A method, system, device and medium for predicting offshore wind power output

Abstract

This invention relates to a method, system, device, and medium for predicting offshore wind power output. The method includes the following steps: acquiring historical data on offshore wind power output and preprocessing it to obtain preprocessed data; inputting the preprocessed data into a Bi-LSTM neural network to obtain a first output; performing time-series decomposition on the preprocessed data to obtain an output trend; modeling the output trend using the Bi-LSTM neural network to obtain a second output; concatenating the first and second outputs to obtain a concatenated result; establishing an MLP model to extract features from meteorological factors affecting offshore wind power output prediction to obtain a third output; performing spatiotemporal attention fusion on the third output and the concatenated result to obtain fused features; and performing prediction based on the fused features to obtain a prediction result. Compared with existing technologies, this invention has advantages such as high accuracy and strong stability.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction technology, and in particular to a method, system, device and medium for predicting offshore wind power output. Background Technology

[0002] Electricity demand forecasting plays a crucial role in power system planning, providing a solid foundation for the economic operation of the power system. The core focus of this forecasting task is wind power. By accurately predicting the spatiotemporal distribution of wind power, reliable data is provided for power system planning and operational decisions. Accurate power forecasting directly impacts the utilization rate of generating equipment and the effectiveness of economic dispatch, further enhancing the security and stability of the power system. Conversely, inaccurate forecasting may lead to increased operating costs, greater power losses, increased economic pressure, and even negatively impact the stable operation of the power system and the supply-demand balance of the electricity market. Therefore, accurate wind power forecasting is of paramount importance.

[0003] However, traditional power forecasting methods often neglect the spatiotemporal correlations in power output data. Power data typically exhibits obvious seasonality, periodicity, and trend characteristics, which traditional methods struggle to fully capture, leading to decreased accuracy in power forecasting. Therefore, an innovative method is needed to better utilize the spatiotemporal correlations in power data to improve the accuracy and stability of power forecasting. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of poor accuracy in the prior art by providing a method, system, device and medium for predicting offshore wind power output.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for predicting offshore wind power output includes the following steps:

[0007] S1: Acquire historical data of offshore wind power output, preprocess it to obtain preprocessed data, input the preprocessed data into the Bi-LSTM neural network to obtain the first output;

[0008] S2: Perform time-series decomposition on the preprocessed data to obtain the power output trend. Model the power output trend using a Bi-LSTM neural network to obtain the second output. Concatenate the first and second outputs to obtain the concatenated result.

[0009] S3: For meteorological factors affecting offshore wind power output prediction, an MLP model is established to extract features and obtain a third output. The third output and the splicing result are then fused by spatiotemporal attention to obtain fused features.

[0010] S4: Make predictions based on fused features to obtain prediction results.

[0011] Furthermore, the calculation expression for preprocessing is as follows:

[0012]

[0013] In the formula, i represents the type of customer-side load, and j represents the time scale.

[0014] Furthermore, the computational expression for the propagation process of the Bi-LSTM neural network is as follows:

[0015] f t =σ(Y) f ·[h t-1 ,x t ]+b f )

[0016] i t =σ(Y) i ·[h t-1 ,x t ]+b i )

[0017]

[0018] o t =σ(Y) o ·[h t-1 ,x t ]+b o )

[0019] h t =o t ·tanh(C t )

[0020] In the formula, matrix Y f Y i Y c Y o Control the weights of the gate states respectively; It represents the state of the cell at time t, o t It is the output of the activation function sigmoid, σ is the S(x) function; the activation function tanh is the hyperbolic tangent function; [·,·] is h t-1 and x t The connection operation, f t It is the Gate of Oblivion, i t It's an input gate, b f b i b c b o It is the bias.

[0021] Furthermore, the time series decomposition yields trend components, seasonal components, and residual terms, and a ternary array of trend components, seasonal components, and residual terms is established as the output trend.

[0022] Furthermore, the computational expression for time series decomposition is:

[0023] X1(t) = H(t) + K(t) + R(t)

[0024] In the formula, X1(t) is the observed value at time t, H(t) represents the trend component at time t, K(t) represents the seasonal component at time t, and R(t) represents the residual term at time t.

[0025] The calculation expressions for the trend component and the seasonal component are as follows:

[0026]

[0027] In the formula, X1(t-1), X1(t-v+1), and X1(tv) represent the observed values ​​at times t-1, t-v+1, and tv, respectively, and X1(j) represents the observed value at the j-th time point in the dataset X1. This represents the average of all observations in dataset X1; θ represents the average value of all time points in the dataset; v represents the seasonal period length; θ0 and θ1 represent the trend constant and trend slope, respectively; t(j) represents the time point corresponding to the j-th observation.

[0028] Furthermore, the MLP model uses the backpropagation algorithm to update and adjust the weights. The output expression of the intermediate hidden layer of the MLP model is as follows:

[0029] H = f(X) h W h +b h )

[0030] In the formula, H is the output of the intermediate hidden layer, and X... h W is the input to the intermediate hidden layer. h b represents the weights of the intermediate hidden layers. h is the threshold of the intermediate hidden layer, and f is the activation function of the intermediate hidden layer;

[0031] The output expression of the output layer of the MLP model is:

[0032] O=g(X o W o +b o )

[0033] In the formula, O represents the output of the output layer, and X represents the output of the output layer. o W is the input to the output layer. ob represents the output layer weights. o denoted as the output layer threshold, and g is the activation function of the output layer.

[0034] Furthermore, in spatiotemporal attention fusion, the computational expression for attention allocation is:

[0035] e i =h(q,k) i )

[0036]

[0037] In the formula, h() is the attention scoring function, e i Let α be the attention value for the i-th key. i The corresponding weight is e, where e is the natural base.

[0038] A second aspect of the present invention provides an offshore wind power output prediction system, comprising:

[0039] Preprocessing module: Acquires historical data of offshore wind power output, performs normalization processing, uses Bi-LSTM to extract the normalization results, and obtains the first output;

[0040] Time-series decomposition module: By decomposing wind power output data over time, the wind power output trend is obtained, and Bi-LSTM is used to extract the wind power output trend to obtain the second output;

[0041] Model Combination Prediction Module: The module uses MLP to obtain the third output, concatenates the first, second, and third outputs, and performs prediction based on the concatenation result to obtain the prediction result.

[0042] Evaluation module: Evaluates the accuracy of the prediction results using mean absolute error, root mean square error, and coefficient of determination.

[0043] A third aspect of the present invention is an offshore wind power output prediction device, comprising a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement any of the offshore wind power output prediction methods described above.

[0044] A fourth aspect of the present invention is a storage medium having a program stored thereon, which, when executed, implements any of the above-mentioned methods for predicting offshore wind power output.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1) This invention adopts a method combining deep learning and machine learning, which extracts the original power features of wind power output data, the trend features obtained by time-series decomposition of wind power output data, and the external factor features extracted by MPL, and fuses them to obtain the prediction results. This can effectively improve the overall prediction performance, enhance the model's anti-interference ability, and improve the prediction accuracy.

[0047] 2) This invention enhances the accuracy and robustness of power prediction through the comprehensive application of multiple models, effectively captures the periodic fluctuations of wind power, and improves the stability of prediction by deeply exploring the randomness of data through two-dimensional spatial modeling. Attached Figure Description

[0048] Figure 1 This is a flowchart of the prediction method of the present invention.

[0049] Figure 2 A schematic diagram illustrating the acquisition of trend components, seasonal components, and residual components for STL decomposition.

[0050] Figure 3 The graph shows a comparison of the prediction curves of the Bi-LSTM combined with the temporal attention mechanism prediction model, and the prediction curves of the LSTM, Bi-LSTM without MLP, KAN, and Auto-form models. Detailed Implementation

[0051] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0052] Example 1

[0053] This invention relates to a method for predicting offshore wind power output, comprising the following steps:

[0054] S1: Acquire historical data of offshore wind power output, preprocess it to obtain preprocessed data, input the preprocessed data into the Bi-LSTM neural network to obtain the first output;

[0055] S1.1: Preprocess the historical power output data X of offshore wind power, mainly including correcting and filling out outliers and missing values ​​in the data, and normalizing the data. The processed dataset is represented as X'.

[0056] The following is the formula for data normalization:

[0057]

[0058] In the formula, i represents the type of customer-side load, and j represents the time scale.

[0059] S1.2: The Bi-LSTM neural network will process these features. The propagation process of the LSTM network is as follows:

[0060] f t =σ(Y) f ·[h t-1 ,x t ]+b f )

[0061] i t =σ(Y) i ·[h t-1 ,x t ]+b i )

[0062]

[0063] o t =σ(Y) o ·[h t-1 ,x t ]+b o )

[0064] h t =o t ·tanh(C t )

[0065] Wherein, matrix Y f Y i Y c Y o Control the weights of the gate states respectively; It represents the state of the cell at time t, o t It is the output of the activation function sigmoid, σ is the S(x) function; the activation function tanh is the hyperbolic tangent function; [·,·] is the output of h. t-1 and x t The connection operation. t It is the Gate of Oblivion, i t It's an input gate. f b i b c b o It is the bias.

[0066] The obtained original wind power characteristics are the first output.

[0067] S2: Perform time-series decomposition on the preprocessed data to obtain the power output trend. Model the power output trend using a Bi-LSTM neural network to obtain the second output. Concatenate the first and second outputs to obtain the concatenated result.

[0068] Offshore wind power output data is decomposed using STL, and time-series decomposition yields trend components, seasonal components, and residual terms. A triple array of trend components, seasonal components, and residual terms is established as the output trend. Bi-LSTM is used to model the output trend, and the resulting output is concatenated with the first output.

[0069] The calculation expression for time series decomposition is:

[0070] X1(t) = H(t) + K(t) + R(t)

[0071] In the formula, X1(t) is the observed value at time t, H(t) represents the trend component at time t, K(t) represents the seasonal component at time t, and R(t) represents the residual term at time t.

[0072] The calculation expressions for the trend component and the seasonal component are as follows:

[0073]

[0074] In the formula, X1(t-1), X1(t-v+1), and X1(tv) represent the observed values ​​at times t-1, t-v+1, and tv, respectively, and X1(j) represents the observed value at the j-th time point in the dataset X1. This represents the average of all observations in dataset X1; θ represents the average value of all time points in the dataset; v represents the seasonal period length; θ0 and θ1 represent the trend constant and trend slope, respectively; t(j) represents the time point corresponding to the j-th observation.

[0075] S3: For meteorological factors affecting offshore wind power output prediction, an MLP model is established to extract features and obtain a third output. The third output and the splicing result are then fused by spatiotemporal attention to obtain fused features.

[0076] The Model-Driven Resource (MLP) model is used to extract features from external factors (such as weather and market fluctuations) affecting offshore wind power output prediction. MLP typically employs the Back Propagation (BP) algorithm for weight updates and adjustments. The output expression of the intermediate hidden layer of the MLP model is as follows:

[0077] H = f(X) h W h +b h )

[0078] In the formula, H is the output of the intermediate hidden layer, and X... h W is the input to the intermediate hidden layer. h b represents the weights of the intermediate hidden layers. his the threshold of the intermediate hidden layer, and f is the activation function of the intermediate hidden layer, which typically includes ReLU, Logistic, and tanh.

[0079] The output expression of the output layer of the MLP model is:

[0080] O=g(X o W o +b o )

[0081] In the formula, O represents the output of the output layer, and X represents the output of the output layer. o W is the input to the output layer. o b represents the output layer weights. o denoted as the output layer threshold, and g is the activation function of the output layer.

[0082] The obtained external factor features are spatiotemporally attention-fused with the concatenated features of S2, with a set of keys k = (k1, k2, ..., k n ) and their corresponding values ​​v = (v1, v2, ..., v n With reference to query q, the final output o is obtained. The expression for calculating attention allocation is:

[0083] e i =h(q,k) i )

[0084]

[0085] In the formula, h() is the attention scoring function, e i Let α be the attention value for the i-th key. i The corresponding weight is e, where e is the natural base.

[0086] S4: Make predictions based on fused features to obtain prediction results.

[0087] Using the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) 2 Three metrics are used to measure the prediction accuracy of the combined prediction model trained in step 3.

[0088] The trained model is used to predict wind power output, and the prediction results are as follows: Figure 2 As shown, the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are used to... 2 The accuracy of the prediction algorithm is measured using the following formula:

[0089]

[0090] In the formula y i and These are the actual value and the predicted value at time i, respectively.

[0091] Example 2

[0092] This embodiment is based on the Python platform, and simulation and optimization analysis are performed under the WIN11 operating system, i9 CPU, 2.20GHz processor environment. The specific solution flowchart is as follows. Figure 1 As shown.

[0093] In this embodiment, the power output data of a certain offshore wind farm is used to predict the power output of the wind farm; 80% of the sample dataset is used as the training set and 20% as the test set. The training set is used to train the model and the test set is used to verify the effectiveness of the model.

[0094] To verify the effectiveness of the method proposed in Example 1, this example sets up the following five scenarios for comparative analysis: Scenario 1: Using the Bi-LSTM combined prediction model described in this invention to predict wind power; Scenario 2: Replacing Bi-LSTM in the method proposed in this paper with LSTM to predict offshore wind power; Scenario 3: Removing the MLP model in the method proposed in this paper and ignoring the influence of external factors on offshore wind power output prediction; Scenario 4: Using the KAN model to predict wind power; Scenario 5: Using the Autoformed model to predict wind power.

[0095] The comparison curves between the predicted and actual values ​​for the five scenarios are as follows: Figure 3 As shown, the wind power curve prediction obtained based on the Bi-LSTM combined prediction model is closer to the actual value. This is because the output data after MLP processing is smoother, and the STL decomposition method is used to decompose it into trend, seasonal and residual components, which improves the performance of the input data of the prediction model. The combined prediction model predicts different components separately, which enhances the model's anti-interference ability and better predicts the wind power curve. Therefore, the prediction curve obtained based on the Bi-LSTM combined prediction model is better than other single prediction models.

[0096] The evaluation parameters of the prediction models for the five scenarios are compared in Table 1 below:

[0097] Table 1 Comparison of Evaluation Parameters for Predictive Models

[0098]

[0099] As shown in Table 1, compared with other single prediction models, the Bi-LSTM-based combined prediction model can effectively improve the accuracy of wind power prediction. The prediction method proposed in this paper outperforms other comparative models in all indicators. Its performance in MAE and RMSE is 12.3157 and 17.2559, respectively, which is at least 0.6002 and 1.2460 lower than that of other models. At the same time, its performance in R2 is 0.9107, which is at least 0.0134 higher than that of other models. Analysis of the above results shows that the MAE and RMSE values ​​of the wind power prediction results are reduced to a certain extent, and the R2 value is increased to a certain extent, verifying that the model proposed in this invention has good applicability in wind power prediction.

[0100] Example 3

[0101] This embodiment provides a system for implementing the above-mentioned offshore wind power output prediction based on Bi-LSTM and time attention mechanism, including:

[0102] The preprocessing module normalizes the historical data of offshore wind power, uses the Bi-LSTM model to capture the correlation information between the time periods before and after the data, models the complex dynamic characteristics in the time series data of offshore wind power, and obtains the original wind power characteristics.

[0103] The STL decomposition module decomposes wind power output data using STL to obtain wind power output trends. It then uses Bi-LSTM to model these trends and obtain trend features.

[0104] The model combination prediction module uses the MLP (Multilayer Perceptron) model to extract features from external factors (such as weather and market fluctuations) that affect offshore wind power output. It then integrates the processed offshore wind power output and wind power output trend with the external influencing factor data. First, it concatenates the original wind power characteristics and trend characteristics, and then performs spatiotemporal attention fusion with the external factor characteristics to obtain the prediction results.

[0105] The evaluation module uses three indicators—mean absolute error, root mean square error, and coefficient of determination—to assess the prediction accuracy of the model combination prediction module.

[0106] Example 4

[0107] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0108] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0109] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for predicting offshore wind power output, characterized in that, Includes the following steps: S1: Acquire historical data of offshore wind power output, preprocess it to obtain preprocessed data, input the preprocessed data into the Bi-LSTM neural network to obtain the first output; S2: Perform time-series decomposition on the preprocessed data to obtain the power output trend. Model the power output trend using a Bi-LSTM neural network to obtain the second output. Concatenate the first and second outputs to obtain the concatenated result. S3: For meteorological factors affecting offshore wind power output prediction, an MLP model is established to extract features and obtain a third output. The third output and the splicing result are then fused by spatiotemporal attention to obtain fused features. S4: Make predictions based on fused features to obtain prediction results; In the spatiotemporal attention fusion, the calculation expression for attention allocation is: In the formula, For attention scoring function, The attention value for the i-th key. For the corresponding weights, It is the natural base.

2. The method for predicting offshore wind power output according to claim 1, characterized in that, The calculation expression for the preprocessing is: In the formula, Category of customer-side load, For time scale.

3. The method for predicting offshore wind power output according to claim 1, characterized in that, The calculation expression for the propagation process of the Bi-LSTM neural network is as follows: In the formula, the matrix , , , Control the weights of the gate states respectively; It represents the state of the cell at time t. It is the output of the activation function sigmoid. yes Function; the activation function tanh is the hyperbolic tangent function; yes and The connection operation, It is the Gate of Oblivion. It's an input gate. , , , It is the bias.

4. The method for predicting offshore wind power output according to claim 1, characterized in that, The time series decomposition yields trend components, seasonal components, and residual terms, and a ternary array of trend components, seasonal components, and residual terms is established as the output trend.

5. The method for predicting offshore wind power output according to claim 4, characterized in that, The calculation expression for the time series decomposition is: In the formula, In time The observed values, express The trend component of time; express The seasonal component of time; express The residual term at time step; The calculation expressions for the trend component and the seasonal component are as follows: In the formula, , , They represent time. , , Observations at that location Represents a data set The Middle Observations at each time point; Represents a data set The average of all observations; This represents the average value across all points in time within the dataset. Indicates the length of seasonal cycles; , These represent the trend constant and the trend slope, respectively. Indicates the first The time points corresponding to each observation.

6. The method for predicting offshore wind power output according to claim 1, characterized in that, The MLP model uses the backpropagation algorithm to update and adjust the weights. The output expression of the intermediate hidden layer of the MLP model is as follows: In the formula, This is the output of the intermediate hidden layer. This is the input to the intermediate hidden layer. The weights of the intermediate hidden layers. The threshold of the intermediate hidden layer. The activation function for the intermediate hidden layer; The output expression of the output layer of the MLP model is: In the formula, For the output of the output layer, For the input of the output layer, For output layer weights, The threshold value for the output layer. This is the activation function for the output layer.

7. An offshore wind power output prediction system that implements the offshore wind power output prediction method as described in any one of claims 1-6, characterized in that, include: Preprocessing module: Acquires historical data of offshore wind power output, performs normalization processing, uses Bi-LSTM to extract the normalization results, and obtains the first output; Time-series decomposition module: By decomposing wind power output data over time, the wind power output trend is obtained, and Bi-LSTM is used to extract the wind power output trend to obtain the second output; Model Combination Prediction Module: The third output is obtained using MLP, the first, second and third outputs are concatenated, and prediction is performed based on the concatenation result to obtain the prediction result; Evaluation module: Evaluates the accuracy of the prediction results using mean absolute error, root mean square error, and coefficient of determination.

8. An offshore wind power output prediction device, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements a method for predicting offshore wind power output as described in any one of claims 1-6.

9. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements a method for predicting offshore wind power output as described in any one of claims 1-6.

Images