A wind power short-term prediction method and system
By extracting features from wind speed forecast data and training a wind power prediction model, the problem of capturing the fluctuations in wind power output has been solved, and the accuracy of short-term wind power prediction has been improved.
Patent Information
- Application Number
- CN202010823570.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-08-17
- Publication Date
- 2026-02-27
- Estimated Expiration
- 2040-08-17
AI Technical Summary
Existing technologies have failed to effectively capture the continuity and volatility of wind power output, resulting in a bottleneck in the short-term prediction of wind power output and making it difficult to achieve accurate wind power output prediction.
Based on the time period to be predicted, the wind speed forecast data is divided into several data segments, the feature values of the wind speed forecast data are extracted, and the data are input into a pre-trained wind power prediction model. A correlation model between wind speed and wind power is established through a BP neural network to predict wind power.
By extracting features from wind speed forecast data and training models, the accuracy of wind power prediction has been improved, with an increase in prediction accuracy of approximately 2 percentage points.
Smart Images
Figure CN114077914B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of new energy power generation, in particular to a wind power short-term prediction method and system. BACKGROUND
[0002] The wind power output has randomness and volatility, and the uncertainty of the output power brings a series of scheduling operation problems. Accurate short-term prediction of wind power is an effective means to reduce the impact of uncertainty. According to the relevant standards promulgated, the wind power short-term prediction requires to predict the power from 0 o'clock of the next day to 72 hours in the future every day, and the time resolution is 15 minutes. In the aspect of short-term prediction, a large number of researches have been carried out, and it has been applied in the scheduling operation of power grid. The current mature technical idea is to establish the correlation model of weather and power at specific time, and output the power prediction value corresponding to the time of weather forecast data, without considering the continuity and volatility of wind power output, which is difficult to effectively capture the output fluctuation process of wind power, and the prediction level reaches a bottleneck. SUMMARY
[0003] In view of the problem that the current wind power short-term prediction does not consider the continuity and volatility of wind power output, and it is difficult to effectively capture the output fluctuation process of wind power, the present application provides a wind power short-term prediction method, comprising:
[0004] Divide the wind speed forecast data into a plurality of data segments according to the wind speed fluctuation process based on the to-be-predicted period, to obtain wind speed forecast data segments;
[0005] Calculate and extract the eigenvalues of the wind speed forecast data segments to obtain wind speed forecast data eigenvalues;
[0006] Input the wind speed forecast data eigenvalues into a pre-trained wind power prediction model to obtain wind power prediction eigenvalues;
[0007] Process the wind power prediction eigenvalues to obtain the predicted power of wind power;
[0008] The wind power prediction model is trained based on the wind speed fluctuation process of historical wind speed forecast data and the wind power fluctuation process of actual power data of the wind farm.
[0009] Preferably, the training of the wind power prediction model comprises:
[0010] Process the historical wind speed forecast data and the actual power data of the wind farm within a certain time;
[0011] Divide the processed historical wind speed forecast data into a plurality of data segments according to the wind speed fluctuation process to obtain historical wind speed forecast data segments;
[0012] The processed wind farm actual power data is divided into several data segments corresponding to the historical wind speed prediction data segments according to the wind speed fluctuation process, to obtain wind farm actual power data segments;
[0013] The historical wind speed prediction data segments are calculated and extracted for feature values, to obtain historical wind speed prediction data feature values;
[0014] The wind farm actual power data segments are calculated for feature values, to obtain wind farm actual power data feature values;
[0015] Based on the BP neural network, the historical wind speed prediction data feature values are taken as inputs, and the corresponding wind farm actual power data feature values are taken as outputs, to perform model training, to obtain a prediction model.
[0016] Preferably, the processing of the historical wind speed prediction data and the wind farm actual power data for a certain time comprises:
[0017] Based on the historical wind speed prediction data and the wind farm actual power data for a certain time, abnormal data are removed and the data time is aligned, to obtain processed historical wind speed prediction data and processed wind farm actual power data;
[0018] Based on the wavelet decomposition method, the processed wind farm actual power data are subjected to multi-layer frequency decomposition, to obtain low-frequency and high-frequency coefficients corresponding to each layer, and the low-frequency coefficients of one layer are used for reconstruction, to obtain a wind farm actual power data low-frequency sequence.
[0019] Preferably, the division into several data segments according to the wind speed fluctuation process comprises:
[0020] The maximum value and the minimum value of the wind speed prediction data or the processed historical wind speed prediction data are found and recorded in the time sequence, and the adjacent "minimum value-maximum value-minimum value" is taken as a data segment, to divide the wind speed prediction data or the processed historical wind speed prediction data into multiple data segments.
[0021] Preferably, the division of the processed wind farm actual power data into several data segments corresponding to the historical wind speed prediction data segments according to the wind speed fluctuation process comprises:
[0022] The historical wind speed prediction data segments are taken as objects, and the data of the same period in the wind farm actual power data low-frequency sequence is extracted, to divide the wind farm actual power data low-frequency sequence into multiple wind farm actual power data segments corresponding to the historical wind speed prediction data segments.
[0023] Preferably, the calculation and extraction for feature values comprise:
[0024] extracting a minimum value, a maximum value, a position of the minimum value, and a position of the maximum value of the wind speed prediction data segment or the historical wind speed prediction data segment;
[0025] calculating a length, a gradient, a maximum gradient, a mean value, and a standard deviation of the wind speed prediction data segment or the historical wind speed prediction data segment.
[0026] Preferably, the calculation of the feature values of the wind farm actual power data segment comprises:
[0027] The processed wind farm actual power data in the wind farm actual power data segment are converted into a plurality of interpolation data by using data interpolation.
[0028] Preferably, the input of the wind speed prediction data feature values into the pre-trained wind power prediction model to obtain wind power prediction feature values comprises:
[0029] The feature values of a plurality of wind speed prediction data segments are input into the wind power prediction model to obtain corresponding groups of wind power prediction feature values.
[0030] Each group of wind power prediction feature values corresponds to a wind power prediction data segment and contains a plurality of wind power prediction feature values.
[0031] Preferably, the processing of the wind power prediction feature values to obtain the predicted power of the wind power comprises:
[0032] For the wind power prediction data segment, a plurality of wind power prediction feature values are reconstructed by using data interpolation to obtain reconstruction data corresponding to the length of the wind speed prediction data segment, and all the reconstruction data are arranged to obtain the predicted power of the wind power.
[0033] Based on the same inventive concept, the present application also provides a wind power short-term prediction system, comprising a data division module, a data processing module, a model processing module, and a final processing module.
[0034] The data division module is configured to divide wind speed prediction data into a plurality of data segments according to wind speed fluctuation processes based on a to-be-predicted time period to obtain wind speed prediction data segments.
[0035] The data processing module is configured to calculate and extract feature values of the wind speed prediction data segments to obtain wind speed prediction data feature values.
[0036] The model processing module is configured to input the wind speed prediction data feature values into a pre-trained wind power prediction model to obtain wind power prediction feature values.
[0037] The final processing module is used for processing the wind power prediction characteristic value to obtain the predicted power of the wind power.
[0038] The wind power prediction model is trained based on the wind speed fluctuation process of the historical wind speed prediction data and the wind power fluctuation process of the actual power data of the wind farm.
[0039] Compared with the prior art, the wind power short-term prediction method has the following beneficial effects:
[0040] 1. The wind power short-term prediction method comprises the following steps: dividing wind speed prediction data into a plurality of data segments according to a wind speed fluctuation process based on a to-be-predicted period to obtain wind speed prediction data segments; calculating and extracting characteristic values of the wind speed prediction data segments to obtain wind speed prediction data characteristic values; inputting the wind speed prediction data characteristic values into a pre-trained wind power prediction model to obtain wind power prediction characteristic values; and processing the wind power prediction characteristic values to obtain predicted power of the wind power; wherein the wind power prediction model is trained based on a wind speed fluctuation process of historical wind speed prediction data and a wind power fluctuation process of actual power data of a wind farm; the wind speed process is divided and the characteristic is extracted, the correlation model of the wind process and the power is established, and the power prediction of the wind power is realized, so that the fluctuation of the wind power can be effectively captured and the prediction level is improved.
[0041] 2. The wind power short-term prediction method and system analyze the fluctuation process of the power, extract the characteristics of the fluctuation process, establish the correlation model of the wind speed and the fluctuation process, the prediction object of the model is the fluctuation process and the characteristics of the wind power, and the prediction result of the wind power is obtained through reconstruction, so that the prediction accuracy is improved; according to statistics, the prediction accuracy of the prediction result of the present application is about 2 percentage points higher than that of the mature method. BRIEF DESCRIPTION OF DRAWINGS
[0042] Figure 1 The method flowchart of the wind power short-term prediction method is provided.
[0043] Figure 2 The wind fluctuation process division schematic diagram of the wind power short-term prediction method is provided.
[0044] Figure 3 The comparison schematic diagram of the prediction effect of the wind power short-term prediction method and the traditional prediction method is provided. DETAILED DESCRIPTION
[0045] Embodiment 1:
[0046] The wind power short-term prediction method is disclosed. Figure 1The method flow chart is introduced, and specifically includes:
[0047] Step 1: based on the wind speed prediction period to be predicted, wind speed prediction data is divided into several data segments according to the wind speed fluctuation process, and wind speed prediction data segments are obtained;
[0048] Step 2: calculating and extracting feature values of the wind speed prediction data segments to obtain wind speed prediction data feature values;
[0049] Step 3: inputting the wind speed prediction data feature values into a pre-trained wind power prediction model to obtain wind power prediction feature values;
[0050] Step 4: processing the wind power prediction feature values to obtain the predicted power of the wind power;
[0051] In the step 1: based on the wind speed prediction period to be predicted, wind speed prediction data is divided into several data segments according to the wind speed fluctuation process, and wind speed prediction data segments are obtained, specifically including:
[0052] Finding the maximum value and the minimum value of the wind speed prediction data N to be predicted and recording the positions of the maximum value and the minimum value in the time sequence, taking the adjacent "minimum value-maximum value-minimum value" as a data segment for division, as shown in the following formula: Figure 2 N = {N pre ,N pre1 ,N pre2 ,…,N prem} is divided into m data segments N pre , and N prem represents the total set of the divided data segments of N .
[0053] Step 2: calculating and extracting feature values of the wind speed prediction data segments to obtain wind speed prediction data feature values, specifically including:
[0054] Calculating and extracting the minimum value Min, the maximum value Max, the position of the minimum value Min_Position, the position of the maximum value Max_Position, the length Len, the gradient Gra, the maximum gradient Max_Gra, the mean value Mean, and the standard deviation Std of each sub-set N pre , a total of 9 feature values.
[0055] Step 3: inputting the wind speed prediction data feature values into a pre-trained wind power prediction model to obtain wind power prediction feature values, specifically including:
[0056] 1. Training of the wind power prediction model, specifically including:
[0057] (1) processing historical wind speed prediction data and actual power data of a wind farm for a certain time:
[0058] Based on the historical one-year wind speed prediction data N0 and the actual power data P of the wind farm, unreasonable data is removed and the data time is aligned to obtain the processed wind speed prediction data N0' and the processed actual power data P' of the wind farm;
[0059] Based on the wavelet decomposition method, the P' is frequency-decomposed to obtain low-frequency and high-frequency coefficients, and the low-frequency coefficients are selected for reconstruction to obtain the low-frequency sequence P l of P'. The decomposition method and the low-frequency coefficients used are selected according to actual conditions and experience in specific use;
[0060] (2) The processed historical wind speed prediction data and the processed actual power data of the wind farm are divided into a plurality of data segments corresponding to each other according to the wind speed fluctuation process:
[0061] The maximum value and the minimum value of N0' are found and the positions of the maximum value and the minimum value in the time sequence are recorded, and the adjacent "minimum value-maximum value-minimum value" is taken as a data segment to divide N0' into n data segments N s ={N s1 ,N s2 ,…,N sn}, N s represents the total set of the divided data segments of N0', and N sn represents the divided data segment of N0';
[0062] Taking each sub-set of N s as an object, the data of the same period in P l is extracted, and P l is divided into n data segments P ls ={P l1 ,P l2 ,…,P ln} corresponding to N s , P ls represents the total set of the divided data segments of P l , P ln represents the divided data segment of P l , and P ln corresponds to N sn ;
[0063] (3) Based on the divided historical wind speed prediction data segments and the actual power data segments of the wind farm, the historical wind speed prediction data characteristic values and the actual power data characteristic values of the wind farm are calculated and extracted:
[0064] For N sCalculate and extract the minimum value Min, maximum value Max, position of the minimum value Min_Position, position of the maximum value Max_Position, length Len, gradient Gra, maximum gradient Max_Gra, mean Mean, and standard deviation Std for each subset, totaling 9 feature values; for P ls Using data interpolation, P ls Each subset is converted into L interpolated data points;
[0065] (4) Based on the BP neural network, with N s The nine feature values of each subset are inputs, P ls The interpolated data of each subset L are used as outputs to train the model and obtain the wind power prediction model M.
[0066] 2. Input the feature values of the wind speed forecast data segment into the wind power prediction model to obtain the corresponding wind power prediction feature values, specifically including:
[0067] N pre The nine eigenvalues of each subset are input into the wind power prediction model M to obtain the wind farm power prediction data P. pre ={P pre1 ,P pre2 ,…,P prem}, P pre P represents the total set of data segments for wind farm power prediction data. prem This represents the data segments divided from the wind farm power prediction data. Each data segment contains L data points, P prem With N prem correspond.
[0068] Step 4: Process the wind power prediction feature values to obtain the predicted wind power, specifically including:
[0069] For the predicted data P pre A subset of L data points is reconstructed using data interpolation to obtain the corresponding N data points. pre The predicted power of wind power is the arrangement of the Len data recorded in the subset and the reconstructed data of all subsets.
[0070] like Figure 3 As shown, the prediction accuracy of the method provided by this invention is improved by approximately 2 percentage points compared with that of traditional methods.
[0071] Example 2:
[0072] Based on the same inventive concept, the present invention also provides a short-term wind power forecasting system, comprising: a data partitioning module, a data processing module, a model processing module, and a final processing module;
[0073] The data division module is configured to divide wind speed prediction data into a plurality of data segments according to a wind speed fluctuation process based on a to-be-predicted period, to obtain wind speed prediction data segments.
[0074] The data processing module is configured to calculate and extract feature values of the wind speed prediction data segments, to obtain wind speed prediction data feature values.
[0075] The model processing module is configured to input the wind speed prediction data feature values into a pre-trained wind power prediction model, to obtain wind power prediction feature values.
[0076] The final processing module is configured to process the wind power prediction feature values, to obtain a predicted power of wind power.
[0077] The wind power prediction model is trained based on a wind speed fluctuation process of historical wind speed prediction data and a wind power fluctuation process of actual power data of a wind farm.
[0078] The model processing module comprises a model training submodule and a wind power prediction submodule.
[0079] The model training submodule is configured to train the wind power prediction model.
[0080] The wind power prediction submodule is configured to input the wind speed prediction data feature values into the wind power prediction model, to obtain corresponding wind power prediction feature values.
[0081] The model training submodule comprises a data preprocessing unit, a wind speed data division unit, a power data division unit, a wind speed data processing unit, a power data processing unit, and a model training unit.
[0082] The data preprocessing unit is configured to process historical wind speed prediction data and actual power data of a wind farm within a certain time.
[0083] The wind speed data division unit is configured to divide the processed historical wind speed prediction data into a plurality of data segments according to a wind speed fluctuation process, to obtain historical wind speed prediction data segments.
[0084] The power data division unit is configured to divide the processed actual power data of the wind farm into a plurality of data segments corresponding to the historical wind speed prediction data segments according to the wind speed fluctuation process, to obtain actual power data segments of the wind farm.
[0085] The wind speed data processing unit is configured to calculate and extract feature values of the historical wind speed prediction data segments, to obtain historical wind speed prediction data feature values.
[0086] The power data processing unit is configured to calculate eigenvalues of the wind farm actual power data segment to obtain wind farm actual power data eigenvalues.
[0087] The model training unit is configured to perform model training based on a BP neural network, with the historical wind speed prediction data eigenvalues as input and the corresponding wind farm actual power data eigenvalues as output, to obtain a prediction model.
[0088] Those skilled in the art will understand that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer usable program code.
[0089] The present application is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions that are executed by the processor of the computer or other programmable data processing apparatus generate an apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flowcharts and / or block diagrams.
[0090] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flowcharts and / or block diagrams.
[0091] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flowcharts and / or block diagrams.
[0092] The above merely illustrates the embodiments of the present application, but should not be taken as limitations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall fall into the protection scope of the present application.
Claims
1. A method for short-term wind power forecasting, characterized in that, include: Based on the period to be predicted, the wind speed forecast data is divided into several data segments according to the wind speed fluctuation process to obtain the wind speed forecast data segments. The feature values of the wind speed forecast data segment are calculated and extracted to obtain the feature values of the wind speed forecast data; The wind speed forecast data feature values are input into a pre-trained wind power prediction model to obtain wind power prediction feature values. The wind power prediction feature value is processed to obtain the predicted wind power. The wind power prediction model is trained based on the wind speed fluctuation process of historical wind speed forecast data and the wind power fluctuation process of actual wind farm power data; the training of the wind power prediction model includes: Processing historical wind speed forecast data and actual power data of wind farms over a certain period of time; The processed historical wind speed forecast data is divided into several data segments according to the wind speed fluctuation process to obtain historical wind speed forecast data segments. The processed actual power data of the wind farm is divided into several data segments corresponding to the historical wind speed forecast data segments according to the wind speed fluctuation process, thus obtaining the actual power data segments of the wind farm. The historical wind speed forecast data segment is subjected to feature value calculation and extraction to obtain the historical wind speed forecast data feature value; The characteristic values of the actual power data segment of the wind farm are calculated to obtain the characteristic values of the actual power data of the wind farm. Based on a BP neural network, the model is trained by taking the feature values of historical wind speed forecast data as input and the feature values of the corresponding actual power data of the wind farm as output, and a prediction model is obtained. The data is divided into several segments based on the wind speed fluctuation process, including: Find the maximum and minimum values of the wind speed forecast data or the processed historical wind speed forecast data and record their positions in the time series. Divide the wind speed forecast data or the processed historical wind speed forecast data into multiple data segments, using adjacent "minimum-maximum-minimum" as a data segment.
2. The method as described in claim 1, characterized in that, The processing of historical wind speed forecast data and actual wind farm power data over a certain period of time includes: Based on historical wind speed forecast data and actual wind farm power data over a certain period of time, outlier data is removed and the data time is aligned to obtain processed historical wind speed forecast data and processed actual wind farm power data. Based on wavelet decomposition, the processed wind farm actual power data is subjected to multi-level frequency decomposition to obtain the corresponding low-frequency and high-frequency coefficients of each layer. The low-frequency coefficients of one layer are used for reconstruction to obtain the low-frequency sequence of the wind farm actual power data.
3. The method as described in claim 2, characterized in that, The processed actual power data of the wind farm is divided into several data segments corresponding to the historical wind speed forecast data segments according to the wind speed fluctuation process, including: Taking the historical wind speed forecast data segment as the object, data from the same period are extracted from the low-frequency sequence of the actual power data of the wind farm, and the low-frequency sequence of the actual power data of the wind farm is divided into multiple wind farm actual power data segments corresponding to the historical wind speed forecast data segment.
4. The method as described in claim 1, characterized in that, The calculation and extraction of feature values includes: Extract the minimum value, maximum value, minimum value location, and maximum value location of the wind speed forecast data segment or the historical wind speed forecast data segment; Calculate the length, gradient, maximum gradient, mean, and standard deviation of the wind speed forecast data segment or the historical wind speed forecast data segment.
5. The method as described in claim 1, characterized in that, The calculation of feature values for the actual power data segment of the wind farm includes: The processed actual power data of the wind farm in the actual power data segment is converted into multiple interpolated data using data interpolation.
6. The method as described in claim 1, characterized in that, The step of inputting the wind speed forecast data feature values into a pre-trained wind power prediction model to obtain wind power prediction feature values includes: The feature values of multiple wind speed forecast data segments are input into the wind power prediction model to obtain multiple sets of corresponding wind power prediction feature values. Each group of wind power prediction feature values corresponds to a wind power prediction data segment, which contains multiple wind power prediction feature values.
7. The method as described in claim 5, characterized in that, The process of processing the wind power prediction feature values to obtain the predicted wind power includes: For the wind power prediction data segment, multiple wind power prediction feature values are reconstructed using data interpolation to obtain reconstructed data corresponding to the length of the wind speed forecast data segment. All the reconstructed data are then arranged to obtain the predicted wind power.
8. A short-term wind power forecasting system, characterized in that, include: The system consists of a data partitioning module, a data processing module, a model processing module, and a final processing module. The data segmentation module is used to divide the wind speed forecast data into several data segments according to the wind speed fluctuation process based on the time period to be predicted, so as to obtain the wind speed forecast data segments. The data processing module is used to calculate and extract feature values from the wind speed forecast data segment to obtain wind speed forecast data feature values. The model processing module is used to input the feature values of the wind speed forecast data into the pre-trained wind power prediction model to obtain the wind power prediction feature values. The final processing module is used to process the wind power prediction feature value to obtain the predicted wind power. The wind power prediction model is trained based on the wind speed fluctuation process of historical wind speed forecast data and the wind power fluctuation process of actual wind farm power data. The training of the wind power prediction model includes: Processing historical wind speed forecast data and actual power data of wind farms over a certain period of time; The processed historical wind speed forecast data is divided into several data segments according to the wind speed fluctuation process to obtain historical wind speed forecast data segments. The processed actual power data of the wind farm is divided into several data segments corresponding to the historical wind speed forecast data segments according to the wind speed fluctuation process, thus obtaining the actual power data segments of the wind farm. The historical wind speed forecast data segment is subjected to feature value calculation and extraction to obtain the historical wind speed forecast data feature value; The characteristic values of the actual power data segment of the wind farm are calculated to obtain the characteristic values of the actual power data of the wind farm. Based on a BP neural network, the model is trained by taking the feature values of historical wind speed forecast data as input and the feature values of the corresponding actual power data of the wind farm as output, and a prediction model is obtained. The data is divided into several segments based on the wind speed fluctuation process, including: Find the maximum and minimum values of the wind speed forecast data or the processed historical wind speed forecast data and record their positions in the time series. Divide the wind speed forecast data or the processed historical wind speed forecast data into multiple data segments, using adjacent "minimum-maximum-minimum" as a data segment.
Citation Information
Patent Citations
Method for forecasting short-term wind-electricity power
CN101741085A