Wind power prediction method based on forecast wind speed correction and multi-model fusion

Abstract

This application discloses a wind power prediction method based on forecast wind speed correction and multi-model fusion, comprising: collecting and preprocessing data to obtain sample data; using historical numerical weather prediction data and historical measured meteorological data as input and output samples respectively to build and train a wind speed correction model; using wind farm wind speed-power curve comparison table data to build and train a naive wind power prediction model; using historical measured meteorological data and the naive wind power prediction model to form new samples, building and training a wind power prediction model based on forecast wind speed correction and multi-model fusion; using numerical weather prediction data, the wind speed correction model, and the naive wind power prediction model to form the input samples of the wind power prediction model, and outputting the final wind power prediction value through the wind power prediction model. This invention, based on the fusion of a wind speed correction model, a physical model, and a statistical model, performs wind power prediction, further improving the accuracy of wind power prediction.

Description

Technical Field

[0001] This application relates to the field of wind power prediction technology, and in particular to a wind power prediction method based on forecast wind speed correction and multi-model fusion. Background Technology

[0002] Currently, mainstream short-term wind power forecasting methods are mainly divided into two categories. One is the traditional physical model method. This method establishes a wind power function based on the characteristics of the wind farm and the wind turbine itself, using numerical weather prediction, topography, and surface roughness as model inputs to predict wind power. The limitations of the physical model algorithm are mainly reflected in: the need to comprehensively consider the influence of wind farm topography, surface roughness, and wind turbine wake effects, requiring many physical parameters and making modeling difficult; and the accuracy of the wind turbine's factory power curve is low because it does not consider the influence of wind direction, humidity, and air pressure. Generally, it cannot be directly used for prediction; another type is the statistical model method. This type of method is generally based on measured meteorological data, numerical weather prediction, and actual wind turbine output as the data foundation for modeling. It learns from the data through artificial intelligence models such as artificial neural networks, support vector machines, and decision trees. The model parameters characterize the characteristics and patterns of wind power. When making predictions, it calls numerical weather prediction and historical wind power data. The limitations of statistical model algorithms are mainly reflected in the fact that they require a certain amount of historical data and a high degree of coupling between the model input and wind power; otherwise, a good model cannot be established.

[0003] Traditional physical and statistical models have significant limitations when applied to wind power prediction alone. Furthermore, due to the influence of numerical weather forecast errors, especially wind speed forecast errors, the prediction accuracy of existing wind power prediction algorithms is relatively low. Summary of the Invention

[0004] This application provides a wind power prediction method based on forecast wind speed correction and multi-model fusion. It integrates a wind turbine physical model with historical big data, establishes a statistical model for forecast wind speed correction based on a deep neural network, establishes a physical model for wind power prediction based on an extreme learning machine, and establishes a statistical model for wind power prediction based on a deep neural network. The three models are then fused together, and the forecast wind speed correction statistical model and the wind power prediction physical model are used as inputs to the wind power prediction statistical model. This reduces the adverse effects of wind speed forecast errors and enriches the identifiable features of wind power prediction. As a result, the fused combined model can achieve better prediction results.

[0005] This application provides a wind power prediction method based on forecast wind speed correction and multi-model fusion, including:

[0006] Collect data and perform normalization preprocessing on the data to obtain sample data for training. The sample data includes historical measured meteorological data of wind farms, historical numerical weather forecast data, and historical measured wind power data.

[0007] The historical numerical weather forecast data is used as input samples and the historical measured meteorological data is used as output samples to build and train a wind speed correction model.

[0008] Using the wind speed-power curve comparison table data of the wind farm, a naive wind power prediction model was built and trained.

[0009] The historical measured meteorological data is called and a new sample is formed by the naive wind power prediction model. A wind power prediction model based on forecast wind speed correction and multi-model fusion is built and trained.

[0010] Other meteorological data, the wind speed correction model, and the naive wind power prediction model are called to perform wind speed correction and naive prediction, forming the input sample of the wind power prediction model. The wind power prediction model based on forecast wind speed correction and multi-model fusion outputs the wind power prediction value.

[0011] Optionally, after the wind power prediction model outputs the predicted wind power value, the method further includes:

[0012] Sample data is updated online, and the wind speed correction model, the naive wind power prediction model, and the wind power prediction model are trained online regularly to calculate the prediction accuracy in order to maintain the prediction accuracy.

[0013] Optionally, the other meteorological data include: forecast wind direction, forecast humidity, forecast temperature, and forecast air pressure.

[0014] The above techniques involve: collecting data and performing normalization preprocessing to obtain sample data; using historical numerical weather prediction data and historical measured meteorological data as input and output samples respectively to build and train a wind speed correction model; using wind farm wind speed-power curve comparison table data to build and train a naive wind power prediction model; using sample data and the naive wind power prediction model to build and train a wind power prediction model; and using other meteorological data, the wind speed correction model, and the naive wind power prediction model respectively to obtain wind speed correction and naive predictions, forming the input samples for the wind power prediction model, which then outputs the predicted wind power value. This invention, based on establishing a wind speed correction model and simultaneously integrating physical and statistical models for wind power prediction, further improves the accuracy of wind power prediction. Attached Figure Description

[0015] Figure 1This is a schematic flowchart of an embodiment of the wind power prediction method based on forecast wind speed correction and multiple models in this application;

[0016] Figure 2 This is a schematic diagram of the wind speed correction model trained and applied in this application;

[0017] Figure 3 This is a schematic diagram of the naive wind power prediction model trained and applied in this application.

[0018] Figure 4 This is a schematic diagram of the wind power prediction model during training in this application;

[0019] Figure 5 This is a schematic diagram of the wind power prediction model used in this application;

[0020] Figure 6 This is a comparison chart of the distribution range of wind speed deviation values ​​in the embodiments of this application;

[0021] Figure 7 This is a comparison chart of wind speed deviation values ​​before and after correction in the embodiments of this application;

[0022] Figure 8 This is a wind power prediction curve for a wind farm in an embodiment of this application;

[0023] Figure 9 This is a wind power prediction curve for another wind farm in an embodiment of this application;

[0024] Figure 10 This is a bar chart comparing the prediction accuracy of different algorithms in the embodiments of this application. Detailed Implementation

[0025] This application provides a wind power prediction method based on forecast wind speed correction and multi-model fusion. It integrates a wind turbine physical model with historical big data, establishes a statistical model for forecast wind speed correction based on a deep neural network, establishes a physical model for wind power prediction based on an extreme learning machine, and establishes a statistical model for wind power prediction based on a deep neural network. The three models are then fused together, and the forecast wind speed correction statistical model and the wind power prediction physical model are used as inputs to the wind power prediction statistical model. This reduces the adverse effects of wind speed forecast errors and enriches the identifiable features of wind power prediction. As a result, the fused combined model can achieve better prediction results.

[0026] Please see Figures 1 to 9 An embodiment of a wind power prediction method based on forecast wind speed correction and multi-model fusion includes:

[0027] 101. Collect data and perform normalization preprocessing to obtain sample data for training. The sample data includes historical measured meteorological data of wind farms, historical numerical weather forecast data, and historical measured wind power data.

[0028] Since wind power forecasting is required, data collection is necessary first. In this embodiment, the collected data includes historical measured meteorological data, historical numerical weather forecast data, and historical measured wind power data. After the data is collected, it is used as sample data. To ensure the validity of the sample data, it is further normalized and preprocessed, and outlier values ​​are removed and missing values ​​are filled in.

[0029] In this embodiment, the collected data are shown in Tables 1 and 2. Table 1 is the wind speed-power comparison table for wind farm A; Table 2 is the wind speed-power comparison table for wind farm B. Tables 1 and 2 are shown below:

[0030] Table 1

[0031]

[0032] Table 2

[0033]

[0034] 102. Using historical numerical weather forecast data as input samples and historical measured meteorological data as output samples, build and train a wind speed correction model;

[0035] This section pertains to a statistical model used for wind speed correction: Since there is an error between predicted and measured wind speeds, and this error exhibits a certain regularity, a wind speed correction model can be used to correct this error. In other words, the wind speed correction model is used to correct predicted wind speeds, making the deviation from the actual wind speed smaller and reducing the adverse effects caused by wind speed forecast errors. In this application, the predicted wind speed correction model is implemented based on a deep neural network.

[0036] The training process of the wind speed correction model is as follows: During training, the input samples of the model are normalized forecast wind speed, forecast wind speed raised to a higher power, forecast wind speed lower power, forecast wind direction, forecast wind direction raised to a higher power, and forecast wind direction lower power; the target samples for training are normalized measured wind speeds. This process is essentially fitting historical numerical weather forecasts and historical measured meteorological data. The training algorithm uses the backpropagation algorithm based on stochastic gradient descent, a fundamental algorithm in deep learning, which will not be elaborated upon here. The training data is shown in Table 3.

[0037] Table 3

[0038]

[0039] Once the wind speed correction model is trained, it is put into use. At this point, the input data for the wind speed correction model is still the normalized forecast wind speed, higher powers of the forecast wind speed, lower powers of the forecast wind speed, forecast wind direction, higher powers of the forecast wind direction, and lower powers of the forecast wind direction; the output data is the normalized corrected wind speed. After inverse normalization, the corrected wind speed value can be obtained. The corrected forecast wind speed will be closer to the posterior measured wind speed. In this embodiment, the model data is shown in Table 4, and the wind speed correction model diagram for training and application is shown in Figure 4. Figure 2 As shown, the distribution of wind speed deviation values ​​and the comparison of deviation values ​​before and after wind speed correction are respectively as follows: Figure 6 and Figure 7 As shown in Table 4:

[0040] Table 4

[0041]

[0042] 103. Use the wind speed-power curve comparison table data of the wind farm to build and train a naive wind power prediction model;

[0043] The naive wind power prediction model belongs to the physical model part of wind power prediction: the wind speed-power curve is used as training data, as shown in Tables 1 and 2 in step 101. The naive wind power prediction model fits the discrete values ​​of the wind speed-power comparison table into a continuous response model to establish a wind speed-power response model for naive wind power prediction ("naive wind power prediction" specifically refers to wind power prediction based solely on wind speed). This naive wind power prediction value is also used as the input of the wind power statistical model. This naive wind power prediction model is implemented based on an extreme learning machine.

[0044] In this embodiment, during training, the input samples of the model are named wind speed values; the target samples are measured named wind speed values. This process is essentially fitting the wind speed-power curve. The training algorithm uses a generalized inverse matrix and the least squares method to solve for the optimal network parameters. Once the naive wind power prediction model is trained, it is put into use. At this time, the input data of the naive wind power prediction model is the wind speed corrected by the wind speed correction model; the output is the naive wind power prediction value. The naive wind power prediction model for training and application is as follows: Figure 3 As shown.

[0045] 104. Use the historical measured meteorological data and the naive wind power prediction model to form a new sample, build and train a wind power prediction model based on forecast wind speed correction and multi-model fusion;

[0046] The wind power prediction model belongs to the statistical model part of wind power prediction. In this embodiment, the wind power prediction model first calls the wind speed correction model to correct the predicted wind speed as the first part of the input, and then calls the naive wind power prediction model to perform naive wind power prediction on the corrected wind speed as the second part of the input. At the same time, it combines the predicted wind direction, temperature and relative humidity as the third part of the input. The final wind power prediction value is output. This wind power prediction model is based on a deep neural network.

[0047] During the training of the wind power prediction model, the input samples are normalized measured wind speed, higher powers of measured wind speed, lower powers of measured wind speed, measured wind direction, measured humidity, measured air pressure, and the naive wind power prediction value corresponding to the measured wind speed; the target sample for training is the normalized measured wind power. This process is essentially fitting the multi-dimensional influencing factors-power curve. The training algorithm uses a backpropagation algorithm based on stochastic gradient descent; the data used during the training of this wind power prediction model is shown in Table 5.

[0048] Table 5

[0049]

[0050] After training, the model is put into use. The input data for this model consists of normalized corrected wind speed, higher powers of the corrected wind speed, lower powers of the corrected wind speed, measured wind direction, measured humidity, measured air pressure, and the naive wind power prediction value corresponding to the corrected wind speed. The predicted output is the wind power prediction value. Table 6 shows the application data of the wind power prediction model. The training and application of the naive wind power prediction model are respectively as follows: Figure 4 and Figure 5 As shown in Table 6:

[0051] Table 6

[0052]

[0053]

[0054] 105. Call numerical weather forecast data, wind speed correction model and naive wind power prediction model to form input samples for wind power prediction model. The wind power prediction model based on forecast wind speed correction and multi-model fusion outputs the final wind power prediction value.

[0055] In this embodiment, after the wind speed correction model, the naive wind power prediction model, and the wind power prediction model are built, these three models are combined and applied. Specifically, the wind speed correction model is used to correct the wind speed and obtain the corrected wind speed. Then, the naive wind power prediction model is used to predict the wind power. Finally, other meteorological data are obtained. Furthermore, the corrected wind speed, the naive wind power prediction value, and other meteorological data are used as inputs to the wind power prediction statistical model. Finally, the wind power prediction model is processed by a deep neural network to obtain the wind power prediction value, and the wind power prediction value is output from the output layer.

[0056] For example, relative to the data in Tables 1 and 2 in step 101 of this embodiment, the wind power prediction curve of wind farm A obtained after calculation is as follows: Figure 8 As shown; the wind power prediction curve for wind farm B is as follows. Figure 9 As shown.

[0057] 106. Update sample data online and regularly train the wind speed correction model, the naive wind power prediction model, and the wind power prediction model online, and calculate the prediction accuracy to maintain the prediction accuracy.

[0058] As historical data continues to accumulate, in order to ensure the accuracy of the wind speed correction model, the naive wind power prediction model, and the wind power prediction model, it is necessary to continuously update the sample data online and regularly train the wind speed correction model, the naive wind power prediction model, and the wind power prediction model online to maintain high prediction accuracy.

[0059] In this embodiment, multiple wind power prediction algorithms based on different models without wind speed correction were compared. The results show that the multi-model fusion wind power prediction algorithm based on predicted wind speed correction outperforms any single-model algorithm or the algorithm without wind speed correction. Furthermore, according to the control variable method, the addition of wind speed correction and model fusion both have a positive impact on the accuracy of wind power prediction. The accuracy comparison of different algorithms is shown in Table 7. Figure 10 As shown in Table 7 below:

[0060] Table 7

[0061]

[0062] This invention acquires sample data through data collection and normalization preprocessing; it uses historical numerical weather prediction data and historical measured meteorological data as input and output samples respectively to build and train a wind speed correction model; it uses wind farm wind speed-power curve comparison table data to build and train a naive wind power prediction model; it uses historical measured meteorological data and the naive wind power prediction model to build and train another wind power prediction model; it uses other meteorological data, the wind speed correction model, and the naive wind power prediction model respectively to form the input samples for the wind power prediction model. After wind speed correction and multi-model fusion, the final wind power prediction model outputs a predicted wind power value. This invention, based on the fusion of a wind speed correction model, a physical model, and a statistical model, performs wind power prediction, further improving the accuracy of wind power prediction.

[0063] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0064] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0065] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0066] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0067] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A wind power prediction method based on forecast wind speed correction and multi-model fusion, characterized in that, include: Collect data and perform normalization preprocessing on the data to obtain sample data for training. The sample data includes historical measured meteorological data of wind farms, historical numerical weather forecast data, and historical measured wind power data. The historical numerical weather forecast data is used as input samples and the historical measured meteorological data is used as output samples to build and train a wind speed correction model. The training process of the wind speed correction model is as follows: During training, the input samples of the model are normalized forecast wind speed, forecast wind speed with higher powers, forecast wind speed with lower powers, forecast wind direction, forecast wind direction with higher powers, and forecast wind direction with lower powers; the target samples for training are normalized measured wind speeds; this process is the process of fitting historical numerical weather forecasts and historical measured meteorological data. The training algorithm employs an error backpropagation algorithm based on stochastic gradient descent; it utilizes the wind farm's wind speed-power curve comparison table data to build and train a naive wind power prediction model, including: During training, the input samples for this model are nominal wind speed values; the target samples are measured nominal wind speed values. This process is essentially fitting the wind speed-power curve. The training algorithm uses a generalized inverse matrix and the least squares method to solve for the optimal network parameters. Once the naive wind power prediction model is trained, it is put into use. At this point, the input data for the naive wind power prediction model is the wind speed corrected by the wind speed correction model; the output is the naive wind power prediction value. The historical measured meteorological data is called and a naive wind power prediction model is used to form new samples. A wind power prediction model based on forecast wind speed correction and multi-model fusion is built and trained. Other meteorological data, the wind speed correction model and the naive wind power prediction model are called to perform wind speed correction and naive prediction to form the input samples of the wind power prediction model. The wind power prediction model based on forecast wind speed correction and multi-model fusion outputs the wind power prediction value.

2. The wind power prediction method according to claim 1, characterized in that, After the wind power prediction model outputs the predicted wind power value, the method further includes: Sample data is updated online, and the wind speed correction model, the naive wind power prediction model, and the wind power prediction model are trained online regularly to maintain prediction accuracy.

3. The wind power prediction method according to claim 1, characterized in that, The other meteorological data include: forecast wind direction, forecast humidity, forecast temperature, and forecast air pressure.

Images