A wind power prediction method based on multi-time gradient guidance and constraint

Abstract

The application discloses a wind power prediction method based on multi-time gradient guidance and constraint, which comprises the following steps: preprocessing wind power data, dividing the wind power data according to minute, hour and day physical scales to obtain decoupled short-term, medium-term and long-term data sequences, and constructing power feature matrices of the short-term, medium-term and long-term data sequences respectively. Then, the long-term feature matrix is input into a BiGRU to obtain a long-term power prediction sequence. Then, a multi-time gradient guidance-constraint function capable of dynamically quantifying cross-scale weights is constructed, and the function is embedded in the BiGRU network as a cross-scale coupling layer to form an MTGC-BiGRU. Finally, cascade prediction and correction are carried out: the long-term prediction sequence is taken as a trend guide, and the long-term prediction sequence and a medium-term feature matrix are input into the MTGC-BiGRU to obtain a medium-term prediction result; and then the medium-term prediction result is taken as a fluctuation envelope, and the medium-term prediction result and a short-term feature matrix are input into the MTGC-BiGRU to obtain a final short-term power prediction sequence.

Description

Technical Field

[0001] This invention relates to the field of offshore wind power prediction, and more specifically, to a wind power prediction method based on multi-time gradient guidance and constraints. Background Technology

[0002] This invention relates to the prediction of offshore wind power, and more particularly to a method and system for predicting offshore wind power using a multi-time gradient guided-constraint function (MTGC) and a bidirectional gated cyclic unit network (BiGRU).

[0003] Offshore wind power has become a core development direction in the renewable energy field due to its advantages such as abundant resources, high power generation efficiency, and no occupation of land resources. Moreover, as an important component of clean energy, the power prediction accuracy of offshore wind power is crucial for grid dispatch, absorption, and safe and stable operation.

[0004] Offshore wind power is affected by complex factors such as sea breeze, waves, and tides, exhibiting characteristics of short-term instantaneous data changes (such as minute-level gust disturbances), medium-term data periodic changes (such as intraday sea and land wind alternation), and long-term data trend evolution (such as seasonal climate migration). By comparison, it can be found that as the sampling period decreases, wind power data is often accompanied by more violent fluctuations and abrupt changes. These instantaneous changes are usually difficult for prediction models to accurately capture and learn, resulting in large errors in the power prediction results.

[0005] Offshore wind power data is typically collected using different sampling periods depending on actual needs, mainly including short-term data at the minute level, medium-term data at the hour level, and long-term data at the daily level. Short-term data at the minute level is primarily dominated by instantaneous gusts and turbulence, with fluctuations reaching 20%–40% of rated power. For example, during a typhoon, power can plummet from full capacity to zero within 10 minutes. These dramatic instantaneous fluctuations are often difficult for prediction models to capture effectively. Medium-term data at the hour level is mainly affected by sea-land wind circulation and solar radiation variations, exhibiting periodic fluctuations. For instance, stronger sea winds during the day lead to increased power, while weaker land winds at night lead to decreased power, and the period length varies seasonally. Prediction models generally have limited effectiveness in capturing this data. Long-term data at the daily level is mainly dominated by atmospheric circulation (such as monsoons) and climate change, with relatively gentle overall changes. It primarily determines the long-term trend of wind power, so prediction models can relatively easily capture and learn the patterns of change. However, traditional error correction models often focus on a single time scale, performing post-processing on a single prediction result (such as residual compensation and moving averages), neglecting the inherent characteristics and interrelationships of data at different time gradients.

[0006] Chinese invention application No. 202411277379.0 discloses "A Multi-Timescale Wind Power Prediction Method and System," which includes: collecting meteorological data, seasonally decomposing the meteorological data, and making seasonal adjustments after decomposition; then constructing a hybrid model using statistical models, physical models, and machine learning models; using a SARIMA model for seasonal time series prediction; and combining the SARIMA model and the hybrid model; further optimizing the prediction model and algorithm parameters; compared with existing technologies, it can effectively reduce the impact of seasonality on prediction and improve prediction accuracy; the hybrid model includes: a physical model prediction model, a statistical model prediction model, and a machine learning model prediction model; the hybrid model combines the advantages of physical models, statistical models, and machine learning models, and improves the overall prediction accuracy of the multi-timescale wind power prediction method and system by integrating the prediction results of different models. Summary of the Invention

[0007] To address the technical problem of large short-to-medium-term prediction errors in existing offshore wind power prediction methods due to neglecting the characteristics of multi-timescale data and cross-scale correlations, this invention provides a wind power prediction method based on multi-time gradient guidance and constraints. The technical solution adopted by this invention is as follows:

[0008] The first aspect of this invention provides a wind power prediction method based on multi-time gradient guidance and constraints, comprising the following steps: S1: Acquire wind power data from the target wind farm and perform preprocessing; S2: Based on the physical time-frequency distribution of wind power data fluctuation characteristics, the processed data is divided into sampling periods according to minute-level turbulence scale, hour-level daily variation scale, and daily-level climate scale to obtain short-term, medium-term, and long-term wind power data with decoupled physical meaning. S3: Construct corresponding power feature matrices for short-term, medium-term, and long-term wind power data respectively; S4: Input the long-term power feature matrix into a bidirectional gated recurrent unit network to capture long-term climate evolution features and obtain a long-term power prediction sequence containing low-frequency trend information; S5: Based on the ratio of sampling frequency and energy transfer characteristics under different time gradients, construct a multi-time gradient guidance-constraint function that can dynamically quantify cross-scale coupled weights; S6: The multi-temporal gradient guidance-constraint function is embedded as a cross-scale coupling layer in the output of the bidirectional gated recurrent unit network to construct an MTGC-BiGRU network with a dual-channel guidance-constraint architecture. S7: The long-term power prediction sequence is used as a global trend guide operator and input into the MTGC-BiGRU network together with the mid-term power feature matrix. The mid-term prediction baseline is corrected using the multi-time gradient guide-constraint function to obtain the corrected mid-term power prediction sequence. S8: The modified medium-term power prediction sequence is used as a local fluctuation envelope constraint and input together with the short-term power feature matrix into the MTGC-BiGRU network. The short-term prediction result is constrained within a reasonable fluctuation range by the multi-time gradient guidance-constraint function to obtain the modified short-term power prediction sequence.

[0009] As a preferred embodiment, in step S1, the wind power data includes wind speed, wind direction, temperature, humidity, and wind power data at a preset altitude.

[0010] As a preferred embodiment, in step S1, the preprocessing method includes: First, the wind power data of the target wind farm is cleaned to remove outliers, and missing values ​​are filled in using linear interpolation. Second, the wind speed, wind direction, temperature, humidity and wind power sequences are normalized using min-max to obtain the corresponding wind speed sequence Ws, wind direction sequence Wd, temperature sequence Tem, humidity sequence Hum and wind power sequence P.

[0011] As a preferred embodiment, in step S3, the method for constructing corresponding power feature matrices for short-term, medium-term, and long-term wind power data includes: Define the following matrix:

[0012] in Indicates the sampling period is T The power characteristic matrix, Indicates the first tn A matrix composed of features at each time step. The specific expression is:

[0013] in, , , , and These represent sampling periods of 1 and 2 respectively. T The wind power data includes power, wind speed, wind direction, temperature, and humidity; based on The expression can then be used to construct the power feature matrix corresponding to short-, medium-, and long-term wind power data. , , .

[0014] As a preferred embodiment, in step S4, the method for inputting the long-term power feature matrix into a bidirectional gated recurrent unit network to capture long-term climate evolution characteristics and obtain a long-term power prediction sequence containing low-frequency trend information includes: The structure of a single gated loop unit is defined as follows:

[0015] in, To reset the door, To update the door, For the Sigmod function, , , This is the weight matrix. , , The bias matrix, This represents the candidate hidden state at the current time step. Hide the state at the current time step. This represents the hidden state of the previous time step, and ⊙ represents element-wise multiplication. The bidirectional gated recurrent unit network consists of forward and reverse gated recurrent units. The forward and reverse gated recurrent units process the forward and reverse information of the sequence respectively, and finally, the hidden states of the two are concatenated to obtain the bidirectional features. The structure of the bidirectional gated recurrent unit network is as follows: in , These are forward and reverse gated loop units, respectively. , These represent the positive hidden states at the current time step and the previous time step, respectively. , These are the reverse hidden states for the current time step and the previous time step, respectively. This is the final hidden state after concatenating the forward and reverse hidden states; After constructing the bidirectional gated cyclic unit network, with Using the input as the input and the activation functions as sigmoid and tanh, the output is the long-term power prediction sequence. .

[0016] As a preferred embodiment, in step S5, the multi-time gradient guided-constraint function is as follows:

[0017] in , , These represent the sampling periods for long-term, medium-term, and short-term data, respectively. , These represent the long-term and medium-term time gradient ratios, respectively, characterizing the energy density decay gradient at different time scales. They are used to quantify the constraint weight of the previous level trend on the current level fluctuation. , , They represent in t Power prediction values ​​for long-term, medium-term, and short-term data. , They represent in t Time-adjusted medium- and short-term power forecasts.

[0018] As a preferred embodiment, in step S6, the structure of the MTGC-BiGRU network with a dual-channel guide-constraint architecture is as follows:

[0019] in, F The power characteristic matrix, For the initial power prediction sequence through a bidirectional gated cyclic cell network, For guidance Corrected guiding power sequence, This is the corrected power prediction sequence.

[0020] As a preferred embodiment, in step S7, the long-term power prediction sequence is used as a global trend guiding operator and input into the MTGC-BiGRU network along with the medium-term power feature matrix. The baseline of the medium-term prediction is corrected using the multi-temporal gradient guiding-constraint function, resulting in the following expression for the corrected medium-term power prediction sequence:

[0021] in, This is a medium-term power prediction sequence. This is the corrected medium-term power prediction sequence.

[0022] As a preferred embodiment, in step S8, the corrected medium-term power prediction sequence is used as a local fluctuation envelope constraint and input into the MTGC-BiGRU network along with the short-term power feature matrix. The multi-temporal gradient guided-constraint function is used to constrain the short-term prediction results within a reasonable fluctuation range, resulting in the following expression for the corrected short-term power prediction sequence:

[0023] in, This is a short-term power prediction sequence. This is the corrected short-term power prediction sequence.

[0024] A second aspect of the present invention provides a computer device, including a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor, wherein the computer program, when executed by the processor, implements the steps of the aforementioned wind power prediction method based on multi-time gradient guidance and constraints.

[0025] Compared with the prior art, the beneficial effects of this invention are: This invention divides wind power data into three scales—minute, hour, and day—based on physical time-frequency distribution and constructs corresponding power feature matrices. This achieves decoupling and accurate modeling of fluctuation characteristics at different scales, such as turbulent abrupt changes, periodic variations, and climate trends, laying a clear data foundation for subsequent hierarchical predictions and overcoming the drawbacks of single models that cannot adequately handle multi-scale features. By designing a multi-temporal gradient-guided constraint function that can dynamically quantify cross-scale constraint weights and embedding it as a "cross-scale coupling layer" within a BiGRU network, an MTGC-BiGRU dual-channel architecture is constructed. This architecture utilizes long-term trends to guide and correct the mid-term prediction baseline, and then uses the mid-term envelope to constrain the short-term prediction fluctuation range, forming a cascaded correction mechanism that effectively suppresses the cumulative prediction errors caused by drastic fluctuations and periodic variations. By cascading the MTGC-BiGRU network to medium- and short-term forecasts, and injecting information from the more predictable higher-level scale as a priori constraints into the lower-level scale, the final medium-term forecasts are calibrated in a trend-based manner, while the short-term forecasts are constrained within a reasonable physical range while retaining necessary high-frequency details. This significantly improves the overall accuracy and robustness of medium- and short-term wind power forecasts under complex sea conditions. Attached Figure Description

[0026] Figure 1 This embodiment provides a flowchart of a wind power prediction method based on multi-time gradient guidance and constraints. Figure 2 This is a comparison chart of the uncorrected interim data prediction results provided in this embodiment; Figure 3 This is a comparison chart of the corrected interim data prediction results provided in this embodiment; Figure 4 A comparison chart of the uncorrected short-term data prediction results provided in this embodiment; Figure 5 This is a comparison chart of the corrected short-term data prediction results provided in this embodiment. Detailed Implementation

[0027] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. It should be understood that the described embodiments are merely some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of the embodiments of this application.

[0028] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0029] In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. In the description of this application, it should be understood that the terms "first," "second," "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0030] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. The invention will be further described below with reference to the accompanying drawings and embodiments.

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] Example 1 Please refer to Figure 1 This embodiment provides a wind power prediction method based on multi-time gradient guidance and constraints, including the following steps: S1: Acquire wind power data from the target wind farm and perform preprocessing; In one specific embodiment, in step S1, the wind power data includes wind speed, wind direction, temperature, humidity, and wind power data at a preset altitude.

[0033] In one specific embodiment, in step S1, the preprocessing method includes: First, the wind power data of the target wind farm is cleaned to remove outliers, and missing values ​​are filled in using linear interpolation. Second, the wind speed, wind direction, temperature, humidity and wind power sequences are normalized using min-max to obtain the corresponding wind speed sequence Ws, wind direction sequence Wd, temperature sequence Tem, humidity sequence Hum and wind power sequence P.

[0034] S2: Based on the physical time-frequency distribution of wind power data fluctuation characteristics, the processed data is divided into sampling periods according to minute-level turbulence scale, hour-level daily variation scale, and daily-level climate scale to obtain short-term, medium-term, and long-term wind power data with decoupled physical meaning. Specifically, the process of dividing the processed data according to different sampling periods is as follows: Set the short-term sampling period to Historical data according to The data is divided into segments to generate short-term time series at the minute level; the intermediate sampling period is set to [value missing]. Historical data according to The data is divided into segments to generate hourly-level medium-term time series; the long-term sampling period is set to [value missing]. Historical data according to The data is divided into segments to generate daily-level long-term time series.

[0035] S3: Construct corresponding power feature matrices for short-term, medium-term, and long-term wind power data respectively; In a specific embodiment, the method for constructing corresponding power feature matrices for short-term, medium-term, and long-term wind power data in step S3 includes: Define the following matrix:

[0036] in Indicates the sampling period is T The power characteristic matrix, Indicates the first tn A matrix composed of features at each time step. The specific expression is:

[0037] in, , , , and These represent sampling periods of 1 and 2 respectively. T The wind power data includes power, wind speed, wind direction, temperature, and humidity; based on The expression can then be used to construct the power feature matrix corresponding to short-, medium-, and long-term wind power data. , , .

[0038] S4: Input the long-term power feature matrix into a bidirectional gated recurrent unit network to capture long-term climate evolution features and obtain a long-term power prediction sequence containing low-frequency trend information; In a specific embodiment, in step S4, the method for inputting the long-term power feature matrix into a bidirectional gated recurrent unit network to capture long-term climate evolution characteristics and obtain a long-term power prediction sequence containing low-frequency trend information includes: Define a single gated loop unit ( GRU The structure of ) is as follows:

[0039] in, To reset the door, To update the door, For the Sigmod function, , , This is the weight matrix. , , The bias matrix, This represents the candidate hidden state at the current time step. Hide the state at the current time step. This represents the hidden state of the previous time step, and ⊙ represents element-wise multiplication. The bidirectional gated recurrent unit network consists of forward and reverse gated recurrent units. The forward and reverse gated recurrent units process the forward and reverse information of the sequence respectively, and finally, the hidden states of the two are concatenated to obtain the bidirectional features. The structure of the bidirectional gated recurrent unit network is as follows: in , These are forward and reverse gated loop units, respectively. , These represent the positive hidden states at the current time step and the previous time step, respectively. , These are the reverse hidden states for the current time step and the previous time step, respectively. This is the final hidden state after concatenating the forward and reverse hidden states; After constructing the bidirectional gated cyclic unit network, with Using the input as the input and the activation functions as sigmoid and tanh, the output is the long-term power prediction sequence. .

[0040] S5: Based on the ratio of sampling frequency and energy transfer characteristics under different time gradients, construct a multi-time gradient guidance-constraint function that can dynamically quantify cross-scale coupled weights; In a specific embodiment, in step S5, the multi-temporal gradient guidance-constraint function is as follows:

[0041] in , , These represent the sampling periods for long-term, medium-term, and short-term data, respectively. , These represent the long-term and medium-term time gradient ratios, respectively, characterizing the energy density decay gradient at different time scales. They are used to quantify the constraint weight of the previous level trend on the current level fluctuation. , , They represent in t Power prediction values ​​for long-term, medium-term, and short-term data. , They represent in t Time-adjusted medium- and short-term power forecasts.

[0042] S6: The multi-temporal gradient guidance-constraint function is embedded as a cross-scale coupling layer in the output of the bidirectional gated recurrent unit network to construct an MTGC-BiGRU network with a dual-channel guidance-constraint architecture. In one specific embodiment, in step S6, the structure of the MTGC-BiGRU network with a dual-channel guide-constraint architecture is as follows:

[0043] in, F The power characteristic matrix, For the initial power prediction sequence through a bidirectional gated cyclic cell network, For guidance Corrected guiding power sequence, This is the corrected power prediction sequence.

[0044] S7: The long-term power prediction sequence is used as a global trend guide operator and input into the MTGC-BiGRU network together with the mid-term power feature matrix. The mid-term prediction baseline is corrected using the multi-time gradient guide-constraint function to obtain the corrected mid-term power prediction sequence. In a specific embodiment, in step S7, the long-term power prediction sequence is used as a global trend guiding operator and input into the MTGC-BiGRU network along with the medium-term power feature matrix. The baseline of the medium-term prediction is corrected using the multi-temporal gradient guiding-constraint function, and the expression of the corrected medium-term power prediction sequence is as follows:

[0045] in, This is a medium-term power prediction sequence. This is the corrected medium-term power prediction sequence.

[0046] S8: The modified medium-term power prediction sequence is used as a local fluctuation envelope constraint and input together with the short-term power feature matrix into the MTGC-BiGRU network. The multi-time gradient guidance-constraint function is used to constrain the short-term prediction result within a reasonable fluctuation range to obtain the modified short-term power prediction sequence. In a specific embodiment, in step S8, the corrected medium-term power prediction sequence is used as a local fluctuation envelope constraint and input into the MTGC-BiGRU network along with the short-term power feature matrix. The multi-temporal gradient guided-constraint function is used to constrain the short-term prediction results within a reasonable fluctuation range, resulting in the following expression for the corrected short-term power prediction sequence:

[0047] in, This is a short-term power prediction sequence. This is the corrected short-term power prediction sequence.

[0048] Example 2 This embodiment provides a wind power prediction method based on multi-time gradient guidance and constraints, including the following steps: S1: Acquire wind power data from the target wind farm and perform preprocessing; In one specific embodiment, in step S1, measured wind power data from a wind farm abroad is acquired. The wind power data has five dimensions, including: wind power output, wind speed, wind direction, temperature, and humidity. S2: Based on the physical time-frequency distribution of wind power data fluctuation characteristics, the processed data is divided into sampling periods according to minute-level turbulence scale, hour-level daily variation scale, and daily-level climate scale to obtain short-term, medium-term, and long-term wind power data with decoupled physical meaning. In one specific embodiment, in step S2, the sampling period for minute-level turbulence scale is set to 15 min, the sampling period for hour-level diurnal variation scale is set to 1 h, and the sampling period for day-level climate scale is set to 24 h.

[0049] S3: Construct corresponding power feature matrices for short-term, medium-term, and long-term wind power data respectively; S4: Input the long-term power feature matrix into a bidirectional gated recurrent unit network to capture long-term climate evolution features and obtain a long-term power prediction sequence containing low-frequency trend information; S5: Based on the ratio of sampling frequency and energy transfer characteristics under different time gradients, construct a multi-time gradient guidance-constraint function that can dynamically quantify cross-scale coupled weights; S6: The multi-temporal gradient guidance-constraint function is embedded as a cross-scale coupling layer in the output of the bidirectional gated recurrent unit network to construct an MTGC-BiGRU network with a dual-channel guidance-constraint architecture. S7: The long-term power prediction sequence is used as a global trend guide operator and input into the MTGC-BiGRU network together with the mid-term power feature matrix. The mid-term prediction baseline is corrected using the multi-time gradient guide-constraint function to obtain the corrected mid-term power prediction sequence. S8: The modified medium-term power prediction sequence is used as a local fluctuation envelope constraint and input together with the short-term power feature matrix into the MTGC-BiGRU network. The short-term prediction result is constrained within a reasonable fluctuation range by the multi-time gradient guidance-constraint function to obtain the modified short-term power prediction sequence.

[0050] Please refer to Figure 2 , Figure 3 , Figure 4 as well as Figure 5 ,from Figure 2 , Figure 3 , Figure 4 as well as Figure 5 As can be seen from the prediction results, the wind power prediction method based on multi-time gradient guidance and constraint provided by this invention can effectively improve the prediction accuracy of wind power.

[0051] Example 3 This embodiment provides a computer device, including a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, it implements the steps of the wind power prediction method based on multi-time gradient guidance and constraint as described in Embodiment 1 or Embodiment 2.

[0052] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A wind power prediction method based on multi-time gradient guidance and constraints, characterized in that, Includes the following steps: S1: Acquire wind power data from the target wind farm and perform preprocessing; S2: Based on the physical time-frequency distribution of wind power data fluctuation characteristics, the processed data is divided into sampling periods according to minute-level turbulence scale, hour-level daily variation scale, and daily-level climate scale to obtain short-term, medium-term, and long-term wind power data with decoupled physical meaning. S3: Construct corresponding power feature matrices for short-term, medium-term, and long-term wind power data respectively; S4: Input the long-term power feature matrix into a bidirectional gated recurrent unit network to capture long-term climate evolution features and obtain a long-term power prediction sequence containing low-frequency trend information; S5: Based on the ratio of sampling frequency and energy transfer characteristics under different time gradients, construct a multi-time gradient guidance-constraint function that can dynamically quantify cross-scale coupled weights; S6: The multi-temporal gradient guidance-constraint function is embedded as a cross-scale coupling layer in the output of the bidirectional gated recurrent unit network to construct an MTGC-BiGRU network with a dual-channel guidance-constraint architecture. S7: The long-term power prediction sequence is used as a global trend guide operator and input into the MTGC-BiGRU network together with the mid-term power feature matrix. The mid-term prediction baseline is corrected using the multi-time gradient guide-constraint function to obtain the corrected mid-term power prediction sequence. S8: The modified medium-term power prediction sequence is used as a local fluctuation envelope constraint and input together with the short-term power feature matrix into the MTGC-BiGRU network. The short-term prediction result is constrained within a reasonable fluctuation range by the multi-time gradient guidance-constraint function to obtain the modified short-term power prediction sequence.

2. The wind power prediction method based on multi-time gradient guidance and constraint according to claim 1, characterized in that, In step S1, the wind power data includes wind speed, wind direction, temperature, humidity, and wind power data at a preset altitude.

3. The wind power prediction method based on multi-time gradient guidance and constraint according to claim 2, characterized in that, In step S1, the preprocessing method includes: First, the wind power data of the target wind farm is cleaned to remove outliers, and missing values ​​are filled in using linear interpolation. Second, the wind speed, wind direction, temperature, humidity and wind power sequences are normalized using min-max to obtain the corresponding wind speed sequence Ws, wind direction sequence Wd, temperature sequence Tem, humidity sequence Hum and wind power sequence P.

4. The wind power prediction method based on multi-time gradient guidance and constraint according to claim 1, characterized in that, In step S3, the method for constructing corresponding power feature matrices for short-term, medium-term, and long-term wind power data includes: Define the following matrix: in Indicates the sampling period is T The power characteristic matrix, Indicates the first tn A matrix composed of features at each time step. The specific expression is: in, , , , and These represent sampling periods of 1 and 2 respectively. T The wind power data includes power, wind speed, wind direction, temperature, and humidity; based on The expression can then be used to construct the power feature matrix corresponding to short-, medium-, and long-term wind power data. , , .

5. The wind power prediction method based on multi-time gradient guidance and constraints according to claim 1, characterized in that, In step S4, the method of inputting the long-term power feature matrix into a bidirectional gated recurrent unit network to capture long-term climate evolution characteristics and obtain a long-term power prediction sequence containing low-frequency trend information includes: The structure of a single gated loop unit is defined as follows: in, To reset the door, To update the door, For the Sigmod function, , , This is the weight matrix. , , The bias matrix, This represents the candidate hidden state at the current time step. Hide the state at the current time step. This represents the hidden state of the previous time step, and ⊙ represents element-wise multiplication. The bidirectional gated recurrent unit network consists of forward and reverse gated recurrent units. The forward and reverse gated recurrent units process the forward and reverse information of the sequence respectively, and finally, the hidden states of the two are concatenated to obtain the bidirectional features. The structure of the bidirectional gated recurrent unit network is as follows: in , These are forward and reverse gated loop units, respectively. , These represent the positive hidden states at the current time step and the previous time step, respectively. , These are the reverse hidden states for the current time step and the previous time step, respectively. This is the final hidden state after concatenating the forward and reverse hidden states; After constructing the bidirectional gated cyclic unit network, with Using the input as the input and the activation functions as sigmoid and tanh, the output is the long-term power prediction sequence. .

6. The wind power prediction method based on multi-time gradient guidance and constraint according to claim 1, characterized in that, In step S5, the multi-time gradient guided-constraint function is as follows: in , , These represent the sampling periods for long-term, medium-term, and short-term data, respectively. , These represent the long-term and medium-term time gradient ratios, respectively, characterizing the energy density decay gradient at different time scales. They are used to quantify the constraint weight of the previous level trend on the current level fluctuation. , , They represent in t Power prediction values ​​for long-term, medium-term, and short-term data. , They represent in t Time-adjusted medium- and short-term power forecasts.

7. The wind power prediction method based on multi-time gradient guidance and constraint according to claim 1, characterized in that, In step S6, the structure of the MTGC-BiGRU network with a dual-channel guide-constraint architecture is as follows: in, F The power characteristic matrix, For the initial power prediction sequence through a bidirectional gated cyclic cell network, For guidance Corrected guiding power sequence, This is the corrected power prediction sequence.

8. The wind power prediction method based on multi-time gradient guidance and constraint according to claim 1, characterized in that, In step S7, the long-term power prediction sequence is used as a global trend guide operator and input into the MTGC-BiGRU network along with the medium-term power feature matrix. The baseline of the medium-term prediction is corrected using the multi-temporal gradient guide-constraint function, resulting in the following expression for the corrected medium-term power prediction sequence: in, This is a medium-term power prediction sequence. This is the corrected medium-term power prediction sequence.

9. The wind power prediction method based on multi-time gradient guidance and constraints according to claim 1, characterized in that, In step S8, the corrected medium-term power prediction sequence is used as a local fluctuation envelope constraint and input into the MTGC-BiGRU network along with the short-term power feature matrix. The multi-temporal gradient guided-constraint function is used to constrain the short-term prediction results within a reasonable fluctuation range, resulting in the following expression for the corrected short-term power prediction sequence: in, This is a short-term power prediction sequence. This is the corrected short-term power prediction sequence.

10. A computer device, characterized in that: The method includes a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor, wherein the computer program, when executed by the processor, implements the steps of a wind power prediction method based on multi-time gradient guidance and constraint as described in any one of claims 1 to 9.

Images