Method and device for predicting power generation data of wind power plant

Abstract

The invention provides a method and a device for predicting power generation data of a wind power plant. When the future expected power generation data of the wind power plant is predicted, the prediction model of the wind power plant power generation data based on data decomposition obtained through training can be used to predict the future expected power generation data of the wind power plant, for example, the prediction model can be used for predicting the power generation data of the wind power plant. The wind power plant power generation data prediction model based on data decomposition comprises a plurality of power generation data prediction sub-models, each power generation data prediction sub-model respectively processes real power generation data of a target wind power plant in a historical time period to obtain respective output prediction components, and then, according to the respective output prediction components, the power generation data of the target wind power plant is predicted. Acquiring expected power generation data of the target wind power plant in a future time period; thus, the prediction accuracy of predicting the future expected power generation data of the wind power plant can be improved.

Description

Technical Field

[0001] The present application relates to the field of wind energy technology, and in particular to a method and device for predicting power generation data of a wind farm. Background Art

[0002] As a clean and renewable energy, wind energy is of great significance in alleviating the shortage of traditional energy and reducing environmental pollution. Therefore, the development and utilization of wind power has received more and more attention.

[0003] The development of wind power requires the use of wind power technology, which refers to the technology of using wind energy to generate electricity. Due to the cleanliness and renewable nature of wind energy, the proportion of wind power in the power system is increasing.

[0004] In the scenario of developing wind power, a wind farm is usually used, in which a plurality of wind turbines are provided to generate electricity.

[0005] However, the power generation data of wind farms is affected by many factors and has strong randomness, intermittency and volatility, which brings great challenges to the stable operation of the power grid and power dispatching.

[0006] Therefore, in this context, it is crucial to accurately predict the power generation data of wind farms. For example, the predicted power generation data of wind farms is of great guiding significance in making long-term dispatching plans for power systems, formulating trading strategies for power markets, and formulating operation and maintenance plans for wind farms. In this way, it is possible to optimize the allocation of power resources, improve the operating efficiency and reliability of the power grid, reduce economic losses caused by power fluctuations, and help improve the market competitiveness of wind power. Summary of the invention

[0007] The present application shows a method and device for predicting wind farm power generation data.

[0008] In a first aspect, the present application shows a method for training a prediction model of wind farm power generation data based on data decomposition, the method comprising: Acquire real power generation data of the sample wind farm in the first historical time period; The fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm is used to decompose the real power generation data of the first historical period of the sample wind farm to obtain multiple intrinsic mode function (IMF) components and a residual component. Acquire real power generation data of a second historical time period of the sample wind farm, where the second historical time period is earlier than the first historical time period; For any one of the multiple IMF components and a residual component, at least using real power generation data of the sample wind farm in the second historical time period as sample data and using the component as labeled data, training a power generation data prediction sub-model corresponding to the component; A prediction model for wind farm power generation data based on data decomposition is obtained at least according to the power generation data prediction sub-models corresponding to the various components.

[0009] In a second aspect, the present application shows a method for predicting wind farm power generation data, the method comprising: Acquire real power generation data of the target wind farm in the second time period; Inputting the real power generation data of the second time period of the target wind farm into the prediction model of wind farm power generation data based on data decomposition, wherein the prediction model of wind farm power generation data based on data decomposition includes a plurality of power generation data prediction sub-models, each power generation data prediction sub-model processes the real power generation data of the second time period of the target wind farm respectively to obtain a respective output power generation data prediction component, and then obtaining the expected power generation data of the first time period of the target wind farm according to the respective output power generation data prediction components; the first time period is later than the second time period; The prediction model of wind farm power generation data based on data decomposition is obtained by training through any one of the methods described in the first aspect.

[0010] In a third aspect, the present application shows a device for training a prediction model of wind farm power generation data based on data decomposition, the device comprising: A first acquisition module is used to acquire real power generation data of a sample wind farm in a first historical time period; A decomposition module is used to decompose the real power generation data of the first historical time period of the sample wind farm by using a fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm to obtain multiple intrinsic mode function (IMF) components and a residual component; A second acquisition module is used to acquire real power generation data of a second historical time period of the sample wind farm, where the second historical time period is earlier than the first historical time period; A training module, for training a power generation data prediction sub-model corresponding to any one of the multiple IMF components and a residual component, using at least the real power generation data of the sample wind farm in the second historical time period as sample data and the component as labeled data; The third acquisition module is used to acquire a prediction model of wind farm power generation data based on data decomposition at least according to the power generation data prediction sub-models corresponding to the various components.

[0011] In a fourth aspect, the present application shows a device for predicting wind farm power generation data, the device comprising: A fourth acquisition module, used to acquire real power generation data of the target wind farm in a second time period; An input module is used to input the real power generation data of the second time period of the target wind farm into the prediction model of wind farm power generation data based on data decomposition, wherein the prediction model of wind farm power generation data based on data decomposition includes a plurality of power generation data prediction sub-models, each power generation data prediction sub-model processes the real power generation data of the second time period of the target wind farm respectively to obtain the power generation data prediction component outputted by each, and then obtain the expected power generation data of the first time period of the target wind farm according to the power generation data prediction component outputted by each; the first time period is later than the second time period; The prediction model of wind farm power generation data based on data decomposition is obtained by training through any one of the methods described in the first aspect.

[0012] Alternatively, the prediction model of wind farm power generation data based on data decomposition is obtained by training through any one of the devices described in the third aspect.

[0013] In a fifth aspect, the present application illustrates an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the method described in any of the above aspects.

[0014] In a sixth aspect, the present application illustrates a non-temporary computer-readable storage medium, which, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute a method as described in any of the above aspects.

[0015] In a seventh aspect, the present application illustrates a computer program product. When instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to execute the method described in any of the above aspects.

[0016] The technical solution provided by this application may have the following beneficial effects: The inventors have found through statistics that the power generation data of a wind farm is affected by the wind speed and wind direction in the area where the wind farm is located. The area where the wind farm is located includes sea areas or land areas. If the wind speed and wind direction in an area of ​​the sea or land area change, the changes in the wind speed and wind direction in the time series are sometimes irregular, and sometimes change suddenly. This will cause the real power generation data to change in the time series sometimes to be irregular, and sometimes change suddenly. In this way, if the real power generation data is directly used as the annotation data to train a model for predicting power generation data, the power generation data will change suddenly due to the sudden change in wind speed and wind direction, and the law of the real power generation data cannot be extracted, resulting in large gradient changes, which in turn causes the model for predicting power generation data to fail to converge, or fail to converge to the optimal solution, which results in the low accuracy of the model for predicting future power generation data.

[0017] In view of this, the inventors have come up with a way to search for the changing patterns of power generation data in the time series, and then reduce the gradient changes so that the model for predicting power generation data can converge, or converge to the optimal solution, so as to improve the prediction accuracy of the model for predicting power generation data in predicting future power generation data.

[0018] For example, the inventors have thought of using the CEEMDAN algorithm to decompose the actual power generation data of the sample wind farm in the first historical time period to obtain multiple IMF components and a residual component. Different components have different frequency domain characteristics of the actual power generation data of the sample wind farm in the first historical time period. Different components can reflect some regularities of the actual power generation data of the sample wind farm in the first historical time period in different frequency domains. In this way, for each component, the corresponding power generation data prediction sub-models can be trained respectively. Since different components can reflect some regularities of the actual power generation data of the sample wind farm in the first historical time period in different frequency domains, different The power generation data prediction sub-models corresponding to the components can learn some rules of the real power generation data of the first historical time period of the sample wind farm in different frequency domains, so that when training the power generation data prediction sub-models corresponding to different components, the characteristics of the different frequency domains of the real power generation data of the first historical time period of the sample wind farm can be better, more comprehensive and more accurately extracted, so that the gradient change is small when training the power generation data prediction sub-models corresponding to different components, so that the trained power generation data prediction sub-models corresponding to different components can converge, or can converge to the optimal solution. In this way, the prediction accuracy of the trained wind farm power generation data prediction model based on data decomposition can be improved.

[0019] Furthermore, since the CEEMDAN algorithm has the ability to eliminate mode mixing and residual noise, it can adaptively decompose the nonlinear data in the real power generation data of the first historical time period of the sample wind farm, thereby eliminating the noise inside the data and improving data availability.

[0020] Secondly, the CEEMDAN algorithm can diversify the data, meet the prerequisite of more complete and diverse data required for the use of interpolation algorithms, and improve the prediction accuracy of the prediction model of wind farm power generation data based on data decomposition.

[0021] In addition, the CEEMDAN algorithm can eliminate modal aliasing and achieve zero reconstruction error of decomposed data, which greatly reduces the consumption of computing resources and improves data decomposition efficiency.

[0022] In this way, when predicting the expected future power generation data of a wind farm, the prediction model of wind farm power generation data based on data decomposition trained by the present application can be used to predict the expected future power generation data of the wind farm. For example, the prediction model of wind farm power generation data based on data decomposition includes multiple power generation data prediction sub-models, and each power generation data prediction sub-model processes the real power generation data of the historical time period of the target wind farm respectively to obtain the prediction components of their respective outputs, and then obtains the expected power generation data of the target wind farm in the future time period based on the prediction components of their respective outputs; in this way, through the present application, the prediction accuracy of predicting the expected future power generation data of the wind farm can be improved.

[0023] Secondly, when predicting the future expected power generation data of the target wind farm, the present application also takes into account the time factor. For example, it takes into account the impact of the historical real power generation data of the target wind farm on the future expected power generation data of the target wind farm. By considering the impact of the historical real power generation data of the target wind farm on the future expected power generation data of the target wind farm, the prediction accuracy of the future expected power generation data of the wind farm can also be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flowchart of the steps of a method for training a prediction model of wind farm power generation data based on data decomposition in the present application.

[0025] Figure 2 It is a schematic diagram of a prediction model of wind farm power generation data based on data decomposition in the present application.

[0026] Figure 3 It is a schematic diagram of a prediction model of wind farm power generation data based on data decomposition in the present application.

[0027] Figure 4It is a schematic diagram of a prediction model of wind farm power generation data based on data decomposition in the present application.

[0028] Figure 5 It is a schematic diagram of a time series model of the present application.

[0029] Figure 6 This is a flowchart of the steps of a method for obtaining real power generation data of a sample wind farm in a second historical time period of the present application.

[0030] Figure 7 This is a flow chart of the steps of a method for predicting wind farm power generation data in the present application.

[0031] Figure 8 It is a structural block diagram of a device for training a prediction model of wind farm power generation data based on data decomposition in the present application.

[0032] Fig. 9 It is a structural block diagram of a device for predicting wind farm power generation data in the present application.

[0033] Fig.10 It is a block diagram of an electronic device of the present application.

[0034] Fig.11 It is a block diagram of an electronic device of the present application. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0036] Reference Figure 1 , shows a flowchart of the steps of a method for training a prediction model of wind farm power generation data based on data decomposition of the present application, the method can be applied to an electronic device, the electronic device can include a terminal or a server, etc., wherein the method can specifically include the following steps: In step S101 , real power generation data of a sample wind farm in a first historical period is obtained.

[0037] The actual power generation data of the sample wind farm in the first historical time period is actual data that occurred during the historical process.

[0038] The length of the first historical time period may be 15 minutes or 20 minutes, etc.

[0039] The first historical time period may be one or more.

[0040] When there are multiple first historical time periods, any two first historical time periods do not overlap, and the multiple first historical time periods may be sequentially adjacent time periods.

[0041] For any first historical time period among multiple first historical time periods, in this step, the actual power generation data of the sample wind farm in the first historical time period can be obtained, and the same is true for each other first historical time period among the multiple first historical time periods, thereby obtaining the actual power generation data of each first historical time period of the sample wind farm.

[0042] That is, the real power generation data of the first historical time period of the sample wind farm may include: the real power generation data of each first historical time period of multiple first historical time periods of the sample wind farm.

[0043] In the present application, a wind farm has multiple wind turbines, or multiple wind turbines constitute a wind farm.

[0044] A wind farm occupies an area, and the wind turbines in the wind farm are located at different positions in the area.

[0045] There are other wind farms around a wind farm, and multiple wind farms can form a wind farm cluster.

[0046] The sample wind farm is one of the wind farms in the wind farm cluster.

[0047] The real power generation data of the sample wind farm in the first historical time period can be understood as: the real power generation power (in kilowatts) or the real power generation (in kilowatt-hours or degrees) of the sample wind farm in the first historical time period.

[0048] The actual power generation (in kilowatts) or the actual power generation (in kilowatt-hours or degrees) of the sample wind farm in the first historical time period can be counted and stored in the electronic device in advance, and can be updated in real time.

[0049] In this way, the real power generation data of the sample wind farm in the first historical time period can be directly obtained in the electronic device.

[0050] The actual power generation power of the sample wind farm in the first historical time period may be the actual average power generation power of the sample wind farm in the first historical time period.

[0051] The actual power generation of the sample wind farm in the first historical time period may be the actual total power generation of the sample wind farm in the first historical time period.

[0052] In step S102, the real power generation data of the first historical time period of the sample wind farm is decomposed by using the fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm to obtain multiple IMF components and a residual component.

[0053] In one example, for example, assuming that the real power generation data of the first historical time period of the sample wind farm is V, the CEEMDAN (Complete Ensemble Empirical Mode Decomposition with AdaptiveNoise Analysis) algorithm is used to decompose the real power generation data V of the first historical time period of the sample wind farm into: several IMF (Intrinsic Mode Function) subsequences and residual R, as shown in the following formula:

[0054] That is, V = IMF 1 +…+IMF k +R.

[0055] Among them, the decomposition process of the CEEMDAN algorithm is illustrated by example.

[0056] Assume E i (·) is the i-th eigenmode component obtained after EMD decomposition, and the i-th eigenmode component obtained by CEEMDAN decomposition is , v j is a Gaussian white noise signal that satisfies the standard normal distribution, j=1,2…,N is the number of times white noise is added, ε is the standard table of white noise, and y(t) is the signal to be decomposed.

[0057] The CEEMDAN decomposition steps are as follows: Add Gaussian white noise to the signal to be decomposed V to obtain a new signal , q=1,2.

[0058] Perform EMD decomposition on the new signal to obtain the first-order eigenmode component C 1 :

[0059] r j It represents the deviation obtained by subtracting each IMF decomposed from the signal to be decomposed after each IMF is decomposed.

[0060] The first eigenmode component of the decomposition is obtained by averaging the N generated modal components:

[0061] Compute the residual after removing the first modal component:

[0062] In r 1 Add positive and negative paired Gaussian white noise to (t) to get a new signal, and perform EMD decomposition on the new signal to get the first-order modal component D 1 , from which the second eigenmode component can be obtained:

[0063] Compute the residual after removing the second modal component:

[0064] Repeat the above steps until the residual signal is a monotonic function and cannot be further decomposed, and the algorithm ends. At this time, the number of intrinsic mode components obtained is K, and the original signal y(t) is decomposed into:

[0065] C k (t) represents IMF k , r k (t) is denoted as R.

[0066] In step S103, the real power generation data of the sample wind farm in a second historical time period is obtained, where the second historical time period is earlier than the first historical time period.

[0067] The actual power generation data of the sample wind farm in the second historical time period is actual data that occurred during the historical process.

[0068] The second historical time period may be 15 minutes or 20 minutes, etc.

[0069] The second historical time period may be one or more.

[0070] When there are multiple second historical time periods, any two second historical time periods do not overlap, and the multiple second historical time periods may be sequentially adjacent time periods.

[0071] For any second historical time period among multiple second historical time periods, in this step, the real power generation data of the second historical time period of the sample wind farm can be obtained, and the same is true for each other second historical time period among the multiple second historical time periods, so as to obtain the real power generation data of each second historical time period of the sample wind farm respectively.

[0072] That is, the real power generation data of the second historical time period of the sample wind farm may include: the real power generation data of each second historical time period in a plurality of second historical time periods of the sample wind farm.

[0073] The real power generation data of the sample wind farm in the second historical time period can be understood as: the real power generation power (in kilowatts) or the real power generation (in kilowatt-hours or degrees) of the sample wind farm in the second historical time period.

[0074] The actual power generation (in kilowatts) or the actual power generation (in kilowatt-hours or degrees) of the sample wind farm in the second historical time period can be counted and stored in the electronic device in advance, and can be updated in real time.

[0075] In this way, the real power generation data of the sample wind farm in the second historical time period can be directly obtained in the electronic device.

[0076] The actual power generation power of the sample wind farm in the second historical time period may be the actual average power generation power of the sample wind farm in the second historical time period.

[0077] The actual power generation of the sample wind farm in the second historical period may be the actual total power generation of the sample wind farm in the second historical period.

[0078] In the aforementioned steps, step S101 and step S103 may be executed in parallel, or may be executed sequentially. If they are executed sequentially, the present application does not limit the execution order of step S101 and step S103.

[0079] Secondly, after step S101 is completed, step S102 may be executed.

[0080] After step S102 and step S103 are executed, step S104 may be executed.

[0081] In step S104, for any one of the multiple IMF components and a residual component, at least the real power generation data of the sample wind farm in the second historical time period is used as sample data and the component is used as labeled data to train a power generation data prediction sub-model corresponding to the component.

[0082] The same is true for each of the multiple IMF wind speed components and one residual component, thereby obtaining the power generation data prediction sub-models corresponding to each component. The number of power generation data prediction sub-models obtained is equal to the sum of the number of multiple IMF components and the number 1 (corresponding to one residual component).

[0083] In one example, following the example of step S102, at least the real power generation data of the sample wind farm in the second historical time period is used as the sample data and the IMF 1 To label data, train IMF 1 The corresponding power generation data prediction sub-model 1 uses at least the real power generation data of the second historical time period of the sample wind farm as sample data and uses IMF 2 To label data, train IMF 2 The corresponding power generation data prediction sub-model 2 ... uses at least the real power generation data of the second historical period of the sample wind farm as sample data and uses IMF k To label data, train IMF k The corresponding power generation data prediction sub-model k uses at least the real power generation data of the sample wind farm in the second historical time period as sample data and R as labeled data to train the power generation data prediction sub-model R corresponding to R.

[0084] In step S105, a prediction model for wind farm power generation data based on data decomposition is obtained at least according to the power generation data prediction sub-models corresponding to the various components.

[0085] In one embodiment, a fully connected layer may be obtained, and the fully connected layer is used to superimpose and reconstruct the data output by the power generation data prediction sub-models corresponding to each component. The superposition and reconstruction includes addition, or includes calculating an average value, etc.

[0086] The output ends of the power generation data prediction sub-models corresponding to each component are respectively connected to the input ends of the fully connected layer.

[0087] The output end of the prediction model of wind farm power generation data based on data decomposition includes the output end of the fully connected layer.

[0088] In addition, the prediction model of wind farm power generation data based on data decomposition also includes a data distribution layer (or multiplexing layer), and the input end of the prediction model of wind farm power generation data based on data decomposition includes the input end of the data distribution layer, and the output end of the data distribution layer is respectively connected to the input end of the power generation data prediction sub-model corresponding to each component. The data distribution layer is used to input the input data to the power generation data prediction sub-model corresponding to each component.

[0089] Among them, the network structure of the prediction model of wind farm power generation data based on data decomposition can be seen in Figure 2 shown.

[0090] The inventors have found through statistics that the power generation data of a wind farm is affected by the wind speed and wind direction in the area where the wind farm is located. The area where the wind farm is located includes sea areas or land areas. If the wind speed and wind direction in an area of ​​the sea or land area change, the changes in the wind speed and wind direction in the time series are sometimes irregular, and sometimes change suddenly. This will cause the real power generation data to change in the time series sometimes to be irregular, and sometimes change suddenly. In this way, if the real power generation data is directly used as the annotation data to train a model for predicting power generation data, the power generation data will change suddenly due to the sudden change in wind speed and wind direction, and the law of the real power generation data cannot be extracted, resulting in large gradient changes, which in turn causes the model for predicting power generation data to fail to converge, or fail to converge to the optimal solution, which results in the low accuracy of the model for predicting future power generation data.

[0091] In view of this, the inventors have come up with a way to search for the changing patterns of power generation data in the time series, and then reduce the gradient changes so that the model for predicting power generation data can converge, or converge to the optimal solution, so as to improve the prediction accuracy of the model for predicting power generation data in predicting future power generation data.

[0092] For example, the inventors have thought of using the CEEMDAN algorithm to decompose the actual power generation data of the sample wind farm in the first historical time period to obtain multiple IMF components and a residual component. Different components have different frequency domain characteristics of the actual power generation data of the sample wind farm in the first historical time period. Different components can reflect some regularities of the actual power generation data of the sample wind farm in the first historical time period in different frequency domains. In this way, for each component, the corresponding power generation data prediction sub-models can be trained respectively. Since different components can reflect some regularities of the actual power generation data of the sample wind farm in the first historical time period in different frequency domains, different The power generation data prediction sub-models corresponding to the components can learn some rules of the real power generation data of the first historical time period of the sample wind farm in different frequency domains, so that when training the power generation data prediction sub-models corresponding to different components, the characteristics of the different frequency domains of the real power generation data of the first historical time period of the sample wind farm can be better, more comprehensive and more accurately extracted, so that the gradient change is small when training the power generation data prediction sub-models corresponding to different components, so that the trained power generation data prediction sub-models corresponding to different components can converge, or can converge to the optimal solution. In this way, the prediction accuracy of the trained wind farm power generation data prediction model based on data decomposition can be improved.

[0093] Furthermore, since the CEEMDAN algorithm has the ability to eliminate mode mixing and residual noise, it can adaptively decompose the nonlinear data in the real power generation data of the first historical time period of the sample wind farm, thereby eliminating the noise inside the data and improving data availability.

[0094] Secondly, the CEEMDAN algorithm can diversify the data, meet the prerequisite of more complete and diverse data required for the use of interpolation algorithms, and improve the prediction accuracy of the prediction model of wind farm power generation data based on data decomposition.

[0095] In addition, the CEEMDAN algorithm can eliminate modal aliasing and achieve zero reconstruction error of decomposed data, which greatly reduces the consumption of computing resources and improves data decomposition efficiency.

[0096] It should be noted that in the scenario of training a prediction model, when the labeled data does not change suddenly, the labeled data can be used directly to train the prediction model. In this way, the regularity of the labeled data that has not changed suddenly can be extracted. The prediction accuracy of the prediction model trained in this way is often higher. In this case, it is not necessary to use the CEEMDAN algorithm to decompose the labeled data. This will consume a lot of computing resources, increase the length of the process of training the prediction model, and will not improve the prediction accuracy. In the process of development, technology will not develop in a direction that has no benefits and will increase costs.

[0097] The scenario of the present application is to train a prediction model for power generation data, and the labeled data used are real historical power generation data. The real historical power generation data may change suddenly due to sudden changes in wind speed and wind direction. It can be seen that the present application is a situation where the labeled data may change suddenly. In this case, if the labeled data is directly used to train the prediction model for power generation data, the prediction accuracy of the trained prediction model for power generation data is often very low because the rules of the suddenly changing labeled data cannot be extracted. Therefore, the inventors thought of using the CEEMDAN algorithm to decompose the labeled data to extract the rules of the labeled data in different frequency domains, which will significantly improve the prediction accuracy of the prediction model for power generation data.

[0098] It can be seen that in the scenario of training a prediction model, if the labeled data does not change suddenly, there is no need or necessity to use the CEEMDAN algorithm.

[0099] In one embodiment of the present application, there are many frameworks that can be used to train the power generation data prediction sub-model, for example, a variety of different time series models, and the various different time series models may include: LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), RNN (Recurrent Neural Network), CNN (Convolutional Neural Networks) and TCN (‌Temporal Convolutional Networks‌), etc.

[0100] In this way, when using at least the actual power generation data of the second historical time period of the sample wind farm as sample data and the component as labeled data to train the power generation data prediction sub-model corresponding to the component, it is possible to use at least the actual power generation data of the second historical time period of the sample wind farm as sample data and the component as labeled data to train the power generation data prediction sub-models corresponding to the component based on a variety of different time series models.

[0101] A total of k+1 power generation data prediction sub-models corresponding to the component are obtained, and each power generation data prediction sub-model is based on a different type of time series model. k is a positive integer greater than or equal to 1.

[0102] For example, the power generation data prediction sub-models corresponding to the component based on LSTM are obtained, the power generation data prediction sub-models corresponding to the component based on GRU are obtained, the power generation data prediction sub-models corresponding to the component based on RNN are obtained, the power generation data prediction sub-models corresponding to the component based on CNN are obtained, and the power generation data prediction sub-models corresponding to the component based on TCN are obtained, etc.

[0103] Accordingly, when obtaining a prediction model for wind farm power generation data based on data decomposition at least according to the power generation data prediction sub-models corresponding to each component, for any type of time series model, a candidate prediction model for power generation data based on the time series model of that type can be obtained according to the power generation data prediction sub-models based on the time series model of that type corresponding to each component.

[0104] The same is done for each other type of time series model, thereby obtaining candidate prediction models for power generation data based on each type of time series model.

[0105] For example, see Figure 3 In one example, for the time series model LSTM, the component IMF1 Corresponding LSTM-based power generation data prediction sub-model 1, component IMF 2 Corresponding LSTM-based power generation data prediction sub-model 2... Component IMF k The corresponding LSTM-based power generation data prediction sub-model k and the LSTM-based power generation data prediction sub-model R corresponding to the component R are used to obtain the candidate prediction model based on LSTM power generation data.

[0106] For example, see Figure 4 In another example, for the time series model TCN, the component IMF 1 Corresponding TCN-based power generation data prediction sub-model 1, component IMF 2 Corresponding TCN-based power generation data prediction sub-model 2... Component IMF k The corresponding TCN-based power generation data prediction sub-model k and the TCN-based power generation data prediction sub-model R corresponding to the component R are used to obtain a candidate prediction model based on TCN power generation data.

[0107] The same is true for the time series models GRU, RNN, and CNN, and we will not give examples one by one.

[0108] Then, a prediction model for wind farm power generation data based on data decomposition may be obtained according to candidate prediction models for power generation data based on a plurality of different types of time series models.

[0109] For example, in one embodiment, the prediction accuracy of each candidate prediction model for power generation data of a wind farm based on power generation data of multiple different types of time series models can be obtained (specifically, it can be obtained through testing methods, which will not be described in detail in this application).

[0110] Among candidate prediction models for power generation data based on a plurality of different types of time series models, a candidate prediction model for power generation data with the highest prediction accuracy is selected as a prediction model for wind farm power generation data based on data decomposition.

[0111] Through this embodiment, the candidate prediction model of power generation data with the highest prediction accuracy is selected as the prediction model of wind farm power generation data based on data decomposition, so that the prediction accuracy of the expected power generation data of a certain time period of the subsequent prediction target wind farm can be made higher as much as possible.

[0112] In another embodiment, a candidate prediction model for power generation data includes power generation data prediction sub-models corresponding to the components, and the time series models based on the power generation data prediction sub-models corresponding to the components in the candidate prediction model for power generation data may be the same (for example, the time series models mentioned above). Figure 3and Figure 4 The embodiments shown in the drawings may also be different from each other, or may be completely different from each other.

[0113] It should be noted that, in the present application, any component may correspond to multiple power generation data prediction sub-models, which are power generation data prediction sub-models based on different time series models, and the same is true for each other component.

[0114] To this end, there may be power generation data prediction sub-models corresponding to respective components and based on respective time series models.

[0115] In this way, in one round, for any component, the power generation data prediction sub-model based on a time series model corresponding to the component is selected from the power generation data prediction sub-models based on various time series models corresponding to the component, and the same is true for each other component. Then, based on the power generation data prediction sub-model of a time series model corresponding to each component, a candidate prediction model for power generation data is obtained.

[0116] In other rounds, similarly, a candidate prediction model for power generation data is obtained in each round, and the time series model based on the power generation data prediction sub-model corresponding to at least one component in the candidate prediction model for power generation data obtained in each round is different.

[0117] Then, among the candidate prediction models of power generation data obtained in each round, the candidate prediction model of power generation data with the highest prediction accuracy is selected and used as the prediction model of wind farm power generation data based on data decomposition.

[0118] In one embodiment, when using at least the real power generation data of the second historical time period of the sample wind farm as sample data and the component as labeled data, and training the power generation data prediction sub-models corresponding to the component and based on a plurality of different time series models respectively, it can be implemented by the following process, including: 11) Acquire the real wind element data of the second historical time period of the sample wind farm, where the real wind element data at least includes wind direction, wind speed, precipitation, temperature, humidity, air pressure and light intensity.

[0119] The real wind element data of the second historical time period of the sample wind farm may be wind element sub-data of each height layer of the sample wind farm, and the height layer may be 100 meters, 90 meters, 80 meters, 70 meters, 60 meters, and other height layers.

[0120] The wind element data of meteorological forecast can be the wind element data of the weather forecast model (WRF model). The weather forecast model (WRF model) represents a new generation of non-hydrostatic equilibrium, high-resolution mesoscale meteorological research and forecasting numerical model, which mainly considers the forecast of important weather from cloud scale to weather scale.

[0121] The wind element data of weather forecast can be understood as: the wind element data given by the weather forecast model (WRF model) in the historical process.

[0122] 12) For any type of time series model, the sample data is input into the input end of the time series model of this type, and the real wind element data is input into the last hidden layer of the time series model of this type. The last hidden layer is used to process the real wind element data and the data after the sample data is processed by the layer before the last hidden layer to obtain the processing result.

[0123] See also Figure 5 , the framework of this type of time series model can be as follows Figure 5 As shown, this type of time series model includes multiple hidden layers connected in series. There may be other layers before and / or after the multiple hidden layers connected in series, which is not limited in this application and will not be described in detail one by one.

[0124] Through this embodiment, the real wind factor data is input into the last hidden layer in this type of time series model, which can minimize the interference with the process of extracting features of sample data by the time series model, so as to improve the prediction accuracy of the prediction model of wind farm power generation data based on data decomposition.

[0125] 13) Obtain the loss value based on the processing result output by the time series model of this type and the component.

[0126] In one embodiment, the loss value may be calculated according to the processing result output by the time series model of this type and the component according to the following formula:

[0127] The above formula is the loss function. In the above formula, error is the loss value, V 1 V is the processing result output by this type of time series model. 2 For this amount.

[0128] 14) According to the loss value, the network parameters in the time series model of this type are adjusted until the network parameters in the time series model of this type converge, thereby obtaining a power generation data prediction sub-model based on the time series model of this type.

[0129] In one embodiment of the present application, see Figure 6, step S103 includes: In step S201, actual power generation data of a sample wind farm in a plurality of candidate historical time periods are respectively obtained, and the plurality of candidate historical time periods are all earlier than the first historical time period.

[0130] Among them, the multiple candidate historical time periods are all time periods earlier than the first historical time period, any two candidate historical time periods do not overlap, and the multiple candidate historical time periods can be consecutively adjacent time periods.

[0131] The actual power generation data of the sample wind farm in the candidate historical time period is actual data that occurred in the historical process.

[0132] The length of the candidate historical time period may be 15 minutes or 20 minutes, etc.

[0133] Any two candidate historical time periods do not overlap, and multiple candidate historical time periods may be consecutively adjacent time periods.

[0134] For any one of the multiple candidate historical time periods, in this step, the actual power generation data of the sample wind farm for the candidate historical time period can be obtained, and the same is true for each of the other candidate historical time periods in the multiple candidate historical time periods, thereby obtaining the actual power generation data of each candidate historical time period of the sample wind farm.

[0135] That is, the actual power generation data of the multiple candidate historical time periods of the sample wind farm may include: the actual power generation data of each candidate historical time period of the multiple candidate historical time periods of the sample wind farm.

[0136] The actual power generation data of the sample wind farm in the candidate historical time period can be understood as: the actual power generation power (in kilowatts) or the actual power generation (in kilowatt-hours or degrees) of the sample wind farm in the candidate historical time period.

[0137] The actual power generation (in kilowatts) or the actual power generation (in kilowatt-hours or degrees) of the sample wind farm in the candidate historical time period can be counted and stored in advance in the electronic device, and can be updated in real time.

[0138] In this way, the real power generation data of the sample wind farm in the candidate historical time period can be directly obtained in the electronic device.

[0139] The actual power generation power of the sample wind farm in the candidate historical time period may be the actual average power generation power of the sample wind farm in the candidate historical time period.

[0140] The actual power generation of the sample wind farm in the candidate historical time period may be the actual total power generation of the sample wind farm in the candidate historical time period.

[0141] In step S202, the correlation relationship between the real power generation data of each candidate historical time period of the sample wind farm and the real power generation data of the first historical time period of the sample wind farm is determined.

[0142] In this way, the contribution degree of the real power generation data of the sample wind farm in multiple candidate historical time periods to the real power generation data of the sample wind farm in the first historical time period can be determined.

[0143] The greater the contribution degree of the candidate historical time period, the greater the correlation between the actual power generation data of the candidate historical time period and the actual power generation data of the first historical time period of the sample wind farm; alternatively, the smaller the contribution degree of the candidate historical time period, the smaller the correlation between the actual power generation data of the candidate historical time period and the actual power generation data of the first historical time period of the sample wind farm.

[0144] In the present application, various methods are supported for determining the correlation between the actual power generation data of each candidate historical time period of the sample wind farm and the actual power generation data of the first historical time period of the sample wind farm. The present application does not limit the specific method for determining the correlation between the actual power generation data of each candidate historical time period of the sample wind farm and the actual power generation data of the first historical time period of the sample wind farm.

[0145] In an optional embodiment, the correlation relationship between the real power generation data of each candidate historical time period of the sample wind farm and the real power generation data of the first historical time period of the sample wind farm may be determined in the following manner.

[0146] For example, the real power generation data of a candidate historical time period of the sample wind farm and the real power generation data of the first historical time period of the sample wind farm are represented as time series vectors, which are recorded as A and B respectively.

[0147] The MI value between A and B is defined as: I(A,B) = H(A) + H(B) - H(A,B) H(A) and H(B) represent the entropy values ​​of variable A and variable B respectively.

[0148] H(A, B) represents the joint entropy between variables A and B.

[0149] They are defined as follows:

[0150]

[0151] p is the probability density of a single variable.

[0152] p joint is the joint probability density between the two variables.

[0153] From the above formula, we can see that the calculation process of I (A, B) is related to the probability density p and the joint probability density p joint Closely related.

[0154] However, in practice, it is often difficult to obtain an accurate probability density function.

[0155] To solve this problem, we can use nonparametric density estimation method to make an approximate estimate, which is more efficient and practical.

[0156] Taking the continuous random variable A as an example, the value range of the variable is divided into K F fragments of equal length, each fragment length is defined as Δ F .

[0157] In order to reduce the error of the estimation method, the number of fragments can be determined according to the following given formula, where the value of n represents the total number of data contained in A:

[0158]

[0159] Based on this, the entropy value of variable A can be expressed as:

[0160]

[0161] p i is the probability that sample A belongs to the i-th segment.

[0162] n i is the number of samples of sample A belonging to the i-th segment.

[0163] Then H(A) can be rewritten as follows:

[0164] Similarly, the entropy value of variable B can be expressed as:

[0165] n j is the number of samples of sample B belonging to the jth segment.

[0166] After determining the entropy values ​​of variables A and B, it is necessary to estimate the joint entropy value of variables A and B.

[0167] The range of its joint space is divided into K parts using the histogram estimation methodF *K V cells of equal size, the cell size is Δ F *Δ V , each cell position is defined by (i, j). By counting the number of samples n belonging to the cell (i, j) ij After statistics, the joint entropy value can be expressed as follows:

[0168] Then the MI values ​​of variables A and B can be calculated as follows:

[0169] In step S203, based on the correlation between the actual power generation data of each candidate historical time period of the sample wind farm and the actual power generation data of the first historical time period of the sample wind farm, the actual power generation data of the second historical time period of the sample wind farm is determined from the actual power generation data of multiple candidate historical time periods of the sample wind farm.

[0170] For example, in one embodiment, the real power generation data of each candidate historical time period of the sample wind farm can be sorted in descending order according to the correlation relationship between the real power generation data of each candidate historical time period of the sample wind farm and the real power generation data of the first historical time period of the sample wind farm. For example, the real power generation data of each candidate historical time period of the sample wind farm is sorted; among the real power generation data of each candidate historical time period of the sample wind farm, the real power generation data of the candidate historical time period with the greater correlation relationship with the real power generation data of the first historical time period of the sample wind farm is ranked higher, and the real power generation data of the candidate historical time period with the smaller correlation relationship with the real power generation data of the first historical time period of the sample wind farm is ranked lower. The correlation relationship can be represented by a numerical value, and the larger the numerical value, the closer the correlation relationship between the two is. Or, the smaller the numerical value, the more distant the correlation relationship between the two is. Then, the real power generation data of at least one candidate historical time period of the sample wind farm can be selected from the real power generation data of multiple candidate historical time periods of the sample wind farm in the sorting order. For example, the top N power generation data can be selected from the real power generation data of multiple candidate historical time periods of the sample wind farm in the sorting order; N is a positive integer greater than or equal to 1, and the specific value of N can be determined according to the actual situation, which is not limited in this application. Then, the real power generation data of the second historical time period of the sample wind farm can be determined based on the selected top N power generation data, for example, the selected top N power generation data are used as the real power generation data of the second historical time period of the sample wind farm.

[0171] Through the present application, the real power generation data of the second historical time period of the sample wind farm with a large correlation relationship with the real power generation data of the first historical time period of the sample wind farm is found and used as training data. The real power generation data of the second historical time period of the sample wind farm with a small correlation relationship with the real power generation data of the first historical time period of the sample wind farm is discarded and not used as training data.

[0172] Reducing useless training data can reduce the computing resources required in the training process, and can avoid the interference of the real power generation data of the second historical time period of the sample wind farm that has a small correlation with the real power generation data of the first historical time period of the sample wind farm on the training process, thereby indirectly improving the prediction accuracy of the prediction model of wind farm power generation data based on data decomposition.

[0173] Afterwards, the prediction model of wind farm power generation data based on data decomposition can be put into use online.

[0174] For example, a prediction model of wind farm power generation data based on data decomposition is used to predict the expected power generation data of the wind farm in the future.

[0175] For example, refer to Figure 7 , shows a flowchart of a method for predicting wind farm power generation data of the present application, the method can be applied to an electronic device, the electronic device can include a terminal or a server, etc., wherein the method can specifically include the following steps: In step S301, real power generation data of a target wind farm in a second time period is obtained.

[0176] In step S302, the real power generation data of the second time period of the target wind farm is input into the prediction model of wind farm power generation data based on data decomposition, wherein the prediction model of wind farm power generation data based on data decomposition includes a plurality of power generation data prediction sub-models, each power generation data prediction sub-model processes the real power generation data of the second time period of the target wind farm respectively, obtains the power generation data prediction component outputted by each, and then obtains the expected power generation data of the first time period of the target wind farm according to the power generation data prediction component outputted by each. The first time period is later than the second time period.

[0177] The second time period is a historical time period, for example, a time period before the current moment, and the first time period is a future time period, for example, a time period after the current moment.

[0178] There may be multiple first time periods, and there may also be multiple second time periods.

[0179] The duration of the first time period may be 15 minutes or 20 minutes, etc.

[0180] The first time period may be one or more.

[0181] When there are multiple first time periods, any two first time periods do not overlap, and the multiple first time periods may be sequentially adjacent time periods.

[0182] For any first time period among multiple first time periods, through the present application, the expected power generation data of the target wind farm in the first time period can be obtained, and the same is true for each other first time period among the multiple first time periods, thereby respectively obtaining the expected power generation data of each first time period of the target wind farm.

[0183] That is, the expected power generation data of the target wind farm in the first time period may include: the expected power generation data of each first time period in a plurality of first time periods of the target wind farm.

[0184] In the present application, a wind farm has multiple wind turbines, or multiple wind turbines are combined into one wind farm.

[0185] A wind farm occupies an area, and the wind turbines in the wind farm are located at different positions in the area.

[0186] There are other wind farms around a wind farm, and multiple wind farms can form a wind farm cluster.

[0187] The target wind farm is one of the wind farms in the wind farm cluster.

[0188] The expected power generation data of the target wind farm in the first time period may be understood as: the expected power generation power (in kilowatts) or the expected power generation (in kilowatt-hours or degrees) of the target wind farm in the first time period.

[0189] The expected power generation power of the target wind farm in the first time period may be the expected average power generation power of the target wind farm in the first time period.

[0190] The expected power generation of the target wind farm in the first time period may be the expected total power generation of the target wind farm in the first time period.

[0191] When predicting the expected future power generation data of a wind farm, the prediction model of wind farm power generation data based on data decomposition trained by the present application can be used to predict the expected future power generation data of the wind farm. For example, the prediction model of wind farm power generation data based on data decomposition includes multiple power generation data prediction sub-models, and each power generation data prediction sub-model processes the real power generation data of the historical time period of the target wind farm respectively to obtain the prediction components of their respective outputs, and then obtains the expected power generation data of the target wind farm in the future time period based on the prediction components of their respective outputs; in this way, through the present application, the prediction accuracy of predicting the expected future power generation data of the wind farm can be improved.

[0192] Secondly, when predicting the future expected power generation data of the target wind farm, the present application also takes into account the time factor. For example, it takes into account the impact of the historical real power generation data of the target wind farm on the future expected power generation data of the target wind farm. By considering the impact of the historical real power generation data of the target wind farm on the future expected power generation data of the target wind farm, the prediction accuracy of the future expected power generation data of the wind farm can also be improved.

[0193] It should be noted that, for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions involved are not necessarily required by the present application.

[0194] Reference Figure 8 , shows a structural block diagram of a device for training a prediction model of wind farm power generation data based on data decomposition in the present application, the device comprising: A first acquisition module 11 is used to acquire real power generation data of a sample wind farm in a first historical time period; The decomposition module 12 is used to decompose the real power generation data of the sample wind farm in the first historical time period by using a fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm to obtain a plurality of intrinsic mode function (IMF) components and a residual component; A second acquisition module 13 is used to acquire real power generation data of a second historical time period of the sample wind farm, where the second historical time period is earlier than the first historical time period; A training module 14 is used for training a power generation data prediction sub-model corresponding to any one of the multiple IMF components and a residual component, using at least the real power generation data of the sample wind farm in the second historical time period as sample data and the component as labeled data; The third acquisition module 15 is used to acquire a prediction model of wind farm power generation data based on data decomposition at least according to the power generation data prediction sub-models corresponding to the various components.

[0195] In an optional implementation, the third acquisition module includes: A first acquisition unit is used to acquire a fully connected layer, where the fully connected layer is used to superimpose and reconstruct data output by the power generation data prediction sub-model corresponding to each component; A connection unit, used to connect the output ends of the power generation data prediction sub-models corresponding to the respective components to the input ends of the fully connected layer respectively; The output end of the prediction model of wind farm power generation data based on data decomposition includes the output end of the fully connected layer.

[0196] In an optional implementation, the training module includes: A training unit, configured to use at least the real power generation data of the sample wind farm in the second historical time period as sample data and the components as labeled data to train power generation data prediction sub-models corresponding to the components and based on a plurality of different time series models respectively; Accordingly, the third acquisition module includes: A second acquisition unit is used for acquiring, for any type of time series model, a candidate prediction model for power generation data based on the type of time series model according to the power generation data prediction sub-models based on the type of time series model corresponding to each component; The third acquisition unit is configured to acquire a prediction model for wind farm power generation data based on data decomposition according to candidate prediction models for power generation data based on a plurality of different types of time series models.

[0197] In an optional implementation, the third acquisition unit includes: A first acquisition subunit is used to obtain prediction accuracy rates of the power generation data of the wind farm by candidate prediction models of power generation data based on multiple different types of time series models; The selection subunit is used to select a candidate prediction model for power generation data with the highest prediction accuracy from among candidate prediction models for power generation data based on multiple different types of time series models, and use it as the prediction model for wind farm power generation data based on data decomposition.

[0198] In an optional implementation, the training unit includes: A second acquisition subunit is used to acquire real wind element data of a second historical time period of the sample wind farm, where the real wind element data at least includes wind direction, wind speed, precipitation, temperature, humidity, air pressure and light intensity; An input subunit is used for inputting sample data into the input end of the time series model of any type, and inputting real wind element data into the last hidden layer in the time series model of the type, wherein the last hidden layer is used for processing the real wind element data and the data processed by the layer before the last hidden layer on the sample data, to obtain a processing result; A third acquisition subunit, configured to acquire a loss value according to a processing result output by the time series model of the type and the component; The adjusting sub-unit is used to adjust the network parameters in the time series model of the type according to the loss value until the network parameters in the time series model of the type converge, thereby obtaining a power generation data prediction sub-model based on the time series model of the type.

[0199] In an optional implementation, the second acquisition module includes: A fourth acquisition unit is used to respectively acquire the real power generation data of a plurality of candidate historical time periods of the sample wind farm, wherein the plurality of candidate historical time periods are all earlier than the first historical time period; A first determining unit, configured to determine a correlation relationship between the real power generation data of each candidate historical time period of the sample wind farm and the real power generation data of the first historical time period of the sample wind farm; The second determination unit is used to determine the real power generation data of the second historical time period of the sample wind farm from the real power generation data of multiple candidate historical time periods of the sample wind farm based on the correlation relationship between the real power generation data of each candidate historical time period of the sample wind farm and the real power generation data of the first historical time period of the sample wind farm.

[0200] In an optional implementation, the second determining unit includes: A sorting subunit is used to sort the real power generation data of each candidate historical time period of the sample wind farm; among the real power generation data of each candidate historical time period of the sample wind farm, the real power generation data of the candidate historical time period with a greater correlation relationship with the real power generation data of the first historical time period of the sample wind farm is sorted higher, and the real power generation data of the candidate historical time period with a smaller correlation relationship with the real power generation data of the first historical time period of the sample wind farm is sorted lower; A selection subunit is used to select TopN power generation data from the real power generation data of multiple candidate historical time periods of the sample wind farm in a sorted order; N is a positive integer greater than or equal to 1; The determination subunit is used to determine the real power generation data of the sample wind farm in the second historical time period according to the selected TopN power generation data.

[0201] The inventors have found through statistics that the power generation data of a wind farm is affected by the wind speed and wind direction in the area where the wind farm is located. The area where the wind farm is located includes sea areas or land areas. If the wind speed and wind direction in an area of ​​the sea or land area change, the changes in the wind speed and wind direction in the time series are sometimes irregular, and sometimes change suddenly. This will cause the real power generation data to change in the time series sometimes to be irregular, and sometimes change suddenly. In this way, if the real power generation data is directly used as the annotation data to train a model for predicting power generation data, the power generation data will change suddenly due to the sudden change in wind speed and wind direction, and the law of the real power generation data cannot be extracted, resulting in large gradient changes, which in turn causes the model for predicting power generation data to fail to converge, or fail to converge to the optimal solution, which results in the low accuracy of the model for predicting future power generation data.

[0202] In view of this, the inventors have come up with a way to search for the changing patterns of power generation data in the time series, and then reduce the gradient changes so that the model for predicting power generation data can converge, or converge to the optimal solution, so as to improve the prediction accuracy of the model for predicting power generation data in predicting future power generation data.

[0203] For example, the inventors have thought of using the CEEMDAN algorithm to decompose the actual power generation data of the sample wind farm in the first historical time period to obtain multiple IMF components and a residual component. Different components have different frequency domain characteristics of the actual power generation data of the sample wind farm in the first historical time period. Different components can reflect some regularities of the actual power generation data of the sample wind farm in the first historical time period in different frequency domains. In this way, for each component, the corresponding power generation data prediction sub-models can be trained respectively. Since different components can reflect some regularities of the actual power generation data of the sample wind farm in the first historical time period in different frequency domains, different The power generation data prediction sub-models corresponding to the components can learn some rules of the real power generation data of the first historical time period of the sample wind farm in different frequency domains, so that when training the power generation data prediction sub-models corresponding to different components, the characteristics of the different frequency domains of the real power generation data of the first historical time period of the sample wind farm can be better, more comprehensive and more accurately extracted, so that the gradient change is small when training the power generation data prediction sub-models corresponding to different components, so that the trained power generation data prediction sub-models corresponding to different components can converge, or can converge to the optimal solution. In this way, the prediction accuracy of the trained wind farm power generation data prediction model based on data decomposition can be improved.

[0204] Furthermore, since the CEEMDAN algorithm has the ability to eliminate mode mixing and residual noise, it can adaptively decompose the nonlinear data in the real power generation data of the first historical time period of the sample wind farm, thereby eliminating the noise inside the data and improving data availability.

[0205] Secondly, the CEEMDAN algorithm can diversify the data, meet the prerequisite of more complete and diverse data required for the use of interpolation algorithms, and improve the prediction accuracy of the prediction model of wind farm power generation data based on data decomposition.

[0206] In addition, the CEEMDAN algorithm can eliminate modal aliasing and achieve zero reconstruction error of decomposed data, which greatly reduces the consumption of computing resources and improves data decomposition efficiency.

[0207] Reference Fig. 9 , shows a structural block diagram of a wind farm power generation data prediction device of the present application, the device comprising: A fourth acquisition module 21 is used to acquire real power generation data of the target wind farm in the second time period; An input module 22 is used to input the real power generation data of the second time period of the target wind farm into the prediction model of wind farm power generation data based on data decomposition, wherein the prediction model of wind farm power generation data based on data decomposition includes a plurality of power generation data prediction sub-models, each power generation data prediction sub-model processes the real power generation data of the second time period of the target wind farm respectively to obtain a respective output power generation data prediction component, and then obtains the expected power generation data of the first time period of the target wind farm according to the respective output power generation data prediction components; the first time period is later than the second time period; Among them, the prediction model of wind farm power generation data based on data decomposition is as follows: Figure 1 obtained by training using any of the methods described above.

[0208] Alternatively, the prediction model of wind farm power generation data based on data decomposition is Figure 8 Obtained by training with any of the above-described devices.

[0209] In this way, when predicting the expected future power generation data of a wind farm, the prediction model of wind farm power generation data based on data decomposition trained by the present application can be used to predict the expected future power generation data of the wind farm. For example, the prediction model of wind farm power generation data based on data decomposition includes multiple power generation data prediction sub-models, and each power generation data prediction sub-model processes the real power generation data of the historical time period of the target wind farm respectively to obtain the prediction components of their respective outputs, and then obtains the expected power generation data of the target wind farm in the future time period based on the prediction components of their respective outputs; in this way, through the present application, the prediction accuracy of predicting the expected future power generation data of the wind farm can be improved.

[0210] Secondly, when predicting the future expected power generation data of the target wind farm, the present application also takes into account the time factor. For example, it takes into account the impact of the historical real power generation data of the target wind farm on the future expected power generation data of the target wind farm. By considering the impact of the historical real power generation data of the target wind farm on the future expected power generation data of the target wind farm, the prediction accuracy of the future expected power generation data of the wind farm can also be improved.

[0211] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0212] Optionally, an embodiment of the present application further provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the various processes of the above-mentioned method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0213] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process of the above method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0214] Fig.10 800 is a block diagram of an electronic device 800 shown in the present application. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0215] Reference Fig.10, the electronic device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .

[0216] The processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0217] The memory 804 is configured to store various types of data to support operations on the device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, images, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0218] The power supply component 806 provides power to the various components of the electronic device 800. The power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the electronic device 800.

[0219] The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0220] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the electronic device 800 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0221] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: home button, volume button, start button, and lock button.

[0222] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the electronic device 800. For example, the sensor assembly 814 can detect the open / closed state of the device 800, the relative positioning of components, such as the display and keypad of the electronic device 800, and the sensor assembly 814 can also detect the position change of the electronic device 800 or a component of the electronic device 800, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration / deceleration of the electronic device 800, and the temperature change of the electronic device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0223] The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 can access a wireless network based on a communication standard, such as WiFi, a carrier network (such as 2G, 3G, 4G or 5G), or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast operation information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0224] In an exemplary embodiment, the electronic device 800 may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above methods.

[0225] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, and the instructions can be executed by a processor 820 of an electronic device 800 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0226] Fig.11 1 is a block diagram of an electronic device 1900 shown in the present application. For example, the electronic device 1900 may be provided as a server.

[0227] Reference Fig.11 , the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions executable by the processing component 1922, such as an application. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0228] The electronic device 1900 may also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input / output (I / O) interface 1958. The electronic device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™ or the like.

[0229] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0230] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0231] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

[0232] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0233] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0234] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0235] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0236] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0237] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard drives, ROM, RAM, magnetic disks, or optical disks.

[0238] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for training a prediction model for wind farm power generation data based on data decomposition, characterized in that: The method comprises: Acquire real power generation data of the sample wind farm in the first historical time period; The fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm is used to decompose the real power generation data of the first historical period of the sample wind farm to obtain multiple intrinsic mode function (IMF) components and a residual component. Acquire real power generation data of a second historical time period of the sample wind farm, where the second historical time period is earlier than the first historical time period; For any one of the multiple IMF components and a residual component, at least using real power generation data of the sample wind farm in the second historical time period as sample data and using the component as labeled data, training a power generation data prediction sub-model corresponding to the component; A prediction model for wind farm power generation data based on data decomposition is obtained at least according to the power generation data prediction sub-models corresponding to the various components.

2. The method according to claim 1, characterized in that The step of obtaining a prediction model for wind farm power generation data based on data decomposition at least according to the power generation data prediction sub-models corresponding to the various components, comprises: A fully connected layer is obtained, where the fully connected layer is used to superimpose and reconstruct the data output by the power generation data prediction sub-model corresponding to each component; Connect the output ends of the power generation data prediction sub-models corresponding to the components to the input ends of the fully connected layer respectively; The output end of the prediction model of wind farm power generation data based on data decomposition includes the output end of the fully connected layer.

3. The method according to claim 1, characterized in that The method of using at least the real power generation data of the sample wind farm in the second historical time period as sample data and the component as labeled data to train the power generation data prediction sub-model corresponding to the component includes: Using at least the real power generation data of the sample wind farm in the second historical time period as sample data and the components as labeled data, training power generation data prediction sub-models corresponding to the components and based on a plurality of different time series models respectively; Accordingly, the step of obtaining a prediction model for wind farm power generation data based on data decomposition at least according to the power generation data prediction sub-models corresponding to the various components, includes: For any type of time series model, according to the power generation data prediction sub-models based on the time series model of the type corresponding to each component, a candidate prediction model for power generation data based on the time series model of the type is obtained; A prediction model for wind farm power generation data based on data decomposition is obtained based on candidate prediction models for power generation data based on a plurality of different types of time series models.

4. The method according to claim 3, characterized in that The step of obtaining a prediction model for wind farm power generation data based on data decomposition according to candidate prediction models for power generation data based on multiple different types of time series models includes: Obtain prediction accuracy rates of the power generation data of the wind farm by candidate prediction models of power generation data based on multiple different types of time series models; Among candidate prediction models for power generation data based on a plurality of different types of time series models, a candidate prediction model for power generation data with the highest prediction accuracy is selected as a prediction model for wind farm power generation data based on data decomposition.

5. The method according to claim 3, characterized in that: The method uses at least the real power generation data of the second historical time period of the sample wind farm as sample data and the components as labeled data to train power generation data prediction sub-models corresponding to the components and based on a plurality of different time series models, including: Acquire real wind factor data of the second historical time period of the sample wind farm, where the real wind factor data at least includes wind direction, wind speed, precipitation, temperature, humidity, air pressure, and light intensity; For any type of time series model, sample data is input into the input end of the time series model of the type, and real wind element data is input into the last hidden layer in the time series model of the type, the last hidden layer is used to process the real wind element data and the data processed by the layer before the last hidden layer on the sample data, and obtain the processing result; Obtaining a loss value based on a processing result output by the time series model of the type and the components; The network parameters in the time series model of the type are adjusted according to the loss value until the network parameters in the time series model of the type converge, thereby obtaining a power generation data prediction sub-model based on the time series model of the type.

6. The method according to claim 1, characterized in that The obtaining of the real power generation data of the sample wind farm in the second historical time period includes: respectively obtaining real power generation data of a plurality of candidate historical time periods of the sample wind farm, wherein the plurality of candidate historical time periods are all earlier than the first historical time period; Determine the correlation relationship between the real power generation data of each candidate historical time period of the sample wind farm and the real power generation data of the first historical time period of the sample wind farm; According to the correlation relationship between the actual power generation data of each candidate historical time period of the sample wind farm and the actual power generation data of the first historical time period of the sample wind farm, the actual power generation data of the second historical time period of the sample wind farm is determined from the actual power generation data of multiple candidate historical time periods of the sample wind farm.

7. The method according to claim 6, characterized in that The determining of the real power generation data of the second historical time period of the sample wind farm from the real power generation data of the candidate historical time periods of the sample wind farm according to the correlation relationship between the real power generation data of each candidate historical time period of the sample wind farm and the real power generation data of the first historical time period of the sample wind farm, includes: The real power generation data of each candidate historical time period of the sample wind farm is sorted; among the real power generation data of each candidate historical time period of the sample wind farm, the real power generation data of the candidate historical time period with a greater correlation relationship with the real power generation data of the first historical time period of the sample wind farm is sorted higher, and the real power generation data of the candidate historical time period with a smaller correlation relationship with the real power generation data of the first historical time period of the sample wind farm is sorted lower; In the sorting order, select the top N power generation data from the real power generation data of multiple candidate historical time periods of the sample wind farm; N is a positive integer greater than or equal to 1; According to the selected TopN power generation data, the real power generation data of the sample wind farm in the second historical time period is determined.

8. A method for predicting wind farm power generation data, characterized in that: The method comprises: Acquire real power generation data of the target wind farm in the second time period; Inputting the real power generation data of the second time period of the target wind farm into the prediction model of wind farm power generation data based on data decomposition, wherein the prediction model of wind farm power generation data based on data decomposition includes a plurality of power generation data prediction sub-models, each power generation data prediction sub-model processes the real power generation data of the second time period of the target wind farm respectively to obtain a respective output power generation data prediction component, and then obtaining the expected power generation data of the first time period of the target wind farm according to the respective output power generation data prediction components; the first time period is later than the second time period; The prediction model of wind farm power generation data based on data decomposition is obtained by training according to any one of the methods described in claims 1 to 7.

9. A device for training a prediction model for wind farm power generation data based on data decomposition, characterized in that: The device comprises: A first acquisition module is used to acquire real power generation data of a sample wind farm in a first historical time period; A decomposition module is used to decompose the real power generation data of the first historical time period of the sample wind farm by using a fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm to obtain multiple intrinsic mode function (IMF) components and a residual component; A second acquisition module is used to acquire real power generation data of a second historical time period of the sample wind farm, where the second historical time period is earlier than the first historical time period; A training module, for training a power generation data prediction sub-model corresponding to any one of the multiple IMF components and a residual component, using at least the real power generation data of the sample wind farm in the second historical time period as sample data and the component as labeled data; The third acquisition module is used to acquire a prediction model of wind farm power generation data based on data decomposition at least according to the power generation data prediction sub-models corresponding to the various components.

10. A device for predicting wind farm power generation data, characterized in that: The device comprises: A fourth acquisition module, used to acquire real power generation data of the target wind farm in a second time period; An input module is used to input the real power generation data of the second time period of the target wind farm into the prediction model of wind farm power generation data based on data decomposition, wherein the prediction model of wind farm power generation data based on data decomposition includes a plurality of power generation data prediction sub-models, each power generation data prediction sub-model processes the real power generation data of the second time period of the target wind farm respectively to obtain the power generation data prediction component outputted by each, and then obtain the expected power generation data of the first time period of the target wind farm according to the power generation data prediction component outputted by each; the first time period is later than the second time period; The prediction model of wind farm power generation data based on data decomposition is obtained by training according to any one of the methods described in claims 1 to 7.

11. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the method according to any one of claims 1 to 8 when executed by the processor.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images