Wind power scene multi-time-sequence wind speed correction method and system based on long-term and short-term memory

Abstract

The invention provides a wind power scene multi-time-sequence wind speed correction method and system based on long and short term memory, and the method comprises the steps: obtaining a forecast wind speed sequence, synchronously collecting the actually measured wind speed, related meteorological auxiliary data and fan operation data of a target wind power plant at the same time-space position, and carrying out the preprocessing of the collected multi-source data, and forming a training sample set; constructing a long-short-term memory network model which takes the forecast wind speed and the multi-source data as input and takes the corrected wind speed as output; inputting forecast data to be corrected into the trained time period adaptive LSTM model to obtain a preliminary correction result; and evaluating the correction effect by using a relative root-mean-square error and a relative mean absolute error index. According to the method, by integrating multi-source information such as fan operation data and performing fine preprocessing, a more comprehensive and high-quality data basis is provided for the model.

Description

Technical Field

[0001] This invention belongs to the field of wind speed correction technology, and specifically relates to a method and system for multi-temporal wind speed correction in wind power scenarios based on long short-term memory. Background Technology

[0002] Accurate wind speed forecasting is a key technology in the field of wind power generation. Its accuracy directly affects the stable operation of the power system, the operating efficiency of wind farms, and the development level of the entire wind power industry.

[0003] The widely used WRF (Weather Research and Forecasting) numerical weather prediction model has significant errors in its wind speed forecasts due to the influence of complex terrain, variable meteorological factors, and the limitations of the model itself, making it difficult to meet the accuracy requirements of actual wind farm applications.

[0004] To improve the forecast error of WRF models, existing technologies mainly employ post-processing methods based on error correction. Among these, traditional artificial neural networks, especially backpropagation (BP) neural networks, are common correction techniques. The basic idea of ​​this method is to use the WRF model's forecast output (such as wind speed and direction) and related meteorological elements as input features, and the actual observed wind speed as the target value. The neural network is then trained to learn the nonlinear mapping relationship between the forecast and the actual values, thereby correcting the initial forecast results.

[0005] However, this correction method based on traditional BP neural networks has significant technical limitations. Firstly, BP networks are essentially static feedforward networks, and their structural characteristics make it difficult for them to effectively learn and remember the long-term dependencies inherent in time-series data, while wind speed changes are precisely a dynamic process with strong temporal correlations. Secondly, this model is prone to getting trapped in local optima, and the training process is unstable with weak generalization ability. Especially when facing complex weather processes or sudden wind speed changes, the correction effect is often unsatisfactory and may even amplify the error. Therefore, existing technical solutions cannot achieve continuous, stable, and high-precision correction of multi-time-series wind speed forecasts. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for multi-temporal wind speed correction in wind power scenarios based on long short-term memory, so as to solve the above-mentioned problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a multi-temporal wind speed correction method for wind power scenarios based on long short-term memory, including: Obtain the forecast wind speed sequence, and simultaneously collect the measured wind speed, relevant meteorological auxiliary data, and wind turbine operation data of the target wind farm at the same time and space location. Preprocess the collected multi-source data to form a training sample set. Construct a long short-term memory network model with forecasted wind speed and multi-source data as input and corrected wind speed as output; The forecast data to be corrected is input into the trained time-adaptive LSTM model to obtain preliminary correction results; The effectiveness of the correction was evaluated using relative root mean square error and relative mean absolute error indices.

[0008] Furthermore, the acquisition of the forecast wind speed sequence, and the simultaneous collection of measured wind speeds, relevant meteorological auxiliary data, and wind turbine operation data of the target wind farm at the same spatiotemporal location, includes: Forecast wind speed data is obtained from the WRF model, covering multiple time steps and different spatial locations; measured wind speed data for the corresponding time period and the same spatial location are collected; auxiliary data related to wind speed are collected, including vertical wind speed, air temperature, air humidity, and air pressure; supplementary data on the operation of wind turbines at the target wind farm during the same period are collected, including real-time turbine speed, pitch angle, nacelle azimuth angle, and generator power output data; during the collection process, it is necessary to ensure that the timestamps of the wind turbine operation data and the wind speed data are completely aligned, and at the same time, data from the fault-free operation period of the wind turbines are selected.

[0009] Furthermore, the preprocessing of the collected multi-source data to form a training sample set includes: Statistical methods are used to identify and remove outliers in wind speed data. Data points that deviate significantly from the mean by more than three times the standard deviation are identified as outliers and corrected or deleted. For data with missing values, if there are few missing values, linear interpolation is used to fill in the missing values ​​by linearly fitting data from adjacent time points; if there are many missing values, a model-based method is used for filling in the missing values. A normalization method is used to map wind speed data and auxiliary data to the interval [0, 1] or [-1, 1]; the formula is used... Normalize.

[0010] Furthermore, the construction of the long short-term memory network model, which takes the forecasted wind speed and multi-source data as input and the corrected wind speed as output, includes: The number of nodes in the input layer is determined based on the number of input features; LSTM hidden layers are set, and the number of neurons in the hidden layers is adjusted through experiments; the output layer has 1 node, which is used to output the corrected wind speed; Choosing the mean squared error (MSE) as the loss function, the formula is as follows: Where n is the number of samples, y i It is the actual value. These are predicted values; Adam is selected as the optimizer.

[0011] Further, model training: The preprocessed data is used as follows: 70%-80% as the training set to train the model to learn patterns in the data; 10%-15% as the validation set to monitor model performance and adjust hyperparameters during training; and 10%-15% as the test set to evaluate the model's final performance. During training, the wind speed forecast by the WRF model and related auxiliary data are organized into samples according to time series and used as model input, with the corresponding measured wind speed as the label; Set the batch size, which is the number of samples input into the model each time it is trained; The model is trained iteratively 50-200 times. In each iteration, the error between the model's predicted value and the true value is calculated based on the loss function, and the model's weights and biases are adjusted through the backpropagation algorithm. During the training process, the model performance is monitored through the validation set. If the validation set loss no longer decreases within a set number of rounds, training is stopped. Based on the time-period characteristics of wind speed changes, the training set was divided into four sub-datasets: daytime group, nighttime group, high-incidence period group, and low-incidence period group. For each sub-dataset, the basic model was fine-tuned by adjusting the weights and learning rate of the hidden layer neurons, so that the model's wind speed correction accuracy at different times better matches the actual wind speed change pattern.

[0012] Furthermore, the step of inputting the forecast data to be corrected into the trained time-adaptive LSTM model to obtain preliminary correction results includes: After training, the wind speed forecast from the WRF model to be corrected and related auxiliary data are preprocessed to meet the model input requirements; the preprocessed data is then input into the trained LSTM model; the model output is the corrected wind speed. Set a reasonable range for wind speed; compare the changes in wind speed before and after correction. If the magnitude of a single correction exceeds the set value, call the correction records under similar historical weather conditions for secondary verification. If the verification results are still abnormal, trigger the manual review process.

[0013] Furthermore, the relative root mean square error and relative mean absolute error indices are used to evaluate the correction effect; If the results are not satisfactory, adjust the LSTM model structure by adding hidden layers and adjusting the number of neurons, and retrain the model until satisfactory correction accuracy is achieved.

[0014] Secondly, the present invention provides a multi-temporal wind speed correction system for wind power scenarios based on long short-term memory, comprising: The data acquisition module is used to acquire the forecast wind speed sequence and simultaneously collect the measured wind speed, relevant meteorological auxiliary data and wind turbine operation data of the target wind farm at the same time and space location. The collected multi-source data is preprocessed to form a training sample set. The model building module is used to build a long short-term memory network model that takes the forecast wind speed and multi-source data as input and the corrected wind speed as output. The correction module is used to input the forecast data to be corrected into the trained time-adaptive LSTM model to obtain preliminary correction results; The evaluation module is used to assess the effectiveness of corrections using relative root mean square error and relative mean absolute error indices.

[0015] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the wind speed correction method for wind power scenarios based on long short-term memory.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the wind speed correction method for wind power scenarios based on long short-term memory.

[0017] Compared with the prior art, the present invention has the following technical effects: This invention provides a more comprehensive and high-quality data foundation for the model by integrating multi-source information such as wind turbine operation data and performing fine preprocessing. Secondly, by constructing a time-adaptive LSTM model and performing grouped fine-tuning, the model's ability to capture wind speed variation patterns at different times (such as day / night, high / low wind speed periods) is significantly enhanced, improving the time-adaptability of the correction results. Furthermore, the introduction of an automatic rationality verification mechanism for the correction results effectively avoids unreasonable outputs caused by extreme data interference, enhancing the reliability and practicality of the results. Finally, this method provides more accurate wind speed input for wind farm power prediction and operation optimization, contributing to improved wind power absorption efficiency and grid operation stability. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings: Example 1, please refer to Figure 1 This invention provides a multi-temporal wind speed correction method for wind power scenarios based on long short-term memory, including: Obtain the forecast wind speed sequence, and simultaneously collect the measured wind speed, relevant meteorological auxiliary data, and wind turbine operation data of the target wind farm at the same time and space location. Preprocess the collected multi-source data to form a training sample set. Construct a long short-term memory network model with forecasted wind speed and multi-source data as input and corrected wind speed as output; The forecast data to be corrected is input into the trained time-adaptive LSTM model to obtain preliminary correction results; The effectiveness of the correction was evaluated using relative root mean square error and relative mean absolute error indices.

[0020] This invention is the first to integrate wind turbine operating data (speed, pitch angle, etc.) into an LSTM wind speed correction model, breaking through the limitations of traditional models that rely solely on meteorological data. By analyzing the correlation between equipment status and wind speed, the correction results are made more consistent with the actual operating scenarios of wind turbines, reducing correction errors during periods of high wind power activity.

[0021] To address the time-varying characteristics of wind speed, a time-segmented fine-tuning strategy is proposed to avoid the problem of insufficient adaptability of traditional single models at different times.

[0022] Example 2: This invention provides a multi-temporal wind speed correction method for wind power scenarios based on long short-term memory, including: Data collection and preprocessing 1. Forecast wind speed data is obtained from the WRF model, covering multiple time steps and different spatial locations. This data reflects the model's prediction of wind speed.

[0023] 2. Collect measured wind speed data for the corresponding time period and the same spatial location to ensure its accuracy and completeness, which serves as the reference benchmark for correction.

[0024] 3. Collect auxiliary data related to wind speed, such as vertical wind speed, air temperature, air humidity, and air pressure. These factors will affect wind speed and provide richer information for the model.

[0025] 4. Collect additional wind turbine operating data from the target wind farm during the same period, including real-time turbine speed, pitch angle, nacelle azimuth angle, and generator power output data. During data collection, ensure that the timestamps of the turbine operating data and wind speed data are perfectly aligned (with an error not exceeding 1 minute). Simultaneously, filter out data from periods of fault-free turbine operation (excluding periods of downtime for maintenance or equipment malfunction) to provide data support for subsequent model optimization and correction based on equipment status.

[0026] 1. Use statistical methods, such as the 3σ criterion, to identify and remove outliers in wind speed data. Data points that deviate significantly from the mean by more than three standard deviations are identified as outliers and corrected or deleted.

[0027] 2. For data with missing values, select an appropriate processing method based on the characteristics of the data. If there are few missing values, linear interpolation can be used to fill in the missing values ​​by linearly fitting data from adjacent time points; if there are many missing values, consider using model-based methods, such as the K-Nearest Neighbors (KNN) algorithm, for filling in the missing values.

[0028] A normalization method is used to map wind speed data and auxiliary data to the interval [0, 1] or [-1, 1]. Taking mapping to the interval [0, 1] as an example, the formula is used... Normalize.

[0029] Building an LSTM model Determine the model structure 1. The number of nodes in the input layer is determined by the number of input features. Assuming there are 5 auxiliary variables (vertical wind speed, air temperature, air humidity, air pressure, and the forecast wind speed itself), the input layer will have 6 nodes.

[0030] 2. Set up 1-2 LSTM hidden layers, and adjust the number of neurons in the hidden layers through experimentation, generally between 32 and 128. More neurons can learn more complex features, but also increase computational cost and the risk of overfitting.

[0031] 3. The output layer consists of one node, used to output the corrected wind speed.

[0032] Choosing the loss function and optimizer 1. Choose the mean squared error (MSE) as the loss function, the formula is as follows: Where n is the number of samples and yi is the true value. It is the predicted value. MSE measures the average squared error between the model's predicted value and the actual value, intuitively reflecting the model's prediction accuracy.

[0033] 2. The optimizer chosen is Adam, which combines the advantages of Adagrad and RMSProp, and can adaptively adjust the learning rate. The learning rate is set between 0.001 and 0.0001, and the optimal value is selected through experimentation. A lower learning rate can make the model training more stable, but the training speed will be slower; a higher learning rate may cause the model to skip the optimal solution during training and fail to converge.

[0034] Model training Split the dataset The preprocessed data is used as follows: 70%-80% is used as the training set to train the model to learn the patterns in the data; 10%-15% is used as the validation set to monitor model performance and adjust hyperparameters during training to prevent overfitting; and 10%-15% is used as the test set to evaluate the final performance of the model.

[0035] Training process 1. During training, the wind speed forecast from the WRF model and related auxiliary data are organized into samples according to time series and used as model input, with the corresponding measured wind speed as the label.

[0036] 2. Set the batch size to 16-64, which is the number of samples input into the model each time it is trained. A larger batch size can make the model training more stable, but it will increase memory consumption and computation time; a smaller batch size can update the model parameters more frequently, but it may lead to larger fluctuations in the training process.

[0037] 3. Iterate the model 50-200 times. In each iteration, calculate the error between the model's predicted value and the true value based on the loss function, and adjust the model's weights and biases using the backpropagation algorithm. During training, monitor the model performance using a validation set. If the validation set loss no longer decreases within a certain number of rounds (e.g., 10 rounds), stop training to prevent overfitting.

[0038] 4. Based on the time-period characteristics of wind speed changes (e.g., daytime and nighttime, high-incidence and low-incidence periods), the training set is divided into four sub-datasets: "Daytime Group (6:00-18:00)," "Nighttime Group (18:00-6:00 the next day)," "High-incidence Group (wind speed ≥ 6 m / s)," and "Low-incidence Group (wind speed < 6 m / s)." For each sub-dataset, the base model is fine-tuned by adjusting the weights and learning rate of hidden layer neurons (the fine-tuning learning rate is set to 1 / 5 of the base learning rate) to make the model's wind speed correction accuracy more accurately reflect actual wind speed variations at different times.

[0039] Wind speed correction 1. After training, the wind speed forecasts and related auxiliary data of the WRF model to be corrected are preprocessed to meet the model input requirements.

[0040] 2. Input the preprocessed data into the trained LSTM model.

[0041] 3. The model output is the corrected wind speed.

[0042] 4. Verification of the reasonableness of the correction results: Set a reasonable wind speed range (determined based on historical wind farm data, such as 0.5-30 m / s; exceeding this range is considered abnormal); compare the wind speed changes before and after correction. If a single correction exceeds 5 m / s (or the relative change rate exceeds 30%), then use correction records under similar historical weather conditions for secondary verification. If the verification results are still abnormal, trigger a manual review process to prevent the model from outputting unreasonable results due to extreme data interference.

[0043] Evaluation and optimization 1. The correction effect is evaluated using indicators such as relative root mean square error (rRMSE) and relative mean absolute error (rMAE). These indicators can intuitively reflect the degree of error between the corrected wind speed and the measured wind speed.

[0044] 2. If the results are not satisfactory, adjust the LSTM model structure, such as adding hidden layers or adjusting the number of neurons; optimize hyperparameters, such as the learning rate; or try adding more relevant feature data and retraining the model until satisfactory correction accuracy is achieved.

[0045] This invention leverages the powerful memory and modeling capabilities of Long Short-Term Memory Networks (LSTM) for time-series data to learn and correct systematic errors in numerical weather prediction (WRF) models. Its innovation lies in not only using WRF-predicted wind speed and conventional meteorological elements as input, but also incorporating data reflecting the actual operating status of wind turbines (such as rotational speed and power). Furthermore, the training data is grouped and fine-tuned according to the temporal characteristics of wind speed changes (such as day / night, high / low wind periods), enabling the model to more precisely capture the dynamic patterns of wind speed evolution under different meteorological conditions, thereby achieving more accurate and physically realistic corrections to predicted wind speeds.

[0046] First, multi-source data collaborative acquisition and refined preprocessing are performed. The system collects WRF forecast data, measured data, relevant meteorological auxiliary data, and key wind turbine operation data (timestamps must be strictly aligned and fault-free periods must be selected). Then, outlier removal and missing value imputation are performed using methods such as the 3σ criterion (e.g., using the K-Nearest Neighbors algorithm), and all data are normalized. Second, an LSTM model structure is constructed, the number of input layer nodes is determined (based on the number of features), 1-2 hidden layers are set (the number of neurons is tuned between 32-128), the output layer has 1 node, and mean square error (MSE) is selected as the loss function and Adam as the optimizer. Third, time-adaptive model training is performed. The preprocessed data is proportionally divided into training, validation, and test sets. Early stopping is introduced during training to prevent overfitting. A key step is to divide the training set into multiple subsets based on day / night and high / low wind speed periods, and fine-tune the basic LSTM model for each subset (e.g., by reducing the learning rate) to improve the model's adaptability in different time periods. Fourth, wind speed correction with reasonableness verification is performed. The data to be corrected is input into the fine-tuned model, and the output results are automatically verified, including checking whether the wind speed is within a reasonable range (e.g., 0.5-30 m / s). For results with excessively large corrections, a secondary verification or manual review process is initiated. Finally, a systematic evaluation and optimization are conducted. Performance is evaluated using metrics such as relative root mean square error (rRMSE), and the model is iteratively optimized by adjusting the model structure and hyperparameters until the accuracy requirements are met.

[0047] The most significant effect of this invention is a substantial improvement in the accuracy and practicality of wind speed correction. By incorporating the strong correlation feature of wind turbine operating data and performing differentiated training on different time periods, the model can more sensitively capture the spatiotemporal variation characteristics of WRF model errors, resulting in correction results that more closely reflect actual wind speed changes. Secondly, by establishing an automatic rationality verification mechanism for the correction results, absurd outputs that the model may produce under extreme conditions are effectively avoided, enhancing the reliability of the entire correction process and its application value in practical operations. Finally, this method provides higher-quality wind speed input for high-precision power prediction and operation control of wind farms, directly serving the stable operation of the power grid and the efficient absorption of wind power.

[0048] In another embodiment of the present invention, a multi-temporal wind speed correction system for wind power scenarios based on long short-term memory is provided, which can be used to implement the above-mentioned multi-temporal wind speed correction method for wind power scenarios based on long short-term memory. Specifically, the system includes: The data acquisition module is used to acquire the forecast wind speed sequence and simultaneously collect the measured wind speed, relevant meteorological auxiliary data and wind turbine operation data of the target wind farm at the same time and space location. The collected multi-source data is preprocessed to form a training sample set. The model building module is used to build a long short-term memory network model that takes the forecast wind speed and multi-source data as input and the corrected wind speed as output. The correction module is used to input the forecast data to be corrected into the trained time-adaptive LSTM model to obtain preliminary correction results; The evaluation module is used to assess the effectiveness of corrections using relative root mean square error and relative mean absolute error indices.

[0049] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0050] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for operating a multi-time-series wind speed correction method for wind power scenarios based on long short-term memory.

[0051] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the multi-time-series wind speed correction method for wind power scenarios based on long short-term memory in the above embodiments.

[0052] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0053] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0054] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A long short-term memory-based wind farm scenario multi-time series wind speed revision method, characterized in that, The application relates to a method for correcting wind speed prediction data, comprising the following steps: acquiring a wind speed prediction sequence and synchronously collecting measured wind speed, related meteorological auxiliary data and wind turbine operation data of a target wind farm at the same space-time position, and preprocessing the collected multi-source data to form a training sample set; constructing a long short-term memory network model taking the wind speed prediction and the multi-source data as input and taking the corrected wind speed as output; inputting the to-be-corrected prediction data into the trained time period adaptive LSTM model to obtain a preliminary correction result; using a relative root mean square error and a relative mean absolute error index to evaluate the correction effect.

2. The long short-term memory based wind power scenario multi-horizon wind speed revision method according to claim 1, characterized in that, The method comprises the following steps: acquiring wind speed prediction data from a WRF model, covering multiple time steps and different spatial position information; collecting measured wind speed data at the same space-time position; collecting auxiliary data related to wind speed, including vertical wind speed, air temperature, air humidity and air pressure; collecting wind turbine operation data of the target wind farm at the same period, including real-time rotating speed, pitch angle, nacelle azimuth angle and generator power output data; and ensuring that the time stamps of the wind turbine operation data and the wind speed data are completely aligned during the collection process, and filtering out the data of the period when the wind turbine is in fault-free operation.

3. The long short-term memory based wind power scenario multi-horizon wind speed revision method according to claim 1, characterized in that, The method comprises the following steps: using a statistical method to identify and remove abnormal values in the wind speed data; for data points obviously deviating from the mean value by more than 3 times the standard deviation, the data points are judged as abnormal values and are corrected or deleted; for data with missing values, if the missing values are few, a linear interpolation method is used to fill in the missing values according to the linear fitting of the data of adjacent time points; if the missing values are many, a model-based method is used to fill in the missing values; The wind speed data and the auxiliary data are mapped to the interval [0, 1] or [-1, 1] by using a normalization method; and the normalization is performed by using the formula .

4. The long short-term memory based wind power scenario multi-horizon wind speed revision method according to claim 1, characterized in that, The method comprises the following steps: the number of input layer nodes is determined according to the number of input features; an LSTM hidden layer is set, and the number of hidden layer neurons is adjusted through experiments; the output layer has one node for outputting the corrected wind speed; The mean square error (MSE) is selected as the loss function, and the formula is where n is the number of samples, y i is the true value, is the predicted value; and the optimizer is Adam.

5. The long short-term memory based wind power scenario multi-horizon wind speed revision method according to claim 1, characterized in that, model training: 70%-80% of the preprocessed data are used as a training set to train the model to learn the rules in the data; 10%-15% are used as a validation set to monitor the model performance during the training process and adjust the hyperparameters; 10%-15% are used as a test set to evaluate the final performance of the model; in the training process, the WRF model prediction wind speed and related auxiliary data are organized into samples in time sequence as model input, and the corresponding measured wind speed is used as a label; the batch size, i.e. the number of samples input into the model during each training, is set; the model is iteratively trained for 50-200 times, and the error between the predicted value and the true value of the model is calculated according to the loss function in each iteration, and the weights and biases of the model are adjusted through the back propagation algorithm; during the training process, the model performance is monitored through the validation set, and if the validation set loss does not decrease within a set number of rounds, the training is stopped. According to the time period characteristics of wind speed changes, the training set is divided into four sub-datasets of day group, night group, high incidence period group and low incidence period group according to time periods; the base model is fine-tuned for each sub-dataset, the hidden layer neuron weight and learning rate are adjusted, so that the wind speed correction accuracy of the model in different time periods is more suitable for the actual wind speed change rule.

6. The long short-term memory based wind power scenario multi-horizon wind speed revision method according to claim 1, characterized in that, The trained period adaptive LSTM model is inputted with the to-be-corrected prediction data to obtain a preliminary correction result, which includes: After the training is completed, the WRF mode prediction wind speed to be corrected and related auxiliary data are preprocessed to meet the model input requirements; the preprocessed data are inputted into the trained LSTM model; and the output result of the model is the corrected wind speed. A reasonable range of wind speed is set; the change amplitude of wind speed before and after correction is compared; if the single correction amplitude exceeds the set value, the correction record under the similar historical meteorological conditions is called for secondary verification; if the verification result is still abnormal, the manual review process is triggered.

7. The long short-term memory based wind power scenario multi-horizon wind speed revision method according to claim 6, characterized in that, The relative root mean square error and the relative average absolute error indicators are used to evaluate the correction effect. If the effect is not up to standard, the LSTM model structure is adjusted by increasing the hidden layer and adjusting the number of neurons, the model is retrained until the satisfactory correction accuracy is achieved.

8. A long short-term memory based wind farm scenario multi-horizon wind speed revision system, characterized in that, It includes: A data acquisition module is configured to acquire a prediction wind speed sequence, and simultaneously collect measured wind speed, related meteorological auxiliary data and wind turbine operation data of a target wind farm at the same space-time position, and preprocess the collected multi-source data to form a training sample set; A model construction module is configured to construct a long short-term memory network model with the prediction wind speed and multi-source data as input and the corrected wind speed as output; A correction module is configured to input the to-be-corrected prediction data into the trained period adaptive LSTM model to obtain a preliminary correction result; An evaluation module is configured to use the relative root mean square error and the relative average absolute error indicators to evaluate the correction effect.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the steps of the long short-term memory-based wind power scene multi-time sequence wind speed correction method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to realize the steps of the long short-term memory-based wind power scene multi-time sequence wind speed correction method according to any one of claims 1 to 7.

Images