the gridded cold wave labels for the future time period are corrected in the time dimension so that the change in cold wave labels of the same grid between adjacent prediction time steps does not exceed a preset time smoothing threshold; the gridded cold wave labels after time dimension correction are corrected in the spatial dimension so that the difference in cold wave labels between adjacent grids within the same prediction time step does not exceed a preset spatial smoothing threshold.

[0014] A further improvement of the present invention is that the spatial dimension correction is performed within the spatial influence area corresponding to the station, so as to constrain the cold wave labels of adjacent grids within the spatial influence area to maintain local spatial consistency.

[0015] A further improvement of the present invention is as follows: after obtaining the gridded cold wave prediction labels for the future period of the area to be predicted, the cold wave impact state of the power station is determined according to the cold wave prediction label corresponding to the grid where the power station is located; the operating mode of the wind power prediction model is determined according to the cold wave impact state of the power station; wherein, when the power station is in a non-cold wave impact state, the constant mode is used for wind power prediction, and when the power station is in a cold wave impact state, the cold wave mode is used for wind power prediction.

[0016] Secondly, the present invention provides a cold wave prediction device based on wind turbine shutdown records, comprising: The data acquisition module is used to obtain numerical weather forecast characteristics of the area to be predicted for future periods. The prediction module is used to input the numerical weather forecast features of the future time period into a pre-trained spatiotemporal prediction model to obtain preliminary prediction results of gridded cold wave labels for the area to be predicted in the future time period. The correction module is used to smooth and correct the preliminary prediction results of the gridded cold wave labels for the future time period, so as to obtain the gridded cold wave prediction labels for the area to be predicted for the future time period. The pre-trained spatiotemporal prediction model is obtained by training based on the historical gridded cold wave labels of the area to be predicted and the historical gridded numerical weather forecast features of the corresponding time period; the historical gridded cold wave labels are determined based on the ratio of the station's shutdown capacity to its installed capacity.

[0017] Thirdly, the present invention provides an electronic device including a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the cold wave prediction method based on wind turbine shutdown records as described above.

[0018] Fourthly, the present invention provides a computer-readable storage medium storing at least one instruction, which, when executed by a processor, implements the cold wave prediction method based on wind turbine shutdown records.

[0019] Compared with the prior art, the present invention has the following beneficial effects: Beneficial effects corresponding to claim 1: This invention establishes a mapping relationship between numerical weather forecast features and cold wave impact labels by inputting future numerical weather forecast features into a spatiotemporal prediction model trained on historical gridded cold wave labels and historical gridded numerical weather forecast features. This enables the prediction of gridded cold wave labels for the area to be predicted in the future. Simultaneously, by smoothing and correcting the preliminary prediction results, abnormal fluctuations in gridded cold wave labels in time or space can be suppressed, improving the stability and reliability of cold wave prediction results. Since historical gridded cold wave labels are determined based on the ratio of station downtime capacity to installed capacity, the prediction results can reflect the actual impact of cold waves on station operation, avoiding prediction bias caused by relying solely on meteorological thresholds.

[0020] Furthermore, this invention divides the area to be predicted into grids and determines the cold wave impact level label corresponding to each grid based on the downtime records of the stations within the area. This allows the station-level downtime records to be converted into cold wave impact labels at the regional grid scale, enabling the degree of cold wave impact to be expressed on a spatial grid. This provides a supervisory label with spatial distribution characteristics for the training of subsequent spatiotemporal prediction models.

[0021] Furthermore, this invention determines the corresponding spatial influence area based on the location of each station and the preset spatial influence radius, and uses the grid within the spatial influence area as the influence grid, which can extend the cold wave impact reflected by the station shutdown records to its surrounding areas; at the same time, the cold wave impact level label is determined by the ratio of shutdown capacity to installed capacity within the spatial influence area, and the grid within the area is assigned, which enables the gridded cold wave label to correspond to the actual shutdown impact of the station, improving the rationality and feasibility of cold wave label construction.

[0022] Furthermore, by setting multiple cold wave impact level labels and sequentially setting the levels according to the ratio of shutdown capacity to installed capacity from low to high, the present invention can classify and characterize the impact of cold waves of different degrees, so that the prediction results can not only reflect whether the cold wave has affected the region, but also reflect the differences in the intensity of the cold wave impact, thereby improving the precision of the cold wave prediction results.

[0023] Furthermore, this invention corrects the preliminary prediction results of the gridded cold wave labels in the time dimension, ensuring that the change in cold wave labels for the same grid between adjacent prediction time steps does not exceed a preset time smoothing threshold, thereby reducing abnormal jumps in the cold wave labels in the time series. By correcting the gridded cold wave labels in the time dimension in the spatial dimension, the difference in cold wave labels between adjacent grids within the same prediction time step does not exceed a preset spatial smoothing threshold, thereby reducing isolated abnormal grids in space and improving the continuity and consistency of cold wave prediction labels in both time and space.

[0024] Furthermore, by correcting the spatial dimension within the spatial influence area corresponding to the station and constraining the cold wave labels of adjacent grids within the spatial influence area to maintain local spatial consistency, the present invention can match the spatial correction process with the actual influence range corresponding to the station shutdown record, reduce the problem of discontinuous or abrupt distribution of cold wave labels within the same station influence range, and improve the spatial rationality of gridded cold wave prediction labels.

[0025] Furthermore, this invention, after obtaining the gridded cold wave prediction labels for the future time period of the area to be predicted, determines the cold wave impact status of the wind farm based on the cold wave prediction label corresponding to the grid where the wind farm is located. This enables the conversion of the regional gridded cold wave prediction results into operational status judgments for specific wind farms. Furthermore, based on the cold wave impact status of the wind farms, the operating mode of the wind power prediction model is determined, and the constant mode and cold wave mode are used for wind power prediction under non-cold wave impact and cold wave impact conditions, respectively. This allows the wind power prediction model to adaptively switch according to the cold wave impact situation, improving the targeting of wind power prediction under cold wave conditions. Attached Figure Description

[0026] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating a cold wave prediction method based on wind turbine shutdown records according to an embodiment of the present invention. Figure 2 This invention describes the wind farm clustering results and their impact range. Figure 3This is a comparison chart showing the accuracy of the cold wave tag of this invention in detecting downtime events; Figure 4 WPF curves for using different cold wave labels during cold waves according to the present invention; Figure 5 This is a correlation diagram showing the characteristics of the entire time period and the cold wave period in this invention; where (a) represents the entire time period and (b) represents the cold wave period. Figure 6 This is a flowchart illustrating another embodiment of the present invention: a cold wave prediction method based on wind turbine shutdown records. Figure 7 This is a schematic diagram of a cold wave prediction device based on wind turbine shutdown records according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0027] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0028] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0029] Convolutional Long Short-Term Memory (ConvLSTM) is an important extension of recurrent neural networks. By introducing convolutional operations into the gating mechanism of traditional LSTM, it constructs a spatiotemporal neural network capable of simultaneously handling temporal memory and spatial structure. This model preserves both the input and hidden states as three-dimensional tensors, using convolutional kernels to extract spatial local features of raster data layer by layer and fusing historical temporal information. It uncovers the deep coupling relationship between spatial neighborhoods and historical moments, achieving efficient analysis and future state prediction of high-dimensional raster field sequences.

[0030] Please see Figure 1 As shown, this embodiment of the invention provides a cold wave prediction method based on wind turbine shutdown records, including the following steps: (1) Construction of historical tags for multi-level cold waves The area to be predicted is divided into several grids based on latitude and longitude; each wind farm site (with coordinates as...) i s , js With )) as the center, and r s Construct a circular influence area for the spatial influence radius. R s Calculate the circular influence area R s The ratio of in-station shutdown capacity to installed capacity α c It is discretized into multiple level labels. L s ∈{0, 1, 2, ..., K} represents the intensity of cold wave impact at different degrees; circular impact area R s All grids use this level label. The intensity of cold wave impacts is classified into four different levels according to the "Standard for Defining Major Weather Processes Affecting Power Supply and Demand Balance," with the following classification criteria: a) Level 4: The shutdown capacity accounts for 10% or more of the total operating capacity, but is less than 30%.

[0031] b) Level 3: The shutdown capacity accounts for 30% or more of the total operating capacity, but is less than 50%.

[0032] c) Level 2: The shutdown capacity accounts for 50% or more of the total operating capacity, but is less than 70%.

[0033] d) Level 1: The shutdown capacity accounts for 70% or more of the total operating capacity.

[0034] Since the cold wave shutdown capacity data used is station-level data, if a station is affected by a cold wave, it indicates that the cold wave has advanced into the surrounding area. Constructing a circular impact area is to infer the degree of cold wave impact on each surrounding grid point from the station's shutdown capacity. The final result is a historical gridded cold wave label, in the format (historical time step × latitude × longitude). T his × H × W ).

[0035] (2) Identification of future cold wave labels based on spatiotemporal prediction models Cold wave events are often accompanied by a gradual expansion of spatial extent and a continuous evolution of temperature gradients. For raster data describing the cold wave trend at various grid points within a region, conventional time-series models struggle to adequately model its spatial structure and correlations. This invention employs a spatiotemporal prediction model, represented by ConvLSTM, which preserves the spatial topology through convolution operators and explicitly learns temporal memory using a gated recurrent mechanism. This allows the modeling of the cold wave's progression, duration, and warming process as a state space with memory properties, enabling the processing of raster features from multiple time points. xt ∈R C×H×W To future raster label output L t ∈R H ×W The sequence mapping. The ConvLSTM structure used in this invention can be uniformly formalized as follows: Input gate i t :

[0036] in, i t It is the input gate at time t. x t W is the gridded feature of the input model at time t. xi It is in the input gate x t The convolution weight matrix, H t-1 It is the hidden state at time t-1, W hi It is in the input gate H t-1 The convolution weight matrix, b i It is the bias term vector of the input gate; σ It is the sigmoid function.

[0037] Forgotten Gate f t :

[0038] in, f t It is the forgetting gate at time t. x t W is the gridded feature of the input model at time t. xf It is within the Gate of Oblivion x t The convolution weight matrix, H t-1 It is the hidden state at time t-1, W hf It is within the Gate of Oblivion H t-1 The convolution weight matrix, b f It is the bias term vector of the forget gate.

[0039] Candidate cell state g t :

[0040] in, g t It represents the candidate cell state at time t.x t W is the gridded feature of the input model at time t. xc It is in this cell state x t The convolution weight matrix, H t-1 It is the hidden state at time t-1, W hc It is in this cell state H t-1 The convolution weight matrix, b c It is the bias term vector of the cell state.

[0041] Cell state C t :

[0042] in, C t It represents the cell state at time t. f t It is the forgetting gate at time t. C t It represents the cell state at time t-1. i t It is the input gate at time t. g t It is the state of the candidate unit at time t.

[0043] Output gate o t :

[0044] in, o t It is the output gate at time t. x t W is the gridded feature of the input model at time t. xo It is the output gate x t The convolution weight matrix, H t-1 It is the hidden state at time t-1, W ho It is the output gate H t-1 The convolution weight matrix, b o It is the bias term vector of the output gate.

[0045] Hidden state H t :

[0046] in, H tIt is the hidden state at time t. o t It is the output gate at time t. C t It represents the cell state at time t.

[0047] Compared to static discrimination paradigms that rely on meteorological thresholds, the end-to-end mapping mechanism based on a spatiotemporal prediction model overcomes the limitation of cold wave definitions that only focus on meteorological elements themselves. It embeds the actual shutdown response of wind farms into the early warning identification process, achieving a shift from meteorological early warning to impact early warning. By capturing the dynamic correlation between the spatiotemporal propagation trajectory of cold waves and the operating status of wind farms, this invention's spatiotemporal prediction model can identify the affected areas and intensity levels of cold waves with actual destructive power in advance. This multi-level identification method oriented towards operational risks not only enhances the relevance of early warning results to the actual operating conditions of wind farms but also provides timely and physically consistent decision inputs for subsequent power prediction model strategy switching.

[0048] This invention utilizes the historical gridded cold wave labels obtained in step (1) and the historical gridded numerical weather forecast (NWP) features of the corresponding time period to train the spatiotemporal prediction model for multi-level cold wave trend identification constructed in step (2), establishes the mapping relationship between cold wave labels and numerical weather forecast features, and obtains the trained spatiotemporal prediction model.

[0049] Input the future weather forecast NWP features for the period to be predicted into the trained spatiotemporal prediction model to obtain preliminary prediction results of gridded cold wave labels for the future period, in the form of (prediction time step × latitude × longitude). T pre × H × W ).

[0050] (3) Smoothing correction mechanism Although spatiotemporal prediction models can effectively capture the spatiotemporal characteristics of cold wave evolution, their output may still contain anomalous abrupt changes, affecting the stability of subsequent predictions. Therefore, this invention addresses the issue of cold wave labels... L t ∈R H×W Smoothing is performed to suppress its drastic fluctuations in the time and space dimensions.

[0051] Time dimension correction: for any grid point ( i , j The label value change at adjacent time steps is constrained to not exceed a preset threshold. max Suppressing drastic fluctuations over time:

[0052] in, yes t Time grid points ( i , j Preliminary forecast results for the cold wave label on the device; Grid points at adjacent time steps ( i , j The cold wave forecast tag; max This is a preset threshold; if the difference in cold wave prediction labels between adjacent time steps exceeds the threshold... max If the change in label value between adjacent time steps exceeds a preset threshold, the time dimension correction mechanism will be triggered. max Grid points ( i , j The time dimension is corrected according to the constraints of formula (7) to obtain the preliminary prediction results of the cold wave label after time correction of grid points; Spatial Dimension Correction: Setting up a Station s Coordinates are ( i s , j s Its radiation range is defined as R s Within the radiation range of any station, the eigenvalues ​​of neighboring grid points are constrained to maintain local spatial consistency, avoiding isolated points.

[0053] in, R s For station s Spatial influence range r s For station s The radius of the spatial influence range, δ max Spatial smoothing threshold; i s , j s ) for station s coordinates, ( i , j () represents the coordinates of the grid points; It is a grid point at a certain moment ( i , j Preliminary prediction results of the cold wave label after time correction; These are the preliminary predictions of cold wave labels for adjacent grid points at the same time, after time correction. δ max This is the spatial smoothing threshold; if the difference in cold wave prediction labels between adjacent grid points exceeds the spatial smoothing threshold... δmax This triggers the spatial dimension correction mechanism: for grid points ( i , j The initial prediction results of the cold wave label are corrected in terms of spatial dimension according to the constraints of formulas (8) and (9) to obtain the final gridded cold wave prediction label for the future time period after temporal and spatial correction. Step (3) ultimately yields a gridded cold wave prediction label for the future time period, in the form of (prediction time step × latitude × longitude) T pre × H × W ).

[0054] This invention provides a cold wave early warning method based on wind turbine shutdown records, including the following steps: using the historical gridded cold wave labels and historical gridded numerical weather prediction (NWP) features obtained in step (1), the multi-level cold wave trend identification and early warning model constructed in step (2) is trained to establish the mapping relationship between cold wave labels and features. The future NWP features are input into the trained model, and after correction in step (3), the gridded cold wave label prediction results for the future period can be obtained. Based on the results, the influence range of the cold wave in the region, the degree of influence on the field stations, and the future evolution trend of the cold wave can be determined, thereby guiding the scheduling decision.

[0055] This invention provides a cold wave prediction method based on wind turbine shutdown records. After obtaining gridded cold wave prediction labels for future periods in the area to be predicted, the cold wave impact status of the wind turbine is determined according to the cold wave prediction label corresponding to the grid where the wind turbine is located. Based on the cold wave impact status of the wind turbine, the operating mode of the wind power prediction model is determined. Specifically, when the wind turbine is not affected by a cold wave, a normal operating mode is used for wind power prediction; when the wind turbine is affected by a cold wave, a cold wave mode is used for wind power prediction. Under normal circumstances, the normal operating mode uses an LSTM model for wind power prediction. When the cold wave prediction enters a cold wave event, the method switches to the cold wave mode and uses a Transformer model for wind power prediction.

[0056] The beneficial effects of the present invention will be illustrated below with specific examples: (1) Data Description Wind farm operation data: The time span is from November 2022 to December 2024, with a time resolution of 1 hour. The data includes the actual power generation, installed capacity, and outage capacity of each wind farm.

[0057] NWP data: sourced from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis dataset, with a spatial resolution of 0.25°×0.25° and a temporal resolution of 1 hour. Meteorological elements include: 10m / 100m wind speed U / V components, 2m temperature, dew point temperature, and total precipitation.

[0058] A sample was constructed based on the distribution of cold wave events: Sample 1: November 1, 2022 to January 31, 2023, used as the training set; Sample 2: November 1-31, 2023, as the validation set; Sample 3: November 1 to December 31, 2023, as the test set. Cold wave identification and early warning require the construction of regional historical labels. Considering the spatial characteristics of cold air movement from north to south, the timing and intensity of the impact of cold waves on wind farms in different geographical locations vary significantly. Therefore, it is necessary to cluster wind farm clusters. The clustering results and the impact range of the wind farms are as follows: Figure 2 As shown, the WPF task adopts a dual-modal combined model. Under normal circumstances, it is in constant mode, using an LSTM model. When the cold wave identification and early warning model determines that a cold wave event has occurred, it switches to the cold wave mode, using a Transformer model.

[0059] (2) Evaluation indicators The effectiveness of cold wave identification and early warning is assessed using weighted precision, weighted recall, weighted F1 score, and accuracy, calculated using the following formulas:

[0060] Where TP, TN, FP, and FN represent the number of true positives, true negatives, false positives, and false negatives, respectively. S i It is the first i Support for each category; n is the total number of samples; Precision is the accuracy rate, Recall is the recall rate, F1-Score is the F1 score, and Accuracy is the precision rate.

[0061] The WPF evaluation metrics selected are Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²). 2 The mathematical definitions of each indicator are as follows:

[0062] in, P i , P i ’ , These are the actual power, the predicted power, and the average predicted power, respectively. C cap N represents the total installed capacity. RMSE It is the normalized root mean square error, N MAE It is the normalized mean absolute error, R 2 It is the coefficient of determination.

[0063] (3) Hyperparameter range The hyperparameters in the model were optimized by traversing the grid search method, and the specific optimization range is shown in Table 1.

[0064] Table 1. Hyperparameter Optimization Range

[0065] (4) Case Analysis To verify the effectiveness of the proposed cold wave warning identification model, this invention compares the following: 1) the ability of the cold wave label output by the model and the cold wave label defined by meteorology to reflect downtime events; 2) the impact of the cold wave label as the basis for switching wind power forecasting modes on the WPF task; and 3) the correlation between the cold wave impact label and the actual wind power output.

[0066] 1) Ability to respond to downtime events The meteorological cold wave label was constructed based on NWP data, according to the cold wave warning standards issued by the China Meteorological Administration. Using actual shutdown events as the evaluation benchmark, the accuracy of the two cold wave labels in capturing shutdown events was compared. The results are shown in Table 2 and... Figure 3 As shown.

[0067] Table 2. Accuracy of Early Warning Identification for Different Cold Wave Labels

[0068] The cold wave labels obtained by the prediction method of this invention improve upon the meteorologically defined labels by 12.53%, 14.99%, 13.41%, and 13.49% in four indicators, respectively. The results indicate that the labels constructed based on cold wave wind turbine shutdown records can more accurately reflect the actual impact of cold waves on wind turbine operation. Some meteorologically significant cold waves did not lead to shutdowns (false alarms), while some shutdowns were not purely caused by temperature drops within meteorological thresholds (missed alarms).

[0069] 2) Impact on WPF tasks Two types of labels were used as input features for the WPF task, guiding the model to switch to the cold wave modality. Prediction performance was compared. Results are shown in Table 3. Figure 4 As shown.

[0070] Table 3. WPF Evaluation Indicators Using Different Cold Wave Labels

[0071] The prediction method of this invention is applicable to all time periods N RMSE Compared to the meteorological definition, it decreased by 3.14%, and by 3.88% during cold wave periods; R 2 In terms of accuracy, the overall accuracy improved by 0.2575 over all periods and by 0.3245 during cold wave periods. This indicates that accurate early warning information can effectively guide the model to adopt the correct prediction strategies during cold wave events. Figure 4 It can be seen that the model can better fit the actual power change process, especially at the end of the cold wave event, where the degree of fit between the predicted results and the actual values ​​is higher. Comparing the performance of the two methods at different times, the method proposed in this invention shows a more significant improvement during the cold wave, indicating that the method of this invention has a stronger adaptability to extreme cold wave conditions and can effectively alleviate the decline in prediction accuracy caused by the cold wave.

[0072] 3) Correlation with actual wind power output Correlation analyses were conducted on samples from the entire time period and samples from the cold wave period to examine the statistical relationship between the cold wave impact labels and main NWP characteristics and wind power. The results are as follows: Figure 5 As shown.

[0073] Analysis shows that during cold wave periods, the absolute value of the correlation coefficient between wind speed and wind power decreases significantly compared to the entire period, indicating that under extreme weather conditions, the stable mapping relationship between wind speed and power based on the power curve is disturbed, weakening the expressive power of traditional wind speed-dominated predictive features. In contrast, the absolute correlation coefficient between the cold wave impact label and power is the largest among all features, demonstrating stronger explanatory power. This result shows that the constructed cold wave label can effectively characterize the changes in the operating status of wind farms impacted by cold waves, providing supplementary information support when wind speed correlation degrades, and statistically validating the necessity of explicitly incorporating the impact of cold waves into the predictive model. Simultaneously, this also provides data support for the subsequent construction of dynamic graph structures, namely, strengthening spatial correlation modeling related to cold wave propagation during cold wave phases to compensate for the predictive performance loss caused by the decreased expressive power of single meteorological elements.

[0074] Please see Figure 6 As shown, this embodiment of the invention provides a cold wave prediction method based on wind turbine shutdown records, including: S1. Obtain numerical weather forecast characteristics of the area to be predicted for future periods; S2. Input the numerical weather forecast features of the future time period into the pre-trained spatiotemporal prediction model to obtain the preliminary prediction results of the gridded cold wave labels of the area to be predicted in the future time period. S3. Smooth and correct the preliminary prediction results of the gridded cold wave labels for the future time period to obtain the gridded cold wave prediction labels for the area to be predicted in the future time period. The pre-trained spatiotemporal prediction model is obtained by training based on the historical gridded cold wave labels of the area to be predicted and the historical gridded numerical weather forecast features of the corresponding time period; the historical gridded cold wave labels are determined based on the ratio of the station's shutdown capacity to its installed capacity.

[0075] In one specific implementation, the historical gridded cold wave label is obtained by dividing the area to be predicted into multiple grids; based on the shutdown records of the stations in the area to be predicted, the cold wave impact level label corresponding to each grid is determined, thus obtaining the historical gridded cold wave label.

[0076] In one specific implementation, based on the shutdown records of the stations within the area to be predicted, the cold wave impact level label corresponding to each grid is determined, including: determining the spatial impact area corresponding to each station based on the location of each station and a preset spatial impact radius, and using the grids within the spatial impact area as the impact grids of the corresponding stations; determining the cold wave impact level label corresponding to the spatial impact area based on the ratio of shutdown capacity to installed capacity within the spatial impact area corresponding to each station, and assigning the cold wave impact level label to the grids within the spatial impact area.

[0077] In one specific implementation, the cold wave impact level label includes multiple labels, and the levels of the multiple cold wave impact level labels are set sequentially from low to high according to the ratio of shutdown capacity to installed capacity.

[0078] In one specific embodiment, both the historical gridded numerical weather forecast features and the numerical weather forecast features for the future period include at least one meteorological element among wind speed component, temperature, dew point temperature, and precipitation.

[0079] In one specific implementation, the pre-trained spatiotemporal prediction model is a convolutional long short-term memory network model; the convolutional long short-term memory network model includes an input gate, a forget gate, candidate unit states, cell states, an output gate, and hidden states.

[0080] In one specific implementation, the preliminary prediction results of the gridded cold wave labels for the future time period are smoothed and corrected, including: correcting the preliminary prediction results of the gridded cold wave labels for the future time period in terms of time dimension, so that the change in cold wave labels of the same grid between adjacent prediction time steps does not exceed a preset time smoothing threshold; and correcting the gridded cold wave labels after time dimension correction in terms of spatial dimension, so that the difference in cold wave labels between adjacent grids within the same prediction time step does not exceed a preset spatial smoothing threshold.

[0081] In one specific implementation, the spatial dimension correction is performed within the spatial influence area corresponding to the station, so as to constrain the cold wave labels of adjacent grids within the spatial influence area to maintain local spatial consistency.

[0082] In one specific implementation, after obtaining the gridded cold wave prediction labels for the future time period of the area to be predicted, the cold wave impact status of the wind farm is determined according to the cold wave prediction label corresponding to the grid where the wind farm is located; the operating mode of the wind power prediction model is determined according to the cold wave impact status of the wind farm; wherein, when the wind farm is in a non-cold wave impact state, the constant mode is used for wind power prediction, and when the wind farm is in a cold wave impact state, the cold wave mode is used for wind power prediction.

[0083] Please see Figure 7 As shown, this embodiment of the invention provides a cold wave prediction device based on wind turbine shutdown records, comprising: The data acquisition module is used to obtain numerical weather forecast characteristics of the area to be predicted for future periods. The prediction module is used to input the numerical weather forecast features of the future time period into a pre-trained spatiotemporal prediction model to obtain preliminary prediction results of gridded cold wave labels for the area to be predicted in the future time period. The correction module is used to smooth and correct the preliminary prediction results of the gridded cold wave labels for the future time period, so as to obtain the gridded cold wave prediction labels for the area to be predicted for the future time period. The pre-trained spatiotemporal prediction model is obtained by training based on the historical gridded cold wave labels of the area to be predicted and the historical gridded numerical weather forecast features of the corresponding time period; the historical gridded cold wave labels are determined based on the ratio of the station's shutdown capacity to its installed capacity.

[0084] Please see Figure 8 As shown, this embodiment of the invention provides an electronic device 100 for implementing a cold wave prediction method based on wind turbine shutdown records; the electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0085] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the cold wave prediction method based on wind turbine shutdown records described in the embodiment by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0086] The at least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor. The processor 102 is the control center of the electronic device 100, connecting various parts of the electronic device 100 via various interfaces and lines.

[0087] The memory 101 in the electronic device 100 stores multiple instructions to implement a cold wave prediction method based on wind turbine shutdown records, and the processor 102 can execute the multiple instructions to achieve the following: Obtain numerical weather forecast characteristics of the area to be predicted for future periods; The numerical weather forecast features of the future time period are input into the pre-trained spatiotemporal prediction model to obtain the preliminary prediction results of the gridded cold wave labels of the area to be predicted in the future time period. The preliminary prediction results of the gridded cold wave labels for the future time period are smoothed and corrected to obtain the gridded cold wave prediction labels for the area to be predicted for the future time period. The pre-trained spatiotemporal prediction model is obtained by training based on the historical gridded cold wave labels of the area to be predicted and the historical gridded numerical weather forecast features of the corresponding time period; the historical gridded cold wave labels are determined based on the ratio of the station's shutdown capacity to its installed capacity.

[0088] If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0089] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A cold wave prediction method based on wind turbine shutdown records, characterized in that, include: Obtain numerical weather forecast characteristics of the area to be predicted for future periods; The numerical weather forecast features of the future time period are input into the pre-trained spatiotemporal prediction model to obtain the preliminary prediction results of the gridded cold wave labels of the area to be predicted in the future time period. The preliminary prediction results of the gridded cold wave labels for the future time period are smoothed and corrected to obtain the gridded cold wave prediction labels for the area to be predicted for the future time period. The pre-trained spatiotemporal prediction model is obtained by training based on the historical gridded cold wave labels of the area to be predicted and the historical gridded numerical weather forecast features of the corresponding time period; the historical gridded cold wave labels are determined based on the ratio of the station's shutdown capacity to its installed capacity.

2. The cold wave prediction method based on wind turbine shutdown records according to claim 1, characterized in that, The historical gridded cold wave labels are obtained by dividing the area to be predicted into multiple grids; based on the shutdown records of the stations in the area to be predicted, the cold wave impact level label corresponding to each grid is determined, thus obtaining the historical gridded cold wave labels.

3. The cold wave prediction method based on wind turbine shutdown records according to claim 2, characterized in that, Based on the shutdown records of the stations within the area to be predicted, the cold wave impact level label corresponding to each grid is determined, including: determining the spatial impact area corresponding to each station based on the location of each station and the preset spatial impact radius, and using the grids within the spatial impact area as the impact grids of the corresponding stations; determining the cold wave impact level label corresponding to the spatial impact area based on the ratio of shutdown capacity to installed capacity within the spatial impact area corresponding to each station, and assigning the cold wave impact level label to the grids within the spatial impact area.

4. The cold wave prediction method based on wind turbine shutdown records according to claim 2, characterized in that, The cold wave impact level label includes multiple labels, and the levels of the multiple cold wave impact level labels are set in ascending order according to the ratio of shutdown capacity to installed capacity.

5. The cold wave prediction method based on wind turbine shutdown records according to claim 1, characterized in that, Both the historical gridded numerical weather forecast features and the numerical weather forecast features for the future period include at least one meteorological element among wind speed component, temperature, dew point temperature, and precipitation.

6. The cold wave prediction method based on wind turbine shutdown records according to claim 1, characterized in that, The pre-trained spatiotemporal prediction model is a convolutional long short-term memory network model; the convolutional long short-term memory network model includes an input gate, a forget gate, candidate unit states, cell states, an output gate, and hidden states.

7. The cold wave prediction method based on wind turbine shutdown records according to claim 1, characterized in that, The preliminary prediction results of the gridded cold wave labels for the future time period are smoothed and corrected, including: correcting the preliminary prediction results of the gridded cold wave labels for the future time period in terms of time dimension, so that the change in cold wave labels of the same grid between adjacent prediction time steps does not exceed a preset time smoothing threshold; and correcting the gridded cold wave labels after time dimension correction in terms of spatial dimension, so that the difference in cold wave labels between adjacent grids within the same prediction time step does not exceed a preset spatial smoothing threshold.

8. The cold wave prediction method based on wind turbine shutdown records according to claim 7, characterized in that, The spatial dimension correction is performed within the spatial influence area corresponding to the station, in order to constrain the cold wave labels of adjacent grids within the spatial influence area to maintain local spatial consistency.

9. The cold wave prediction method based on wind turbine shutdown records according to claim 1, characterized in that, After obtaining the gridded cold wave prediction labels for the future period of the area to be predicted, the cold wave impact status of the wind farm is determined according to the cold wave prediction label corresponding to the grid where the wind farm is located; the operating mode of the wind power prediction model is determined according to the cold wave impact status of the wind farm; wherein, when the wind farm is in a non-cold wave impact state, the constant mode is used for wind power prediction, and when the wind farm is in a cold wave impact state, the cold wave mode is used for wind power prediction.

10. A cold wave prediction device based on wind turbine shutdown records, characterized in that, include: The data acquisition module is used to obtain numerical weather forecast characteristics of the area to be predicted for future periods. The prediction module is used to input the numerical weather forecast features of the future time period into a pre-trained spatiotemporal prediction model to obtain preliminary prediction results of gridded cold wave labels for the area to be predicted in the future time period. The correction module is used to smooth and correct the preliminary prediction results of the gridded cold wave labels for the future time period, so as to obtain the gridded cold wave prediction labels for the area to be predicted for the future time period. The pre-trained spatiotemporal prediction model is obtained by training based on the historical gridded cold wave labels of the area to be predicted and the historical gridded numerical weather forecast features of the corresponding time period; the historical gridded cold wave labels are determined based on the ratio of the station's shutdown capacity to its installed capacity.

11. The cold wave prediction device based on wind turbine shutdown records according to claim 10, characterized in that, The historical gridded cold wave labels are obtained by dividing the area to be predicted into multiple grids; based on the shutdown records of the stations in the area to be predicted, the cold wave impact level label corresponding to each grid is determined, thus obtaining the historical gridded cold wave labels.

12. The cold wave prediction device based on wind turbine shutdown records according to claim 11, characterized in that, Based on the shutdown records of the stations within the area to be predicted, the cold wave impact level label corresponding to each grid is determined, including: determining the spatial impact area corresponding to each station based on the location of each station and the preset spatial impact radius, and using the grids within the spatial impact area as the impact grids of the corresponding stations; determining the cold wave impact level label corresponding to the spatial impact area based on the ratio of shutdown capacity to installed capacity within the spatial impact area corresponding to each station, and assigning the cold wave impact level label to the grids within the spatial impact area.

13. The cold wave prediction device based on wind turbine shutdown records according to claim 11, characterized in that, The cold wave impact level label includes multiple labels, and the levels of the multiple cold wave impact level labels are set in ascending order according to the ratio of shutdown capacity to installed capacity.

14. The cold wave prediction device based on wind turbine shutdown records according to claim 10, characterized in that, Both the historical gridded numerical weather forecast features and the numerical weather forecast features for the future period include at least one meteorological element among wind speed component, temperature, dew point temperature, and precipitation.

15. The cold wave prediction device based on wind turbine shutdown records according to claim 10, characterized in that, The pre-trained spatiotemporal prediction model is a convolutional long short-term memory network model; the convolutional long short-term memory network model includes an input gate, a forget gate, candidate unit states, cell states, an output gate, and hidden states.

16. The cold wave prediction device based on wind turbine shutdown records according to claim 10, characterized in that, The preliminary prediction results of the gridded cold wave labels for the future time period are smoothed and corrected, including: correcting the preliminary prediction results of the gridded cold wave labels for the future time period in terms of time dimension, so that the change in cold wave labels of the same grid between adjacent prediction time steps does not exceed a preset time smoothing threshold; and correcting the gridded cold wave labels after time dimension correction in terms of spatial dimension, so that the difference in cold wave labels between adjacent grids within the same prediction time step does not exceed a preset spatial smoothing threshold.

17. The cold wave prediction device based on wind turbine shutdown records according to claim 16, characterized in that, The spatial dimension correction is performed within the spatial influence area corresponding to the station, in order to constrain the cold wave labels of adjacent grids within the spatial influence area to maintain local spatial consistency.

18. The cold wave prediction device based on wind turbine shutdown records according to claim 1, characterized in that, It also includes a power prediction module, which, after obtaining the gridded cold wave prediction labels for the future period of the area to be predicted, determines the cold wave impact status of the wind farm based on the cold wave prediction label corresponding to the grid where the wind farm is located; and determines the operating mode of the wind power prediction model based on the cold wave impact status of the wind farm; wherein, when the wind farm is in a non-cold wave impact state, the constant mode is used for wind power prediction, and when the wind farm is in a cold wave impact state, the cold wave mode is used for wind power prediction.

19. An electronic device, characterized in that, It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement the cold wave prediction method based on wind turbine shutdown records as described in any one of claims 1 to 9.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the cold wave prediction method based on wind turbine shutdown records as described in any one of claims 1 to 9.

Images