Offshore wind plant generation power prediction method and system based on big data

Abstract

The invention relates to the technical field of big data processing, in particular to an offshore wind plant generation power prediction method and system based on big data. According to the method, an ultra-short-term generated power prediction model based on data and physical hybrid driving is adopted to predict the generated power of the offshore wind plant, and the ultra-short-term generated power prediction model based on data and physical hybrid driving comprises a deep learning model, a physical model module, a reinforcement learning module and a dynamic switching mechanism; the dynamic switching mechanism is used for dynamically selecting the deep learning model module or the physical model module according to a wind speed fluctuation threshold value and feedback of the reinforcement learning module; according to the model selection logic provided by the invention, the prediction precision can be remarkably improved; meanwhile, in combination with feedback of a wind speed fluctuation threshold value and reinforcement learning, a dynamic switching mechanism improves the robustness and generalization ability of a prediction model.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and in particular to a method and system for predicting the generated power of an offshore wind farm based on big data. Background Art

[0002] In recent years, with the increasing global demand for clean energy, offshore wind power has rapidly developed as an efficient, clean, and sustainable energy source. Offshore wind farms are typically located in waters far from land, offering advantages such as abundant wind energy resources and a small footprint. However, the complex operating environment of offshore wind farms makes predicting their power generation difficult.

[0003] Existing technologies for offshore wind farm power generation forecasting employ hybrid models consisting of physical and data-driven models. However, these solutions lack a clear switching mechanism, resulting in low power generation forecast accuracy. Furthermore, existing technologies often lack time stamp alignment for offshore wind farm big data, further impacting power generation forecast accuracy. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides a method and system for predicting the power generation of an offshore wind farm based on big data, which are used to solve the problems existing in the prior art.

[0005] The present invention provides a method for predicting the generated power of an offshore wind farm based on big data, comprising the following steps: S1: Obtain big data for offshore wind farm power generation prediction; S2: performing a data preprocessing operation on the big data to obtain preprocessed big data; S3: Establish an offshore wind farm power generation prediction model; The offshore wind farm power generation prediction model is an ultra-short-term power generation prediction model based on data-physics hybrid drive, and the ultra-short-term power generation prediction model includes: A deep learning model for extracting time series features from the preprocessed big data; A physical model module, configured to perform power prediction based on the physical characteristics of the wind farm; A reinforcement learning module, configured to dynamically adjust a switching strategy of the ultra-short-term power generation prediction model; A dynamic switching mechanism is used to dynamically select the deep learning model module or the physical model module based on the wind speed fluctuation threshold and the feedback of the reinforcement learning module.

[0006] S4: Inputting the real-time wind power data in the pre-processed big data into the offshore wind farm power generation prediction model to obtain a power generation prediction result.

[0007] Preferably, in step S3, the input of the deep learning model module is the preprocessed big data; the network architecture of the deep learning model module includes: a multi-layer convolutional neural network, a long short-term memory network, a Transformer encoder and an output layer.

[0008] Preferably, the multi-layer convolutional neural network is used to extract spatial features from the preprocessed big data, and its specific structure includes: Convolutional layer: uses multiple convolution kernels of different sizes to extract multi-scale features; Pooling layer: Use maximum pooling or average pooling after each convolutional layer to reduce the feature dimension; Feature fusion layer: fuses the feature maps extracted by the convolutional layer.

[0009] Preferably, the long short-term memory network is used to capture the long-term dependencies of time series from the preprocessed big data, and its specific structure includes: Input layer: takes the feature map extracted by the multi-layer convolutional neural network as input; Multi-layer LSTM: Using a multi-layer LSTM structure, each LSTM layer performs deeper feature extraction on the input time series; each LSTM unit of the multi-layer LSTM contains an input gate, a forget gate, and an output gate to control the inflow, retention, and outflow of information; Dropout layer: A Dropout layer is added between LSTM layers to prevent overfitting.

[0010] Preferably, the Transformer encoder is used to further enhance the modeling capability of the deep learning model module for time series, and its specific structure includes: Input embedding layer: embeds the output feature map of each LSTM layer into the input space of the Transformer encoder; Multi-head self-attention mechanism: enables the Transformer encoder to capture features from different subspaces; Feedforward neural network: Each Transformer encoder contains a feedforward neural network for further processing features; Residual connection and normalization: used to improve model stability and training efficiency.

[0011] Preferably, in step S3, the function of the physical model module is to predict the generated power by utilizing the physical characteristics of the wind farm; the physical model module includes a wind speed-power curve fitting layer and a dynamic system model; wherein the wind speed-power curve fitting layer is a physical model that describes the relationship between wind speed and generated power; the dynamic system model is used to describe the dynamic relationship between wind speed changes and generated power output.

[0012] Preferably, the implementation steps of the dynamic switching mechanism are: Calculate wind speed fluctuations. At each time step t, calculate the difference between the wind speed at the current time step and the wind speed at the previous time step. Set wind speed fluctuation threshold; If the wind speed fluctuation exceeds the wind speed fluctuation threshold, the physical model module is selected to perform power generation prediction; if the wind speed fluctuation is less than or equal to the wind speed fluctuation threshold, the deep learning model module or the physical model module is selected based on the feedback from the reinforcement learning module.

[0013] Preferably, in step S2, data preprocessing includes timestamp alignment; the process of timestamp alignment is: Sa: Select the base time resolution based on the acquisition frequency of different types of data; Sb: Multi-level resampling is used to adjust the large data of different acquisition frequencies to the reference time resolution.

[0014] Preferably, the step Sb is specifically: Performing a first-level resampling on the large data of different frequencies to achieve preliminary alignment; performing a second-level resampling on the preliminarily aligned large data of different frequencies; The second-level resampling includes: performing multidimensional interpolation on sensor data that has undergone preliminary alignment and whose acquisition frequency is lower than the reference time resolution; and performing dynamic weight aggregation on sensor data that has undergone preliminary alignment and whose acquisition frequency is higher than the reference time resolution.

[0015] On the other hand, the present invention further provides an offshore wind farm power generation prediction system based on big data, which adopts the above-mentioned offshore wind farm power generation prediction method based on big data. The offshore wind farm power generation prediction system includes: A data acquisition module, used to acquire big data for offshore wind farm power generation prediction; A preprocessing module, used for performing data preprocessing operations on the big data; Model building module, used to build offshore wind farm power generation prediction model and obtain pre-processed big data; The power prediction module is used to input the real-time wind power data in the pre-processed big data into the offshore wind farm power generation prediction model to obtain a power generation prediction result.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention uses an ultra-short-term power generation prediction model driven by a data-physics hybrid to predict offshore wind farm power generation. The model includes a deep learning model module, a physical model module, a reinforcement learning module, and a dynamic switching mechanism. The dynamic switching mechanism dynamically selects either the deep learning model module or the physical model module based on a wind speed fluctuation threshold and feedback from the reinforcement learning module. The model selection logic provided by the present invention significantly improves the prediction accuracy of offshore wind farm power generation, reduces the average normalized root mean square error (RMSE), and verifies the effectiveness and practicality of the reinforcement learning module. The model selection logic provided by the present invention can dynamically adjust the prediction model based on real-time wind speed fluctuations and prediction errors to adapt to different operating conditions. Furthermore, the combination of wind speed fluctuation thresholds, reinforcement learning feedback, and the dynamic switching mechanism improves the robustness and generalization capabilities of the prediction model.

[0017] At the same time, the present invention uses a two-level resampling approach to achieve timestamp alignment for big data. In the second level of resampling, the timestamp alignment of the data is further optimized through two key steps: multidimensional interpolation and dynamic weight aggregation. The multidimensional interpolation method can simultaneously consider the time and data feature dimensions to generate more accurate interpolation results, while the dynamic weight aggregation method dynamically adjusts the aggregation weight according to the actual data acquisition frequency and timestamp deviation to generate more accurate aggregation results. The combination of these two steps can effectively handle the dynamic changes in acquisition frequency and timestamp deviation, further improving the temporal consistency of the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of a method for predicting offshore wind farm power generation based on big data provided by an embodiment of the present invention; Figure 2 This is a flowchart of timestamp alignment provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are also within the scope of protection of the present invention.

[0021] Example 1 like Figure 1 As shown, a method for predicting the power generation of an offshore wind farm based on big data includes the following steps: S1: Obtain big data for offshore wind farm power generation prediction; The big data includes current wind speed data, historical wind speed data and historical power generation data.

[0022] S2: performing a data preprocessing operation on the big data to obtain preprocessed big data; Data preprocessing directly affects the accuracy and reliability of subsequent data analysis and early warning models. In this step, data preprocessing includes data cleaning, timestamp alignment, data noise reduction, data normalization, and feature extraction: Data cleaning is used to remove abnormal and duplicate data to ensure data quality and reliability. Specifically, the big data from step S1 is checked by comparing the data timestamps and sensor measurements to identify identical duplicate data records. Data points outside the normal range are removed by setting reasonable thresholds, thus achieving pre-processing of abnormal data. For example, the normal range of a wind speed sensor is 0-30m / s, and the normal range of vibration frequency is 0-10Hz. Data points outside these ranges are considered abnormal data.

[0023] Timestamp alignment is to adjust inconsistent data of different types of timestamps to a unified time resolution while ensuring the integrity and accuracy of the data.

[0024] In this embodiment, if Figure 2 As shown in the figure, the timestamp alignment process includes: Sa: Select the base time resolution based on the acquisition frequency of different types of data; Among them, the least common multiple (LCM) of the acquisition frequencies of all types of data is selected as the benchmark time resolution. For types of data with lower acquisition frequencies, the time resolution is appropriately reduced, while for types of data with higher acquisition frequencies, a higher time resolution is maintained to retain more details.

[0025] Sb: Multi-level resampling is used to adjust the big data with different acquisition frequencies to the benchmark time resolution. The specific process includes: Perform the first level of resampling on big data of different frequencies to achieve preliminary alignment; The above-mentioned first-level resampling includes the first upsampling and the first downsampling; the first upsampling is to upsample the sensor data with an acquisition frequency lower than the benchmark time resolution, and use the linear interpolation method to generate data of the missing time points; the first downsampling is to downsample the sensor data with an acquisition frequency higher than the benchmark time resolution, and use the average aggregation method to merge multiple data points into one data point.

[0026] In practice, the hardware accuracy of different sensors may vary, which may cause the actual acquisition frequency to deviate from the nominal value. For example, the acquisition frequency of some sensors may become unstable during actual operation due to hardware aging, failure, or environmental interference. Or the clock mechanism within the sensor may drift, causing the acquisition frequency to gradually deviate from the nominal value. All of these situations make it difficult for the initial alignment step to fully unify all large data to the reference time resolution. Based on the above findings, this embodiment also includes a second-level resampling of the large data of different frequencies that have undergone initial alignment to further improve the accuracy of the time resolution alignment.

[0027] The second-level resampling includes: multi-dimensional interpolation of sensor data that have undergone preliminary alignment and whose acquisition frequency is lower than the benchmark time resolution; dynamic weight aggregation of sensor data that have undergone preliminary alignment and whose acquisition frequency is higher than the benchmark time resolution; Among them, multidimensional interpolation takes into account both the time and data feature dimensions, thereby improving the accuracy of interpolation. Specifically, in the big data of different frequencies after preliminary alignment, the time points that need to be interpolated are selected, and the semivariance function between the time points that need to be interpolated is calculated. The semivariance function describes the spatial correlation between data points; based on the semivariance function, a Kriging model is constructed. The Kriging model generates the values ​​of missing data points by minimizing the variance of the prediction error.

[0028] Dynamic weighted aggregation specifically involves calculating the time deviation of each data point in the preliminarily aligned large data sets of different frequencies. This time deviation refers to the difference between the actual timestamp of the data point and the ideal timestamp. Then, based on the time deviation, the aggregation weight is dynamically determined. The ideal timestamp is a standard time series generated according to the baseline time resolution. Furthermore, the size of the aggregation window is dynamically adjusted based on the statistical characteristics of the time deviation; specifically: Perform statistical analysis on the time deviation of all data points, calculate the mean and standard deviation, and dynamically determine the aggregation weight wi of each data point based on the mean and standard deviation; the calculation formula of the aggregation weight wi is: ; Where α is a tuning parameter used to control the dynamic range of the weight wi. When α is large, the weight changes more significantly; when α is small, the weight changes more slowly. μΔt is the mean of the time deviation, σΔt is the standard deviation of the time deviation, and Δti is the time deviation of the i-th data point. The data points are weightedly aggregated according to the aggregation weight of each data point, thereby achieving dynamic weight aggregation.

[0029] In the second level of resampling, data timestamp alignment is further optimized through two key steps: multidimensional interpolation and dynamic weighted aggregation. Multidimensional interpolation considers both time and data feature dimensions to produce more accurate interpolation results, while dynamic aggregation dynamically adjusts aggregation weights based on the actual data acquisition frequency and timestamp deviations, producing even more accurate aggregation results. The combination of these two steps effectively handles dynamic changes in acquisition frequency and timestamp deviations, further improving data temporal consistency.

[0030] S3: Establish an offshore wind farm power generation prediction model; In this step, the offshore wind farm power generation prediction model is an ultra-short-term power generation prediction model based on data-physics hybrid drive. The ultra-short-term power generation prediction model architecture combines deep learning, reinforcement learning and dynamic switching mechanism, which can more accurately capture the spatiotemporal characteristics of power generation and dynamically adjust the early warning strategy according to real-time wind speed fluctuations.

[0031] Specifically, the ultra-short-term power generation prediction model includes: Deep learning model module for extracting time series features from preprocessed big data; Physical model module, used for power prediction based on the physical characteristics of the wind farm; A reinforcement learning module is used to dynamically adjust the switching strategy of the ultra-short-term power generation forecast model; A dynamic switching mechanism is used to dynamically select a deep learning model module or a physical model module based on the wind speed fluctuation threshold and the feedback of the reinforcement learning module.

[0032] Furthermore, the input of the above-mentioned deep learning model module is preprocessed big data (preprocessed big data); the network architecture of the deep learning model module includes: a multi-layer convolutional neural network, a long short-term memory network, a Transformer encoder and an output layer.

[0033] Among them, the multi-layer convolutional neural network is used to extract spatial features from preprocessed big data. Its specific structure includes: Convolutional layer: Uses multiple convolution kernels of different sizes (such as 3×3, 5×5, and 7×7) to extract multi-scale features. Each convolutional layer is followed by a ReLU activation function.

[0034] Pooling layer: Use max pooling or average pooling after each convolutional layer to reduce feature dimensions and computation while retaining important features.

[0035] Feature fusion layer: The feature maps extracted by the convolutional layer are fused by using weighted summation, concatenation, or attention mechanism, so that the multi-layer convolutional neural network can simultaneously utilize feature information of different scales.

[0036] The Long Short-Term Memory Network is used to capture the long-term dependencies of time series in preprocessed big data. Its specific structure includes: Input layer: takes the feature map extracted by the multi-layer convolutional neural network as input; Multi-layer LSTM: Using a multi-layer LSTM structure, each LSTM layer performs deeper feature extraction on the input time series. Each LSTM unit of the multi-layer LSTM contains an input gate, a forget gate, and an output gate to control the inflow, retention, and outflow of information.

[0037] Dropout layer: A Dropout layer is added between LSTM layers to prevent overfitting.

[0038] The role of the Transformer encoder is to further enhance the deep learning model module's ability to model time series, especially by capturing global dependencies through the self-attention mechanism. Its specific structure includes: Input embedding layer: embeds the output feature map of each LSTM layer into the input space of the Transformer encoder; Multi-head self-attention mechanism: Through the multi-head self-attention mechanism, the Transformer encoder can capture features from different subspaces; Feedforward neural network: Each Transformer encoder contains a feedforward neural network for further processing features; Residual connection and normalization: Residual connection and layer normalization are used after each sub-layer to improve the stability of the model and training efficiency.

[0039] The output layer includes a fully connected layer and an output module, wherein the fully connected layer is used to integrate the output feature vector of the Transformer encoder, and the output module is used to generate the final power generation prediction value through a linear layer or a Softmax layer.

[0040] Specifically, the role of the above-mentioned physical model module is to use the physical characteristics of the wind farm to predict the generated power, especially when the wind speed fluctuates greatly, which can effectively make up for the shortcomings of the data-driven model.

[0041] Furthermore, the physical model module includes a wind speed-power curve fitting layer and a dynamic system model; wherein, the wind speed-power curve fitting layer is a physical model that describes the relationship between wind speed and generated power; the steps for establishing the wind speed-power curve of the wind speed-power curve fitting layer are: collecting historical data of the wind farm, including wind speed and corresponding generated power; and using polynomial regression to fit the relationship between wind speed and generated power.

[0042] The dynamic system model is used to describe the dynamic relationship between wind speed changes and power output. This dynamic system model takes into account the physical characteristics and dynamic response of the wind farm and can more accurately predict power generation. The specific steps for predicting power generation include: Establish a state-space model between wind speed and power to describe the dynamic behavior of offshore wind farm systems; Model parameters are estimated through historical data. Common methods include least squares method and maximum likelihood estimation. Model parameters are polynomial coefficients. Dynamic prediction: Based on the current state and model parameters, the power generation at the next moment is predicted; the current state includes wind speed, current output power, etc.

[0043] Reinforcement learning is a machine learning method that learns optimal behavioral strategies through the interaction between an agent and its environment. The agent selects actions based on its current state, and the environment provides rewards based on these actions. The agent learns the optimal strategy by maximizing the cumulative rewards. In an ultra-short-term power generation forecasting model based on a hybrid data-physics approach, the reinforcement learning module is used to dynamically adjust the model's switching strategy to select the optimal forecasting model (data-driven or physical model) based on real-time wind speed fluctuations and forecast errors.

[0044] Among them, the reinforcement learning module includes state space, action space, reward function, policy network, experience replay, and dynamic switching mechanism.

[0045] Furthermore, the state space is used to describe the observation information of the agent at each time step; in power generation prediction, the state space includes: Current wind speed fluctuation: the difference between the wind speed at the current moment and the wind speed at the previous moment; Historical prediction error: the prediction error over the past N time steps (the difference between the actual power and the predicted power); Current power output: the power output at the current moment; Historical prediction error: the prediction error of the past few time steps (the difference between the actual power and the predicted power); Current power output: the power output at the current moment.

[0046] Furthermore, the action space is used to describe the actions that the agent can take. In power generation prediction, the action space includes the following two options: Select the Deep Learning Model module: Use the deep learning model for prediction; Select the Physical Model module: Use the physical model to make predictions.

[0047] Furthermore, the reward function is used to evaluate the quality of the agent's actions. In power generation forecasting, the reward function is designed based on the prediction error. In this embodiment, the reward function is a negative error reward function, meaning that the reward value is the negative of the prediction error. The more accurate the prediction, the higher the reward.

[0048] Furthermore, the policy network is the core part of reinforcement learning, which is used to select the optimal action based on the current state. In this embodiment, the policy network is a deep Q-network, which uses a deep neural network to approximate the Q-value function and learn the optimal policy by maximizing the cumulative reward.

[0049] Furthermore, experience replay is a technology that improves the efficiency of reinforcement learning training. The intelligent agent stores the experience (state, action, reward, next state) of each interaction in the experience replay cache, and then randomly samples small batches of data from the cache for training to break the correlation between the data and improve the stability and convergence speed of training.

[0050] Furthermore, the dynamic switching module dynamically selects a data-driven model or a physical model for prediction according to the output of the policy network.

[0051] The dynamic switching mechanism dynamically selects the appropriate prediction module (data-driven model or physical model) based on the wind speed fluctuation threshold and the feedback from the reinforcement learning module. The implementation steps of the dynamic switching mechanism are as follows: Calculate wind speed fluctuations. At each time step t, calculate the difference between the current wind speed and the wind speed at the previous time step; Set wind speed fluctuation threshold; In this step, a wind speed fluctuation threshold is set based on historical data and experience. Specifically, the statistical characteristics of all wind speed fluctuations, such as the mean, standard deviation, maximum, and minimum, are calculated. Then, based on the wind farm's forecast accuracy requirements, an appropriate percentile is selected as the threshold. Common choices include the 90th, 95th, or 99th percentile. For example, selecting the 95th percentile means that in most cases (95% of the time), wind speed fluctuations will not exceed this value.

[0052] In this embodiment, the threshold is set to the 95% percentile of historical wind speed fluctuations.

[0053] According to the wind speed fluctuation threshold and the feedback of the reinforcement learning module, the data-driven model or the physical model is dynamically selected; the specific logic is: if the wind speed fluctuation exceeds the wind speed fluctuation threshold, the physical model is selected for power generation prediction; if the wind speed fluctuation is less than or equal to the wind speed fluctuation threshold, the deep learning model module or the physical model module is selected based on the feedback of the reinforcement learning module; wherein, the feedback of the reinforcement learning module is the output result of the reinforcement learning module, including two results, namely, selecting the deep learning model and selecting the physical model module.

[0054] Experimental validation was conducted on data from actual wind farms, comparing the prediction performance of a data-driven model alone with that of a physical model. The experimental results showed that the introduction of the reinforcement learning module enabled the model to dynamically adjust its prediction strategy based on real-time wind speed fluctuations. At the same time, compared with only using data-driven models or physical models to judge wind speed fluctuations, the model selection logic provided by this embodiment can significantly improve prediction accuracy and reduce the average normalized root mean square error (RMSE), verifying the effectiveness and practicality of the reinforcement learning module; at the same time, the model selection logic provided by this embodiment can dynamically adjust the prediction model according to real-time wind speed fluctuations and prediction errors to adapt to different operating conditions; at the same time, combined with the wind speed fluctuation threshold and reinforcement learning feedback, the dynamic switching mechanism improves the robustness and generalization ability of the prediction model.

[0055] S4: Input the real-time wind power data in the preprocessed big data into the offshore wind farm power generation prediction model to obtain the power generation prediction result.

[0056] The above power prediction was carried out at a wind farm in Shandong Province. One year of historical data was collected and power prediction was carried out based on the historical data. The power was then compared with the actual power in the historical data, which showed that the solution of this application has a good prediction effect.

[0057] At the same time, the average normalized root mean square error (RMSE) of different prediction methods (this embodiment, data-driven model only, and physical model only) was tested. This solution: 0.1589, data-driven model only: 0.2056, physical model only: 0.2234; by combining the advantages of the data-driven model and the physical model, the prediction error was significantly reduced and the prediction accuracy was improved; at the same time, the performance of this embodiment under different (seasonal) meteorological conditions was compared. In spring: this solution: RMSE = 0.1654, data-driven model only: RMSE =0.2102, physical model only: RMSE=0.2256; summer: this scheme: RMSE=0.1523, data-driven model only: RMSE=0.2034, physical model only: RMSE=0.2189; autumn: this scheme: RMSE=0.1487, data-driven model only: RMSE=0.1956, physical model only: RMSE=0.2145; winter: this scheme: RMSE=0.1712, data-driven model only: RMSE=0.2201, physical model only: RMSE =0.2312; It can be seen that in different seasons and meteorological conditions, the scheme of this embodiment shows higher prediction accuracy. Especially in winter, when wind speed fluctuates greatly, the advantages of the physical model are more obvious. The solution of this embodiment can effectively utilize the advantages of the physical model and improve prediction accuracy through a dynamic switching mechanism. At the same time, the model was also subjected to a long-term operation test for one year, and the monthly RMSE and favorable switching ratio were recorded. The results showed that the average RMSE for the whole year was 0.1623, and the monthly RMSE fluctuation range was 0.1456 to 0.1897; the average favorable switching ratio for the whole year was 89.54%, and the monthly favorable switching ratio fluctuation range was 85.23% to 92.15%, demonstrating that the solution of this embodiment exhibits high stability and robustness in long-term operation, can continuously provide high-precision wind power forecasts, and is suitable for the long-term operation needs of actual wind farms.

[0058] Example 2 The present invention further provides an offshore wind farm power generation prediction system based on big data, which can adopt an offshore wind farm power generation prediction method based on big data according to Example 1, comprising: A data acquisition module, used to acquire big data for offshore wind farm power generation prediction; A preprocessing module, used for performing data preprocessing operations on the big data; Model building module, which builds a power prediction model for offshore wind farms; The power prediction module is used to input the real-time wind power data in the big data into the offshore wind farm power generation prediction model to obtain the power generation prediction result.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than 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 they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting offshore wind farm power generation based on big data, characterized in that: The following steps are involved: S1: Obtain big data for offshore wind farm power generation prediction; S2: performing a data preprocessing operation on the big data to obtain preprocessed big data; S3: Establish an offshore wind farm power generation prediction model; The offshore wind farm power generation prediction model is an ultra-short-term power generation prediction model based on data-physics hybrid drive, and the ultra-short-term power generation prediction model includes: A deep learning model module, used to extract time series features from the preprocessed big data; A physical model module, configured to perform power prediction based on the physical characteristics of the wind farm; A reinforcement learning module, configured to dynamically adjust a switching strategy of the ultra-short-term power generation prediction model; A dynamic switching mechanism for dynamically selecting the deep learning model module or the physical model module based on the wind speed fluctuation threshold and feedback from the reinforcement learning module; S4: Inputting the real-time wind power data in the pre-processed big data into the offshore wind farm power generation prediction model to obtain a power generation prediction result.

2. The method for predicting offshore wind farm power generation based on big data according to claim 1, characterized in that: In step S3, the input of the deep learning model module is the pre-processed big data; The network architecture of the deep learning model module includes: a multi-layer convolutional neural network, a long short-term memory network, a Transformer encoder and an output layer.

3. The method for predicting offshore wind farm power generation based on big data according to claim 2, characterized in that: The multi-layer convolutional neural network is used to extract spatial features from the pre-processed big data, and its specific structure includes: Convolutional layer: Use multiple convolution kernels of different sizes to extract multi-scale features and obtain feature maps; Pooling layer: Use maximum pooling or average pooling after each convolutional layer to reduce the feature dimension; Feature fusion layer: fuses the feature maps extracted by the convolutional layer.

4. The method for predicting offshore wind farm power generation based on big data according to claim 2, characterized in that: The long short-term memory network is used to capture the long-term dependencies of time series from the preprocessed big data. Its specific structure includes: Input layer: takes the feature map extracted by the multi-layer convolutional neural network as input; Multi-layer LSTM: Using a multi-layer LSTM structure, each LSTM layer performs deeper feature extraction on the input time series; each LSTM unit of the multi-layer LSTM contains an input gate, a forget gate, and an output gate to control the inflow, retention, and outflow of information; Dropout layer: A Dropout layer is added between LSTM layers to prevent overfitting.

5. The method for predicting offshore wind farm power generation based on big data according to claim 4, characterized in that: The Transformer encoder is used to further enhance the modeling capability of the deep learning model module for the time series. Its specific structure includes: Input embedding layer: embeds the output feature map of each LSTM layer into the input space of the Transformer encoder; Multi-head self-attention mechanism: enables the Transformer encoder to capture features from different subspaces; Feedforward neural network: Each Transformer encoder contains a feedforward neural network for further processing features; Residual connection and normalization: used to improve model stability and training efficiency.

6. The method for predicting offshore wind farm power generation based on big data according to claim 1, characterized in that: In S3, the function of the physical model module is to predict the generated power by utilizing the physical characteristics of the wind farm; the physical model module includes a wind speed-power curve fitting layer and a dynamic system model; wherein the wind speed-power curve fitting layer is a physical model that describes the relationship between wind speed and generated power; the dynamic system model is used to describe the dynamic relationship between wind speed changes and generated power output.

7. The method for predicting offshore wind farm power generation based on big data according to claim 1, characterized in that: The implementation steps of the dynamic switching mechanism are: Calculate wind speed fluctuations. At each time step t, calculate the difference between the wind speed at the current time step and the wind speed at the previous time step. Set wind speed fluctuation threshold; If the wind speed fluctuation exceeds the wind speed fluctuation threshold, the physical model module is selected to perform power generation prediction; if the wind speed fluctuation is less than or equal to the wind speed fluctuation threshold, the deep learning model module or the physical model module is selected based on the feedback from the reinforcement learning module.

8. The method for predicting offshore wind farm power generation based on big data according to claim 1, characterized in that: In step S2, data preprocessing includes timestamp alignment; the process of timestamp alignment is as follows: Sa: Select the base time resolution based on the acquisition frequency of different types of data; Sb: Multi-level resampling is used to adjust the large data of different acquisition frequencies to the reference time resolution.

9. The method for predicting offshore wind farm power generation based on big data according to claim 8, characterized in that: The step Sb is specifically as follows: Performing a first-level resampling on the large data of different frequencies to achieve preliminary alignment; performing a second-level resampling on the preliminarily aligned large data of different frequencies; The second-level resampling includes: performing multi-dimensional interpolation on the sensor data that has been preliminarily aligned and whose acquisition frequency is lower than the reference time resolution; Dynamic weight aggregation is performed on sensor data that has been preliminarily aligned and has an acquisition frequency higher than the reference time resolution.

10. A big data-based offshore wind farm power generation prediction system, characterized in that: The offshore wind farm power generation prediction system adopts the offshore wind farm power generation prediction method based on big data according to any one of claims 1 to 9, which comprises: A data acquisition module, used to acquire big data for offshore wind farm power generation prediction; A preprocessing module, configured to perform a data preprocessing operation on the big data to obtain preprocessed big data; Model building module, used to build offshore wind farm power generation prediction model; The power prediction module is used to input the real-time wind power data in the pre-processed big data into the offshore wind farm power generation prediction model to obtain a power generation prediction result.

Images