Multi-model integrated offshore wind power dynamic prediction method and system

Abstract

The invention discloses a multi-model integrated offshore wind power dynamic prediction method and system, and relates to the technical field of wind power prediction, and the method comprises the steps: driving a first collection channel to collect first offshore wind power data; performing wind power prediction based on the first prediction model to generate a first prediction sequence, and performing deletion detection and noise detection to obtain deviation data; performing wind power prediction of each prediction model to generate a second prediction sequence, a third prediction sequence and a fourth prediction sequence; and performing deviation correction on the deviation data to obtain a corrected prediction result. The technical problems that in the prior art, the offshore wind power data acquisition quality is low, and wind power prediction is inaccurate due to the fact that high-quality data are depended and low-quality data robustness processing capacity is lacked are solved, and the technical effect of remarkably improving the accuracy and reliability of offshore wind power dynamic prediction is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power prediction, in particular to a multi-model integrated offshore wind power dynamic prediction method and system. BACKGROUND

[0002] The prediction of offshore wind power is facing many severe challenges. On the one hand, the environment of offshore wind power is complex and harsh, and is affected by many factors such as sea wind, sea wave, sea current and weather conditions. These factors are intertwined, making it extremely difficult to obtain offshore wind power data. On the other hand, the existing wind power prediction technology highly depends on high-quality, complete and accurate data. However, the actual collected offshore wind power data often has a large number of missing values and noise interference, which is caused by sensor failure, communication transmission problems and harsh marine environment. The low quality of data seriously affects the accuracy and reliability of the prediction model, resulting in a large deviation between the prediction result and the actual wind power. In addition, the traditional prediction method usually only relies on a single data source or a single prediction model, lacks effective fusion of multi-source data and collaborative use of multi-model, and is difficult to fully tap the potential information in the data, and cannot adapt to the complex and changeable characteristics of offshore wind power.

[0003] The existing technology has the technical problems of low quality of offshore wind power data acquisition, and inaccurate wind power prediction due to the dependence on high-quality data and the lack of robustness processing ability for low-quality data. SUMMARY

[0004] The present application provides a multi-model integrated offshore wind power dynamic prediction method and system, which is used to solve the technical problems of low quality of offshore wind power data acquisition in the prior art, and inaccurate wind power prediction due to the dependence on high-quality data and the lack of robustness processing ability for low-quality data.

[0005] In view of the above problems, the present application provides a multi-model integrated offshore wind power dynamic prediction method and system.

[0006] In a first aspect of the present application, a multi-model integrated offshore wind power dynamic prediction method is provided, which comprises:

[0007] The first acquisition channel is driven to collect first offshore wind power data; wind power prediction based on a first prediction model is performed through the first offshore wind power data to generate a first prediction sequence; missing detection and noise detection are performed on the first offshore wind power data to obtain deviation data; second offshore wind power data of a second acquisition channel, third offshore wind power data of a third acquisition channel and fourth offshore wind power data of a fourth acquisition channel are extracted to perform wind power prediction of each prediction model to generate a second prediction sequence, a third prediction sequence and a fourth prediction sequence; and the deviation data is corrected in deviation by using the second prediction sequence, the third prediction sequence and the fourth prediction sequence to obtain a corrected prediction result.

[0008] In a second aspect of the present application, a multi-model integrated offshore wind power dynamic prediction system is provided, and the system comprises:

[0009] The offshore wind power data acquisition module is configured to drive a first acquisition channel to collect first offshore wind power data; the first prediction sequence generation module is configured to perform wind power prediction based on a first prediction model through the first offshore wind power data to generate a first prediction sequence; the deviation data acquisition module is configured to perform missing detection and noise detection on the first offshore wind power data to obtain deviation data; the wind power prediction module is configured to extract second offshore wind power data of a second acquisition channel, third offshore wind power data of a third acquisition channel and fourth offshore wind power data of a fourth acquisition channel to perform wind power prediction of each prediction model to generate a second prediction sequence, a third prediction sequence and a fourth prediction sequence; and the corrected prediction result acquisition module is configured to correct the deviation data in deviation by using the second prediction sequence, the third prediction sequence and the fourth prediction sequence to obtain a corrected prediction result.

[0010] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0011] The first acquisition channel is driven to collect first offshore wind power data; wind power prediction based on a first prediction model is performed to generate a first prediction sequence; missing detection and noise detection are performed on the first offshore wind power data to obtain deviation data; second offshore wind power data of a second acquisition channel, third offshore wind power data of a third acquisition channel and fourth offshore wind power data of a fourth acquisition channel are extracted to perform wind power prediction of each prediction model; and the deviation data is corrected in deviation to obtain a corrected prediction result. The technical effect of significantly improving the accuracy and reliability of offshore wind power dynamic prediction is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0013] Figure 1 A flowchart of a multi-model integrated offshore wind power dynamic prediction method provided by the embodiments of the present application is shown.

[0014] Figure 2 A structural diagram of a multi-model integrated offshore wind power dynamic prediction system provided by the embodiments of the present application is shown.

[0015] The figure mark explanation: offshore wind power data acquisition module 10, first prediction sequence generation module 20, deviation data acquisition module 30, wind power prediction module 40, corrected prediction result acquisition module 50. DETAILED DESCRIPTION

[0016] The present application provides a multi-model integrated offshore wind power dynamic prediction method and system, which is used to solve the technical problem of low quality of offshore wind power data acquisition in the prior art, and the inaccuracy of wind power prediction caused by the dependence on high-quality data and the lack of robustness processing capability for low-quality data.

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] Embodiment one, as shown in the present application provides a multi-model integrated offshore wind power dynamic prediction method, which comprises: Figure 1

[0019] Step S100: driving the first acquisition channel to acquire first offshore wind power data.

[0020] ​Specifically, in the offshore wind farm, the first acquisition channel is composed of various sensors, data transmission devices and control units. First, the installation and debugging of hardware devices need to be completed. Wind speed sensors are installed at different heights of the fan tower, which are used to measure the wind speed at different levels; wind vanes are installed in the nacelle to accurately obtain wind direction information; temperature sensors and vibration sensors are installed at key positions such as generators to monitor the running state of the equipment in real time; power sensors are connected in the circuit system to collect wind power data. These sensors are connected to the data acquisition terminal, which has analog-to-digital conversion function and can convert the analog signals from the sensors into digital signals. After completing the hardware preparation, the control unit plays a key role. The control unit can be an industrial computer running a specially developed acquisition control program. After the program starts, the control unit sends initialization instructions to the data acquisition terminal to activate each sensor to start working and set the acquisition frequency of the sensors, such as collecting data once every second for the wind speed sensor. The data collected by the sensors is converted by the data acquisition terminal and transmitted to the control unit through wired or wireless communication. For wind turbines close to the central control room, optical fiber is used for wired transmission to ensure the stability and high speed of data transmission; for wind turbines far away, 5G and other wireless communication technologies are used for transmission. After receiving the data, the control unit will perform preliminary processing and storage on the data. On the one hand, the data is checked for its integrity and accuracy, and obviously incorrect data is removed; on the other hand, the data is stored in the local database. The database uses a time series database that can efficiently store wind power data arranged in chronological order. These stored data constitute the first offshore wind power data, which will be used for wind power prediction and analysis in the future, providing strong support for the stable operation and efficient management of offshore wind farms.

[0021] Step S200: performing wind power prediction based on a first prediction model through the first offshore wind power data to generate a first prediction sequence.

[0022] Specifically, after obtaining the first offshore wind power data, first determine the value of K (for example, set K = 10 according to the historical data rule and model training effect), and divide the continuous time of a preset time length (such as the past 24 hours) according to a preset window division granularity (assuming that the initial window length is 10 minutes) to obtain 10 preliminary continuous windows. Calculate the parameter fluctuation rate of the first offshore wind power data. If the wind speed fluctuation rate is large, appropriately reduce the window division granularity, otherwise increase, and then slide update the preliminary continuous windows to obtain accurate 10 continuous windows. Through the 10 continuous windows, 10 sample sensing parameters (wind speed, wind direction, etc.), 10 sample first wind turbine operating parameters (generator temperature, blade speed, etc.) and the next 10 sample first wind power are extracted from the first offshore wind power data to form 10 first continuous window samples. Select a neural network model architecture (such as long short-term memory network LSTM), train the 10 continuous window neural network model architecture using the 10 first continuous window samples, and obtain 10 continuous window prediction models. Again, use the 10 continuous windows to cut the first offshore wind power data to obtain 10 first cut wind power sequences. After matching the cut sequences with the continuous windows, input them into the corresponding continuous window prediction models for wind power prediction to generate 10 first continuous window wind power prediction powers corresponding to the next 10 continuous windows, and finally paste these prediction powers in order to successfully generate a first prediction sequence, providing basic data support for subsequent prediction analysis.

[0023] Step S300: detecting missing data and noise in the first offshore wind power data to obtain deviation data.

[0024] Specifically, the time interval of the first offshore wind power data is first accurately counted. Based on the time stamp, the time difference between adjacent data records is calculated, and a reasonable excessive time interval threshold (such as 15 minutes as the excessive time interval threshold when the normal collection interval is 10 minutes) is set. If a time interval exceeds the threshold, the data in this part is identified as a part that may have missing data, forming the first missing detection result. At the same time, the data is continuously detected for null values, and each data field in the data set is traversed. If there are consecutive null values in a field, and the number of null values reaches or exceeds the pre-set valid missing threshold (such as 5 consecutive null values), it is determined that there is a missing problem in this part of data, and the second missing detection result is obtained. The two are combined as the final missing detection result. In the noise detection link, the statistical characteristics (such as mean, standard deviation, etc.) of the data are first used to determine whether the first offshore wind power data is abnormal. If a data point deviates from the mean by more than 3 times the standard deviation, it is preliminarily determined to be an outlier, and is recorded as the first noise detection result. Next, a sliding window (such as a window size of 10 data points) is pre-set, and the window is slid in the data set in segments. The local statistical characteristics (such as local mean, local standard deviation) of the data in each window are calculated, and a sliding window threshold is set. If the data in the window does not meet the threshold range (such as too large or too small data fluctuations), the data in this part is marked as noise data, and the second noise detection result is obtained. In addition, the data is subjected to change rate anomaly recognition, and the change rate between adjacent data points is calculated. If the change rate exceeds the normal fluctuation range (such as wind speed changing by more than a certain percentage in a short time), it is identified as an anomaly, and is the third noise detection result. The three noise detection results are combined to obtain the final noise detection result. Finally, the missing data is extracted according to the missing detection result, and the noise data is extracted according to the noise detection result. The two parts of data are combined to obtain the deviation data, which provides a key basis for subsequent data correction and prediction model optimization.

[0025] Step S400: Extracting the second offshore wind power data of the second collection channel, the third offshore wind power data of the third collection channel, and the fourth offshore wind power data of the fourth collection channel to perform wind power prediction of each prediction model, and generating a second prediction sequence, a third prediction sequence, and a fourth prediction sequence.

[0026] Specifically, first, the second offshore wind power data is extracted from the second collection channel, which is, for example, from a sea buoy. Using the previously determined K consecutive windows, K sample buoy parameters (such as sea state data measured by the buoy, surrounding wind speed and direction, etc.), K sample second wind turbine operating parameters (blade angle, generator speed, etc. of the second wind turbine) and the next K sample second wind power are extracted from the second offshore wind power data to form K second consecutive window samples. Similar modeling methods (using the same type of long short-term memory network LSTM neural network architecture) are used to train the K second consecutive window samples to obtain K second prediction models. The second offshore wind power data is cut into K second cut wind power sequences using K consecutive windows, and these sequences are input into the corresponding second prediction models for wind power prediction to generate K second consecutive window wind power predictions corresponding to the next K consecutive windows. These predicted powers are pasted in order to generate a second prediction sequence. The third offshore wind power data (from satellite monitoring) of the third collection channel is directly input into the third prediction model (trained based on the relationship between satellite data characteristics and wind power) for wind power prediction, and the output is the third prediction sequence. The fourth offshore wind power data (from unmanned aerial vehicle inspection) of the fourth collection channel is also input into the corresponding fourth prediction model (trained according to the characteristics of unmanned aerial vehicle collected data) for prediction to generate a fourth prediction sequence. These prediction sequences provide prediction results of wind power from different data sources and model perspectives, providing multi-dimensional data support for subsequent comprehensive correction of prediction results to improve the accuracy and reliability of offshore wind power prediction.

[0027] Step S500: bias correction of the bias data using the second prediction sequence, the third prediction sequence and the fourth prediction sequence to obtain a corrected prediction result.

[0028] Specifically, a correction model set containing multiple correction models is established, for example, a model containing multiple different algorithms such as a linear regression model, a decision tree model, etc. A first correction model is extracted from the correction model set, and the model is used to simulate and correct the deviation of the deviation data on the second prediction sequence. Specifically, the deviation data is combined with the second prediction sequence and input into the first correction model, and the model processes the data according to its own algorithm and outputs the first simulated correction prediction result. To evaluate the accuracy of the result, a correction sample with known accurate wind power values is used to compare the first simulated correction prediction result with the actual value in the correction sample, and by calculating the sum of squares of the difference between the two, the first simulated correction error coefficient is obtained, which reflects the error size of this simulation correction. Then, the correction model set is traversed, and other models in the set are used to simulate and correct the deviation of the deviation data on the second prediction sequence in turn to obtain the second simulated correction prediction result to the Nth simulated correction prediction result, and the corresponding second simulated correction error coefficient to the Nth simulated correction error coefficient is calculated. All simulated correction error coefficients are arranged in descending order, and the correction model corresponding to the minimum error coefficient is selected as the second correction model for the second prediction sequence. The same method is used to select the third correction model suitable for the third prediction sequence and the fourth correction model suitable for the fourth prediction sequence from the correction model set. After selecting the correction model, the second correction model is used to formally correct the deviation of the deviation data on the second prediction sequence to obtain the second correction prediction result, and the correction error of the result is calculated again to generate the second correction error coefficient. Similarly, the third correction model and the fourth correction model are used to correct the deviation of the deviation data on the third prediction sequence and the fourth prediction sequence, respectively, to obtain the third correction prediction result and the fourth correction prediction result, and the corresponding third correction error coefficient and fourth correction error coefficient are calculated. Finally, the second, third and fourth correction error coefficients are arranged in descending order, and according to the arrangement result, the prediction result with the smallest correction error coefficient is preferentially selected to correct the deviation data, if the second correction error coefficient is the smallest, the second correction prediction result is used as the main correction result for the deviation data, and finally the correction prediction result is obtained, improving the accuracy of offshore wind power prediction.

[0029] In one possible implementation manner, the step S200 further includes:

[0030] Step S210: Extract K sample sensing parameters, K sample first wind turbine operating parameters and next K sample first wind power through K continuous windows to obtain K first continuous window samples.

[0031] Step S220: training K continuous window neural network model architectures with the K first continuous window samples to generate K continuous window prediction models, wherein the K continuous window neural network model architectures are the same neural network model architecture.

[0032] Step S230: performing data cutting on the first offshore wind power data with the K continuous windows to obtain K first cut wind power sequences, wherein the first offshore wind power data includes sensing parameters, first wind turbine operating parameters, and first wind power.

[0033] Step S240: matching the K first cut wind power sequences with the K continuous windows, and inputting the K first cut wind power sequences into the K continuous window prediction models corresponding to the matched continuous windows to perform wind power prediction, thereby generating K first continuous window wind power prediction powers corresponding to the next K continuous windows.

[0034] Step S250: sequentially pasting the K first continuous window wind power prediction powers corresponding to the next K continuous windows to generate the first prediction sequence.

[0035] Specifically, assuming that the value of K is set to 10, the time span and interval of the continuous windows are determined according to the pre-set window division rule. For example, each window has a duration of 15 minutes, and the interval between adjacent windows is 5 minutes. Through the 10 continuous windows, relevant parameters are extracted from the first offshore wind power data. Using a sensor data acquisition system, sensing parameters such as wind speed, wind direction, temperature, and air pressure are obtained within each window, which reflect the meteorological environment in which the wind turbine is located; at the same time, first wind turbine operating parameters such as blade speed, generator temperature, and gearbox pressure are collected, which are used to represent the operating state of the wind turbine itself; in addition, first wind power data for the next time period (also 10 continuous window durations) is obtained. The sensing parameters, first wind turbine operating parameters, and corresponding first wind power of the next window within each window are integrated to obtain 10 first continuous window samples.

[0036] A suitable neural network model architecture is selected, such as a recurrent neural network (RNN) or a long short-term memory network (LSTM), because they are good at processing time series data and can effectively capture the time dependence relationship in wind power data. The obtained 10 first continuous window samples are used as training data to train 10 continuous window neural network models with the same architecture. During the training process, the weights and biases of the model are continuously adjusted through the backpropagation algorithm, so that the model can learn the complex relationship between the input parameters (sensing parameters and wind turbine operating parameters) and the output (wind power), until the model converges, thereby generating 10 continuous window prediction models.

[0037] The first offshore wind power data is cut according to the determined time range and interval of 10 continuous windows. From the data starting point, the original data containing the sensing parameters, the first wind turbine operating parameters and the first wind power are sequentially cut into 10 first cut wind power sequences according to the window setting, and each sequence contains complete wind turbine operation and environment related data in a specific time period.

[0038] The 10 first cut wind power sequences are matched with the corresponding 10 continuous windows to ensure that each cut sequence can correspond to the window from which it originates. Then, each first cut wind power sequence is input into the continuous window prediction model corresponding to the continuous window matched therewith. For example, the first cut sequence is input into the prediction model corresponding to the first window, the second cut sequence is input into the prediction model corresponding to the second window, and so on. Each model analyzes and predicts the input cut sequence based on the learned knowledge from training to generate 10 first continuous window wind power predictions corresponding to the next 10 continuous windows.

[0039] The 10 first continuous window wind power predictions are sequentially connected in time order. For example, the power value predicted by the first window is placed at the front, and then the predicted power values of the subsequent windows are sequentially connected to form a complete sequence, which is the first prediction sequence finally generated, providing important basic data for subsequent analysis and prediction of offshore wind power.

[0040] In one possible implementation manner, step S210 further includes:

[0041] Step S211: dividing the continuous time of the preset time length with a preset window division granularity to obtain K preliminary continuous windows.

[0042] Step S212: calculating the parameter fluctuation rate of the first offshore wind power data, updating the window division granularity, and slidingly updating the K preliminary continuous windows to obtain the K continuous windows.

[0043] Specifically, according to the requirements of the prediction task and the preliminary analysis of the data features, the window division granularity and the time length are set. Assuming that the preset window division granularity is 10 minutes and the preset time length is the past 24 hours. The continuous time of 24 hours is divided at intervals of 10 minutes, and a window boundary is determined every 10 minutes from the starting time point, obtaining 144 (24x60÷10=144) preliminary continuous windows, laying a foundation for subsequent data extraction and analysis.

[0044] The parameter fluctuation rate of the first offshore wind power data is calculated. Representative parameters such as wind speed and fan blade speed are selected. Taking wind speed as an example, the standard deviation of the wind speed data in each preliminary continuous window is calculated. The larger the standard deviation, the more intense the wind speed fluctuation in the window, and the higher the parameter fluctuation rate. The window division granularity is updated according to the calculated parameter fluctuation rate. If the wind speed standard deviation exceeds a certain threshold (such as 5 meters / second), it indicates that the data fluctuation is large, and the window division granularity is shortened to 5 minutes to capture the data changes more finely; if the standard deviation is small, the window division granularity is appropriately increased, such as adjusting to 15 minutes. After updating the window division granularity, the 144 preliminary continuous windows obtained previously are updated in a sliding manner. Taking the example of shortening the window division granularity to 5 minutes, starting from the first preliminary continuous window, sliding according to the new 5-minute granularity, if the original first window is 0-10 minutes, it may be updated to 0-5 minutes after updating, and the second window is 5-10 minutes, and so on. In this way, the window range is re-determined, and K continuous windows that can more accurately reflect the data characteristics are finally obtained, providing better data window selection for subsequent sample data extraction and model training, and improving the accuracy of offshore wind power prediction

[0045] In one possible implementation manner, step S400 further includes:

[0046] Step S410: based on the K continuous windows, K sample buoy parameters, K sample second wind turbine operating parameters and next K sample second wind power are extracted, and K second continuous window samples are obtained.

[0047] Step S420: training K second prediction models with the K second continuous window samples, wherein the K second prediction models are the same kind of models.

[0048] Step S430: performing data cutting on the second offshore wind power data with the K continuous windows to obtain K second cut wind power sequences, wherein the second offshore wind power data includes buoy parameters, second wind turbine operating parameters and second wind power.

[0049] Step S440: inputting the K second cut wind power sequences into the matched K second prediction models corresponding to the continuous windows to perform wind power prediction, and generating K second continuous window wind power prediction powers corresponding to the next K continuous windows.

[0050] Step S450: sequentially pasting the K second continuous window wind power prediction powers corresponding to the next K continuous windows to obtain the second prediction sequence.

[0051] Step S460: inputting the third offshore wind power data into a third prediction model to perform wind power prediction, and generating the third prediction sequence; and inputting the fourth offshore wind power data into a fourth prediction model to perform wind power prediction, and generating the fourth prediction sequence, wherein the third offshore wind power data comprises satellite parameters, third wind turbine operating parameters and third wind power, and the fourth offshore wind power data comprises unmanned aerial vehicle parameters, fourth wind turbine operating parameters and fourth wind power.

[0052] Specifically, based on the determined K continuous windows, required parameters are extracted from the second offshore wind power data. The buoys, as data collection sources, are equipped with sensors for measuring parameters such as wind speed, wind direction, sea current speed, and seawater temperature. Through the data collection system, these buoy parameters are obtained within each continuous window, forming K sample buoy parameters. At the same time, for the second wind turbine, second wind turbine operating parameters such as blade angle, generator speed, and torque are collected in each window, obtaining K sample second wind turbine operating parameters. In addition, second wind power data corresponding to the next K continuous windows is collected, and the buoy parameters, second wind turbine operating parameters in each window, and the corresponding second wind power of the next window are integrated together, thereby obtaining K second continuous window samples.

[0053] A model suitable for processing wind power data is selected, such as a support vector machine (SVM) model, and K second continuous window samples are trained. The K sample data is divided into a training set and a validation set. By adjusting the kernel function type, penalty parameter, and other hyperparameters of the SVM model, the training set data is used to train the model, and the model performance is evaluated on the validation set. The model is continuously optimized until it converges, generating K second prediction models.

[0054] According to the time range and interval of the K continuous windows, the second offshore wind power data is cut. Starting from the data starting position, the original data containing the buoy parameters, second wind turbine operating parameters, and second wind power are divided into K second cut wind power sequences according to the window setting, each sequence containing complete related data within a specific time period.

[0055] The K second cut wind power sequences are respectively input into the second prediction models corresponding to the continuous windows matched thereto. For example, the first cut sequence is input into the prediction model corresponding to the first window, the second cut sequence is input into the prediction model corresponding to the second window, and so on. Each model analyzes and predicts the input cut sequence according to the learned parameter relationship, generating K second continuous window wind power prediction powers corresponding to the next K continuous windows.

[0056] The K second continuous window wind power prediction powers are sequentially arranged in time order and pasted together to form a complete sequence, which is the second prediction sequence.

[0057] For the third offshore wind power data, it is obtained by satellite monitoring, including satellite observed cloud thickness, light intensity, sea surface temperature and other satellite parameters, and the third wind turbine operating parameters and the third wind power. These data are directly input into the third prediction model (such as a model based on convolutional neural network CNN, which can effectively process the relationship between satellite image derived data and wind power) pre-trained to predict wind power, and the third prediction sequence is obtained. For the fourth offshore wind power data, it is derived from unmanned aerial vehicle inspection, including local wind speed changes, wind turbine appearance states and other unmanned aerial vehicle parameters measured by unmanned aerial vehicle, and the fourth wind turbine operating parameters and the fourth wind power. These data are input into the fourth prediction model (such as a model based on recurrent neural network RNN, which can process time series data collected by unmanned aerial vehicle) specially trained to generate the fourth prediction sequence. These second, third and fourth prediction sequences provide multi-dimensional data support for subsequent comprehensive correction of prediction results, which helps to improve the accuracy of offshore wind power prediction.

[0058] In a possible implementation manner, the step S300 further includes:

[0059] Step S310: The time interval of the first offshore wind power data is counted, and an excessive time interval is identified to obtain a first missing detection result;

[0060] Step S320: Continuous null detection is performed on the first offshore wind power data, and an effective missing threshold is used to judge continuous null to obtain a second missing detection result;

[0061] Step S330: The first missing detection result and the second missing detection result are combined as a missing detection result.

[0062] Specifically, by analyzing the time stamp information attached to each data record in the data set, the time interval between adjacent data records is calculated. An excessive time interval standard is pre-set, which is assumed to be 15 minutes. The calculated time interval is compared with the excessive time interval one by one, and if a certain time interval is greater than 15 minutes, the time period corresponding to the time interval is marked as a region where data missing may exist, and the collection of these marked regions constitutes the first missing detection result.

[0063] Continuous null detection is performed on the first offshore wind power data. Each data field (such as wind speed, power and other fields) in the data set is traversed, and when the value of a certain field is empty, counting begins. An effective missing threshold is set, such as 5 consecutive nulls. If the number of consecutive nulls of a certain field reaches or exceeds 5 during counting, the data region containing the consecutive nulls is marked, and these marked regions constitute the second missing detection result.

[0064] The first missing detection result and the second missing detection result are combined. The possible missing area identified through the excessive time interval is integrated with the missing area marked by the continuous null detection, and the repeatedly marked part is removed to form a unified missing detection result. This result clearly indicates all possible data missing positions and ranges in the first offshore wind power data, and provides an important basis for subsequent data processing and prediction model optimization, so as to discover and handle the data missing problem in time and improve the accuracy of offshore wind power prediction.

[0065] In a possible implementation manner, the step S300 further includes:

[0066] Step S340: judging whether the first offshore wind power data is abnormal by using statistical characteristics to obtain a first noise detection result.

[0067] Step S350: presetting a sliding window to calculate the local statistical characteristics of the first offshore wind power data, judging whether each sliding window data meets a sliding window threshold value, and obtaining a second noise detection result.

[0068] Step S360: performing data change rate abnormality identification on the first offshore wind power data to obtain a third noise detection result.

[0069] Step S370: combining the first noise detection result, the second noise detection result and the third noise detection result as a noise detection result.

[0070] Step S380: extracting missing detection data through the missing detection result, extracting noise detection data through the noise detection result, and combining the missing detection data and the noise detection data as first bias data.

[0071] Specifically, whether the data is abnormal is judged by using statistical characteristics. The statistical quantities such as the mean value and the standard deviation of each parameter (such as the wind speed, the fan speed, the power generation power and the like) in the first offshore wind power data are calculated. For the wind speed parameter, it is assumed that the mean value of the historical data is 10 meters / second, and the standard deviation is 2 meters / second. A reasonable abnormal range is set, for example, when the wind speed data is greater than the mean value plus 3 times the standard deviation (i.e. 10+3×2=16 meters / second) or less than the mean value minus 3 times the standard deviation (i.e. 10-3×2=4 meters / second), the wind speed data is determined as an abnormal value. Similar judgment is performed on all data parameters, and the data records in which these abnormal values are located are marked to form a first noise detection result.

[0072] A sliding window is preset, for example, the window size contains 10 consecutive data points. The window is slid on the first offshore wind power data from the starting position, and the local statistical characteristics of the data in each window are calculated, such as the local mean and the local standard deviation. For each sliding window, a corresponding sliding window threshold is set. Taking wind speed data as an example, assuming that the local mean of the wind speed data in the window is 12 m / s, the sliding window threshold of the local mean is set to be floating up and down by 2 m / s, i.e. the normal range is 10-14 m / s. If the local mean of the wind speed data in a certain sliding window exceeds this range, or the local standard deviation is too large (indicating that the data fluctuation is abnormal), it is determined that the data in the window has noise, and the original data records corresponding to these noisy windows are marked to obtain a second noise detection result.

[0073] The data change rate anomaly of the first offshore wind power data is identified. The data change rate between adjacent data points is calculated. Taking wind speed as an example, if the current wind speed is 10 m / s, the wind speed of the next data point is 15 m / s, and the time interval is 1 minute, then the wind speed change rate is (15-10) ÷ 1 = 5 m / (s·min). A reasonable data change rate threshold is set, such as the normal wind speed change rate is between 0-3 m / (s·min). When the data change rate exceeds this threshold, the corresponding wind speed data is marked as abnormal change data, and these marked data constitute a third noise detection result.

[0074] The first noise detection result, the second noise detection result and the third noise detection result are combined. Remove the repeatedly marked data records, integrate all the data judged to have noise, and form a unified noise detection result.

[0075] According to the previously obtained missing detection result, the data records marked as missing are extracted from the original first offshore wind power data, i.e. missing detection data. Similarly, according to the noise detection result, the data records marked as noise are extracted, i.e. noise detection data. The missing detection data and the noise detection data are combined together to obtain the first deviation data. These first deviation data reflect the missing and noise problems in the first offshore wind power data, provide key information for subsequent data correction and optimization of prediction model, and help to improve the accuracy and reliability of offshore wind power prediction.

[0076] In one possible implementation manner, the step S500 further includes:

[0077] Step S510: Obtain a correction model set, and extract a first correction model of the correction model set.

[0078] Step S520: simulate the deviation of the deviation data to the second prediction sequence by the first correction model, obtain a first simulated correction prediction result, calculate the correction error of the first simulated correction prediction result by the correction sample, and generate a first simulated correction error coefficient.

[0079] Step S530: simulate the deviation of the deviation data to the second prediction sequence by the correction model set, obtain a second simulated correction prediction result to an Nth simulated correction prediction result, and calculate a second simulated correction error coefficient to an Nth simulated correction error coefficient.

[0080] Step S540: arrange the first simulated correction error coefficient, the second simulated correction error coefficient to the Nth simulated correction error coefficient in descending order, and select the correction model corresponding to the minimum value of the simulated correction error coefficient as the second correction model of the second prediction sequence.

[0081] Step S550: select a third correction model corresponding to the third prediction sequence and a fourth correction model corresponding to the fourth prediction sequence from the correction model set.

[0082] Step S560: correct the deviation of the deviation data to the second prediction sequence by the second correction model, obtain a second correction prediction result, calculate the correction error of the second correction prediction result based on the correction sample, generate a second correction error coefficient, correct the deviation of the deviation data to the third prediction sequence by the third correction model, obtain a third correction prediction result, calculate the correction error of the third correction prediction result based on the correction sample, generate a third correction error coefficient, correct the deviation of the deviation data to the fourth prediction sequence by the fourth correction model, obtain a fourth correction prediction result, calculate the correction error of the fourth correction prediction result based on the correction sample, and generate a fourth correction error coefficient.

[0083] Step S570: arrange the second correction error coefficient, the third correction error coefficient, and the fourth correction error coefficient in descending order, and prioritize the correction prediction result corresponding to the arrangement result to correct the deviation data, to obtain the correction prediction result.

[0084] Specifically, first, a correction model set is constructed. Based on the deep analysis of offshore wind power data and the experience accumulated in the past wind power prediction practice, a variety of models with different characteristics and advantages are selected to form the correction model set. Among them, the linear regression model can effectively fit the linear relationship between the wind power prediction deviation and the related factors according to the least square method principle, providing basic support for preliminary deviation correction. The decision tree model is also integrated into the model set, which uses tree structure to recursively divide the data and make decisions based on different values of data characteristics, so as to accurately analyze and process complex data structures and nonlinear deviation relationships. In addition, neural network models also become an important part of the model set, especially multilayer perceptron (MLP), recurrent neural network (RNN) and its variants such as long short-term memory network LSTM and gated recurrent unit GRU. These neural network models have strong nonlinear mapping ability through a large number of neuron connections and complex weight adjustment mechanisms, which can effectively capture the complex internal patterns and potential laws in the data, and show good correction performance. After the construction of the correction model set, the first correction model needs to be extracted from it. The correction model set is usually stored in a specific data structure, such as an ordered list, hash table or database table, etc. Taking the ordered list storage as an example, the model with index value 0 in the list is determined as the first correction model through the pre-set index rule. Through the call of the index value "0", the first model in the list, i.e. the first correction model, is accurately obtained. This first correction model will be used as the starting model for the subsequent simulation correction process, and its correction result will be used as an important reference basis to compare and evaluate the correction effect of other models, and then select the optimal correction model for different prediction sequences to realize the accurate deviation correction of offshore wind power prediction results and improve the accuracy and reliability of the prediction.

[0085] The first correction model (such as linear regression model) is used to simulate the deviation of the deviation data on the second prediction sequence. The deviation data is combined with the second prediction sequence as the input data of the linear regression model. The model operates according to the input data and outputs the first simulation correction prediction result. In order to evaluate the accuracy of this result, a correction sample is needed. The correction sample is a data set with known accurate wind power values. By calculating the difference between the first simulation correction prediction result and the corresponding actual value in the correction sample, such as the sum of the squares of the difference, and then according to certain calculation rules, the first simulation correction error coefficient is obtained, which can reflect the error size of this simulation correction.

[0086] The other models in the correction model set are traversed, and the deviation of the deviation data on the second prediction sequence is simulated and corrected by using these models in turn. Assuming that there are 5 models in the correction model set, after completing the simulation and correction by using the first linear regression model, the simulation and correction is performed by using the second decision tree model to obtain the second simulation and correction prediction result, and the second simulation and correction error coefficient is calculated by using the correction sample. In this way, the simulation and correction is performed by using the fifth model to obtain the fifth simulation and correction prediction result, and the fifth simulation and correction error coefficient is calculated.

[0087] All obtained simulation and correction error coefficients, i.e., the first simulation and correction error coefficient to the fifth simulation and correction error coefficient, are arranged in descending order. Among these coefficients, the simulation and correction error coefficient corresponding to the model with the smallest value is selected as the second correction model for formally correcting the second prediction sequence. For example, if it is found after sorting that the simulation and correction error coefficient corresponding to the third model (assuming that it is a neural network model) is the smallest, the neural network model is determined as the second correction model.

[0088] The third correction model suitable for the third prediction sequence and the fourth correction model suitable for the fourth prediction sequence are selected from the correction model set according to the same method. These two models are determined through the simulation and correction and error coefficient calculation and comparison processes.

[0089] The second correction model determined is used to formally correct the deviation of the deviation data on the second prediction sequence to obtain the second correction prediction result. The second correction error coefficient is generated again by calculating the error between the second correction prediction result and the actual value in the correction sample. Similarly, the third correction model is used to correct the deviation of the deviation data on the third prediction sequence to obtain the third correction prediction result, and the third correction error coefficient is calculated; the fourth correction model is used to correct the deviation of the deviation data on the fourth prediction sequence to obtain the fourth correction prediction result, and the fourth correction error coefficient is calculated.

[0090] The second correction error coefficient, the third correction error coefficient and the fourth correction error coefficient are arranged in descending order. According to the arrangement result, the prediction result with the smallest correction error coefficient is preferentially selected to correct the deviation data. For example, if the second correction error coefficient is the smallest, the second correction prediction result is mainly used to correct the deviation data. The corrected data is integrated to finally obtain the correction prediction result. Compared with the previous prediction result without correction, the correction prediction result can more accurately reflect the actual situation of the offshore wind power, and the accuracy of the offshore wind power prediction is improved.

[0091] In the second embodiment, the same inventive concept as the offshore wind power dynamic prediction method based on multi-model integration in the foregoing embodiments is used, as Figure 2As shown, the application provides a multi-model integrated offshore wind power dynamic prediction system, and the system and method embodiments in the application are based on the same inventive concept. The system comprises:

[0092] The offshore wind power data acquisition module 10 is configured to drive the first acquisition channel to acquire first offshore wind power data.

[0093] The first prediction sequence generation module 20 is configured to perform wind power prediction based on a first prediction model through the first offshore wind power data to generate a first prediction sequence.

[0094] The bias data acquisition module 30 is configured to perform missing detection and noise detection on the first offshore wind power data to obtain bias data.

[0095] The wind power prediction module 40 is configured to extract second offshore wind power data of a second acquisition channel, third offshore wind power data of a third acquisition channel, and fourth offshore wind power data of a fourth acquisition channel to perform wind power prediction of each prediction model to generate a second prediction sequence, a third prediction sequence, and a fourth prediction sequence.

[0096] The corrected prediction result acquisition module 50 is configured to perform bias correction on the bias data with the second prediction sequence, the third prediction sequence, and the fourth prediction sequence to obtain a corrected prediction result.

[0097] Further, the first prediction sequence generation module 20 further comprises:

[0098] The first continuous window sample acquisition unit is configured to extract K sample sensing parameters, K sample first wind turbine operating parameters, and next K sample first wind power through K continuous windows to acquire K first continuous window samples.

[0099] The continuous window prediction model generation unit is configured to train K continuous window neural network model architectures with the K first continuous window samples to generate K continuous window prediction models, wherein the K continuous window neural network model architectures are the same neural network model architectures.

[0100] The first cut wind power sequence acquisition unit is configured to perform data cutting on the first offshore wind power data with the K continuous windows to obtain K first cut wind power sequences, wherein the first offshore wind power data comprises sensing parameters, first wind turbine operating parameters, and first wind power.

[0101] The continuous window wind power prediction power generation unit is configured to match the K first cut wind power sequences with the K continuous windows, and input the K first cut wind power sequences into K continuous window prediction models corresponding to the matched continuous windows to perform wind power prediction, thereby generating K first continuous window wind power prediction powers corresponding to the next K continuous windows.

[0102] The first prediction sequence acquisition unit is configured to sequentially paste the K first continuous window wind power prediction powers corresponding to the next K continuous windows to generate the first prediction sequence.

[0103] Further, the first continuous window sample acquisition unit further comprises:

[0104] The preliminary continuous window acquisition unit is configured to divide a continuous time of a preset time length in a preset window division granularity to obtain K preliminary continuous windows.

[0105] The sliding update unit is configured to calculate a parameter fluctuation rate of the first offshore wind power data, update the window division granularity, and perform sliding update on the K preliminary continuous windows to obtain the K continuous windows.

[0106] Further, the wind power prediction module 40 further comprises:

[0107] The second continuous window sample acquisition unit is configured to extract K sample buoy parameters, K sample second wind turbine operation parameters and next K sample second wind power based on the K continuous windows to obtain K second continuous window samples.

[0108] The second prediction model training unit is configured to train K second prediction models by using the K second continuous window samples, wherein the K second prediction models are the same type of models.

[0109] The second cut wind power sequence acquisition unit is configured to perform data cutting on the second offshore wind power data by using the K continuous windows to obtain K second cut wind power sequences, wherein the second offshore wind power data comprises buoy parameters, second wind turbine operation parameters and second wind power.

[0110] The second continuous window wind power prediction power generation unit is configured to input the K second cut wind power sequences into K second prediction models corresponding to the matched continuous windows to perform wind power prediction, thereby generating K second continuous window wind power prediction powers corresponding to the next K continuous windows.

[0111] The second prediction sequence acquisition unit is configured to sequentially paste the K second continuous window wind power prediction powers corresponding to the next K continuous windows to obtain the second prediction sequence.

[0112] A fourth prediction sequence generation unit is configured to input the third offshore wind power data into a third prediction model to perform wind power prediction and generate a third prediction sequence, and input the fourth offshore wind power data into a fourth prediction model to perform wind power prediction and generate a fourth prediction sequence, wherein the third offshore wind power data comprises satellite parameters, third wind turbine operation parameters and third wind power, and the fourth offshore wind power data comprises unmanned aerial vehicle parameters, fourth wind turbine operation parameters and fourth wind power.

[0113] Further, the deviation data acquisition module 30 further comprises:

[0114] A first missing detection result acquisition unit is configured to count the collection time interval of the first offshore wind power data, identify the collection time interval according to an excessive time interval, and acquire a first missing detection result.

[0115] A second missing detection result acquisition unit is configured to perform continuous null value detection on the first offshore wind power data, judge the continuous null value according to an effective missing threshold, and obtain a second missing detection result.

[0116] A missing detection result generation unit is configured to combine the first missing detection result and the second missing detection result as a missing detection result.

[0117] Further, the deviation data acquisition module 30 further comprises:

[0118] A first noise detection result acquisition unit is configured to judge whether the first offshore wind power data is abnormal by using statistical characteristics, and obtain a first noise detection result.

[0119] A second noise detection result acquisition unit is configured to preset a sliding window to calculate the local statistical characteristics of the first offshore wind power data, judge whether each sliding window data meets a sliding window threshold, and obtain a second noise detection result.

[0120] A third noise detection result acquisition unit is configured to perform data change rate abnormality identification on the first offshore wind power data, and obtain a third noise detection result.

[0121] A noise detection result generation unit is configured to combine the first noise detection result, the second noise detection result and the third noise detection result as a noise detection result.

[0122] A detection data merging unit is configured to extract missing detection data by using the missing detection result, extract noise detection data by using the noise detection result, and combine the missing detection data and the noise detection data as first deviation data.

[0123] Further, the correction prediction result acquisition module 50 further comprises:

[0124] a first correction model obtaining unit, configured to obtain a correction model set, and extract a first correction model from the correction model set.

[0125] a first simulated correction error coefficient generating unit, configured to simulate correction of the deviation of the deviation data to the second prediction sequence by the first correction model to obtain a first simulated correction prediction result, calculate a correction error of the first simulated correction prediction result based on the correction sample, and generate a first simulated correction error coefficient.

[0126] an Nth simulated correction error coefficient obtaining unit, configured to simulate correction of the deviation of the deviation data to the second prediction sequence by the correction model set to obtain a second simulated correction prediction result to an Nth simulated correction prediction result, and calculate a second simulated correction error coefficient to an Nth simulated correction error coefficient.

[0127] a second correction model obtaining unit, configured to arrange the first simulated correction error coefficient, the second simulated correction error coefficient to the Nth simulated correction error coefficient in descending order, and select a correction model corresponding to a minimum value of the simulated correction error coefficient as a second correction model of the second prediction sequence.

[0128] a correction model selecting unit, configured to select a third correction model corresponding to the third prediction sequence and a fourth correction model corresponding to the fourth prediction sequence from the correction model set.

[0129] a correction error coefficient generating unit, configured to correct the deviation of the deviation data to the second prediction sequence by the second correction model to obtain a second correction prediction result, calculate a correction error of the second correction prediction result based on the correction sample, and generate a second correction error coefficient, correct the deviation of the deviation data to the third prediction sequence by the third correction model to obtain a third correction prediction result, calculate a correction error of the third correction prediction result based on the correction sample, and generate a third correction error coefficient, correct the deviation of the deviation data to the fourth prediction sequence by the fourth correction model to obtain a fourth correction prediction result, calculate a correction error of the fourth correction prediction result based on the correction sample, and generate a fourth correction error coefficient.

[0130] a priority correction unit, configured to arrange the second correction error coefficient, the third correction error coefficient and the fourth correction error coefficient in descending order, and correct the deviation data by correction prediction results corresponding to arrangement results to obtain the correction prediction result.

[0131] It should be noted that the above-mentioned embodiment sequences of the present application are merely for description only, but not for representing the advantages and disadvantages of the embodiments. And the above-mentioned embodiments of the present specification have been described. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0132] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0133] The specification and drawings are merely exemplary of the present application, and any and all modifications, variations, combinations or equivalents that are within the scope of the present application should be included. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A multi-model integrated offshore wind power dynamic prediction method, characterized in that, The method comprises the steps of: driving the first acquisition channel to acquire first offshore wind power data; performing wind power prediction based on a first prediction model through the first offshore wind power data to generate a first prediction sequence; detecting missing data and noise in the first offshore wind power data to obtain bias data; extracting second offshore wind power data of a second acquisition channel, third offshore wind power data of a third acquisition channel and fourth offshore wind power data of a fourth acquisition channel to perform wind power prediction of each prediction model to generate a second prediction sequence, a third prediction sequence and a fourth prediction sequence; performing bias correction on the bias data by using the second prediction sequence, the third prediction sequence and the fourth prediction sequence to obtain a corrected prediction result.

2. The multi-model integrated offshore wind power dynamic prediction method of claim 1, wherein, performing wind power prediction based on a first prediction model through the first offshore wind power data to generate a first prediction sequence, comprising: extracting K sample sensing parameters, K sample first wind turbine operating parameters and next K sample first wind power through K continuous windows to obtain K first continuous window samples; training K continuous window neural network model architectures with the K first continuous window samples to generate K continuous window prediction models, wherein the K continuous window neural network model architectures are the same kind of neural network model architectures; performing data cutting on the first offshore wind power data with the K continuous windows to obtain K first cut wind power sequences, wherein the first offshore wind power data comprises sensing parameters, first wind turbine operating parameters and first wind power; matching the K first cut wind power sequences with the K continuous windows, and inputting the K first cut wind power sequences into the K continuous window prediction models corresponding to the matched continuous windows to perform wind power prediction, thereby generating K first continuous window wind power prediction powers corresponding to next K continuous windows; sequentially pasting the K first continuous window wind power prediction powers corresponding to next K continuous windows to generate the first prediction sequence.

3. The multi-model integrated offshore wind power dynamic prediction method of claim 2, wherein, obtaining K continuous windows, comprising: dividing a continuous time of a preset time length with a preset window division granularity to obtain K preliminary continuous windows; calculating a parameter fluctuation rate of the first offshore wind power data, updating the window division granularity, and slidingly updating the K preliminary continuous windows to obtain the K continuous windows.

4. The multi-model integrated offshore wind power dynamic prediction method of claim 2, wherein, extracting second offshore wind power data of a second acquisition channel, third offshore wind power data of a third acquisition channel and fourth offshore wind power data of a fourth acquisition channel to perform wind power prediction of each prediction model to generate a second prediction sequence, a third prediction sequence and a fourth prediction sequence, comprising: extracting K sample buoy parameters, K sample second wind turbine operating parameters and next K sample second wind power based on the K continuous windows to obtain K second continuous window samples; training K second prediction models with the K second continuous window samples, wherein the K second prediction models are the same kind of models; performing data cutting on the second offshore wind power data with the K continuous windows to obtain K second cut wind power sequences, wherein the second offshore wind power data comprises buoy parameters, second wind turbine operating parameters and second wind power; inputting the K second cut wind power sequences into K second prediction models corresponding to matched continuous windows to perform wind power prediction, to generate K second continuous window wind power prediction powers corresponding to next K continuous windows; sequentially pasting the K second continuous window wind power prediction powers corresponding to the next K continuous windows to obtain the second prediction sequence; inputting third offshore wind power data into a third prediction model to perform wind power prediction to generate the third prediction sequence, and inputting fourth offshore wind power data into a fourth prediction model to perform wind power prediction to generate the fourth prediction sequence, wherein the third offshore wind power data comprises satellite parameters, third wind turbine operation parameters and third wind power, and the fourth offshore wind power data comprises unmanned aerial vehicle parameters, fourth wind turbine operation parameters and fourth wind power.

5. The multi-model integrated offshore wind power dynamic prediction method of claim 1, wherein, The first offshore wind power data is subjected to missing detection, comprising: statistically analyzing time intervals of the first offshore wind power data to identify the time intervals with excessive time intervals, to obtain a first missing detection result; performing continuous null value detection on the first offshore wind power data to determine continuous null values with an effective missing threshold, to obtain a second missing detection result; combining the first missing detection result and the second missing detection result as a missing detection result.

6. The multi-model integrated offshore wind power dynamic prediction method of claim 5, wherein, Obtaining bias data, comprising: judging whether the first offshore wind power data is abnormal by using statistical characteristics to obtain a first noise detection result; presetting a sliding window to calculate local statistical characteristics of the first offshore wind power data, judging whether each sliding window data meets a sliding window threshold to obtain a second noise detection result; performing data change rate anomaly recognition on the first offshore wind power data to obtain a third noise detection result; combining the first noise detection result, the second noise detection result and the third noise detection result as a noise detection result; extracting missing detection data through the missing detection result and extracting noise detection data through the noise detection result, and combining the missing detection data and the noise detection data as first bias data.

7. The multi-model integrated offshore wind power dynamic prediction method of claim 1, wherein, Performing bias correction on the second prediction sequence, the third prediction sequence and the fourth prediction sequence with the bias data to obtain a corrected prediction result, comprising: obtaining a correction model set and extracting a first correction model of the correction model set; simulating correction of the bias data on the second prediction sequence through the first correction model to obtain a first simulated correction prediction result, calculating correction errors of the first simulated correction prediction result through correction samples to generate a first simulated correction error coefficient; iterating the correction model set to simulate correction of the bias data on the second prediction sequence to obtain a second simulated correction prediction result to an Nth simulated correction prediction result, and to calculate a second simulated correction error coefficient to an Nth simulated correction error coefficient; sequentially arranging the first simulated correction error coefficient, the second simulated correction error coefficient to the Nth simulated correction error coefficient in descending order, and selecting a correction model corresponding to a minimum simulated correction error coefficient as a second correction model of the second prediction sequence; select a third correction model corresponding to the third prediction sequence and a fourth correction model corresponding to the fourth prediction sequence from the correction model set; correct the deviation of the second prediction sequence caused by the deviation data through the second correction model to obtain a second correction prediction result, calculate a correction error of the second correction prediction result based on the correction sample, generate a second correction error coefficient, correct the deviation of the third prediction sequence caused by the deviation data through the third correction model to obtain a third correction prediction result, calculate a correction error of the third correction prediction result based on the correction sample, generate a third correction error coefficient, correct the deviation of the fourth prediction sequence caused by the deviation data through the fourth correction model to obtain a fourth correction prediction result, calculate a correction error of the fourth correction prediction result based on the correction sample, and generate a fourth correction error coefficient; arrange the second correction error coefficient, the third correction error coefficient, and the fourth correction error coefficient in descending order, and correct the deviation data with the correction prediction result corresponding to the arrangement result in priority to obtain the correction prediction result.

8. A multi-model integrated offshore wind power dynamic prediction system, characterized in that, The system is used to implement the multi-model integrated offshore wind power dynamic prediction method of any one of claims 1-7, and the system comprises: an offshore wind power data acquisition module configured to drive a first acquisition channel to acquire first offshore wind power data; a first prediction sequence generation module configured to perform wind power prediction based on a first prediction model through the first offshore wind power data to generate a first prediction sequence; a deviation data acquisition module configured to perform missing detection and noise detection on the first offshore wind power data to obtain deviation data; a wind power prediction module configured to extract second offshore wind power data of a second acquisition channel, third offshore wind power data of a third acquisition channel, and fourth offshore wind power data of a fourth acquisition channel to perform wind power prediction of each prediction model to generate a second prediction sequence, a third prediction sequence, and a fourth prediction sequence; a correction prediction result acquisition module configured to correct the deviation data with the second prediction sequence, the third prediction sequence, and the fourth prediction sequence to obtain a correction prediction result.

Images