Wind power prediction method and system based on event-driven variable state space

Abstract

The application discloses a wind power prediction method and system based on an event-driven variable state space, and the method comprises the following steps: constructing a reference set based on normal samples, performing counterfactual re-labeling, and expanding the measured power into enhanced supervision information containing available power, suppression coefficients and confidence; constructing an event-driven variable state space prediction model, the model comprising an event condition parameter generator and a variable state space recursion module, generating state modulation parameters according to an event condition sequence, and adaptively updating the hidden state; performing multi-task training by using the enhanced supervision information, so that the model simultaneously outputs the predicted values of the observed power, the available power and the suppression coefficients; and inputting to-be-predicted data into the model to obtain a prediction result. The label distortion problem is solved through counterfactual re-labeling, and the complex working condition modeling capability is enhanced through the event-driven variable state space, so that the precision and interpretability of wind power prediction are improved.

Description

Technical Field

[0001] This invention belongs to the field of wind power prediction technology, and particularly relates to a wind power prediction method and system based on event-driven variable state space. Background Technology

[0002] With the continuous expansion of new energy grid connection, wind power, as an important form of clean energy, has developed rapidly. However, wind power output is affected by various factors such as wind speed, wind direction, air density, turbulence, wake effect, and unit operating status, exhibiting strong randomness, volatility, and non-stationarity. Accurate prediction of wind power output is of great significance for grid dispatch, reserve capacity allocation, and new energy consumption.

[0003] Current deep learning-based wind power prediction methods still have shortcomings. In terms of data preprocessing, existing methods typically directly remove outliers, interpolate, or smooth them, failing to differentiate the impact of operating conditions such as power curtailment, derating, icing, and wake limitation on measured power. This results in a mixture of observed and normally available power, and distorted training labels. Regarding model structure, traditional recurrent neural networks, Transformers, or ordinary Mamba-type time series models often employ fixed state updates or attention modeling methods, usually treating operating events only as ordinary input features. They lack the ability to dynamically adjust the state transition process according to different operating conditions, making it difficult to effectively characterize the dynamic changes in wind power under complex operating conditions.

[0004] Therefore, there is an urgent need for a wind power prediction method that can effectively handle the impact of operating conditions and adaptively adjust the prediction model according to the operating conditions. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies by providing a wind power prediction method and system based on event-driven variable state space. It aims to solve the problems of training label distortion and the inability of models to adaptively adjust the state transition process according to operating conditions in existing technologies, thereby improving the accuracy, robustness, and interpretability of wind power prediction under complex operating conditions.

[0006] In a first aspect, the present invention provides a wind power prediction method based on event-driven variable state space, comprising: Obtain raw wind power operation data, identify the operation status at each sampling time, and obtain the status code; A normal reference set is constructed based on normal state samples, and counterfactual relabeling is performed on each sampling time according to the state code to obtain enhanced supervision information containing available power, suppression coefficient, and confidence level, specifically including: Under normal conditions, the measured power is directly taken as the usable power. For abnormal states, the counterfactual available power is reconstructed based on the normal reference set; The suppression coefficient is determined based on the measured power and the counterfactual available power, the confidence level is determined based on the operating status and the nearest neighbor distance, and the measured power is expanded into the enhanced supervision information based on the suppression coefficient and the confidence level. A wind power prediction model is constructed to receive historical input sequences, future meteorological driving sequences, and event condition sequences extracted from raw data. The wind power prediction model includes an event condition parameter generator and a variable state space recursion module. The event condition parameter generator generates state modulation parameters based on the event condition sequences, and the variable state space recursion module uses the state modulation parameters to adaptively recursively update the hidden states. The enhanced supervision information is used to train the wind power prediction model in multiple tasks to obtain the target wind power prediction model, so that the wind power prediction model can simultaneously output the predicted values ​​of observed power, available power and suppression coefficient. Meteorological data and operational event data for the period to be predicted are input into the target wind power prediction model, and the target wind power prediction model outputs the wind power prediction result.

[0007] Secondly, the present invention provides a wind power prediction system based on event-driven variable state space, comprising: The acquisition module is configured to acquire raw wind power operation data, identify the operation status at each sampling time, and obtain the status code; The annotation module is configured to construct a normal reference set based on normal state samples, and perform inverse fact re-annotation on each sampling time according to the state code to obtain enhanced supervision information containing available power, suppression coefficient, and confidence level, specifically including: Under normal conditions, the measured power is directly taken as the usable power. For abnormal states, the counterfactual available power is reconstructed based on the normal reference set; The suppression coefficient is determined based on the measured power and the counterfactual available power, the confidence level is determined based on the operating status and the nearest neighbor distance, and the measured power is expanded into the enhanced supervision information based on the suppression coefficient and the confidence level. The construction module is configured to build a wind power prediction model for receiving historical input sequences, future meteorological driving sequences, and event condition sequences extracted from raw data. The wind power prediction model includes an event condition parameter generator and a variable state space recursion module. The event condition parameter generator generates state modulation parameters based on the event condition sequences, and the variable state space recursion module uses the state modulation parameters to adaptively recursively update the hidden states. The training module is configured to perform multi-task training on the wind power prediction model using the enhanced supervision information to obtain the target wind power prediction model, so that the wind power prediction model can simultaneously output the predicted values ​​of observed power, available power and suppression coefficient. The output module is configured to input meteorological data and operational event data for the period to be predicted into the target wind power prediction model, and the target wind power prediction model outputs the wind power prediction result.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the wind power prediction method based on event-driven variable state space according to any embodiment of the present invention.

[0009] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the processor performs the steps of the wind power prediction method based on event-driven variable state space according to any embodiment of the present invention.

[0010] This application presents a wind power prediction method and system based on event-driven variable state space. By relabeling the original measured power using the counterfactual method of operating conditions, it decouples the original measured power into available power, suppression coefficient, and confidence level. This effectively reduces the pollution of training labels by abnormal operating conditions such as power curtailment, derating, icing, and wake, thus solving the label distortion problem. Through the event-driven variable state space model, the model's state update process can adaptively change with operating conditions, enhancing the modeling capability for complex wind power operating conditions. The multi-task joint prediction method simultaneously outputs observed power, available power, and suppression coefficient, which not only improves prediction accuracy but also enhances the interpretability and engineering application value of the prediction results. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a wind power prediction method based on an event-driven variable state space, as provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of a wind power prediction system based on an event-driven variable state space, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Please see Figure 1 The diagram shows a flowchart of a wind power prediction method based on event-driven variable state space according to this application.

[0015] like Figure 1 As shown, the wind power prediction method based on event-driven variable state space specifically includes the following steps: Step S101: Obtain raw wind power operation data, identify the operation status at each sampling time, and obtain the status code.

[0016] In this step, raw wind power operation data is first acquired from the wind farm data acquisition and monitoring system. The system records various operating parameters of the wind turbines at a fixed sampling frequency (typically every 10 or 15 minutes). The acquired raw data includes multiple feature dimensions and measured power sequences. Feature dimensions include wind speed, wind direction, ambient temperature, yaw error, active power setpoint, curtailment warning, derating warning, icing warning, and wake-related characteristics. The curtailment warning, derating warning, and icing warning are typically written into the wind farm data acquisition and monitoring system by the wind turbine's main control system based on internal logic; these are discrete variables with values ​​of 0 or 1.

[0017] Then, key features closely related to the operation status determination are selected from the raw data to form a status feature matrix, and the operation status is determined one by one at each sampling time. The operation status is identified according to a preset priority order, in the following order: abnormal measurement status, icing status, power limiting status, derating status, wake limiting status, and normal status.

[0018] When identifying abnormal measurement states, if the sensor disconnection flag or communication timeout flag is 1 at the sampling time, then that time is considered an abnormal measurement state; if the measured power is negative or exceeds 105% of the rated power, then it is considered an abnormal measurement state; if the wind speed value exceeds the physically reasonable range, then it is considered an abnormal measurement state; if the power change rate is abnormal, that is, the absolute value of the change in measured power at the current time relative to the measured power at the previous time divided by the rated power is greater than 80%, and this phenomenon continues for more than 3 sampling points, then it is considered an abnormal measurement state.

[0019] When determining the icing state, if the icing flag is 1, then the current moment is considered to be in an icing state. If the ambient temperature remains below 0 degrees Celsius and the measured power at the current moment is less than 60% of the normal operating power at the same wind speed, then the current moment is considered to be in an icing state. The normal operating power at the same wind speed is determined as follows: using wind speed as the horizontal axis and power as the vertical axis, a piecewise linear fit is performed on the historical normal operating data to establish a benchmark wind speed-power mapping table. Then, the corresponding benchmark power is obtained by looking up the table based on the current wind speed.

[0020] When determining the power restriction status, if the power restriction flag is 1 and the measured power is less than 70% of the base power at the same wind speed, then the moment is determined to be a power restriction state.

[0021] When determining the derating status, if the derating flag is 1, then the moment is determined to be in a derating status; if the gearbox temperature or generator winding temperature exceeds the preset threshold, causing the unit to actively reduce its output, and the measured power is 50% to 85% of the base power at the same wind speed, then the moment is determined to be in a derating status.

[0022] When determining wake-limited conditions, if the measured wind speed at the current location is lower than the average wind speed at the upwind reference location, and the difference between the two is greater than 15% of the average wind speed at the upwind reference location, then it is determined that the location may be affected by a wake. If the ratio of the standard deviation of the wind speed within 10 minutes to the average wind speed within 10 minutes exceeds 12%, and the measured power is lower than 70% to 90% of the baseline power at the same wind speed, then it is determined that the location is in a wake-limited state.

[0023] If the sampling time does not meet any of the above-mentioned abnormal state discrimination conditions, then the sampling time is determined to be a normal state.

[0024] After the above discrimination, a state code is assigned to each sampling time. The state code is assigned according to the following rules: normal state is assigned 0, power-limited state is assigned 1, derating state is assigned 2, icing state is assigned 3, wake-limited state is assigned 4, and abnormal measurement state is assigned 5.

[0025] In the above manner, step S101 transforms the original wind farm data acquisition and monitoring system data into a time-series state coding sequence with clear operating status labels, providing accurate state prior information for subsequent counterfactual relabeling and model training.

[0026] Step S102: Construct a normal reference set based on normal state samples, and perform counterfactual relabeling on each sampling time according to the state code to obtain enhanced supervision information containing available power, suppression coefficient, and confidence level, specifically including: Under normal conditions, the measured power is directly taken as the usable power. For abnormal states, the counterfactual available power is reconstructed based on the normal reference set; The suppression coefficient is determined based on the measured power and the counterfactual available power, the confidence level is determined based on the operating status and the nearest neighbor distance, and the measured power is extended into the enhanced supervision information based on the suppression coefficient and the confidence level.

[0027] In this step, the Euclidean distance between the state features at the current sampling time and each sample in the normal reference set is calculated, and the closest sample is selected. The counterfactual available power is obtained by weighted summation of the nearest neighbors, expressed as: , In the formula, Power available for counterfactual purposes For the number of nearest neighbors, For the current sampling time and the first Euclidean distance between the nearest neighbors For the current sampling time and the first Euclidean distance between the nearest neighbors From the current sampling time to The average Euclidean distance of the nearest neighbors For the current sampling time and the first The time interval between nearest neighbor samples For time decay scale parameters, For the first The normal reference power of the nearest neighbor, for The average of the nearest neighbor normal reference power, The wind speed at the current sampling time. For the first The wind speed corresponding to the nearest neighbor sample This is the wind speed power-law adaptive adjustment coefficient. This is the local power residual compensation coefficient. The upper limit of the counterfactual usable power. It is a very small positive number. It is the hyperbolic tangent function. This is the amplitude limiting function.

[0028] The expression for calculating the inhibition coefficient is as follows: , In the formula, The suppression coefficient, For actual measured power, Power available for counterfactual purposes Encoding the state at the current sampling time. For state coding The corresponding reference inhibition level, From the current sampling time to The average distance of the nearest neighbors These are the state prior correction coefficients. The power deviation response coefficient is... The power deviation trigger threshold, The nearest neighbor matching attenuation coefficient, For distance scale parameters, It is a very small positive number. This is a limiting function; The expression for calculating the confidence level is: , In the formula, For confidence level, Encoding the state at the current sampling time The basis for the decision is the credibility coefficient. From the current sampling time to The average distance of the nearest neighbors For the number of nearest neighbors, For the current sampling time and the first Euclidean distance between the nearest neighbors The suppression coefficient, Encoding the state at the current sampling time The corresponding reference inhibition level, The average nearest neighbor distance is the scale parameter. The nearest neighbor distance dispersion scale parameter, To suppress the uniformity response coefficient, To suppress the allowable deviation threshold for consistency.

[0029] Step S103: Construct a wind power prediction model for receiving historical input sequences, future meteorological driving sequences, and event condition sequences extracted from raw data. The wind power prediction model includes an event condition parameter generator and a variable state space recursion module. The event condition parameter generator generates state modulation parameters based on the event condition sequences, and the variable state space recursion module uses the state modulation parameters to adaptively recursively update the hidden states.

[0030] In this step, the state modulation parameters include state gating parameters. State scaling parameters and state bias parameters ; The variable state-space recursive module performs the following adaptive update at the t-th prediction step: , , In the formula, For the future weather input at the t-th prediction step, For the first The candidate hidden states generated in each prediction step are used to represent the temporary state update results obtained under the influence of current and future meteorological inputs and state modulation parameters. For the first The updated hidden state output from each prediction step is used to propagate to the next prediction step and for subsequent power prediction. For the first The historical hidden states output by each prediction step Let be the state gating parameters for the t-th prediction step. The state scaling parameter for the t-th prediction step. , For learnable parameters, Let be the state bias parameter for the t-th prediction step.

[0031] The event condition parameter generator generates state gating parameters. The expression is: , In the formula, For the Sigmoid function, For event condition sequence, , These are learnable parameters; The state scaling parameters are generated using a learnable parameter generation network based on the event condition sequence. and the state bias parameters The expression is: , In the formula, Generate networks for learnable parameters.

[0032] Step S104: Use the enhanced supervision information to perform multi-task training on the wind power prediction model to obtain the target wind power prediction model, so that the wind power prediction model can simultaneously output the predicted values ​​of observed power, available power and suppression coefficient.

[0033] In this step, multi-task training uses a confidence-weighted loss function: , In the formula, The confidence-weighted loss function, For the confidence tensor, , , These are the predicted values ​​for observed power, available power, and suppression coefficient, respectively. , , These are the true values ​​of observed power, available power, and suppression coefficient, respectively. Encoding the state at the current sampling time. Encoding the state at the current sampling time The defined task coupling weight matrix is ​​used to characterize the coupling relationship between observed power, available power, suppression coefficient, and power consistency constraints. The confidence level threshold is used for the decision-making process. This is the confidence level adjustment coefficient. , where is the confidence level gating steepness coefficient. The robust smoothing coefficient is... It is a very small positive number. For element-wise multiplication, To calculate the average of the sample dimension and the prediction step dimension.

[0034] Step S105: Input the meteorological data and operational event data for the period to be predicted into the target wind power prediction model, and the target wind power prediction model outputs the wind power prediction result.

[0035] In summary, the method of this application acquires raw wind power operation data and identifies the operating state to obtain a state code; constructs a reference set based on normal samples, performs counterfactual relabeling, and expands the measured power into enhanced supervision information including available power, suppression coefficient, and confidence level; constructs an event-driven variable state space prediction model, which includes an event condition parameter generator and a variable state space recursion module, generates state modulation parameters based on the event condition sequence, and adaptively recursively updates the hidden states; utilizes enhanced supervision information for multi-task training, enabling the model to simultaneously output predicted values ​​of observed power, available power, and suppression coefficient; inputs the data to be predicted into the model to obtain the prediction result; solves the label distortion problem through counterfactual relabeling, and enhances the modeling capability for complex operating conditions through event-driven variable state space, thereby improving the accuracy and interpretability of wind power prediction.

[0036] Please see Figure 2 The diagram shows a structural block diagram of a wind power prediction system based on an event-driven variable state space according to this application.

[0037] like Figure 2 As shown, the wind power prediction system 200 includes an acquisition module 210, a labeling module 220, a construction module 230, a training module 240, and an output module 250.

[0038] The acquisition module 210 is configured to acquire raw wind power operation data, identify the operation status at each sampling time, and obtain a status code. The annotation module 220 is configured to construct a normal reference set based on normal state samples, and perform counterfactual re-annotation on each sampling time according to the status code to obtain enhanced supervision information containing available power, suppression coefficient, and confidence level. Specifically, this includes: for normal states, directly using the measured power as available power; for abnormal states, reconstructing the counterfactual available power based on the normal reference set; determining the suppression coefficient based on the measured power and the counterfactual available power; determining the confidence level based on the operation status and the nearest neighbor distance; and expanding the measured power into the enhanced supervision information based on the suppression coefficient and the confidence level. The construction module 230 is configured to construct a system for receiving historical input sequences and future meteorological driving sequences. The system includes a wind power prediction model based on event condition sequences extracted from raw data. The wind power prediction model comprises an event condition parameter generator and a variable state space recursion module. The event condition parameter generator generates state modulation parameters based on the event condition sequences, and the variable state space recursion module uses the state modulation parameters to adaptively update the hidden states. A training module 240 is configured to perform multi-task training on the wind power prediction model using the enhanced supervision information to obtain a target wind power prediction model, which simultaneously outputs predicted values ​​for observed power, available power, and suppression coefficient. An output module 250 is configured to input meteorological data and operational event data for the period to be predicted into the target wind power prediction model, and the target wind power prediction model outputs the wind power prediction result.

[0039] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0040] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the wind power prediction method based on event-driven variable state space in any of the above method embodiments. In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows: Obtain raw wind power operation data, identify the operation status at each sampling time, and obtain the status code; A normal reference set is constructed based on normal state samples, and counterfactual relabeling is performed on each sampling time according to the state code to obtain enhanced supervision information containing available power, suppression coefficient, and confidence level, specifically including: Under normal conditions, the measured power is directly taken as the usable power. For abnormal states, the counterfactual available power is reconstructed based on the normal reference set; The suppression coefficient is determined based on the measured power and the counterfactual available power, the confidence level is determined based on the operating status and the nearest neighbor distance, and the measured power is expanded into the enhanced supervision information based on the suppression coefficient and the confidence level. A wind power prediction model is constructed to receive historical input sequences, future meteorological driving sequences, and event condition sequences extracted from raw data. The wind power prediction model includes an event condition parameter generator and a variable state space recursion module. The event condition parameter generator generates state modulation parameters based on the event condition sequences, and the variable state space recursion module uses the state modulation parameters to adaptively recursively update the hidden states. The enhanced supervision information is used to train the wind power prediction model in multiple tasks to obtain the target wind power prediction model, so that the wind power prediction model can simultaneously output the predicted values ​​of observed power, available power and suppression coefficient. Meteorological data and operational event data for the period to be predicted are input into the target wind power prediction model, and the target wind power prediction model outputs the wind power prediction result.

[0041] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the event-driven variable state space-based wind power prediction system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely configured relative to a processor, which can be connected to the event-driven variable state space-based wind power prediction system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0042] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, memory 320 is the computer-readable storage medium described above. Processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in memory 320, thereby implementing the wind power prediction method based on event-driven variable state space described in the above embodiment. Input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the wind power prediction system based on event-driven variable state space. Output device 340 may include a display screen or other display device.

[0043] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0044] In one implementation, the above-described electronic device is applied to a wind power forecasting system based on event-driven variable state space, for a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Obtain raw wind power operation data, identify the operation status at each sampling time, and obtain the status code; A normal reference set is constructed based on normal state samples, and counterfactual relabeling is performed on each sampling time according to the state code to obtain enhanced supervision information containing available power, suppression coefficient, and confidence level, specifically including: Under normal conditions, the measured power is directly taken as the usable power. For abnormal states, the counterfactual available power is reconstructed based on the normal reference set; The suppression coefficient is determined based on the measured power and the counterfactual available power, the confidence level is determined based on the operating status and the nearest neighbor distance, and the measured power is expanded into the enhanced supervision information based on the suppression coefficient and the confidence level. A wind power prediction model is constructed to receive historical input sequences, future meteorological driving sequences, and event condition sequences extracted from raw data. The wind power prediction model includes an event condition parameter generator and a variable state space recursion module. The event condition parameter generator generates state modulation parameters based on the event condition sequences, and the variable state space recursion module uses the state modulation parameters to adaptively recursively update the hidden states. The enhanced supervision information is used to train the wind power prediction model in multiple tasks to obtain the target wind power prediction model, so that the wind power prediction model can simultaneously output the predicted values ​​of observed power, available power and suppression coefficient. Meteorological data and operational event data for the period to be predicted are input into the target wind power prediction model, and the target wind power prediction model outputs the wind power prediction result.

[0045] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0046] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wind power prediction method based on event-driven variable state space, characterized in that, include: Obtain raw wind power operation data, identify the operation status at each sampling time, and obtain the status code; A normal reference set is constructed based on normal state samples, and counterfactual relabeling is performed on each sampling time according to the state code to obtain enhanced supervision information containing available power, suppression coefficient, and confidence level, specifically including: Under normal conditions, the measured power is directly taken as the usable power. For abnormal states, the counterfactual available power is reconstructed based on the normal reference set; The suppression coefficient is determined based on the measured power and the counterfactual available power, the confidence level is determined based on the operating status and the nearest neighbor distance, and the measured power is expanded into the enhanced supervision information based on the suppression coefficient and the confidence level. A wind power prediction model is constructed to receive historical input sequences, future meteorological driving sequences, and event condition sequences extracted from raw data. The wind power prediction model includes an event condition parameter generator and a variable state space recursion module. The event condition parameter generator generates state modulation parameters based on the event condition sequences, and the variable state space recursion module uses the state modulation parameters to adaptively recursively update the hidden states. The enhanced supervision information is used to train the wind power prediction model in multiple tasks to obtain the target wind power prediction model, so that the wind power prediction model can simultaneously output the predicted values ​​of observed power, available power and suppression coefficient. Meteorological data and operational event data for the period to be predicted are input into the target wind power prediction model, and the target wind power prediction model outputs the wind power prediction result.

2. The wind power prediction method based on event-driven variable state space according to claim 1, characterized in that, The available power for reconstructing counterfacts based on the normal reference set includes: Calculate the Euclidean distance between the state features at the current sampling time and each sample in the normal reference set, and select the closest one. The nearest neighbors are used to reconstruct the counterfactual available power based on Euclidean distance, time interval, wind speed physical correction relationship, and local power residual compensation term. The expression is as follows: ， In the formula, Power available for counterfactual purposes For the number of nearest neighbors, For the current sampling time and the first Euclidean distance between the nearest neighbors For the current sampling time and the first Euclidean distance between the nearest neighbors From the current sampling time to The average Euclidean distance of the nearest neighbors For the current sampling time and the first The time interval between nearest neighbor samples For time decay scale parameters, For the first The normal reference power of the nearest neighbor, for The average of the nearest neighbor normal reference power, The wind speed at the current sampling time. For the first The wind speed corresponding to the nearest neighbor sample This is the wind speed power-law adaptive adjustment coefficient. This is the local power residual compensation coefficient. The upper limit of the counterfactual usable power. It is a very small positive number. It is the hyperbolic tangent function. This is the amplitude limiting function.

3. The wind power prediction method based on event-driven variable state space according to claim 1, characterized in that, The expression for calculating the inhibition coefficient is as follows: ， In the formula, The suppression coefficient, For actual measured power, Power available for counterfactual purposes The state code at the current sampling time, For state coding The corresponding reference inhibition level, From the current sampling time to The average distance of the nearest neighbors These are the state prior correction coefficients. The power deviation response coefficient is... The power deviation trigger threshold, The nearest neighbor matching attenuation coefficient, For distance scale parameters, It is a very small positive number. This is a limiting function; The expression for calculating the confidence level is: ， In the formula, For confidence level, Encoding the state at the current sampling time The basis for the decision is the credibility coefficient. From the current sampling time to The average distance of the nearest neighbors For the number of nearest neighbors, For the current sampling time and the first Euclidean distance between the nearest neighbors The suppression coefficient, The average nearest neighbor distance is the scale parameter. The nearest neighbor distance dispersion scale parameter, To suppress the uniformity response coefficient, To suppress the allowable deviation threshold for consistency.

4. The wind power prediction method based on event-driven variable state space according to claim 1, characterized in that, The state modulation parameters include state gating parameters. State scaling parameters and state bias parameters ; The variable state-space recursive module performs the following adaptive update at the t-th prediction step: ， ， In the formula, For the future weather input at the t-th prediction step, For the first The candidate hidden states generated in each prediction step are used to represent the temporary state update results obtained under the influence of current and future meteorological inputs and state modulation parameters. For the first The updated hidden state output from each prediction step is used to propagate to the next prediction step and for subsequent power prediction. For the first The historical hidden states output by each prediction step Let be the state gating parameters for the t-th prediction step. The state scaling parameter for the t-th prediction step. , For learnable parameters, Let be the state bias parameter for the t-th prediction step.

5. The wind power prediction method based on event-driven variable state space according to claim 4, characterized in that, The event condition parameter generator generates state gating parameters. The expression is: ， In the formula, For the Sigmoid function, For event condition sequence, , These are learnable parameters; The state scaling parameters are generated using a learnable parameter generation network based on the event condition sequence. and the state bias parameters The expression is: ， In the formula, Generate networks for learnable parameters.

6. The wind power prediction method based on event-driven variable state space according to claim 3, characterized in that, The multi-task training employs a confidence-weighted loss function: ， In the formula, The confidence-weighted loss function, For the confidence tensor, , , These are the predicted values ​​for observed power, available power, and suppression coefficient, respectively. , , These are the true values ​​of observed power, available power, and suppression coefficient, respectively. Encoding the state at the current sampling time. Encoding the state at the current sampling time The defined task coupling weight matrix is ​​used to characterize the coupling relationship between observed power, available power, suppression coefficient, and power consistency constraints. The confidence level threshold is used for the decision-making process. This is the confidence level adjustment coefficient. , where is the confidence level gating steepness coefficient. The robust smoothing coefficient is... It is a very small positive number. For element-wise multiplication, To calculate the average of the sample dimension and the prediction step dimension.

7. A wind power prediction system based on event-driven variable state space, characterized in that, include: The acquisition module is configured to acquire raw wind power operation data, identify the operation status at each sampling time, and obtain the status code; The annotation module is configured to construct a normal reference set based on normal state samples, and perform inverse fact re-annotation on each sampling time according to the state code to obtain enhanced supervision information containing available power, suppression coefficient, and confidence level, specifically including: Under normal conditions, the measured power is directly taken as the usable power. For abnormal states, the counterfactual available power is reconstructed based on the normal reference set; The suppression coefficient is determined based on the measured power and the counterfactual available power, the confidence level is determined based on the operating status and the nearest neighbor distance, and the measured power is expanded into the enhanced supervision information based on the suppression coefficient and the confidence level. The construction module is configured to build a wind power prediction model for receiving historical input sequences, future meteorological driving sequences, and event condition sequences extracted from raw data. The wind power prediction model includes an event condition parameter generator and a variable state space recursion module. The event condition parameter generator generates state modulation parameters based on the event condition sequences, and the variable state space recursion module uses the state modulation parameters to adaptively recursively update the hidden states. The training module is configured to perform multi-task training on the wind power prediction model using the enhanced supervision information to obtain the target wind power prediction model, so that the wind power prediction model can simultaneously output the predicted values ​​of observed power, available power and suppression coefficient. The output module is configured to input meteorological data and operational event data for the period to be predicted into the target wind power prediction model, and the target wind power prediction model outputs the wind power prediction result.

8. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method described in any one of claims 1 to 6.

Images