Wind power supply guarantee probability prediction method based on QR-Transform model

Abstract

The invention discloses a QR-Transform model-based wind power supply guarantee probability prediction method. The method comprises the following steps of: based on a Transform model architecture, fusing a multi-head attention mechanism and a quantile regression layer to establish a wind power probability prediction model; the model is trained through Pinball loss functions under different quantile levels, corresponding loss values are adopted to evaluate the performance of the model, and a multi-quantile prediction model is output; inputting gridding wind speed and wind direction features to a multi-quantile prediction model to obtain a wind power probability prediction result, constructing wind power discrete probability distribution, performing fitting by using a pChip interpolation method, and performing stacking according to time scales to obtain a multi-time scale cumulative distribution function set; combining a power grid power generation plan and a typical operation scene, determining a guarantee power section threshold of each cluster, constructing a power supply guarantee demand section, and calculating a cumulative probability, namely a power supply guarantee probability, of the cumulative distribution function curve at the section; and a power supply guarantee probability prediction evaluation index system is constructed based on an expected correction error ECE, a Brier score and a Pinball loss function, and comprehensive evaluation of a prediction result is realized.

Description

Technical Field

[0001] The present invention relates to a wind power supply guarantee probability prediction method based on a QR-Transformer model, and belongs to the field of new energy power prediction. Background Art

[0002] With the introduction of the "dual carbon" goals and the continued growth of energy demand, large-scale wind power has been connected to the power system, and the role of wind power in ensuring the security of the system's power supply and promoting the consumption of renewable energy has become increasingly prominent. However, wind power output is significantly affected by meteorological conditions and has characteristics such as volatility and intermittency, which poses great challenges to grid scheduling, load balancing, and backup resource allocation. Current wind power forecasting research focuses on improving the accuracy of single-point or time series forecasts, which is difficult to meet the demand for output reliability in actual grid operation. It may face problems such as insufficient security due to overestimation of forecasts and the inability of backup power sources to respond in a timely manner. If a conservative reduction strategy is adopted for safety reasons, it may result in a large amount of wind power waste and affect the effective utilization of wind energy resources. Therefore, how to reasonably express its uncertainty and measure the power supply guarantee capability based on accurate wind power forecasting has become a key issue that needs to be urgently addressed in the current scheduling and operation of new energy.

[0003] Currently, there is a lack of modeling methods for predicting power supply security capabilities, especially the prediction of wind power supply security probability driven by multi-quantile information, which is still a research gap. Summary of the Invention

[0004] The purpose of the present invention is to provide a wind power supply guarantee probability prediction method based on the QR-Transformer model to solve the problems existing in the above-mentioned background technology. The present invention uses quantile modeling to predict the probability distribution of wind power, and then constructs a wind power supply guarantee probability curve to achieve quantitative analysis of the power supply guarantee capability, providing decision support for scheduling strategy formulation and system operation.

[0005] The purpose of the present invention is achieved by the following technical measures: A wind power supply security probability prediction method based on the QR-Transformer model is characterized by the following steps: S1: Build a wind power probability prediction model based on the Transformer architecture, introduce attention mechanism and position encoding to enhance high-dimensional feature extraction and time series modeling capabilities; integrate a quantile regression layer in the output layer to output multiple target quantiles for each prediction moment 1. 2… The prediction results are obtained by using Pinball loss function and Adam optimizer to train the wind power probability prediction model until the loss function converges and output QR-Transformer wind power probability prediction model at different quantile levels; S2: Based on the wind power probability prediction model of the QR-Transformer at different quantiles obtained in S1, the gridded wind speed and direction characteristics are input to obtain the wind power probability multi-quantile prediction results, and the cumulative distribution function at each moment is obtained through pChip interpolation fitting; the power supply demand level of the power grid is determined by combining the power grid dispatch plan, load forecast curve and reserve capacity demand, and a power supply guarantee demand section is constructed for the power grid demand at each moment. According to the guaranteed power section threshold of each cluster, the cumulative distribution function is stacked at multiple time scales to obtain a multi-time scale cumulative distribution function set; the power supply guarantee demand section and the multi-time scale cumulative distribution function set are combined to obtain the wind power power supply guarantee probability prediction result, evaluate the power supply shortage risk, and optimize the power grid dispatch decision; S3: Based on the guaranteed power section thresholds of each cluster and the wind power supply guarantee probability prediction results in S2, the Brier score is measured to measure the difference between the guarantee probability corresponding to the wind power supply guarantee demand section and the actual power supply situation. The wind power supply guarantee probability prediction results are divided into intervals according to the expected guarantee probability, and the average power supply guarantee probability in each interval is calculated. This is then compared with the probability that the samples in the interval meet the power supply guarantee demand. The absolute difference between the average expected guarantee probability and the statistical probability in each interval is statistically calculated to obtain the expected calibration error (ECE). A Pinball loss function comprehensive evaluation indicator is proposed by combining the coverage rate and average interval width indicators of the wind power supply guarantee probability. The three evaluation indicators mentioned above, Brier score, ECE, and Pinball loss function, are combined to form a multi-dimensional evaluation system for wind power power supply guarantee probability prediction. The power supply guarantee capability is comprehensively evaluated based on the wind power power supply guarantee probability prediction results.

[0006] Furthermore, the specific steps of S1 include: S1.1: Build a wind power probability prediction model based on the Transformer architecture. Obtain the historical power feature sequence of wind farms and transform it into a feature space sequence of uniform dimension through linear mapping. Add a learnable positional encoding to this feature space sequence, with the position encoding dimension consistent with the input feature dimension. This is then sequentially input into a Transformer encoder. Each encoder layer includes a multi-head self-attention mechanism, summing and normalization layers, and a feedforward neural network. The output is a coding matrix that extracts high-dimensional features of the temporal information of the feature space sequence. S1.2: For each predicted future time point, a corresponding timestamp is constructed and embedded into the future wind power data sequence, which serves as the decoder input. The decoder consists of a multi-layer structure, each layer of which includes a multi-head attention mechanism to capture the correlation between predicted time points, an addition and normalization layer, and a feed-forward neural network. The feature sequence output by the decoder is input into a quantile regression layer, corresponding to different confidence levels, to output the wind power probability prediction results at different quantiles at each moment of the future wind power data. S1.3: Set multiple target quantiles 1. 2… For each prediction quantile output by the wind power probability prediction model based on the Transformer architecture, the Pinball loss function is used to measure the degree of deviation between the predicted value and the true value; the Adam optimizer is used to update the parameters of the wind power probability prediction model based on the Transformer architecture, and the quantile loss is minimized through back propagation; the maximum number of iterations is set as the training termination condition, and the performance of the wind power probability prediction model based on the Transformer architecture is evaluated on the test set; when the training convergence condition is met, the trained The quantile prediction models are denoted as

[0007] Furthermore, the specific steps of S2 include: S2.1: Based on the different quantile prediction models obtained in S1.3, input the gridded wind speed and direction characteristics and output the prediction time Multiple sets of prediction results; for each future moment, the quantile values ​​are regarded as anchor points on the discrete cumulative distribution function, and the pChip interpolation method is used to fit these quantiles to construct a continuous cumulative distribution function; S2.2: Based on the grid's original power generation plan and network topology, combined with load forecast curves and reserve capacity requirements, a DC power flow model is used to calculate the minimum power supply demand under the system's minimum operating mode, while meeting grid safety operation constraints. Considering the grid's power demand level and grid dispatch plan, the guaranteed power section thresholds for each cluster are obtained, and a wind power supply guarantee demand section is constructed. S2.3: Based on the cumulative distribution function constructed in S2.1, different time scales are set in combination with the dispatching side demand; the cumulative distribution functions within each time scale are stacked in chronological order to form a multi-time-scale wind power cumulative distribution function set; the wind power supply guarantee demand sections of each time period are input in parallel, and for each time scale and prediction moment, the cumulative probability of the current cumulative distribution function at the section is calculated as the wind power supply guarantee probability; the wind power supply guarantee probability prediction results are coupled with the power grid dispatch plan for analysis; it is assessed whether there is a shortage of power grid security, and the power grid is guided to optimize dispatching decisions.

[0008] Furthermore, the specific steps of S3 include: S3.1: Based on the guaranteed power section thresholds of each cluster in S2.2 and the wind power supply guarantee probability prediction results in S2.3, the corresponding wind power supply guarantee power section is given according to different actual needs. The difference between the guarantee probability corresponding to the wind power supply demand section and the actual power supply situation is measured to evaluate the accuracy of the power supply guarantee probability estimation as the Brier score evaluation indicator : Where: for The probability that the actual output at any moment meets the power supply demand; for The actual output at any moment meets the power supply demand, which is 1 if satisfied and 0 if not; T is the total number of samples, and the Brier score evaluation index The closer it is to 0, the better the prediction performance of wind power supply guarantee probability is; the closer it is to 1, the less accurate the prediction performance of wind power supply guarantee probability is. S3.2: Divide the wind power supply guarantee probability prediction results under different wind power supply demand sections at each time into several intervals according to the expected guarantee probability. Then calculate the average wind power supply guarantee probability predicted by each quantile prediction model in each interval, and then compare it with the probability that the samples in the interval meet the wind power supply guarantee demand. Count the absolute difference between the average expected wind power supply guarantee probability and the statistical probability in each interval and calculate the weighted average as the expected correction error. : Where: For the An estimated guarantee probability interval; is the total number of expected guarantee probability intervals; For the The number of samples of the estimated guaranteed probability interval; N is the total number of samples of all expected guarantee probability intervals; For the The frequency of samples meeting the wind power supply security demand within the estimated security probability interval; For the The average expected wind power supply security probability in each interval; The closer it is to 0, the better the performance of the wind power supply security probability prediction model; S3.3: Based on The loss function proposes a comprehensive evaluation index: In the formula To ensure the probability of wind power supply, for Moment The predicted power under the probability level of wind power supply guarantee, for The actual power at the moment, is the total number of power supply moments participating in the evaluation, 、 is the decision variable; the formula is divided into two parts, The part is the sharpness part, which is the distance between the power under the proposed wind power supply guarantee probability and the actual power; and They are respectively the reliability part, that is, compared with the case where the guaranteed output of wind power supply is lower than the actual output, the penalty caused by the error during the period when the guaranteed output of wind power supply is higher than the actual output is increased; S3.4: The Brier score measures the difference between the wind power supply guarantee probability and the actual power supply situation, the expected corrected error (ECE) measures the degree of deviation between the wind power supply guarantee probability and the probability of meeting the actual situation, and the Pinball loss function integrates the deviation of the probability prediction and the asymmetric penalty of the upper and lower bounds, reflecting the fitting effect of the wind power supply guarantee probability prediction model on the multi-quantile power supply guarantee probability; based on the above three evaluation indicators of Brier score, ECE and Pinball loss function, a multidimensional evaluation system for wind power supply guarantee probability prediction is established, and the prediction results are evaluated from the three dimensions of prediction accuracy, calibration and comprehensive error; the wind power supply guarantee probability prediction result obtained in S2.3 is input into the wind power supply guarantee probability prediction multidimensional evaluation system, and the evaluation result of the wind power supply guarantee probability prediction is output.

[0009] The method of the present invention can effectively predict the probability of wind power supply security, improve the dispatch rationality and operation stability of wind power grid connection, provide scientific decision-making support for power dispatching departments, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a structural block diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0011] The following is a detailed description of the embodiments of the present invention with reference to the accompanying drawings. The purpose of the present invention is to provide a method for predicting the probability of wind power supply security based on the QR-Transformer model.

[0012] like Figure 1 As shown, a wind power supply guarantee probability prediction method based on the QR-Transformer model is characterized by the following steps: S1: Build a wind power probability prediction model based on the Transformer architecture, introduce attention mechanism and position encoding to enhance high-dimensional feature extraction and time series modeling capabilities; integrate a quantile regression layer in the output layer to output multiple target quantiles for each prediction moment 1. 2… The prediction results are obtained by using Pinball loss function and Adam optimizer to train the wind power probability prediction model until the loss function converges and output The QR-Transformer wind power probability prediction model at different quantile levels; its specific steps include: S1.1: Build a wind power probability prediction model based on the Transformer architecture. Obtain the historical power feature sequence of wind farms and transform it into a feature space sequence of uniform dimension through linear mapping. Add a learnable positional encoding to this feature space sequence, with the position encoding dimension consistent with the input feature dimension. This is then sequentially input into a Transformer encoder. Each encoder layer includes a multi-head self-attention mechanism, summing and normalization layers, and a feedforward neural network. The output is a coding matrix that extracts high-dimensional features of the temporal information of the feature space sequence. S1.2: For each predicted future time point, a corresponding timestamp is constructed and embedded into the future wind power data sequence, which serves as the decoder input. The decoder consists of a multi-layer structure, each layer of which includes a multi-head attention mechanism to capture the correlation between predicted time points, an addition and normalization layer, and a feed-forward neural network. The feature sequence output by the decoder is input into a quantile regression layer, corresponding to different confidence levels, to output the wind power probability prediction results at different quantiles at each moment of the future wind power data. S1.3: Set multiple target quantiles 1. 2… For each prediction quantile output by the wind power probability prediction model based on the Transformer architecture, the Pinball loss function is used to measure the degree of deviation between the predicted value and the true value; the Adam optimizer is used to update the parameters of the wind power probability prediction model based on the Transformer architecture, and the quantile loss is minimized through back propagation; the maximum number of iterations is set as the training termination condition, and the performance of the wind power probability prediction model based on the Transformer architecture is evaluated on the test set; when the training convergence condition is met, the trained The quantile prediction models are denoted as

[0013] S2: Based on the wind power probability prediction model of the QR-Transformer at different quantiles obtained in S1, the gridded wind speed and direction characteristics are input to obtain the wind power probability multi-quantile prediction results, and the cumulative distribution function at each moment is obtained through pChip interpolation fitting; the power supply demand level of the power grid is determined by combining the power grid dispatch plan, load forecast curve and reserve capacity demand, and a power supply guarantee demand section is constructed for the power grid demand at each moment. The cumulative distribution function is stacked at multiple time scales according to the guaranteed power section threshold of each cluster to obtain a multi-time scale cumulative distribution function set; the power supply guarantee demand section and the multi-time scale cumulative distribution function set are combined to obtain the wind power power supply guarantee probability prediction result, evaluate the power supply shortage risk, and optimize the power grid dispatch decision; the specific steps include: S2.1: Based on the different quantile prediction models obtained in S1.3, input the gridded wind speed and direction characteristics and output the prediction time Multiple sets of prediction results; for each future moment, the quantile values ​​are regarded as anchor points on the discrete cumulative distribution function, and the pChip interpolation method is used to fit these quantiles to construct a continuous cumulative distribution function; S2.2: Based on the grid's original power generation plan and network topology, combined with load forecast curves and reserve capacity requirements, a DC power flow model is used to calculate the minimum power supply demand under the system's minimum operating mode, while meeting grid safety operation constraints. Considering the grid's power demand level and grid dispatch plan, the guaranteed power section thresholds for each cluster are obtained, and a wind power supply guarantee demand section is constructed. S2.3: Based on the cumulative distribution function constructed in S2.1, different time scales are set in combination with dispatch-side requirements. The cumulative distribution functions within each time scale are stacked in chronological order to form a multi-time-scale wind power cumulative distribution function set. The wind power supply guarantee demand sections for each time period are input in parallel. For each time scale and prediction moment, the cumulative probability of the current cumulative distribution function at the section is calculated as the wind power supply guarantee probability. The wind power supply guarantee probability prediction results are coupled with the grid dispatch plan for analysis. The assessment of whether there is insufficient grid security is carried out, and the grid optimization dispatch decision is guided. S3: Based on the guaranteed power section thresholds of each cluster and the wind power supply guarantee probability prediction results in S2, the difference between the guarantee probability corresponding to the wind power supply guarantee demand section and the actual power supply situation is measured to obtain the Brier score; the wind power supply guarantee probability prediction results are divided into intervals according to the expected guarantee probability, and the average power supply guarantee probability in each interval is calculated. This is then compared with the probability that the samples in the interval meet the power supply guarantee demand. The absolute difference between the average expected guarantee probability and the statistical probability in each interval is statistically calculated to obtain the expected calibration error (ECE); combining the coverage rate and average interval width indicators of the wind power supply guarantee probability, a Pinball loss function comprehensive evaluation indicator is proposed; the three evaluation indicators of the Brier score, ECE, and Pinball loss function are combined to form a multi-dimensional evaluation system for wind power supply guarantee probability prediction. Based on the wind power supply guarantee probability prediction results, the power supply guarantee capability is comprehensively evaluated; the specific steps include: S3.1: Based on the guaranteed power section thresholds of each cluster in S2.2 and the wind power supply guarantee probability prediction results in S2.3, the corresponding wind power supply guarantee power section is given according to different actual needs. The difference between the guarantee probability corresponding to the wind power supply demand section and the actual power supply situation is measured to evaluate the accuracy of the power supply guarantee probability estimation as the Brier score evaluation indicator : Where: for The probability that the actual output at any moment meets the power supply demand; for The actual output at any moment meets the power supply demand, which is 1 if satisfied and 0 if not; T is the total number of samples, and the Brier score evaluation index The closer it is to 0, the better the prediction performance of wind power supply guarantee probability is; the closer it is to 1, the less accurate the prediction performance of wind power supply guarantee probability is. S3.2: Divide the wind power supply guarantee probability prediction results under different wind power supply demand sections at each time into several intervals according to the expected guarantee probability. Then calculate the average wind power supply guarantee probability predicted by each quantile prediction model in each interval, and then compare it with the probability that the samples in the interval meet the wind power supply guarantee demand. Count the absolute difference between the average expected wind power supply guarantee probability and the statistical probability in each interval and calculate the weighted average as the expected correction error. : Where: For the An estimated guarantee probability interval; is the total number of expected guarantee probability intervals; For the The number of samples of the estimated guaranteed probability interval; N is the total number of samples of all expected guarantee probability intervals; For the The frequency of samples meeting the wind power supply security demand within the estimated security probability interval; For the The average expected wind power supply security probability in each interval; The closer it is to 0, the better the performance of the wind power supply security probability prediction model; S3.3: Based on The loss function proposes a comprehensive evaluation index: In the formula To ensure the probability of wind power supply, for Moment The predicted power under the probability level of wind power supply guarantee, for The actual power at the moment, is the total number of power supply moments participating in the evaluation, 、 is the decision variable; the formula is divided into two parts, The part is the sharpness part, which is the distance between the power under the proposed wind power supply guarantee probability and the actual power; and They are respectively the reliability part, that is, compared with the case where the guaranteed output of wind power supply is lower than the actual output, the penalty caused by the error during the period when the guaranteed output of wind power supply is higher than the actual output is increased; S3.4: The Brier score measures the difference between the wind power supply guarantee probability and the actual power supply situation, the expected corrected error (ECE) measures the degree of deviation between the wind power supply guarantee probability and the probability of meeting the actual situation, and the Pinball loss function integrates the deviation of the probability prediction and the asymmetric penalty of the upper and lower bounds, reflecting the fitting effect of the wind power supply guarantee probability prediction model on the multi-quantile power supply guarantee probability; based on the above three evaluation indicators of Brier score, ECE and Pinball loss function, a multidimensional evaluation system for wind power supply guarantee probability prediction is established, and the prediction results are evaluated from the three dimensions of prediction accuracy, calibration and comprehensive error; the wind power supply guarantee probability prediction result obtained in S2.3 is input into the wind power supply guarantee probability prediction multidimensional evaluation system, and the evaluation result of the wind power supply guarantee probability prediction is output.

Claims

1. A wind power supply guarantee probability prediction method based on the QR-Transformer model, characterized by Follow these steps: S1: Build a wind power probability prediction model based on the Transformer architecture, introduce attention mechanism and position encoding to enhance high-dimensional feature extraction and time series modeling capabilities; integrate a quantile regression layer in the output layer to output multiple target quantiles for each prediction moment 1. 2… The prediction results are obtained by using Pinball loss function and Adam optimizer to train the wind power probability prediction model until the loss function converges and output QR-Transformer wind power probability prediction model at different quantile levels; S2: Based on the wind power probability prediction model of the QR-Transformer at different quantiles obtained in S1, the gridded wind speed and direction characteristics are input to obtain the wind power probability multi-quantile prediction results, and the cumulative distribution function at each moment is obtained through pChip interpolation fitting; the power supply demand level of the power grid is determined by combining the power grid dispatch plan, load forecast curve and reserve capacity demand, and a power supply guarantee demand section is constructed for the power grid demand at each moment. According to the guaranteed power section threshold of each cluster, the cumulative distribution function is stacked at multiple time scales to obtain a multi-time scale cumulative distribution function set; the power supply guarantee demand section and the multi-time scale cumulative distribution function set are combined to obtain the wind power power supply guarantee probability prediction result, evaluate the power supply shortage risk, and optimize the power grid dispatch decision; S3: Based on the guaranteed power section thresholds of each cluster and the wind power supply guarantee probability prediction results in S2, the Brier score is measured to measure the difference between the guarantee probability corresponding to the wind power supply guarantee demand section and the actual power supply situation. The wind power supply guarantee probability prediction results are divided into intervals according to the expected guarantee probability, and the average power supply guarantee probability in each interval is calculated. This is then compared with the probability that the samples in the interval meet the power supply guarantee demand. The absolute difference between the average expected guarantee probability and the statistical probability in each interval is statistically calculated to obtain the expected calibration error (ECE). A Pinball loss function comprehensive evaluation indicator is proposed by combining the coverage rate and average interval width indicators of the wind power supply guarantee probability. The three evaluation indicators mentioned above, Brier score, ECE, and Pinball loss function, are combined to form a multi-dimensional evaluation system for wind power power supply guarantee probability prediction. The power supply guarantee capability is comprehensively evaluated based on the wind power power supply guarantee probability prediction results.

2. According to the QR-Transformer model-based wind power supply security probability prediction method of claim 1, it is characterized in that: The S1 step specifically includes: S1.1: Build a wind power probability prediction model based on the Transformer architecture. Obtain the historical power feature sequence of wind farms and transform it into a feature space sequence of uniform dimension through linear mapping. Add a learnable positional encoding to this feature space sequence, with the position encoding dimension consistent with the input feature dimension. This is then sequentially input into a Transformer encoder. Each encoder layer includes a multi-head self-attention mechanism, summing and normalization layers, and a feedforward neural network. The output is a coding matrix that extracts high-dimensional features of the temporal information of the feature space sequence. S1.2: For each predicted future time point, a corresponding timestamp is constructed and embedded into the future wind power data sequence, which serves as the decoder input. The decoder consists of a multi-layer structure, each layer of which includes a multi-head attention mechanism to capture the correlation between predicted time points, an addition and normalization layer, and a feed-forward neural network. The feature sequence output by the decoder is input into a quantile regression layer, corresponding to different confidence levels, to output the wind power probability prediction results at different quantiles at each moment of the future wind power data. S1.3: Set multiple target quantiles 1. 2… For each prediction quantile output by the wind power probability prediction model based on the Transformer architecture, the Pinball loss function is used to measure the degree of deviation between the predicted value and the true value; the Adam optimizer is used to update the parameters of the wind power probability prediction model based on the Transformer architecture, and the quantile loss is minimized through back propagation; the maximum number of iterations is set as the training termination condition, and the performance of the wind power probability prediction model based on the Transformer architecture is evaluated on the test set; when the training convergence condition is met, the trained The quantile prediction models are denoted as 3. According to the QR-Transformer model-based wind power supply security probability prediction method of claim 2, it is characterized in that: The S2 step specifically includes: S2.1: Based on the different quantile prediction models obtained in S1.3, input the gridded wind speed and direction characteristics and output the prediction time Multiple sets of prediction results; for each future moment, the quantile values ​​are regarded as anchor points on the discrete cumulative distribution function, and the pChip interpolation method is used to fit these quantiles to construct a continuous cumulative distribution function; S2.2: Based on the grid's original power generation plan and network topology, combined with load forecast curves and reserve capacity requirements, a DC power flow model is used to calculate the minimum power supply demand under the system's minimum operating mode, while meeting grid safety operation constraints. Considering the grid's power demand level and grid dispatch plan, the guaranteed power section thresholds for each cluster are obtained, and a wind power supply guarantee demand section is constructed. S2.3: Based on the cumulative distribution function constructed in S2.1, different time scales are set in combination with the dispatching side demand; the cumulative distribution functions within each time scale are stacked in chronological order to form a multi-time-scale wind power cumulative distribution function set; the wind power supply guarantee demand sections of each time period are input in parallel, and for each time scale and prediction moment, the cumulative probability of the current cumulative distribution function at the section is calculated as the wind power supply guarantee probability; the wind power supply guarantee probability prediction results are coupled with the power grid dispatch plan for analysis; it is assessed whether there is a shortage of power grid security, and the power grid is guided to optimize dispatching decisions.

4. The method for predicting wind power supply security probability based on the QR-Transformer model according to claim 3 is characterized in that: The S3 step specifically includes: S3.1: Based on the guaranteed power section thresholds of each cluster in S2.2 and the wind power supply guarantee probability prediction results in S2.3, the corresponding wind power supply guarantee power section is given according to different actual needs. The difference between the guarantee probability corresponding to the wind power supply demand section and the actual power supply situation is measured to evaluate the accuracy of the power supply guarantee probability estimation as the Brier score evaluation indicator : Where: for The probability that the actual output at any moment meets the power supply demand; for The actual output at any moment meets the power supply demand, which is 1 if satisfied and 0 if not; T is the total number of samples, and the Brier score evaluation index The closer it is to 0, the better the prediction performance of wind power supply guarantee probability is; the closer it is to 1, the less accurate the prediction performance of wind power supply guarantee probability is. S3.2: Divide the wind power supply guarantee probability prediction results under different wind power supply demand sections at each time into several intervals according to the expected guarantee probability. Then calculate the average wind power supply guarantee probability predicted by each quantile prediction model in each interval, and then compare it with the probability that the samples in the interval meet the wind power supply guarantee demand. Count the absolute difference between the average expected wind power supply guarantee probability and the statistical probability in each interval and calculate the weighted average as the expected correction error. : Where: For the An estimated guarantee probability interval; is the total number of expected guarantee probability intervals; For the The number of samples of the estimated guaranteed probability interval; N is the total number of samples of all expected guarantee probability intervals; For the The frequency of samples meeting the wind power supply security demand within the estimated security probability interval; For the The average expected wind power supply security probability in each interval; The closer it is to 0, the better the performance of the wind power supply security probability prediction model; S3.3: Based on The loss function proposes a comprehensive evaluation index: In the formula To ensure the probability of wind power supply, for Moment The predicted power under the probability level of wind power supply guarantee, for The actual power at the moment, is the total number of power supply moments participating in the evaluation, 、 is the decision variable; the formula is divided into two parts, The part is the sharpness part, which is the distance between the power under the proposed wind power supply guarantee probability and the actual power; and They are respectively the reliability part, that is, compared with the case where the guaranteed output of wind power supply is lower than the actual output, the penalty caused by the error during the period when the guaranteed output of wind power supply is higher than the actual output is increased; S3.4: The Brier score measures the difference between the wind power supply guarantee probability and the actual power supply situation, the expected corrected error (ECE) measures the degree of deviation between the wind power supply guarantee probability and the probability of meeting the actual situation, and the Pinball loss function integrates the deviation of the probability prediction and the asymmetric penalty of the upper and lower bounds, reflecting the fitting effect of the wind power supply guarantee probability prediction model on the multi-quantile power supply guarantee probability; based on the above three evaluation indicators of Brier score, ECE and Pinball loss function, a multidimensional evaluation system for wind power supply guarantee probability prediction is established, and the prediction results are evaluated from the three dimensions of prediction accuracy, calibration and comprehensive error; the wind power supply guarantee probability prediction result obtained in S2.3 is input into the wind power supply guarantee probability prediction multidimensional evaluation system, and the evaluation result of the wind power supply guarantee probability prediction is output.

Images