Dynamic operation mechanism knowledge-based BiGRU-embedded wind power prediction method and system
By embedding wind speed-power dynamic operating curves and decision boundaries into the BiGRU model, the problem of neglecting physical constraints in traditional wind power prediction methods is solved, achieving higher prediction accuracy and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- QINGDAO UNIV OF TECH
- Filing Date
- 2026-01-14
- Publication Date
- 2026-04-21
AI Technical Summary
Traditional wind power prediction methods neglect the physical constraints and mechanistic characteristics of wind turbines, resulting in a lack of physical consistency and interpretability in the prediction results, and insufficient prediction accuracy under complex operating conditions.
By constructing dynamic wind speed-power operation curves and decision boundaries as dynamic operation mechanism knowledge, embedding them into the BiGRU model, constructing a composite loss function, and combining data-driven and operation mechanism constraints, implicit constraints of physical mechanisms are realized.
It improves the accuracy and physical interpretability of wind power prediction, enhances the model's adaptability under complex operating conditions, and improves the stability of prediction results.
Smart Images
Figure CN121901646A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of wind power prediction technology, and specifically relates to a wind power prediction method and system based on dynamic operation mechanism knowledge embedded in BiGRU. Background Technology
[0002] In recent years, renewable energy has developed rapidly. As a promising clean energy source, wind power plays an indispensable role in the renewable energy system. Despite the rapid development of wind power technology, the high randomness of wind speed leads to significant uncertainty in the power output of wind turbines, posing a serious challenge to the safe and stable operation of the power grid. Wind power forecasting plays a crucial role in effectively addressing these challenges. By forecasting power in advance, the negative impact of power fluctuations on grid stability can be reduced.
[0003] Traditional time-series wind power forecasting methods are mostly based on linear models, such as autoregressive, moving average, and autoregressive moving average models. These methods typically predict future power by characterizing the statistical dependencies between historical power data. However, wind turbines are inherently strong nonlinear systems, and factors such as wind speed disturbances, meteorological conditions, and operating status can all cause the output power to exhibit significant nonlinear characteristics. Therefore, the prediction accuracy of linear models under complex operating conditions is clearly limited.
[0004] With the development of machine learning technology, more and more nonlinear models are being introduced into the field of wind power prediction, such as artificial neural networks, convolutional neural networks, and recurrent neural networks. These models can capture nonlinear mapping relationships in historical data to a certain extent, and their prediction performance is significantly improved compared to traditional linear models. However, such methods are essentially data-driven 'black box' models. During training, they usually only aim to minimize the mean squared error (MSE), often ignoring the physical constraints and mechanistic characteristics of wind turbine operation. The power output of wind turbines is mainly determined by wind speed and its dynamic changes. If the model fails to consider this physical law, its prediction results may numerically approximate the real data, but they often lack physical consistency and interpretability, and may violate the operating characteristics of the turbine (including positive power below the cut-in wind speed, or power exceeding the rated wind speed without limitation). Therefore, purely data-driven prediction methods still have certain shortcomings in improving the physical credibility of the model, prediction accuracy, and maintaining the stability of grid operation.
[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0006] To address or at least alleviate one or more of the above problems, a wind power prediction method and system based on BiGRU embedded with dynamic operating mechanism knowledge is provided. The fitted dynamic power operation curve and wind speed-power decision boundary can be regarded as dynamic operating mechanism knowledge. By improving the loss function, the implicit constraints of the physical mechanism are realized, achieving an organic combination of data-driven and operating mechanism constraints. This effectively solves the problem that the prediction results of traditional models violate physical laws, and significantly improves the generalization ability and physical rationality of the model.
[0007] To achieve the above objectives, according to the first aspect of this application, a wind power prediction method based on dynamically embedded BiGRU knowledge is provided, comprising the following steps: Acquire the operating data of the wind turbine SCADA system, including wind speed and power; divide the preprocessed operating data into training set, validation set and test set in chronological order; Based on the operational data in the training set, a wind speed-power scatter plot is established. By using the data consistency principle, key trend points between wind speed and power are identified, and a dynamic wind speed-power operation curve is constructed. Based on the aforementioned wind speed-power dynamic operation curve, a wind speed-power decision boundary is constructed. The constructed wind speed-power dynamic operation curve and wind speed-power decision boundary are regarded as dynamic operation mechanism knowledge, and the dynamic operation mechanism knowledge is embedded into the BiGRU prediction model. The dynamic operation curve and decision boundary are used as physical loss to construct a composite loss function. The embedded BiGRU prediction model is trained using the training set, and the network parameters are optimized through backpropagation until the loss of the BiGRU prediction model converges. The trained BiGRU prediction model was applied to the test set for wind power prediction.
[0008] To achieve the above objectives, according to a second aspect of this application, a wind power prediction system based on dynamically embedded BiGRU knowledge is provided, the wind power prediction system comprising: The acquisition module is used to acquire the operating data of the wind turbine SCADA system, including wind speed and power; the preprocessed operating data is divided into training set, validation set and test set in chronological order; The module is used to build a wind speed-power scatter plot based on the running data in the training set, identify key trend points between wind speed and power through the data consistency principle, and build a dynamic wind speed-power running curve. Based on the aforementioned wind speed-power dynamic operation curve, a wind speed-power decision boundary is constructed. The training module is used to treat the constructed wind speed-power dynamic operating curve and wind speed-power decision boundary as dynamic operating mechanism knowledge, and to embed the dynamic operating mechanism knowledge into the BiGRU prediction model. The dynamic operating curve and decision boundary are used as physical loss to construct a composite loss function. The embedded BiGRU prediction model is trained using the training set, and the network parameters are optimized through backpropagation until the loss of the BiGRU prediction model converges. The application module is used to apply the trained BiGRU prediction model to the test set for wind power prediction.
[0009] To achieve the above objectives, according to a third aspect of this application, a computer-readable storage medium is provided, storing a computer program, characterized in that, when executed by a processor, the computer program is used to implement the wind power prediction method based on dynamically operating mechanism knowledge embedded in BiGRU as described above.
[0010] By adopting the above technical solution, this application has the following beneficial effects compared with the prior art: This application fully considers the underlying mechanisms of wind turbine operation and achieves an organic combination of data learning and dynamic operational mechanism constraints through joint optimization of data-driven losses and operational mechanism constraint losses. Compared with traditional BiGRU models without dynamic operational mechanism knowledge embedding, this method exhibits superior fitting ability and generalization performance when dealing with complex nonlinear conditions, thereby improving the accuracy and physical interpretability of wind power prediction.
[0011] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. Attached Figure Description
[0012] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application, but do not constitute an undue limitation of this application. Obviously, the drawings described below are merely some embodiments, and those skilled in the art can obtain other drawings based on these drawings without creative effort.
[0013] In the attached diagram: Figure 1 This is a flowchart illustrating the wind power prediction method based on embedding BiGRU with dynamic operating mechanism knowledge in this specific embodiment. Figure 2 This is a schematic diagram of the wind power data of an actual wind turbine unit over six months in this specific embodiment. Figure 3 This is a schematic diagram of the wind power training set after data preprocessing in this specific embodiment; Figure 4 This is a schematic diagram of the wind power test set after data preprocessing in this specific embodiment; Figure 5 This is a scatter plot of wind speed-power on the training set in this specific embodiment; Figure 6 This is a schematic diagram of the wind speed-power dynamic operation curve identified on the training set in this specific embodiment; Figure 7 This is a schematic diagram of the wind speed-power decision boundary identified on the training set in this specific embodiment; Figure 8 The prediction effect diagram of embedding dynamic operation mechanism knowledge into BiGRU in this specific implementation embodiment; Figure 9 This is a graph showing the prediction performance of a traditional BiGRU in this specific implementation. Detailed Implementation
[0014] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments will be clearly and completely described below with reference to the accompanying drawings. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.
[0015] Please see Figure 1 This application provides a wind power prediction method based on dynamic operating mechanism knowledge embedded in BiGRU, including the following steps: Acquire the operating data of the wind turbine SCADA system, including wind speed and power; divide the preprocessed operating data into training set, validation set and test set in chronological order; Based on the operational data in the training set, a wind speed-power scatter plot is established. By using the data consistency principle, key trend points between wind speed and power are identified, and a dynamic wind speed-power operation curve is constructed. Based on the aforementioned wind speed-power dynamic operation curve, a wind speed-power decision boundary is constructed. The constructed wind speed-power dynamic operation curve and wind speed-power decision boundary are regarded as dynamic operation mechanism knowledge, and the dynamic operation mechanism knowledge is embedded into the BiGRU prediction model. The dynamic operation curve and decision boundary are used as physical loss to construct a composite loss function. The embedded BiGRU prediction model is trained using the training set, and the network parameters are optimized through backpropagation until the loss of the BiGRU prediction model converges. The trained BiGRU prediction model was applied to the test set for wind power prediction.
[0016] It should be noted that the execution subject of the wind power prediction method based on dynamically embedded BiGRU based on operational mechanism knowledge in this embodiment is a wind power prediction system based on dynamically embedded BiGRU based on operational mechanism knowledge. This device can be an electronic device, a component in an electronic device, an integrated circuit, or a chip. The electronic device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, vehicle-mounted electronic devices, wearable devices, etc., and non-mobile electronic devices can be servers and personal computers, etc., which are not specifically limited in this application. The following description uses a server as the execution subject to illustrate the wind power prediction method based on dynamically embedded BiGRU based on operational mechanism knowledge in this embodiment.
[0017] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0018] In some embodiments, the power generation of wind turbines is used as the prediction target. Traditional prediction methods often rely on pure data-driven approaches, ignoring the inherent physical coupling between wind speed and power, resulting in limited model generalization ability and a lack of physical interpretability in the prediction results.
[0019] Therefore, this embodiment proposes a method for identifying key trend points in the wind speed-power curve based on the principle of data consistency. It uses PCHIP to fit the identified key trend points and identifies the dynamic operating curve of wind speed-power. Based on this, a decision boundary for wind speed-power is further constructed. The dynamic operating curve of wind speed-power and the decision boundary are considered as dynamic operating mechanism knowledge.
[0020] This embodiment aims to embed dynamic operating mechanism knowledge into the prediction model, enhancing the physical interpretability of the prediction process and the accuracy of the results. Introducing physically meaningful wind speed-power dynamic operating curves and decision boundaries as constraints helps guide the model to learn dynamic characteristics consistent with the wind turbine operating mechanism.
[0021] However, since measured data from SCADA systems often contain noise and outliers, directly fitting a robust and reasonable wind speed-power dynamic operating curve presents a significant challenge. This embodiment divides wind speed data into intervals, extracts representative trend points using the median within each interval, and verifies these trend points using data consistency principles to obtain key trend points. A smooth operating curve is fitted using PCHIP, while a constant power segment is set for intervals above the rated wind speed, thereby statistically characterizing a stable mapping relationship between wind speed and power and maintaining physical consistency. Based on the constructed dynamic power operating curve, decision boundaries are further constructed. Specifically, the residuals from the power values near each key trend point to the dynamic power operating curve are statistically analyzed using the quartile method to obtain decision points (including upper and lower decision points). These decision points are then fitted again using PCHIP to construct the upper and lower boundaries.
[0022] This embodiment further embeds the established dynamic operating mechanism knowledge into BiGRU. This method achieves an organic integration of data-driven learning and dynamic operating mechanism constraints, which not only improves the model's adaptability to complex working conditions but also enhances the rationality and stability of the prediction results.
[0023] Please see Figure 1 and Figure 2 To obtain the operating data of the wind turbine SCADA system, including wind speed and power.
[0024] Wind power data such as Figure 2 As shown, first check the integrity of the running data, and record missing values as Nan. Then, use the sliding window quartile method to detect outliers in the running data and mark them as Nan. Finally, use XGBoost to correct all Nan values.
[0025] For example, the sliding window quartile method used refers to: ; ; ; ; ; ; ; in, It is the window size, taken as 1% of the data volume. It is the wind speed value at time t. It is the set of wind speed values within the t window. It is the 25th value within the t window. This is the 75th value within the t window, where k is a scaling constant, set to 1.5. It is the interquartile range within the t window. It is the upper boundary within the t window. It is the lower boundary of the t window.
[0026] In some embodiments, see Figure 1 , Figure 3 and Figure 4 The preprocessed runtime data is divided into training set, validation set and test set in chronological order.
[0027] The preprocessed data is divided according to time sequence to ensure the continuity of the time series data and the scientific rationality of the prediction process. For example, such as... Figure 3 As shown, the first 60% of the samples are used as the training set to identify dynamic operating mechanisms and train the prediction model; the last 20% of the samples are used as the validation set, mainly for monitoring and adjusting the model's hyperparameters; as shown... Figure 4 As shown, the last 20% of the samples were used as a test set to evaluate the generalization ability of the proposed model.
[0028] In some embodiments, see Figure 1 and Figure 5 Based on the operational data in the training set, a wind speed-power scatter plot is established. By using the data consistency principle, key trend points between wind speed and power are identified, and a dynamic wind speed-power operation curve is constructed.
[0029] like Figure 5 As shown, a wind speed-power scatter plot is created, and key trend points between wind speed and power are identified based on the principle of data consistency; then, as... Figure 6 As shown, the PCHIP method is used to construct the dynamic operating curve of wind speed-power.
[0030] For example, the data consistency principle includes: Within each 1 m / s wind speed interval, the median of all power values within the wind speed interval is first calculated, and the median and the center value of the wind speed interval corresponding to the median are used as preliminary representative trend points. For each wind speed range, more refined local data sampling is performed near the ±0.1 m / s wind speed value corresponding to the representative trend point of the wind speed range, and the median power in the local range is recalculated to achieve local correction and refinement of the trend point. Based on the locally corrected power value, a power tolerance range of ±10% is constructed, and all data points whose power falls within the power tolerance range are screened out from the global data of the training set, and then the median wind speed corresponding to these data points is calculated. Determine whether the difference between the median wind speed obtained from the global reverse lookup and the wind speed value of the preliminary representative trend point is less than 1 m / s; if yes, then determine the locally corrected trend point as the final key trend point; if not, then retain the preliminary representative trend point as the key trend point.
[0031] For example, the PCHIP used refers to: The data consistency principle identifies key trend points on the training set. PCHIP interpolation was used to fit the dynamic operating curve of wind speed-power.
[0032] Specifically, the mathematical construction of PCHIP interpolation is as follows: Define the interval length and the difference quotient: ; ; Define local normalization variables: ; Hermite cubic basis functions: ; In the interval The interpolation function is as follows: ; This function satisfies: ; ; ; ; in, It is the wind speed value at the i-th key trend point. It is the power value of the i-th key trend point. It is the interval length between two adjacent key trend points. It is in the interval The difference in price It is a locally normalized variable. , , , These are Hermite cubic basis functions, used to construct fixed mathematical expressions for smooth piecewise cubic functions. It is in the i-th interval Piecewise interpolation function, At key trend points ( The first derivative of the interpolation function at point ().
[0033] In some embodiments, see Figure 7 Based on the dynamic power operation curve, a wind speed-power decision boundary is established. First, the power residuals from points near the key trend points to the dynamic power operation curve are calculated. Then, the quartile method is used on the residuals to obtain the decision points, including the upper and lower decision points. However, the decision points need to be verified to ensure that the decision points are monotonically constant. Finally, PCHIP is used to fit the decision boundary.
[0034] For example, the decision boundary construction method refers to: Calculate local residuals: Traverse the key trend points of the dynamic power operation curve. For the wind speed value corresponding to each trend point, select data points in the training set whose wind speed values are within ±0.25 m / s of the trend point, and calculate the residual between the actual power of these data points and the predicted power of the dynamic power operation curve. Determine the decision points at the top and bottom: For the local residual set corresponding to each key trend point, the upper and lower bounds of the residuals are determined using the quartile method to ensure that the decision points are monotonically non-decreasing. The PCHIP interpolation method is then used to smooth these decision points, generating a continuous and non-negative decision curve.
[0035] Decision point monotonicity verification and adjustment: Check the generated decision point sequence to ensure that the power values of both the upper and lower decision points remain monotonically unchanged as the wind speed increases; Decision boundary fitting: The PCHIP method was used to interpolate and fit the adjusted upper and lower decision point sequences to generate continuous and smooth upper and lower decision boundary curves, which together constitute the wind speed-power decision boundary.
[0036] In some embodiments, to enhance the physical interpretability of the model, a loss function is further designed, comprising a dynamic operating mechanism knowledge data loss term, a dynamic operating curve constraint term, a wind speed-power decision boundary constraint term, and a composite loss function; enabling the model to not only fit the observed data but also satisfy the physical laws of the wind speed-power curve. The dynamic operation mechanism knowledge data loss term is represented as follows: ; The dynamic running curve constraint term is represented as follows: ; The wind speed-power decision boundary constraint term is expressed as follows: ; The composite loss function: ; in, , , , These represent the loss term for dynamic operating mechanism knowledge data, the constraint term for dynamic operating curve, the boundary constraint term for wind speed-power decision, and the composite loss function, respectively. This represents the total amount of training data. It is the weight of the constraint term of the dynamic running curve. These are the weights of the boundary constraint terms in the wind speed-power decision. Is the model in the first Power prediction values at each time step. It is the first The actual power value at each time step It is the first The power value corresponding to the dynamic power curve at each time step. and It is the first The power values corresponding to the upper and lower bounds of wind speed-power at each time step.
[0037] In this experiment, Constraints are used to guide model predictions closer to the dynamic power operating curve. Because wind turbines, due to inertia, will cause power distribution to occur near the dynamic power operating curve, these constraints should not be set too high in applications. This experiment... It is 0.02; while To ensure that the predicted value lies within the physically feasible range, a larger weight can be assigned. In this experiment... It is 0.8.
[0038] In some embodiments, the established dynamic operating mechanism knowledge is embedded into the BiGRU prediction model using the training set, and the network parameters are optimized through backpropagation until the model's loss converges. Predictions are then made on the test set, and the prediction results on the test set are as follows: Figure 8 As shown.
[0039] For example, BiGRU refers to: Update Gate: ; Reset Door: ; Candidate cell status: ; Current hidden state: ; Bidirectional GRU: ; in, This is the input data for the current time step. It is the update gate value at time step t. It's a time-step reset gate. Here, σ represents the candidate cell state, and σ is the sigmoid activation function. , , Different gates for the current input The weight matrix, , , Different doors correspond to a previous hidden state. The weight matrix, It is the hidden state of the previous time step. , , These are the corresponding bias terms for different gates. It is the hyperbolic tangent activation function. It is the Hadamard product, which represents the element-wise multiplication of matrices. The current hidden state at time step t represents the output of that time step. It is a forward-facing GRU. It is a backward GRU. It is the hidden state of the forward GRU at time step t. It is the hidden state of the backward GRU at time step t. It is a vector concatenation operation.
[0040] In some embodiments, the prediction results are evaluated for performance. RMSE, MAE, and R² metrics are used to quantitatively assess the model's prediction accuracy.
[0041] For example, RMSE, MAE, R 2 This refers to: RMSE definition: ; MAE Definition: ; R 2 definition: ; Where m is the amount of data in the test set. It is the average power on the test set.
[0042] In some embodiments, the dynamic operating mechanism knowledge embedded in the BiGRU prediction model of this method is compared and analyzed with the traditional BiGRU prediction model without dynamic operating mechanism knowledge embedded. The traditional BiGRU prediction results are as follows: Figure 9As shown in Table 1, the evaluation metrics for both methods are presented. The method verifies the improvement in accuracy and stability achieved by embedding dynamic operational mechanism knowledge, thus demonstrating its feasibility and superiority.
[0043] Table 1: Evaluation Table of Traditional BiGRU and Dynamic Operating Mechanism Knowledge Embedded BiGRU Indicators
[0044] As shown in Table 1, the proposed method reduces RMSE by approximately 16.8% and MAE by approximately 14.9%, demonstrating that the proposed method significantly reduces outlier predictions that deviate from the true values, indicating that physical knowledge plays a good constraining role. Although the baseline model's R... 2 The results were already high, but this application still achieved further improvements, demonstrating that physical knowledge embedding can explain more data variance.
[0045] Based on the same inventive concept, this application also provides a wind power prediction system based on dynamic operating mechanism knowledge embedded in a BiGRU, the wind power prediction system comprising: The acquisition module is used to acquire the operating data of the wind turbine SCADA system, including wind speed and power; the preprocessed operating data is divided into training set, validation set and test set in chronological order; The module is used to build a wind speed-power scatter plot based on the running data in the training set, identify key trend points between wind speed and power through the data consistency principle, and build a dynamic wind speed-power running curve. Based on the aforementioned wind speed-power dynamic operation curve, a wind speed-power decision boundary is constructed. The training module is used to treat the constructed wind speed-power dynamic operating curve and wind speed-power decision boundary as dynamic operating mechanism knowledge, and to embed the dynamic operating mechanism knowledge into the BiGRU prediction model. The embedded BiGRU prediction model is trained using the training set, and the network parameters are optimized through backpropagation until the loss of the BiGRU prediction model converges. The application module is used to apply the trained BiGRU prediction model to the test set for wind power prediction.
[0046] Based on the same inventive concept, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the wind power prediction method based on dynamically operating mechanism knowledge embedded in a BiGRU as described above.
[0047] The program product of this application for implementing the above method may employ a portable compact disk read-only memory and include program code, and may run on a terminal device, such as a personal computer. However, the program product of this application is not limited thereto. In this application, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.
[0048] It should be noted that a computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.
[0049] The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Although this application has disclosed preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make some modifications or alterations to the above-mentioned technical content to create equivalent embodiments without departing from the scope of the technical solution of this application. The implementation schemes in the above embodiments can also be further combined or replaced. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of this application without departing from the content of the technical solution of this application shall still fall within the scope of this application.
Claims
1. A wind power prediction method based on dynamic operation mechanism knowledge embedded in BiGRU, characterized in that, Includes the following steps: Acquire the operating data of the wind turbine SCADA system, including wind speed and power; divide the preprocessed operating data into training set, validation set and test set in chronological order; Based on the operational data in the training set, a wind speed-power scatter plot is established. By using the data consistency principle, key trend points between wind speed and power are identified, and a dynamic wind speed-power operation curve is constructed. Based on the aforementioned wind speed-power dynamic operation curve, a wind speed-power decision boundary is constructed. The constructed wind speed-power dynamic operation curve and wind speed-power decision boundary are regarded as dynamic operation mechanism knowledge, and the dynamic operation mechanism knowledge is embedded into the BiGRU prediction model. The dynamic operation curve and decision boundary are used as physical loss to construct a composite loss function. The embedded BiGRU prediction model is trained using the training set, and the network parameters are optimized through backpropagation until the loss of the BiGRU prediction model converges. The trained BiGRU prediction model was applied to the test set for wind power prediction.
2. The method according to claim 1, characterized in that, The preprocessing includes using the sliding window quartile method to detect outliers in the running data, and using extreme gradient boosting extrapolation to correct outliers and missing values in the running data.
3. The method according to claim 2, characterized in that, The data consistency principle includes: Within each 1 m / s wind speed interval, the median of all power values within the wind speed interval is first calculated, and the median and the center value of the wind speed interval corresponding to the median are used as preliminary representative trend points. For each wind speed range, more refined local data sampling is performed near the ±0.1 m / s wind speed value corresponding to the representative trend point of the wind speed range, and the median power in the local range is recalculated to achieve local correction and refinement of the trend point. Based on the locally corrected power value, a power tolerance range of ±10% is constructed, and all data points whose power falls within the power tolerance range are screened out from the global data of the training set, and then the median wind speed corresponding to these data points is calculated. Determine whether the difference between the median wind speed obtained from the global reverse lookup and the wind speed value of the preliminary representative trend point is less than 1 m / s; if so, determine the locally corrected trend point as the final key trend point; if not, retain the preliminary representative trend point as the key trend point.
4. The method according to claim 3, characterized in that, The process of identifying key trend points between wind speed and power through data consistency principles and constructing a dynamic wind speed-power curve includes: Key trend points identified on the training set using the data consistency principle. ,in, The wind speed value represents the trend point. The power value at the trend point is used to fit the wind speed-power dynamic operating curve using PCHIP interpolation, in order to identify the mapping relationship between power and wind speed.
5. The method according to claim 4, characterized in that, The method of constructing the dynamic wind speed-power operating curve using the PCHIP method includes: PCHIP is defined as follows: Define the interval length and the difference quotient as follows: ； ； Define a local normalization variable as follows: ； Hermite cubic basis functions are expressed as: ； In the interval The interpolation function is as follows: ； The interpolation function satisfies: ； ； ； ； in, It is the first Wind speed values at key trend points This represents the power value of the i-th key trend point, where n represents the total number of key trend points, and i represents the index number of the key trend point. It is the interval length between two adjacent key trend points. It is in the interval The difference in price It is a locally normalized variable. , , , These are Hermite cubic basis functions, used to construct fixed mathematical expressions for smooth piecewise cubic functions. It is in the i-th interval Piecewise interpolation function, At key trend points ( The first derivative of the interpolation function at point () Indicates in The first derivative at that point.
6. The method according to claim 4, characterized in that, The construction of the wind speed-power decision boundary based on the wind speed-power dynamic operation curve includes: Iterate through the selected trend points, select data with wind speeds close to each key trend point within a range of ±0.25 m / s, and calculate the residual between the corresponding power and the dynamic power operation curve. Based on the calculated residuals, upper and lower decision points are determined to ensure that the decision points remain monotonically constant. The PCHIP interpolation method is used to smooth these decision points, generating a continuous and non-negative decision curve.
7. The method according to claim 1, characterized in that, Dynamic operating mechanism knowledge includes wind speed-power dynamic operating curves and wind speed-power decision boundaries; among which, The wind speed-power dynamic operation curve is represented as follows: ； The wind speed-power decision boundary includes: The upper boundary is represented as: ； The lower boundary is represented as: ； in, and These are the power values of the dynamic operating curves, upper decision boundary, and lower decision boundary corresponding to the wind speed, respectively; It's the wind speed value. It's the cut-in wind speed. That is the rated wind speed. These are the turning wind speeds corresponding to the upper and lower boundary curves; It is the rated power of the dynamic power operation curve. , , These represent the trend line, upper boundary, and lower boundary identified by PCHIP at the corresponding wind speed, respectively.
8. The method according to claim 1, characterized in that, The composite loss function includes a dynamic operating mechanism knowledge data loss term, a dynamic operating curve constraint term, and a wind speed-power decision boundary constraint term; wherein... The dynamic operation mechanism knowledge data loss term is represented as follows: ； The dynamic running curve constraint term is represented as follows: ； The wind speed-power decision boundary constraint term is expressed as follows: ； The composite loss function: ； in, , , , These represent the loss term for dynamic operating mechanism knowledge data, the constraint term for dynamic operating curve, the boundary constraint term for wind speed-power decision, and the composite loss function, respectively. This represents the total amount of training data. It is the weight of the constraint term of the dynamic running curve. These are the weights of the boundary constraint terms in the wind speed-power decision. Is the model in the first Power prediction values at each time step. It is the first The actual power value at each time step It is the first The power value corresponding to the dynamic power curve at each time step. and It is the first The power values corresponding to the upper and lower bounds of wind speed-power at each time step.
9. A wind power prediction system based on dynamic operation mechanism knowledge embedded in BiGRU, characterized in that, The wind power prediction system includes: The acquisition module is used to acquire the operating data of the wind turbine SCADA system, including wind speed and power; the preprocessed operating data is divided into training set, validation set and test set in chronological order; The module is used to build a wind speed-power scatter plot based on the running data in the training set, identify key trend points between wind speed and power through the data consistency principle, and build a dynamic wind speed-power running curve. Based on the aforementioned wind speed-power dynamic operation curve, a wind speed-power decision boundary is constructed. The training module is used to treat the constructed wind speed-power dynamic operating curve and wind speed-power decision boundary as dynamic operating mechanism knowledge, and to embed the dynamic operating mechanism knowledge into the BiGRU prediction model. The dynamic operating curve and decision boundary are used as physical loss to construct a composite loss function. The embedded BiGRU prediction model is trained using the training set, and the network parameters are optimized through backpropagation until the loss of the BiGRU prediction model converges. The application module is used to apply the trained BiGRU prediction model to the test set for wind power prediction.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it is used to implement the wind power prediction method based on dynamic operating mechanism knowledge embedded in BiGRU as described in any one of claims 1-8.