Regional power grid wind power generation power prediction optimization method and system

Abstract

The invention discloses a regional power grid wind power generation power prediction optimization method and system, and particularly relates to the technical field of power grid wind power generation. Multi-source environment data is collected, wind speed change characteristics are extracted, a time series data set is constructed, a hybrid prediction model is constructed in combination with a long-short term memory network and a gradient boosting decision tree, a time series trend and a nonlinear wind speed-power mapping relation are captured respectively, and prediction weights of the time series trend and the nonlinear wind speed-power mapping relation are adaptively adjusted through an attention mechanism. According to the method, the model hyper-parameters are adaptively adjusted by further combining Bayesian optimization, short-term prediction errors are corrected by using Kalman filtering, the real-time adaptability and stability of prediction are enhanced, the problems of prediction misalignment and the like caused by high nonlinearity and data scarcity of a wind speed mode in a complex terrain environment are effectively solved, and the prediction accuracy is improved. The wind power prediction precision is improved, a more stable and reliable scheduling reference is provided for a regional power grid, and the standby capacity demand and the operation cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of grid-connected wind power generation, and particularly to an optimization method and system for wind power generation power prediction in a regional power grid. Background Art

[0002] With the global energy structure transformation towards renewable energy, the penetration rate of wind power generation in the regional power grid is continuously increasing. However, due to the significant influence of meteorological factors such as wind speed, temperature, humidity, etc. on wind power generation, its output power has volatility and intermittency, posing challenges to the safe and stable operation of the power grid. In order to improve the predictability of wind power and the reliability of power grid dispatching, wind power generation power prediction technology has been widely studied. Current prediction methods mainly include physical model methods, statistical methods, and machine learning methods based on artificial intelligence. These methods have improved the prediction accuracy of wind power to a certain extent. However, there are still problems such as limited prediction accuracy in complex environments and difficulty in accurately capturing short-term power fluctuations, which urgently need to be optimized and improved.

[0003] The existing technologies have the following deficiencies:

[0004] In high-altitude or complex terrain areas, due to terrain shielding effects and local air flow disturbances, the wind speed change pattern is highly non-linear, resulting in the failure or significant reduction in accuracy of existing wind power generation power prediction models in such environments. Traditional prediction methods are difficult to effectively capture the local turbulence characteristics of wind speed, especially in areas with frequent wind shear and wind energy backflow. There is a lack of high-quality data samples during the model training process, further reducing the prediction reliability. Such problems will not only affect the power generation plan of the wind farm itself, but may also have an adverse impact on the stability of the regional power grid, increasing the demand for reserve capacity and dispatching costs. Summary of the Invention

[0005] The purpose of the present invention is to provide an optimization method and system for wind power generation power prediction in a regional power grid to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions: An optimization method for wind power generation power prediction in a regional power grid, including the following steps:

[0007] S1: Collect multi-source environmental data of the target wind farm, including wind speed, wind direction, temperature, humidity, atmospheric pressure, terrain features, and historical power output data;

[0008] S2: Extract the characteristics of the wind speed change pattern in high-altitude or complex terrain areas based on the collected multi-source environmental data, including extracting the turbulence intensity characteristics and wind energy backflow characteristics in the multi-source environmental data, and generating a time series data set;

[0009] S3: Combine the long short-term memory network and the gradient boosting decision tree to establish a hybrid prediction model. Among them, the long short-term memory network is used to capture the long-term trends and short-term fluctuation characteristics in the time series, and the gradient boosting decision tree is used to learn the non-linear wind speed-power mapping relationship in the time series, and dynamically allocate the prediction weights of the two through the attention mechanism;

[0010] S4: Based on the real-time monitoring data of the regional power grid, use the Bayesian optimization method to adaptively adjust the hyperparameters of the hybrid prediction model, and correct the error of the short-term power prediction result through the Kalman filtering method to improve the prediction accuracy and the reliability of power grid dispatching.

[0011] Preferably, in S2, the sliding window method is used to calculate the change of turbulence intensity over time: set the sliding window size w; calculate the mean and standard deviation of the wind speed within each time window; calculate the turbulence intensity and store it in the time series data set; calculate the wind direction change rate. If the wind direction change rate Δθ reverses by 180° in a short time, it is judged as a backflow phenomenon, and the wind direction change rate calculated when the backflow phenomenon occurs is stored in the time series data set.

[0012] Preferably, in S3, the input data of the hybrid prediction model is derived from the extracted multi-source environmental data time series, including: wind speed time series U(t), wind direction time series θ(t), turbulence intensity time series TI(t), wind energy backflow feature ΔU(t), and historical power output P(t);

[0013] Adopt the Min-Max normalization method to normalize the input data to the range of [0, 1] or [-1, 1], use the sliding window method to create training samples, and predict the wind power in the next 10 - 30 minutes; the task of the LSTM network is to: learn the time-dependent relationship of wind speed, wind direction, and turbulence characteristics, and predict the future wind speed change trend; output the predicted wind speed value and the wind speed change trend characteristics;

[0014] The LSTM structure includes: Input layer: time series data;

[0015] LSTM layer: multiple LSTM cells to extract time series features;

[0016] Fully connected layer: convert the LSTM output into the predicted wind power value;

[0017] Output layer: the predicted wind speed trend feature F LSTM and the short-term power prediction value

[0018] LSTM prediction formula:

[0019] Preferably, the gradient boosting decision tree learns the complex wind speed-power relationship through multiple decision trees. The task of GBDT is: input the wind speed feature F predicted by LSTM LSTM and environmental features, and learn the non-linear mapping relationship from wind speed to power; use historical data to train the decision tree and optimize the wind speed prediction value to the conversion of wind power generation; output the power prediction value as another branch of the hybrid prediction model; GBDT prediction formula:

[0020] Preferably, the attention mechanism is used to dynamically adjust the prediction weights of the two to improve the final prediction accuracy, specifically including:

[0021] When the short-term wind speed fluctuates greatly, increase the LSTM prediction weight; under stable wind speed conditions, increase the GBDT prediction weight; calculate the confidence level α of the LSTM prediction value L : α L =σ(W att ·F LSTM +b att ); calculate the confidence level α of the GBDT prediction value G : α G =1 - α L ; where, W att and b att are the parameters of the attention network, σ is the Sigmoid activation function, and calculate the final power prediction value The expression is:

[0022] Preferably, in S4, the goal is to minimize the prediction error: where, P true (t) is the actual wind power, is the predicted power of the hybrid prediction model, N is the total of the predicted power; set the hyperparameter search range of LSTM and GBDT, and construct the Gaussian process regression GP model; set the loss function L as the optimization goal; use the historical training data to initialize the GP surrogate model and estimate the prediction errors of different hyperparameter configurations; calculate the prediction error of the current hyperparameter configuration; select the hyperparameter combination x next : x next =argmaxα(x); where, α(x) is the expected improvement criterion; update the GP surrogate model and repeat the iteration until convergence to the optimal hyperparameters.

[0023] Preferably, in S4, under the Kalman filter framework, the predicted value of wind power is modeled as a linear dynamic system: X t =AX t-1 +BUt +W t ; Z t = HX t +V t ; where: X t is the true value of wind power; U t is the external environmental factor; Z t is the prediction result of the hybrid prediction model W t , V t respectively represent process noise and measurement noise;

[0024] Predict the wind power at the next moment: In the formula, is the predicted state, that is, the wind power predicted according to historical data, A is the state transition matrix, describing the change relationship of wind power between adjacent time steps, B is the control matrix, describing the influence of the external input U t on the wind power; Calculate the prediction error covariance: is the prediction error covariance matrix, indicating the uncertainty of the current state prediction, and T is the matrix transpose;

[0025] Calculate the Kalman gain: K t is the Kalman gain, used to adjust the predicted value, H is the observation matrix, mapping the system state X t to the measurement space, and R is the measurement noise covariance matrix; Correct the predicted value: is the updated state estimate value, that is, the wind power prediction value after error correction; Update the error covariance: P t is the updated error covariance matrix, indicating the uncertainty of the current state estimate, and I is the identity matrix, used to calculate the new error matrix.

[0026] The present invention also provides an optimized system for predicting the wind power of a regional power grid, including a data acquisition module, a feature extraction module, a hybrid prediction model module, and an error correction module;

[0027] Data acquisition module: Collect multi-source environmental data of the target wind farm, including wind speed, wind direction, temperature, humidity, atmospheric pressure, terrain features, and historical power output data;

[0028] Feature extraction module: Extract the characteristics of the wind speed change pattern in high-altitude or complex terrain areas based on the collected multi-source environmental data, including extracting the turbulence intensity characteristics and wind energy return characteristics in the multi-source environmental data, and generating a time series data set;

[0029] Hybrid Prediction Model Module: A hybrid prediction model is established by combining the Long Short-Term Memory network and the Gradient Boosting Decision Tree. Among them, the Long Short-Term Memory network is used to capture the long-term trends and short-term fluctuation characteristics in the time series, and the Gradient Boosting Decision Tree is used to learn the non-linear wind speed-power mapping relationship in the time series, and the prediction weights of the two are dynamically allocated through the attention mechanism;

[0030] Error Correction Module: Based on the real-time monitoring data of the regional power grid, the Bayesian optimization method is used to adaptively adjust the hyperparameters of the hybrid prediction model, and the Kalman filtering method is used to correct the errors of the short-term power prediction results to improve the prediction accuracy and the reliability of power grid dispatching.

[0031] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0032] 1. By collecting multi-source environmental data such as wind speed, wind direction, temperature, humidity, atmospheric pressure, and terrain features, the present invention extracts the turbulence intensity and wind energy return characteristics, and constructs a time series data set to accurately depict the wind speed change pattern under complex terrains. A hybrid prediction model is constructed using the Long Short-Term Memory (LSTM) network and the Gradient Boosting Decision Tree (GBDT) to capture the long-term trends, short-term fluctuation characteristics of the time series, and the non-linear mapping relationship between wind speed and power respectively, and the weights of the two are dynamically adjusted through the attention mechanism to improve the accuracy of wind power prediction. In addition, the present invention uses the Bayesian optimization method to adaptively adjust the model hyperparameters to ensure that the model can maintain high prediction accuracy under different meteorological conditions, and at the same time uses the Kalman filter to correct the errors of short-term power prediction to enhance the reliability and stability of real-time prediction.

[0033] 2. The optimization method of the present invention effectively overcomes the limitations of the prior art in wind power prediction under complex terrains, significantly improves the prediction accuracy, and reduces the power grid dispatching deviation caused by prediction errors. Through the combination of deep learning, machine learning, and real-time optimization algorithms, it not only improves the accuracy of the power generation plan of the wind farm, but also reduces the reserve capacity demand and dispatching cost of the regional power grid, and enhances the safety and stability of the power grid operation. This method is applicable to complex wind farm environments, has strong generality, and can be widely applied to fields such as smart grids, wind farm optimal dispatching, and new energy consumption, which is of great significance for promoting the efficient utilization of wind power. Description of the Drawings

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0035] Figure 1 This is the flowchart of the method of the present invention.

[0036] Figure 2 This is the system module diagram of the present invention. Detailed implementation manners

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0038] Example 1. Refer to Figure 1 As shown, the method for optimizing the prediction of wind power generation power in a regional power grid in this embodiment includes the following steps:

[0039] S1: Collect multi-source environmental data of the target wind farm, including wind speed, wind direction, temperature, humidity, atmospheric pressure, terrain features, and historical power output data;

[0040] S2: Extract the characteristics of the wind speed change pattern in high-altitude or complex terrain areas based on the collected multi-source environmental data, including extracting the turbulence intensity characteristics and wind energy recirculation characteristics in the multi-source environmental data, and generating a time series data set;

[0041] S3: Combine a long short-term memory network and a gradient boosting decision tree to establish a hybrid prediction model. Among them, the long short-term memory network is used to capture the long-term trend and short-term fluctuation characteristics in the time series, and the gradient boosting decision tree is used to learn the non-linear wind speed-power mapping relationship in the time series, and dynamically allocate the prediction weights of the two through an attention mechanism;

[0042] S4: Based on the real-time monitoring data of the regional power grid, use the Bayesian optimization method to adaptively adjust the hyperparameters of the hybrid prediction model, and correct the error of the short-term power prediction result through the Kalman filtering method to improve the prediction accuracy and the reliability of power grid dispatching.

[0043] In S1, in order to improve the accuracy of the prediction of wind power generation power in the regional power grid, it is necessary to comprehensively collect the multi-source environmental data of the wind farm. These data cover multiple key factors affecting the wind power generation power output, including meteorological conditions, terrain features, and the historical operation data of wind turbines. Specifically, it includes the following content:

[0044] Wind Speed is one of the key parameters that mainly affect the power output of wind power generation. The higher the wind speed, the stronger the power generation capacity of the wind turbine. However, when the wind speed exceeds the rated wind speed, the wind turbine may enter a state of load reduction or shutdown.

[0045] Data acquisition method: Use wind speed sensors (such as ultrasonic anemometers, three-cup anemometers) or remote sensing devices (such as lidar LIDAR, SODAR) for measurement, and collect wind speed data at different heights to capture the wind shear effect.

[0046] Wind Direction affects the windward angle of the wind turbine and the wake effect (Wake Effect) between wind turbines in the wind farm, thereby affecting the wind power output.

[0047] Data acquisition method: Wind direction sensors (such as wind vanes or wind direction meters) monitor the change of wind direction, and at the same time, a high-resolution meteorological model is combined for wind farm reconstruction.

[0048] Temperature affects air density, and the change of air density will directly affect the conversion efficiency of wind energy. Too low temperature may cause icing on the blades, affecting the normal operation of the wind turbine.

[0049] Data acquisition method: Obtain temperature information from the meteorological station in the wind farm or remote meteorological data services (such as NOAA, ECMWF).

[0050] Humidity affects air density and wind energy conversion efficiency. At the same time, a high-humidity environment is prone to blade icing or corrosion.

[0051] Data acquisition method: Meteorological station sensors or remote sensing devices monitor relative humidity.

[0052] Atmospheric Pressure affects air density, and air density is one of the important parameters for calculating wind energy power conversion.

[0053] Data acquisition method: Pressure sensors (such as MEMS barometers) or meteorological stations obtain real-time barometric data, and analyze the trend changes in combination with historical data.

[0054] Topographical Features determine the wind speed distribution pattern of the wind farm. Especially in complex terrains (such as mountains, canyons, high-altitude areas), the terrain shielding effect, wind tunnel effect and turbulence characteristics will have a significant impact on the wind speed.

[0055] Data acquisition method: Use digital elevation models (DEM), lidar (LiDAR), satellite remote sensing images or unmanned aerial vehicle aerial survey data to construct a topographic model of the wind farm.

[0056] The historical power output data records the actual power generated by the wind turbine under different environmental conditions to establish a wind speed-power mapping relationship.

[0057] Data collection method: SCADA (supervisory control and data acquisition) system of wind turbines, smart meters or data storage system of regional power grid dispatching center.

[0058] S2: Based on the collected multi-source environmental data, feature extraction of wind speed change patterns in high-altitude or complex terrain areas is performed, including extracting turbulence intensity characteristics and wind energy return characteristics from multi-source environmental data, and generating a time series data set.

[0059] Turbulence intensity TI reflects the degree of fluctuation of wind speed and is usually defined as: Where TI is the turbulence intensity, σ u is the standard deviation of wind speed, indicating the fluctuation amplitude of wind speed; is the average value of wind speed. The greater the turbulence intensity, the more violent the wind speed fluctuation, which may cause a greater load impact on the wind turbine blades and affect the stable operation of the wind turbine.

[0060] Wind speed data U(t): from SCADA system, ultrasonic anemometer, LiDAR (laser radar) or SODAR (sound radar) and other equipment. Wind speed at different heights Uh(t): obtain wind speed data at different heights of the wind turbine impeller to analyze the impact of wind shear and turbulence. Environmental factor data: temperature, humidity, atmospheric pressure, etc., used to analyze the causes of turbulence.

[0061] In order to generate time series, the sliding window method is used to calculate the temporal variation of turbulence intensity: set the sliding window size w (for example, 10 minutes, 30 minutes or 1 hour); calculate the mean and standard deviation of wind speed in each time window; calculate the turbulence intensity TI and store it in the time series dataset.

[0062] Wind energy backflow refers to the reversal or significant decrease in the direction of local wind speed, resulting in a reduction in the wind energy available to the wind turbine. It usually occurs in: the nighttime inversion layer in the valley: cold air settles, resulting in a decrease in wind speed or reverse airflow; on the leeward side of the ridge: the wind speed drops sharply, forming a low-speed vortex area; the wind turbine wake area: a low-speed area is formed behind the wind turbine, affecting the power output of downstream wind turbines.

[0063] Wind direction data is used to analyze wind direction reversal (180° change). Wind speed data is used to analyze abnormal wind speed drop. Wind speeds at different heights are used to analyze vertical profile changes of wind speed and determine the recirculation area. CFD simulation data: used to simulate wind speed recirculation patterns under specific terrain.

[0064] Calculate the wind direction change rate Δθ = θ(t) - θ(t - 1); if Δθ undergoes a 180° reversal within a short period (e.g., 5 minutes), it is judged as a backflow phenomenon; and store the wind direction change rate calculated when the backflow phenomenon occurs into the time series dataset.

[0065] S3: Combine the long short - term memory network and the gradient - boosting decision tree to establish a hybrid prediction model. Among them, the long short - term memory network is used to capture the long - term trend and short - term fluctuation characteristics in the time series, and the gradient - boosting decision tree is used to learn the non - linear wind speed - power mapping relationship in the time series, and dynamically allocate the prediction weights of the two through the attention mechanism.

[0066] The input data of the hybrid prediction model comes from the extracted multi - source environmental data time series, including: wind speed time series U(t), wind direction time series θ(t), turbulence intensity time series TI(t), wind energy backflow characteristic ΔU(t), and historical power output P(t);

[0067] Adopt the Min - Max normalization method to normalize the input data to the range of [0, 1] or [-1, 1] to improve the model convergence speed. Use the sliding window method to create training samples. For example, the wind speed, wind direction, and turbulence characteristics in the past 60 minutes are used as inputs; predict the wind power in the next 10 - 30 minutes; ensure that the timestamps of data from different sources are aligned to avoid errors caused by data mismatch.

[0068] LSTM is suitable for processing time - series data, can remember information with a long time span, and suppress the problem of gradient disappearance. In this method, the tasks of the LSTM network are:

[0069] Learn the time - dependent relationship of wind speed, wind direction, and turbulence characteristics, and predict the future wind speed change trend; analyze the short - term wind speed fluctuation characteristics and identify abnormal wind speed change patterns; output the wind speed prediction value and the wind speed change trend characteristics for use by GBDT and the attention mechanism.

[0070] The LSTM structure includes:

[0071] Input layer: Time - series data (wind speed, wind direction, turbulence characteristics, wind energy backflow).

[0072] LSTM layer: Multiple LSTM units to extract time - series characteristics.

[0073] Fully - connected layer: Convert the LSTM output into a wind power prediction value.

[0074] Output layer: Predicted wind speed trend characteristic F LSTM and short - term power prediction value

[0075] LSTM prediction formula:

[0076] Gradient Boosting Decision Tree (GBDT) is a powerful non - linear regression model that can learn the complex wind speed - power relationship through multiple decision trees. In this method, the tasks of GBDT are:

[0077] Input the wind speed feature F predicted by LSTM LSTM and environmental features, and learn the non - linear mapping relationship from wind speed to power;

[0078] Use historical data to train the decision tree and optimize the conversion from the wind speed prediction value to wind power generation;

[0079] Output the power prediction value as another branch of the hybrid prediction model.

[0080] GBDT prediction formula:

[0081] Since the prediction performances of LSTM and GBDT are different in different situations, an attention mechanism is needed to dynamically adjust the prediction weights of the two to improve the final prediction accuracy.

[0082] The role of the attention mechanism: When the short - term wind speed fluctuates greatly, increase the prediction weight WL of LSTM; Under stable wind speed conditions, increase the prediction weight WG of GBDT;

[0083] Calculate the final power prediction value

[0084] Calculate the confidence level α of the LSTM prediction value L : α L =σ(W att ·F LSTM +b att ); Calculate the confidence level α of the GBDT prediction value G : α G =1 - α L ; Among them, W att and b att are the parameters of the attention network, and σ is the Sigmoid activation function. Calculate the final power prediction value:

[0085] S4: In wind power prediction, the hyperparameters of LSTM and GBDT (such as learning rate, number of LSTM layers, GBDT tree depth) have an important impact on prediction performance. Traditional optimization methods like Grid Search or Random Search have high computational costs and are difficult to handle complex high-dimensional search spaces. Bayesian optimization uses Gaussian process (GP) to model the relationship between hyperparameters and prediction error, and selects the best parameters through the Expected Improvement (EI) or Upper Confidence Bound (UCB) strategy, thus improving search efficiency and optimization effect.

[0086] The goal is to minimize the prediction error, such as Mean Squared Error (MSE) or Mean Absolute Percentage Error (MAPE): where, P true (t) is the true wind power, is the predicted power of the hybrid prediction model (LSTM + GBDT + attention), and N is the total number of predicted powers. Set the search range of hyperparameters for LSTM and GBDT, construct a Gaussian process regression GP model; set the loss function L as the optimization goal; initialize the GP surrogate model using historical training data, estimate the prediction error of different hyperparameter configurations; calculate the prediction error of the current hyperparameter configuration; select the hyperparameter combination x next : x next = argmaxα(x); where α(x) is the Expected Improvement criterion; update the GP surrogate model and repeat the iteration until convergence to the optimal hyperparameters.

[0087] Due to the uncertainty of wind speed, the short-term power prediction of the hybrid prediction model may be affected by noise. Kalman filtering can use real-time monitoring data to dynamically correct the prediction error, thereby reducing short-term prediction bias and improving the reliability of power grid scheduling.

[0088] Under the Kalman filtering framework, the predicted value of wind power is modeled as a linear dynamic system: X t = AX t-1 + BU t + W t ; Z t = HX t + V t ; where: X t is the true value of wind power (to be estimated); U t are external environmental factors such as wind speed, turbulence intensity, and wind energy recirculation; Z t is the prediction result of the hybrid prediction model W t 、V t : represent process noise and measurement noise respectively.

[0089] Predict the wind power at the next moment: In the formula, is the predicted state, that is, the wind power predicted based on historical data. A is the state transition matrix, which describes the change relationship of wind power between adjacent time steps. It is usually determined by a physical model or a data-driven model. For example, A = 1 represents a simple Markov assumption (that is, the current power is mainly affected by the power at the previous moment). If environmental factors such as wind speed are considered, A may be a more complex matrix.

[0090] B is the control matrix, which describes the influence of the external input U t on the wind power. In this problem, Ut represents external influencing factors such as wind speed, turbulence intensity, and wind energy return. If these factors have a greater impact on the power, B may be a non-zero matrix.

[0091] Calculate the prediction error covariance: is the prediction error covariance matrix, which represents the uncertainty of the current state prediction. T is the matrix transpose;

[0092] Calculate the Kalman gain: K t is the Kalman gain, which is used to adjust the predicted value to make it closer to the true observed value. The larger K t , the more trust in the measured value; the smaller K t , the more trust in the predicted value.

[0093] H is the observation matrix, which maps the system state X t to the measurement space. In this problem, H = 1 (assuming that the measured wind power directly reflects the system state).

[0094] R is the measurement noise covariance matrix, which describes the uncertainty of wind power measurement (such as sensor error, data acquisition noise, etc.).

[0095] Correct the predicted value: is the updated state estimate value, that is, the wind power prediction value after error correction.

[0096] Update the error covariance: P t is the updated error covariance matrix, which represents the uncertainty of the current state estimate. I is the identity matrix, which is used to calculate the new error matrix.

[0097] In this embodiment, first, multi-source environmental data of the target wind farm is collected, including wind speed, wind direction, temperature, humidity, atmospheric pressure, terrain features, and historical power data. Subsequently, based on the collected data, key wind speed change features are extracted, such as turbulence intensity and wind energy return features, and a time series dataset is generated to characterize the non-linear change pattern of the wind speed. On this basis, a hybrid prediction model is constructed, which combines the long short-term memory network (LSTM) to capture the long-term trend and short-term fluctuation features in the time series, while using the gradient boosting decision tree (GBDT) to learn the non-linear mapping relationship between wind speed and power, and dynamically adjusts the prediction weights of the two through an attention mechanism. Finally, based on the real-time monitoring data of the regional power grid, the Bayesian optimization method is used to adaptively adjust the hyperparameters of the hybrid prediction model, and the Kalman filter is used to correct the error of the short-term power prediction, improving the prediction accuracy and the stability of the power grid scheduling. This method realizes more accurate and efficient wind power prediction through multi-source data fusion, the combination of deep learning and machine learning, real-time optimization, and error correction.

[0098] Example 2, please refer to Figure 2 As shown in the figure, the regional power grid wind power generation power prediction optimization system in this embodiment includes a data acquisition module, a feature extraction module, a hybrid prediction model module, and an error correction module;

[0099] Data acquisition module: Collect multi-source environmental data of the target wind farm, including wind speed, wind direction, temperature, humidity, atmospheric pressure, terrain features, and historical power output data;

[0100] Feature extraction module: Extract the wind speed change pattern features of high-altitude or complex terrain areas based on the collected multi-source environmental data, including extracting the turbulence intensity features and wind energy return features in the multi-source environmental data, and generating a time series dataset;

[0101] Hybrid prediction model module: Combine the long short-term memory network and the gradient boosting decision tree to establish a hybrid prediction model. Among them, the long short-term memory network is used to capture the long-term trend and short-term fluctuation features in the time series, and the gradient boosting decision tree is used to learn the non-linear wind speed-power mapping relationship in the time series, and dynamically allocate the prediction weights of the two through an attention mechanism;

[0102] Error correction module: Based on the real-time monitoring data of the regional power grid, use the Bayesian optimization method to adaptively adjust the hyperparameters of the hybrid prediction model, and correct the error of the short-term power prediction result through the Kalman filter method to improve the prediction accuracy and the reliability of the power grid scheduling.

[0103] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0104] It should be understood that the term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood by referring to the context before and after.

[0105] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0106] As described above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. An optimization method for wind power generation power prediction in regional power grids, characterized in that: It includes the following steps: S1: Collect multi-source environmental data of the target wind farm, including wind speed, wind direction, temperature, humidity, atmospheric pressure, terrain features, and historical power output data; S2: Extract the characteristics of the wind speed change pattern in high-altitude or complex terrain areas based on the collected multi-source environmental data, including extracting the turbulence intensity characteristics and wind energy recirculation characteristics in the multi-source environmental data, and generating a time series data set; S3: Combine the long short-term memory network and the gradient boosting decision tree to establish a hybrid prediction model. Among them, the long short-term memory network is used to capture the long-term trend and short-term fluctuation characteristics in the time series, and the gradient boosting decision tree is used to learn the non-linear wind speed-power mapping relationship in the time series, and dynamically allocate the prediction weights of the two through the attention mechanism; S4: Based on the real-time monitoring data of the regional power grid, adaptively adjust the hyperparameters of the hybrid prediction model using the Bayesian optimization method, and correct the error of the short-term power prediction result through the Kalman filter method to improve the prediction accuracy and the reliability of power grid scheduling.

2. The optimization method for predicting the wind power generation power of a regional power grid according to claim 1, wherein: In S2, the sliding window method is used to calculate the change of turbulence intensity over time: set the sliding window size w; calculate the mean and standard deviation of the wind speed within each time window; Calculate the turbulence intensity and store it in the time series data set; Calculate the wind direction change rate. If the wind direction change rate Δθ reverses by 180° in a short time, it is judged as a recirculation phenomenon, and the wind direction change rate calculated when the recirculation phenomenon occurs is stored in the time series data set.

3. The optimized method for predicting the wind power generation of a regional power grid according to claim 2, characterized in that: In S3, the input data of the hybrid prediction model comes from the time series of the extracted multi-source environmental data, including: wind speed time series U(t), wind direction time series θ(t), turbulence intensity time series TI(t), wind energy recirculation characteristic ΔU(t), and historical power output P(t); Using the Min-Max normalization method, the input data is normalized to the range of [0, 1] or [-1, 1]. The sliding window method is used to create training samples to predict the wind power in the next 10 - 30 minutes. The task of the LSTM network is to learn the time-dependent relationships of wind speed, wind direction, and turbulence characteristics and predict the future wind speed change trend; output the predicted wind speed value and the characteristics of the wind speed change trend; The LSTM structure includes: Input layer: time series data; LSTM layer: multiple LSTM units to extract time series features; Fully connected layer: convert the LSTM output into a wind power prediction value; Output layer: Predicted wind speed trend feature F LSTM and short-term power prediction value LSTM prediction formula: F LSTM (t) = LSTM(U(t - n), θ(t - n), TI(t - n), …, U(t), θ(t), TI(t)); 4. The method for optimizing the wind power generation power prediction of a regional power grid according to claim 3, wherein: Gradient Boosting Decision Tree learns the complex wind speed-power relationship through multiple decision trees. The task of GBDT is: input the wind speed features F predicted by LSTM LSTM and environmental features, and learn the non-linear mapping relationship from wind speed to power; use historical data to train decision trees and optimize the wind speed prediction value for the conversion to wind power generation; output the power prediction value as another branch of the hybrid prediction model; GBDT prediction formula:

5. The optimization method for wind power generation power prediction in a regional power grid according to claim 4, wherein: Use the attention mechanism to dynamically adjust the prediction weights of the two to improve the final prediction accuracy, specifically including: When the short-term wind speed fluctuates greatly, increase the LSTM prediction weight; under stable wind speed conditions, increase the GBDT prediction weight; calculate the confidence level α of the LSTM prediction value L : α L =σ(W att ·F LSTM +b att ); calculate the confidence level α of the GBDT prediction value G : α G =1 - α L ; where, W att and b att are the parameters of the attention network, σ is the Sigmoid activation function, calculate the final power prediction value The expression is:

6. The optimized method for predicting the wind power generation power of a regional power grid according to claim 1, characterized in that: In S4, the goal is to minimize the prediction error: where P true (t) is the true wind power, is the predicted power of the hybrid prediction model, N is the total number of predicted powers; set the hyperparameter search ranges of LSTM and GBDT, construct a Gaussian process regression GP model; set the loss function L as the optimization objective; use historical training data to initialize the GP surrogate model, estimate the prediction errors of different hyperparameter configurations; calculate the prediction error of the current hyperparameter configuration; select the hyperparameter combination x next : x next = argmax α(x); where α(x) is the expected improvement criterion; update the GP surrogate model and repeat the iteration until convergence to the optimal hyperparameters.

7. The optimization method for predicting the wind power generation of a regional power grid according to claim 6, wherein: In S4, under the Kalman filter framework, the predicted value of wind power is modeled as a linear dynamic system: X t = AX t-1 + BU t + W t ; Z t = HX t + V t ; Where: X t is the true value of wind power; U t is the external environmental factor; Z t is the prediction result of the hybrid prediction model W t , V t represent process noise and observation noise respectively; Predict the wind power at the next moment: In the formula, is the predicted state, that is, the wind power predicted based on historical data. A is the state transition matrix, which describes the change relationship of wind power between adjacent time steps. B is the control matrix, which describes the influence of the external input U t on the wind power; Calculate the prediction error covariance: is the prediction error covariance matrix, which represents the uncertainty of the current state prediction. T is the matrix transpose; Calculate the Kalman gain: K t is the Kalman gain, which is used to adjust the predicted value. H is the observation matrix that maps the system state X t to the measurement space, and R is the measurement noise covariance matrix; Correct the predicted value: is the updated state estimate, that is, the wind power prediction value after error correction; Update the error covariance: P t is the updated error covariance matrix, which represents the uncertainty of the current state estimate. I is the identity matrix used to calculate the new error matrix.

8. A regional power grid wind power generation power prediction optimization system for implementing the regional power grid wind power generation power prediction optimization method according to any one of claims 1-7, characterized in that: It includes a data acquisition module, a feature extraction module, a hybrid prediction model module, and an error correction module; Data acquisition module: Collect multi-source environmental data of the target wind farm, including wind speed, wind direction, temperature, humidity, atmospheric pressure, terrain features, and historical power output data; Feature extraction module: Extract the characteristics of the wind speed change pattern in high-altitude or complex terrain areas based on the collected multi-source environmental data, including extracting the turbulence intensity characteristics and wind energy recirculation characteristics in the multi-source environmental data, and generating a time series data set; Hybrid prediction model module: Combine the long short-term memory network and the gradient boosting decision tree to establish a hybrid prediction model. Among them, the long short-term memory network is used to capture the long-term trend and short-term fluctuation characteristics in the time series, and the gradient boosting decision tree is used to learn the non-linear wind speed-power mapping relationship in the time series, and dynamically allocate the prediction weights of the two through the attention mechanism; Error correction module: Based on the real-time monitoring data of the regional power grid, the Bayesian optimization method is used to adaptively adjust the hyperparameters of the hybrid prediction model, and the Kalman filtering method is used to correct the errors of the short-term power prediction results, so as to improve the prediction accuracy and the reliability of power grid scheduling.

Images