Wind power prediction method and device, computer equipment and storage medium
By optimizing the time series decomposition and diffusion model, the problem of wind power prediction models being unable to capture time dynamics was solved, thus improving accuracy and reliability.
Patent Information
- Application Number
- CN202510536533.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-27
- Publication Date
- 2025-08-01
AI Technical Summary
Existing wind power prediction models struggle to capture the dynamic time of real-world time series, leading to inaccurate predictions.
By decomposing time series data to obtain component features at different time scales, and combining this with a diffusion model to perform stepwise noise addition and denoising, the deterministic prediction results are made probabilistic.
It improves the accuracy and reliability of wind power forecasting, and can provide forecast results within any confidence interval, effectively assessing model uncertainty.
Smart Images
Figure CN120414510A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of clean energy, and particularly to a wind power prediction method, device, computer device and storage medium. Background Art
[0002] Wind energy is one of the most important renewable energy sources, and an accurate and reliable wind power prediction method can greatly promote the planning and scheduling of power systems.
[0003] However, the wind power curve often exhibits intermittency and significant uncertainty in the time domain, which is not conducive to accurate and reliable prediction; the wind power prediction models in the related art are based on the assumption that the statistical characteristics in time are constant, and often it is difficult to capture the time dynamics in the real time series, thus resulting in biases and errors in time series analysis.
[0004] Therefore, there is an urgent need to propose a wind power prediction method to solve the problem that the wind power prediction models in the related art are difficult to capture the time dynamics in the real time series, resulting in inaccurate wind power prediction. Summary of the Invention
[0005] In view of this, the present invention provides a wind power prediction method, device, computer device and storage medium to solve the problem that the wind power prediction models in the related art are difficult to capture the time dynamics in the real time series, resulting in inaccurate wind power prediction.
[0006] In a first aspect, the present invention provides a wind power prediction method, and the wind power prediction method includes: obtaining historical power data of a target wind farm; performing time series decomposition on the historical power data to obtain a historical first component feature corresponding to a first preset time scale and a historical second component feature corresponding to a second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale; reconstructing the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature; predicting based on the feature similarity of the historical second component feature to obtain a predicted second component feature; fusing the predicted first component feature and the predicted second component feature to obtain a deterministic wind power prediction result; inputting the deterministic wind power prediction result into a pre-constructed diffusion model, and performing step-by-step noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result.
[0007] As an exemplary embodiment, reconstructing the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature includes: performing multi-level wavelet decomposition on the historical first component feature based on the granularity level of the historical first component feature to obtain low-rank approximate sub-historical first component features corresponding to each granularity level; and reconstructing based on the low-rank approximate sub-historical first component features to obtain the predicted first component feature.
[0008] As an exemplary embodiment, performing multi-level decomposition on the historical first component feature based on the granularity level of the historical first component feature to obtain low-rank approximate sub-historical first component features corresponding to each granularity level includes: splitting the historical first component feature to obtain a plurality of non-overlapping initial even-index components and initial odd-index components corresponding to each granularity level; estimating the odd-index components based on the even-index components of each granularity level to obtain detail coefficients; updating the even-index components based on the detail coefficients of each granularity level to obtain approximation coefficients; and combining the approximation coefficients and the detail coefficients of each granularity level to obtain low-rank approximate sub-historical first component features corresponding to each granularity level.
[0009] As an exemplary embodiment, reconstructing based on the low-rank approximate sub-historical first component features to obtain the predicted first component feature includes: performing channel attention mechanism processing on the low-rank approximate sub-historical first component feature of the target granularity level finally obtained by multi-level decomposition to obtain a target approximate sub-historical first component feature; and reconstructing the target approximate sub-historical first component feature using transposed convolution to obtain the predicted first component feature.
[0010] As an exemplary embodiment, predicting a predicted second component feature based on the feature similarity of the historical second component features includes: clustering the historical second component features based on the feature similarity of each historical second component feature to obtain a plurality of clustering clusters; determining a sub-linear transformation rule based on the feature similarity of each clustering cluster, where the sub-linear transformation rule is related to the feature similarity; performing linear transformation on the corresponding clustering cluster based on each sub-linear transformation rule to obtain sub-predicted second component features corresponding to each clustering cluster; and fusing the sub-predicted second component features to obtain the predicted second component feature.
[0011] As an exemplary embodiment, inputting the deterministic wind power prediction result into a pre-constructed diffusion model, and performing step-by-step noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result includes: determining an actual conditional distribution based on the historical power data and the deterministic wind power prediction result; inputting the actual conditional distribution into the diffusion model, and performing forward diffusion on the actual conditional distribution based on the diffusion model to obtain multiple intermediate conditional distribution samples after noise addition; performing a reverse process on the intermediate conditional distribution samples based on the diffusion model to obtain multiple target samples after denoising; and obtaining the probability prediction result based on the target samples.
[0012] As an exemplary embodiment, the method for training the diffusion model includes: training the diffusion model based on minimizing the evidence lower bound of variational inference to obtain the diffusion model.
[0013] In a second aspect, the present invention provides a wind power prediction device, which includes: an acquisition module for acquiring historical power data of a target wind farm; a time series decomposition module for performing time series decomposition on the historical power data to obtain a historical first component feature corresponding to a first preset time scale and a historical second component feature corresponding to a second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale; a reconstruction module for reconstructing the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature; a prediction module for performing prediction based on the feature similarity of the historical second component feature to obtain a predicted second component feature; a fusion module for fusing the predicted first component feature and the predicted second component feature to obtain a deterministic wind power prediction result; and a wind power prediction result determination module for inputting the deterministic wind power prediction result into a pre-constructed diffusion model, and performing step-by-step noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result.
[0014] In a third aspect, the present invention provides a computer device, which includes: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the method according to the first aspect or any corresponding embodiment thereof.
[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the method according to the first aspect or any corresponding embodiment thereof.
[0016] The wind power prediction method provided in this embodiment obtains the historical power data of the target wind farm; performs time series decomposition on the historical power data to obtain the historical first component features corresponding to the first preset time scale and the historical second component features corresponding to the second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale; reconstructs the historical first component features based on the granularity level of the historical first component features to obtain the predicted first component features; predicts based on the feature similarity of the historical second component features to obtain the predicted second component features; fuses the predicted first component features and the predicted second component features to obtain the deterministic wind power prediction result; inputs the deterministic wind power prediction result into the pre-constructed diffusion model, and performs step-by-step noise addition and denoising based on the diffusion model to obtain the probability prediction result as the wind power prediction result; for the above method, firstly, it can consider the fluctuation characteristics of historical power data affected by intermittency and uncertainty on a multi-scale basis of granularity level in a short time scale, so as to capture the complex dynamic characteristics of the time series corresponding to the historical power data; and, it can consider the similar or different long-term trends of historical power data affected by intermittency and uncertainty based on the feature similarity of the historical second component features for power prediction on the trend scale, so as to capture the complex dynamic characteristics of the time series corresponding to the historical power data on another time scale; on the other hand, the setting of the diffusion model based on the deterministic prediction result extends the deterministic prediction to the probability prediction, can provide prediction results within any confidence interval, effectively evaluate the uncertainty of the prediction model, and enhance the reliability of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0018] Figure 1 It is the overall flowchart of the wind power prediction method provided by an embodiment of the present invention.
[0019] Figure 2 It is the structural block diagram of the wind power prediction device of the wind power prediction method provided by an embodiment of the present invention.
[0020] Figure 3 It is the schematic hardware structure diagram of the computer device of the wind power prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0021] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0022] According to an embodiment of the present invention, an embodiment of a wind power prediction method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.
[0023] Embodiment 1 provides a wind power prediction method in this embodiment. Figure 1 It is a flowchart of the wind power prediction method according to an embodiment of the present invention, as Figure 1 shown. The process includes the following steps:
[0024] Step S101: Obtain the historical power data of the target wind farm.
[0025] In this embodiment, the historical wind power data can be obtained by collecting the historical data of the target wind farm.
[0026] Step S102: Perform time series decomposition on the historical power data to obtain a historical first component feature corresponding to a first preset time scale and a historical second component feature corresponding to a second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale.
[0027] Wind power curves often exhibit intermittency and significant uncertainty in the time domain. The wind power prediction models in the related art are based on the assumption of constant statistical characteristics over time and often have difficulty capturing the time dynamics in the real time series. Therefore, it is necessary to perform time series decomposition on the historical power data at a relatively short time scale to capture the intermittency and uncertainty of the wind power time series at a short time scale. On the other hand, although the wind power curve exhibits intermittency and uncertainty in the time domain, it still has certain trend characteristics at a relatively long time scale. Therefore, it is necessary to perform time series decomposition on the historical power data at a relatively long time scale to capture the trend characteristics of the wind power time series at a long time scale.
[0028] Based on this, in this embodiment, time series decomposition is performed on the historical power data to obtain historical component features that characterize the fluctuations of the power data at different time scales, so as to further process the historical component features at different time scales in the future, thereby capturing the intermittency and uncertainty in the historical power data for power prediction.
[0029] Exemplarily, time series decomposition is performed on the historical power data to obtain a historical first component feature corresponding to a first preset time scale and a historical second component feature corresponding to a second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale.
[0030] Exemplarily, the additive time series decomposition method can be used to decompose the historical power data; for the historical first component feature corresponding to the first preset time scale, it can be obtained by smoothing the historical power data using a moving average filter; specifically, for each time point t, the average value of m time points around it is calculated to obtain a plurality of historical first component features; where m is the length of the seasonal period.
[0031] For the historical second component feature corresponding to the second preset time scale, a moving average is performed using a sliding window corresponding to the second preset time scale that is greater than the first preset time scale to smooth the time series and extract the trend component; specifically, for each time point t, the average value of n time points around it is calculated to obtain a plurality of historical second component features; where n is greater than m.
[0032] Step S103, reconstruct the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature.
[0033] In this embodiment, for the decomposed historical first component feature, it has various different granularity features affected by intermittency and uncertainty. The time series of the historical first component feature at different granularity levels can characterize the power performance when the wind power is affected by intermittency and uncertainty; therefore, reconstructing the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature can consider the fluctuation characteristics of the historical power data affected by intermittency and uncertainty at a short time scale in a multi-scale manner based on the granularity level, thereby capturing the complex dynamic characteristics of the time series corresponding to the historical power data.
[0034] Exemplarily, different granularity level time series can be obtained by performing wavelet decomposition or empirical mode decomposition on the historical first component feature, and further reconstructing the time series obtained by wavelet decomposition to obtain a predicted first component feature.
[0035] In one embodiment, after obtaining time series at different granularity levels, the time series obtained by wavelet decomposition is reconstructed based on the temporal variation information of the time series to obtain the first component feature; wherein, as a possible implementation, a recurrent neural network (RNN), a long short-term memory neural network (LSTM), etc. can be used to learn the temporal variation features of the time series, and further reconstruction is performed according to the learned temporal variation features, so as to obtain the predicted first component feature.
[0036] In one embodiment, after obtaining time series at different granularity levels, an adaptive wavelet transform method is used to implement the reconstruction of the time series to obtain the predicted first component feature.
[0037] The method of this embodiment can, by reconstructing the historical first component feature based on the granularity level of the historical first component feature, consider the fluctuation characteristics of historical power data affected by intermittency and uncertainty on a short time scale in a multi-scale manner based on the granularity level, so as to capture the complex dynamic characteristics of the time series corresponding to the historical power data.
[0038] Step S104, perform prediction based on the feature similarity of the historical second component feature to obtain the predicted second component feature.
[0039] In this embodiment, the historical second component feature is used to characterize the similar trends and different trends existing in wind power on a relatively long time scale under the influence of factors such as geographical conditions and seasons; in order to achieve separate predictions for the similar trends and different trends, in this embodiment, prediction is performed based on the feature similarity of the historical second component feature.
[0040] Exemplarily, after obtaining the historical second component feature, first perform clustering processing on the historical second component feature based on the feature similarity of the historical second component feature to obtain multiple clustering clusters; among the multiple clustering clusters, each clustering cluster can characterize the performance of wind power under similar trends, and the differences between the clustering clusters can characterize the performance of different trends between wind powers.
[0041] Further, after obtaining multiple clustering clusters, perform predictions according to each clustering cluster respectively to perform prediction based on the feature similarity of the historical second component feature to obtain the predicted second component feature.
[0042] In one embodiment, when making a prediction based on the feature similarity of the historical second component features, different prediction models can be trained according to the feature attribute adaptability of each type of clustering cluster, so as to select different prediction models for prediction based on the feature attributes of the historical second component features, and obtain the predicted second component features; among them, exemplarily, the prediction model can be constructed based on RNN, LSTM, etc.
[0043] In one embodiment, when making a prediction based on the feature similarity of the historical second component features, a linear transformation method can be adopted, and the linear transformation rule can be adaptively selected according to the feature attributes of each type of clustering cluster to obtain the predicted second component features.
[0044] The above method of this embodiment makes a prediction based on the feature similarity of the historical second component features to obtain the predicted second component features, and can perform power prediction on the trend scale considering the similar or different long-term trends affected by intermittency and uncertainty including historical power data based on the feature similarity of the historical second component features, so as to capture the complex dynamic features of the time series corresponding to the historical power data on another time scale.
[0045] Step S105: Fuse the predicted first component features and the predicted second component features to obtain a deterministic wind power prediction result.
[0046] Exemplarily, after obtaining the predicted first component features and the predicted second component features, the predicted first component features and the predicted second component features are fused to obtain a deterministic wind power prediction result.
[0047] Exemplarily, the predicted first component features and the predicted second component features can be superimposed based on time series to obtain a deterministic wind power prediction result.
[0048] Step S106: Input the deterministic wind power prediction result into a pre-constructed diffusion model, and perform step-by-step noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result.
[0049] After obtaining the preliminary prediction result, this embodiment further optimizes the wind power prediction result. Specifically, the deterministic wind power prediction result is input into a pre-constructed diffusion model, and step-by-step noise addition and denoising are performed based on the diffusion model to obtain a probability prediction result as the wind power prediction result.
[0050] In the above implementation method, the diffusion model is based on the deterministic prediction result, and optimizes the prediction result through a process of gradually adding noise and denoising; the noise addition step simulates the uncertainty of data by introducing random noise, while the subsequent denoising step gradually restores a more accurate prediction result by learning the internal distribution of the data; the above process not only captures the deep features of the data, but also significantly improves the robustness of the prediction result, enabling it to maintain a high prediction accuracy when facing the intermittency and uncertainty of wind power.
[0051] Embodiment 2, the wind power prediction method provided in this embodiment obtains the historical power data of the target wind farm; performs time series decomposition on the historical power data to obtain the historical first component feature corresponding to the first preset time scale and the historical second component feature corresponding to the second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale; reconstructs the historical first component feature based on the granularity level of the historical first component feature to obtain the predicted first component feature; predicts based on the feature similarity of the historical second component feature to obtain the predicted second component feature; fuses the predicted first component feature and the predicted second component feature to obtain a deterministic wind power prediction result; inputs the deterministic wind power prediction result into a pre-constructed diffusion model, and performs gradual noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result; the above method, firstly, can consider the fluctuation characteristics of historical power data affected by intermittency and uncertainty on a multi-scale based on the granularity level, so as to capture the complex dynamic characteristics of the time series corresponding to the historical power data; and, can consider the similar or different long-term trends of historical power data affected by intermittency and uncertainty based on the feature similarity of the historical second component feature to perform power prediction on the trend scale, so as to capture the complex dynamic characteristics of the time series corresponding to the historical power data on another time scale; on the other hand, the setting of the diffusion model based on the deterministic prediction result expands the deterministic prediction to the probability prediction, can provide prediction results within any confidence interval, effectively evaluate the uncertainty of the prediction model, and enhance the reliability of the prediction.
[0052] As an exemplary embodiment, the reconstructing the historical first component feature based on the granularity level of the historical first component feature to obtain the predicted first component feature includes: performing multi-level wavelet decomposition on the historical first component feature based on the granularity level of the historical first component feature to obtain the low-rank approximate sub-historical first component features corresponding to each granularity level; reconstructing based on the low-rank approximate sub-historical first component features to obtain the predicted first component feature.
[0053] In this embodiment, a wavelet decomposition method is adopted to reconstruct the historical first component feature based on the granularity level of the historical first component feature, and a predicted first component feature is obtained.
[0054] Among them, an adaptive wavelet transform method can be adopted to implement the historical first component feature; specifically, as an exemplary embodiment, the historical first component feature is decomposed at multiple levels based on the granularity level of the historical first component feature to obtain low-rank approximate sub-historical first component features corresponding to each granularity level, including: splitting the historical first component feature to obtain a plurality of non-overlapping initial even-index components and initial odd-index components corresponding to each granularity level; estimating the odd-index components based on the even-index components of each granularity level to obtain detail coefficients; updating the even-index components based on the detail coefficients of each granularity level to obtain approximation coefficients; combining the approximation coefficients and the detail coefficients of each granularity level to obtain low-rank approximate sub-historical first component features corresponding to each granularity level.
[0055] In this embodiment, an adaptive wavelet transform method is used to decompose the historical first component feature that exhibits a non-stationary time series to obtain low-rank approximate sub-historical first component features at different granularity levels.
[0056] Exemplarily, an adaptive wavelet block of an adaptive wavelet network is used to implement the decomposition process of the first component feature; specifically, when the adaptive wavelet block processes the first component feature, it first splits the historical first component feature into a plurality of non-overlapping initial even-index components and initial odd-index components corresponding to each granularity level; further, the odd-index components are estimated based on the even-index components of each granularity level to calculate detail coefficients; further, the even-index components are updated based on the detail coefficients of each granularity level to obtain approximation coefficients; the approximation coefficients and the detail coefficients of each granularity level are combined to obtain low-rank approximate sub-historical first component features corresponding to each granularity level.
[0057] After obtaining the low-rank approximate sub-historical first component feature, the low-rank approximate sub-historical first component feature is reconstructed to reconstruct the low-rank approximate sub-historical first component feature into the original component feature.
[0058] Moreover, for the rank approximate sub-historical first component feature corresponding to the final granularity level finally obtained by the multi-level decomposition, a channel attention mechanism is applied for processing to enhance the feature representation of the rank approximate sub-historical first component feature corresponding to the final granularity level; by processing through the channel attention mechanism, focusing on the global context information between channels, thereby improving the representation ability for complex time series.
[0059] Specifically, as an exemplary embodiment, the reconstruction based on the first component feature of the low-rank approximated sub-history to obtain the predicted first component feature includes: performing channel attention mechanism processing on the first component feature of the low-rank approximated sub-history at the target granularity level finally obtained by multi-level decomposition to obtain the target approximated first component feature of the sub-history; using transposed convolution to reconstruct the target approximated first component feature of the sub-history to obtain the predicted first component feature.
[0060] In this embodiment, the first component feature of the sub-history is reconstructed into the original component feature by reconstructing the target approximated first component feature of the sub-history using transposed convolution.
[0061] Exemplarily, the inverse adaptive wavelet block of the adaptive wavelet network is used to implement the reconstruction of the target approximated first component feature of the sub-history.
[0062] As an exemplary embodiment, the prediction based on the feature similarity of the second component feature of the history to obtain the predicted second component feature includes: clustering the second component features of the history based on the feature similarity of each of the second component features of the history to obtain a plurality of clustering clusters; determining a sub-linear transformation rule based on the feature similarity of each of the clustering clusters; wherein the sub-linear transformation rule is related to the feature similarity; performing a linear transformation on the corresponding clustering cluster based on each of the sub-linear transformation rules to obtain a sub-predicted second component feature corresponding to each of the clustering clusters; fusing each of the sub-predicted second component features to obtain the predicted second component feature.
[0063] In this embodiment, an appropriate sub-linear transformation rule is selected based on the feature similarity of each second component feature of the history for linear transformation to consider the similar trends and different trends existing in the historical power data at a longer time scale for prediction respectively.
[0064] Exemplarily, the k-clustering method can be used to implement the clustering based on the feature similarity pair of each of the second component features of the history.
[0065] Exemplarily, the grouped linear module of the adaptive wavelet network is used to implement the processing of the second component feature of the history; specifically, the grouped linear module applies a linear transformation W k and a bias b k to each clustering cluster k to obtain the prediction result of this clustering:
[0066]
[0067] In formula (1), x trend，k represents the sub-predicted second component feature of the channel group belonging to clustering k, W k represents the adaptive linear transformation based on this channel group, bk Indicates the bias based on this channel group.
[0068] Furthermore, based on Equation (2), the results of all clusters are fused to obtain the final predicted second component feature:
[0069]
[0070] In Equation (2), represents the predicted second component feature, K represents the total number of clustering clusters, and x trend，k characterizes the sub-predicted second component feature of the channel group belonging to cluster k.
[0071] After obtaining the preliminary prediction result, the present invention further optimizes the wind power prediction result. Specifically, the deterministic wind power prediction result is input into a pre-constructed diffusion model, and based on the diffusion model, gradual noise addition and denoising are performed to obtain a probability prediction result as the wind power prediction result.
[0072] Based on this, as an exemplary embodiment, the inputting the deterministic wind power prediction result into a pre-constructed diffusion model, and performing gradual noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result includes: determining the actual conditional distribution based on the historical power data and the deterministic wind power prediction result; inputting the actual conditional distribution into the diffusion model, and performing forward diffusion on the actual conditional distribution based on the diffusion model to obtain multiple intermediate conditional distribution samples after noise addition; performing an inverse process on the intermediate conditional distribution samples based on the diffusion model to obtain multiple target samples after denoising; and obtaining the probability prediction result based on the target samples.
[0073] In this embodiment, first, the actual conditional distribution p(Y0|X) is determined based on the historical power data and the deterministic wind power prediction result. Further, a diffusion model that pre-learns the conditional distribution pφ(Y0|X) is used to obtain the probability prediction result.
[0074] Among them, the conditional distribution p(Y0|X) refers to the probability distribution of the true value Y0 given the historical data X; this distribution describes the different values that the future wind power may take and their corresponding probabilities under the condition of known historical data X; the conditional distribution p(Y0|X) describes the probability distribution of the future wind power Y0 under the condition of known historical data X. This distribution considers the influence of historical data on future values and reflects the uncertainty of future values.
[0075] Among them, the diffusion model learns the conditional distribution pφ(Y0∣X) through a process of gradually adding noise and denoising; specifically, the diffusion model generates multiple intermediate samples Y1, Y2, ……, YT by gradually adding noise, and finally makes the distribution p(YT∣X) approach the standard normal distribution N(0, I).
[0076] Exemplarily, after obtaining the actual conditional distribution, the actual conditional distribution is input into the diffusion model, and forward diffusion is performed on the actual conditional distribution based on the diffusion model to obtain multiple intermediate conditional distribution samples after adding noise.
[0077] Specifically, the process of forward diffusion can be represented by Equation (3):
[0078]
[0079] In Equation (3), β t represents the noise scheduling parameter that controls the amount of noise added in each step, and P θ (X) represents the actual conditional distribution.
[0080] After obtaining the intermediate conditional distribution samples, an inverse process is performed on the intermediate conditional distribution samples based on the diffusion model to obtain multiple target samples after denoising; the probability prediction result is obtained based on the target samples
[0081] Specifically, the inverse process can be represented by Equation (4):
[0082]
[0083] In Equation (4), represents the mean of the inverse process, represents the variance of the inverse process.
[0084] As an exemplary embodiment, the method for training the diffusion model includes: training the diffusion model based on minimizing the evidence lower bound of variational inference to obtain the diffusion model.
[0085] Specifically, Equation (5) can be used to train the diffusion model by minimizing the evidence lower bound of variational inference:
[0086]
[0087] In Equation (5), is the loss of the first step of the inverse process, is the loss of the intermediate steps, is the loss of the final step of the forward process.
[0088] In the above-described embodiment, by means of the trained diffusion model, noise is gradually removed from YT, and finally multiple samples of Y0 are obtained. These samples can be used to construct a probability distribution. Specifically, by generating multiple samples of Y0, statistical quantities such as the mean and variance of the probability distribution can be calculated, thereby obtaining a probability prediction result.
[0089] Embodiment 3. This embodiment provides a wind power prediction device, as Figure 2 shown, including:
[0090] An acquisition module 501, configured to acquire historical power data of a target wind farm.
[0091] A time series decomposition module 502, configured to perform time series decomposition on the historical power data to obtain a historical first component feature corresponding to a first preset time scale and a historical second component feature corresponding to a second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale.
[0092] A reconstruction module 503, configured to reconstruct the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature.
[0093] A prediction module 504, configured to perform prediction based on the feature similarity of the historical second component feature to obtain a predicted second component feature.
[0094] A fusion module 505, configured to fuse the predicted first component feature and the predicted second component feature to obtain a deterministic wind power prediction result.
[0095] A wind power prediction result determination module 506, configured to input the deterministic wind power prediction result into a pre-constructed diffusion model, and perform step-by-step noise addition and noise removal based on the diffusion model to obtain a probability prediction result as the wind power prediction result.
[0096] It should be noted here that the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in the above embodiments.
[0097] It should be noted that the above modules, as part of the device, can be implemented by software or by hardware, where the hardware environment includes a network environment.
[0098] This embodiment of the present invention further provides a computer device, including a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus. The memory is used to store a computer program; the processor is configured to execute the method in any one of the above embodiments by running the computer program stored on the memory.
[0099] Figure 3 is a structural block diagram of an optional computer device according to an embodiment of the present application. As Figure 3 shown, it includes a processor 10, a communication interface 20, a memory 30, and a communication bus 40. Among them, the processor 10, the communication interface 20, and the memory 30 complete mutual communication through the communication bus 40. Among them,
[0100] The memory 30 is used to store computer programs;
[0101] When the processor 10 is used to execute the computer program stored on the memory 30, the method of any of the above embodiments is implemented.
[0102] Optionally, in this embodiment, the above communication bus may be a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 3 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.
[0103] The communication interface is used for communication between the above computer device and other devices.
[0104] The memory may include a RAM, or may also include a non-volatile memory, for example, at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.
[0105] The above processor may be a general-purpose processor, which may include but is not limited to: a CPU (Central Processing Unit), an NP (Network Processor), etc.; it may also be a DSP (Digital Signal Processing), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
[0106] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and will not be repeated here.
[0107] Those of ordinary skill in the art can understand that Figure 3 The structures shown are merely illustrative. The device for implementing the method of any one of the above embodiments may be a terminal device, which may be a smart phone (such as an Android phone, an IOS phone, etc.), a tablet computer, a handheld computer, and a mobile Internet device (MID), a PAD, or other terminal devices. Figure 3 It does not limit the structure of the above electronic device. For example, the terminal device may further include more or fewer components (such as a network interface, a display device, etc.) than those shown Figure 3 in the figure, or have a different configuration from that shown Figure 3 in the figure.
[0108] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware of the terminal device through a program, and the program can be stored in a computer-readable storage medium. The storage medium may include: a flash drive, a ROM, a RAM, a magnetic disk, or an optical disc, etc.
[0109] As an exemplary embodiment, the present application further provides a computer-readable storage medium, in which a computer program is stored, and the computer program is configured to execute the method steps of any one of the embodiments when running.
[0110] Optionally, in this embodiment, the above storage medium may be used for the program code for executing the method steps of the embodiments of the present application.
[0111] Optionally, in this embodiment, the above storage medium may be located on at least one of the multiple network devices in the network shown in the above embodiments.
[0112] Optionally, in this embodiment, the storage medium is configured to store the program for executing the method in the above embodiments.
[0113] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and details are not described herein again.
[0114] Optionally, in this embodiment, the above storage medium may include but is not limited to: a USB flash drive, a ROM, a RAM, a mobile hard disk, a magnetic disk, or an optical disc, etc., which can store program code.
[0115] The serial numbers of the embodiments of the present application above are only for description and do not represent the advantages and disadvantages of the embodiments.
[0116] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage media. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable one or more computer devices (which can be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods in the above embodiments.
[0117] In several embodiments provided by this application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, 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 displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of units or modules can be in electrical or other forms.
[0118] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution provided in this embodiment.
[0119] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.
[0120] In the above embodiments of this application, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0121] The above is only the preferred embodiment of this application. It should be noted that for those of ordinary skill in the art, without departing from the principle of this application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this application.
Claims
1. A wind power prediction method, characterized in that: Including, Obtaining historical power data of a target wind farm; Performing time series decomposition on the historical power data to obtain a historical first component feature corresponding to a first preset time scale and a historical second component feature corresponding to a second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale; Reconstructing the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature; Performing prediction based on the feature similarity of the historical second component feature to obtain a predicted second component feature; Fusing the predicted first component feature and the predicted second component feature to obtain a deterministic wind power prediction result; Inputting the deterministic wind power prediction result into a pre-constructed diffusion model, and performing step-by-step noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result.
2. The wind power prediction method according to claim 1, characterized in that: The reconstructing the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature includes: Performing multi-level wavelet decomposition on the historical first component feature based on the granularity level of the historical first component feature to obtain low-rank approximate sub-historical first component features corresponding to each granularity level; Reconstructing based on the low-rank approximate sub-historical first component features to obtain the predicted first component feature.
3. The wind power prediction method according to claim 2, wherein: The performing multi-level decomposition on the historical first component feature based on the granularity level of the historical first component feature to obtain low-rank approximate sub-historical first component features corresponding to each granularity level includes: Splitting the historical first component feature to obtain multiple non-overlapping initial even-index components and initial odd-index components corresponding to each granularity level; Estimating the odd-index components based on the even-index components of each granularity level to obtain detail coefficients; Updating the even-index components based on the detail coefficients of each granularity level to obtain approximation coefficients; Combining the approximation coefficients and the detail coefficients of each granularity level to obtain low-rank approximate sub-historical first component features corresponding to each granularity level.
4. The wind power prediction method according to claim 3, wherein: The reconstructing based on the low-rank approximate sub-historical first component features to obtain the predicted first component feature includes: Performing channel attention mechanism processing on the low-rank approximate sub-historical first component feature of the target granularity level finally obtained by multi-level decomposition to obtain a target approximate sub-historical first component feature; Using transposed convolution to reconstruct the target approximate sub-historical first component feature to obtain the predicted first component feature.
5. The wind power prediction method according to claim 4, wherein: The performing prediction based on the feature similarity of the historical second component feature to obtain a predicted second component feature includes: Clustering the historical second component features based on the feature similarity of each historical second component feature to obtain multiple clustering clusters; Determining a sub-linear transformation rule based on the feature similarity of each clustering cluster; wherein, the sub-linear transformation rule is related to the feature similarity; Performing linear transformation on the corresponding clustering cluster based on each sub-linear transformation rule to obtain sub-predicted second component features corresponding to each clustering cluster; Fuse each of the sub-predicted second component features to obtain the predicted second component feature.
6. The wind power prediction method according to claim 5, wherein: The step of inputting the deterministic wind power prediction result into a pre-constructed diffusion model, and performing progressive noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result includes: Determine the actual conditional distribution based on the historical power data and the deterministic wind power prediction result; Input the actual conditional distribution into the diffusion model, and perform forward diffusion on the actual conditional distribution based on the diffusion model to obtain multiple intermediate conditional distribution samples after noise addition; Perform an inverse process on the intermediate conditional distribution samples based on the diffusion model to obtain multiple target samples after denoising; Obtain the probability prediction result based on the target samples.
7. The wind power prediction method according to claim 6, wherein: The method for training the diffusion model includes: Train the diffusion model based on minimizing the evidence lower bound of variational inference to obtain the diffusion model.
8. A wind power prediction device applying the wind power prediction method according to any one of claims 1 to 7, characterized in that It includes: An acquisition module for acquiring historical power data of a target wind farm; A time series decomposition module for performing time series decomposition on the historical power data to obtain a historical first component feature corresponding to a first preset time scale and a historical second component feature corresponding to a second preset time scale; wherein, the first preset time scale is shorter than the second preset time scale; A reconstruction module for reconstructing the historical first component feature based on the granularity level of the historical first component feature to obtain a predicted first component feature; A prediction module for predicting based on the feature similarity of the historical second component feature to obtain a predicted second component feature; A fusion module for fusing the predicted first component feature and the predicted second component feature to obtain a deterministic wind power prediction result; A wind power prediction result determination module for inputting the deterministic wind power prediction result into a pre-constructed diffusion model, and performing progressive noise addition and denoising based on the diffusion model to obtain a probability prediction result as the wind power prediction result.
9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the wind power prediction method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the wind power prediction method according to any one of claims 1 to 7.