A wind speed correction method based on upper and lower envelope approximation

Abstract

The present invention provides a wind speed correction method based on upper and lower envelope approximation, comprising: step 1, collecting multi-day wind speed time series data NWP and multi-day measured wind speed time series data obs of wind field numerical weather forecast data; step 2, extracting trend items of multi-day wind speed time series data NWP; step 3, extracting trend items of multi-day measured wind speed time series data obs; step 4, dynamic time warping; step 5, dynamic fitting inversion; the present invention adopts the upper and lower envelope approximation method, performs dynamic time warping DTW processing on the upper and lower envelopes of the trend items of the multi-day wind speed time series data NWP, approximates the upper and lower envelopes of the trend items of the historical wind speed data after empirical mode decomposition, and inverts the wind speed through the obtained new envelope to realize wind speed correction, thereby improving the accuracy of wind power prediction.

Description

Technical Field

[0001] The invention belongs to the field of wind speed correction, and in particular relates to a wind speed correction method based on upper and lower envelope approximation. Background Art

[0002] Wind power generation is a renewable, clean and environmentally friendly form of energy. Wind energy is a renewable resource and its power generation process does not produce greenhouse gas emissions or air pollutants, which helps alleviate energy shortages and reduce air pollution.

[0003] Wind power generation has many advantages and broad application prospects, but it also faces problems caused by the randomness and volatility of wind energy. The randomness and volatility of wind energy will lead to the instability of wind power output, which not only affects the frequency and voltage stability of the power grid when wind power is connected to the grid, increases the difficulty of peak load regulation of the power grid, but also makes wind power prediction complicated, requiring the use of advanced prediction models and technologies to improve prediction accuracy.

[0004] Since the near-surface wind field is easily affected by complex underlying surfaces and complex turbulence processes, it has significant volatility and uncertainty characteristics. The distribution of wind speed in space and time is uneven and uncertain. There may be large differences in wind speed data at different locations and times, which limits the applicability of existing wind speed correction models in different regions or time periods. There are problems with data quality and representativeness. The correction methods have different applicability in different wind speed ranges, and the correction capabilities for low and high wind speed ranges are also limited. Summary of the invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to provide a wind speed correction method based on upper and lower envelope approximation in view of the deficiencies in the prior art. It mainly aims at the problem that there is a deviation between the wind speed data of the current numerical weather forecast and the actual wind speed. The numerical weather forecast wind speed data is corrected through the daily fluctuation characteristics of the actual wind speed, which helps to improve the wind power prediction.

[0006] The present invention comprises the following steps:

[0007] Step 1: Collect data: Select the multi-day wind speed time series data of wind field numerical weather prediction data NWP (numerical weather prediction, NWP) And the time series data of multi-day measured wind speed obs (observed wind speed, obs) ,in, It is a collection of NWP time series data of multi-day wind speed time series data. obs is the time series data set of wind speed measured over multiple days; d is the date serial number, It is the serial number of the time point of daily wind speed measurement; is the date number, is the total number of wind speed measurement time points per day;

[0008] Step 2: Extract the trend item of the multi-day wind speed time series data NWP: Perform EMD (Empirical Mode Decomposition) on the multi-day wind speed time series data NWP to obtain the trend item of each day's wind speed data, and extract the upper envelope of the trend item of the multi-day wind speed time series data NWP. and lower envelope , , They are the upper and lower envelope data sets of the NWP trend item of multi-day wind speed time series data;

[0009] Step 3, extract the trend item of the time series data obs of the multi-day measured wind speed: perform EMD processing on the time series data obs of the multi-day measured wind speed to obtain the trend item of the wind speed data of each day, and use the same method as step 2 to extract the upper envelope of the trend item of the time series data obs of the multi-day measured wind speed and lower envelope , , They are the upper and lower envelope data sets of the obs trend item of the time series data of the wind speed measured over multiple days;

[0010] Step 4, dynamic time warping: correspond the upper envelope of the NWP trend item of the multi-day wind speed time series data to the upper envelope of the obs trend item of the multi-day measured wind speed time series data, correspond the upper envelope of the NWP trend item of the multi-day wind speed time series data to the lower envelope of the obs trend item of the multi-day measured wind speed time series data, and process it by the optimized dynamic time warping DTW (Dynamic Time Warping, DTW) method, so that the upper and lower envelopes of the NWP trend item of the multi-day wind speed time series data are close to the upper and lower envelopes of the obs trend item of the multi-day measured wind speed time series data, and obtain a new upper envelope and lower envelope , , are the data sets for obtaining the new upper and lower envelopes respectively;

[0011] Step 5, dynamic fitting inversion: The time series data of each date is inverted by using the dynamic adaptive graph attention fitting algorithm for the new upper and lower envelopes to obtain the time series data , as the corrected multi-day wind speed time series data NWP, is the inverted time series data set.

[0012] Step 4 includes:

[0013] Step 4-1, define two sets of time series, namely:

[0014] Upper envelope time series combination: and ;

[0015] Lower envelope time series combination: and ;

[0016] The length of the time series is ,remember For parameters ;

[0017] Step 4-2, construct the distance matrix: calculate the Manhattan distance between each pair of points in each time series to form two The distance matrix and ,

[0018] Distance Matrix Each element in represents an element of the NWP trend item sequence of multi-day wind speed time series data in the upper envelope time series combination The Manhattan distance between the element j of the obs trend term sequence of the multi-day measured wind speed time series data;

[0019] Distance Matrix Each element in represents an element of the NWP trend item sequence of multi-day wind speed time series data in the lower envelope time series combination The Manhattan distance between the element j of the obs trend term sequence of the multi-day measured wind speed time series data;

[0020] On the basis of Manhattan distance calculation, the derivative is extended to include the gradient and curvature of the signal into the distance calculation. The calculation formula is:

[0021] ,

[0022] in, is the distance matrix, and They are the elements of the NWP trend item sequence of multi-day wind speed time series data. and the element j of the trend term sequence of the obs time series data of the measured wind speed for multiple days, , , is the weight coefficient, , , They are and Manhattan distance, first derivative distance and second derivative distance between them; , They are The first-order difference of The first-order difference of , ; , They are The second-order difference of The second-order difference of

[0023] , ;

[0024] , , The calculation process is:

[0025] remember , , The Pearson correlation coefficients are , , , the calculation formula is:

[0026] ,

[0027] ,

[0028] ,

[0029] Among them, the intermediate parameters , intermediate parameters , intermediate parameters , intermediate parameters , intermediate parameters , intermediate parameters ;

[0030] The Pearson correlation coefficient is normalized to obtain , , :

[0031] ,

[0032] ,

[0033] ;

[0034] Step 4-3, dynamic programming to calculate the cumulative distance: define the cumulative distance matrix and , The former represents the NWP trend term sequence of multi-day wind speed time series data in the upper envelope time series combination. elements to the obs trend term sequence of the time series data of the measured wind speed for multiple days The minimum cumulative distance of elements, The front of the NWP trend term sequence of multi-day wind speed time series data in the lower envelope time series combination elements to the obs trend term sequence of the time series data of the measured wind speed for multiple days The minimum cumulative distance of elements;

[0035] The recursive formula for dynamic programming is:

[0036] ,

[0037] in is the minimum cumulative distance between the elements in the two time series. If the NWP trend item sequence of the multi-day wind speed time series data and the obs trend item sequence of the multi-day measured wind speed time series data are used as the x-axis and y-axis to construct the coordinate system, then Indicates from Arrive on the left. Indicates from Arrived above, Indicates from The upper left corner reaches diagonally;

[0038] The initialization condition is , and Can be initialized according to boundary conditions;

[0039] Step 4-4, find the optimal path: From the lower right corner of the cumulative distance matrix Start backtracking and find the optimal path with the smallest cumulative distance. The sum of the cumulative distances on the optimal path is the dynamic time warping DTW distance between the NWP trend item sequence of the multi-day wind speed time series data and the obs trend item sequence of the multi-day measured wind speed time series data.

[0040] According to the optimal path, the upper and lower envelope data of the NWP trend item of the multi-day wind speed time series data are corrected, so that the upper and lower envelope data of the NWP trend item of the multi-day wind speed time series data gradually approach the upper and lower envelope data of the obs trend item of the multi-day measured wind speed time series data, and finally a new upper envelope is obtained. and lower envelope .

[0041] Step 5 includes:

[0042] Step 5-1, construct a dynamic graph: model the local features of the time series data obs of the multi-day measured wind speed into a graph structure, where the nodes represent time points and the edge weights reflect local similarities;

[0043] Step 5-2, enhance the graph attention feature: use the multi-head graph attention network (GAT) that aggregates neighborhood information through the attention mechanism, strengthens the key node representation, dynamically adjusts the importance of different time points through attention weights, and focuses on mutation areas and key phases. The expression is:

[0044] ,

[0045] in, is the number of attention heads, For the The attention weight of each head, For the The independent parameter matrix of each head, For the concatenation feature operation, the final output dimension is .

[0046] Step 5-3, perform multi-scale dynamic alignment;

[0047] Step 5-4, online incremental update.

[0048] Step 5-1 includes:

[0049] Step 5-1-1, extract node features: for two time series and Extract multidimensional features at each time point, including amplitude, first-order derivative, second-order derivative and local entropy;

[0050] Step 5-1-2, calculate edge weights: construct time series based on feature similarity dynamics The adjacency matrix of and The adjacency matrix of , the formula is:

[0051] ,

[0052] ,

[0053] Where exp is the natural exponential function, , are the elements in the data set of the new upper envelope respectively , ,

[0054] , are the elements in the data set of the new lower envelope , ,

[0055] , The time series Information entropy and time series at each time point Information entropy at each time point; is a parameter that controls the similarity decay rate, is a parameter that balances temporal proximity, usually derived from empirical values ​​(e.g. ) and gradually adjust. You can also choose an appropriate method to determine the specific task and data characteristics. and If further optimization is needed, cross-validation or Bayesian optimization can be combined for automatic parameter adjustment.

[0056] Information Entropy The calculation method is:

[0057] ,

[0058] in, It is a window Middle The probability of each category is calculated with a 4-hour time interval as a window in the time series. The value range is .

[0059] Step 5-3 includes:

[0060] Step 5-3-1, coarse-grained alignment: The multi-day measured wind speed time series data obs is used as the original signal for wavelet downsampling to obtain a low-frequency approximate signal, and the improved dynamic time warping DTW method is applied to the graph attention feature space to generate a global alignment path;

[0061] Step 5-3-2, fine-grained correction: In the high-frequency sub-band, based on the global path constraint, a local adaptive window (the window size is dynamically adjusted by the signal gradient) is used for fine-tuning, and a derivative consistency loss is introduced to ensure that the alignment path remains smooth in amplitude and first-order / second-order derivatives.

[0062] Step 5-4 includes: using a sliding window mechanism to maintain only the most recent Y 1 The graph structure and attention weights of (the value is 96 in this invention) time points.

[0063] The present invention also provides an electronic device, comprising a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor executes the steps of the described method.

[0064] The present invention also provides a storage medium storing a computer program or instruction. When the computer program or instruction is run on a computer, the steps of the method described are executed.

[0065] The present invention adopts the upper and lower envelope approximation method to perform dynamic time warping (DTW) processing on the upper and lower envelopes of the NWP trend items of multi-day wind speed time series data, approximates the upper and lower envelopes of the trend items of the historical wind speed data after empirical mode decomposition, and inverts the wind speed through the obtained new envelope to achieve wind speed correction, thereby improving the accuracy of wind power prediction.

[0066] The present invention has the following beneficial effects: First, it optimizes the dynamic time warping DTW algorithm. The present invention processes the upper and lower envelopes of the NWP trend item of the multi-day wind speed time series data through the optimized dynamic time warping DTW algorithm to approximate the upper and lower envelopes of the obs trend item of the multi-day measured wind speed time series data, and obtains new upper and lower envelopes. The optimized dynamic time warping DTW algorithm takes into account the trend changes of non-stationary signals and needs to be combined with derivative information, and incorporates the gradient and curvature of the signal into the distance calculation. While calculating the Manhattan distance, the first-order derivative distance and the second-order derivative distance are calculated, which effectively solves the problem that the traditional dynamic time warping DTW algorithm relies on amplitude differences.

[0067] Second, inversion using a dynamic fitting algorithm. The present invention uses a dynamic adaptive graph attention fitting algorithm to invert time series data. The algorithm combines graph neural networks (GNNs), attention mechanisms, and multi-scale dynamic programming. Its core idea is dynamic graph modeling, adaptive attention mechanisms, and multi-scale hierarchical alignment, which improves the ability to capture non-stationary features while ensuring robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 is a flow chart of the method of the present invention.

[0069] Figure 2 It is a schematic diagram comparing the envelope matching results of the method of the present invention in January.

[0070] Figure 3 It is a schematic diagram comparing the envelope matching results of the method of the present invention in April.

[0071] Figure 4 It is a schematic diagram of the revised wind speed results in January according to the method of the present invention.

[0072] Figure 5 It is a schematic diagram of the revised wind speed results in April of the method of the present invention. DETAILED DESCRIPTION

[0073] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods, and the advantages of the method of the present invention will become more clear.

[0074] The embodiment of the present invention provides a wind speed correction method based on upper and lower envelope approximation, such as Figure 1 As shown, the following steps are included:

[0075] Step 1: Data collection. The present invention selects the multi-day wind speed time series data NWP from April 2020 to January 2021 of an offshore wind farm located in Rudong, Jiangsu, with longitude and latitude coordinates (32˚65'N, 121˚26'E) and the time series data of wind speed measured over multiple days obs ,in, is the NWP collection of multi-day wind speed time series data, is the obs set of time series data of wind speed measured over multiple days. is the date serial number, The wind speed is measured at the same time as the daily wind speed. The wind farm area belongs to the typical subtropical monsoon climate zone, with rain and heat occurring at the same time. In summer and autumn, it is often affected by extreme weather such as tropical cyclones. The spatiotemporal distribution of wind speed in this area shows significant seasonal fluctuations.

[0076] Step 2: Extract the trend item of the multi-day wind speed time series data NWP. Perform EMD processing on the multi-day wind speed time series data NWP to obtain the trend item of each day's wind speed data, and extract the upper envelope and lower envelope of the trend item of the multi-day wind speed time series data NWP, that is, and ,in, , They are the upper and lower envelope data sets of the NWP trend item of multi-day wind speed time series data; is the date serial number, It is the serial number of the time point for measuring wind speed every day.

[0077] Step 3: Extract the trend item of the multi-day measured wind speed time series data obs. Perform the empirical mode decomposition (EMD) process on the multi-day measured wind speed time series data obs to obtain the trend item of each day's wind speed data. Use the same method as S2 to extract the upper and lower envelopes of the trend item of the multi-day measured wind speed time series data obs, that is, and ,in, , They are the upper and lower envelope data sets of the obs trend item of the time series data of the wind speed measured over multiple days; is the date serial number, It is the serial number of the time point for measuring wind speed every day.

[0078] Step 4, dynamic time warping. The upper envelope of the NWP trend item of the multi-day wind speed time series data corresponds to the upper envelope of the obs trend item of the multi-day measured wind speed time series data, and the lower envelopes of the two correspond to each other. Through the optimized dynamic time warping (Dynamic Time Warping, hereinafter referred to as dynamic time warping DTW) method, the upper and lower envelopes of the NWP trend item of the multi-day wind speed time series data are close to the upper and lower envelopes of the obs trend item of the multi-day measured wind speed time series data, and the new upper and lower envelopes are obtained, that is, and , , are the data sets for obtaining the new upper and lower envelopes respectively; is the date serial number, It is the serial number of the time point for measuring wind speed every day.

[0079] The dynamic time warping (DTW) algorithm is an algorithm that searches and estimates the minimum possible distance (maximum possible similarity) between two time series after dynamic time domain warping based on a dynamic programming strategy. The core idea is to find the best alignment between the two series by nonlinearly stretching or compressing the time series, thereby minimizing the cumulative distance between them. This method allows a point in one sequence to correspond to multiple points in another sequence, thereby achieving time warping. The specific steps of the dynamic time warping (DTW) algorithm are as follows:

[0080] Step 401: define two groups of time series, namely, the upper envelope time series combination of the NWP trend item of the multi-day wind speed time series data and the obs trend item of the multi-day measured wind speed time series data. and The lower envelope time series combination of the NWP trend item of the multi-day wind speed time series data and the obs trend item of the multi-day measured wind speed time series data and The lengths of the four time series are .

[0081] in is the date number; is the total number of wind speed measurement time points per day, for .

[0082] Step 402: Construct a distance matrix. Calculate the Manhattan distance between each pair of points in each sequence to form two The distance matrix and Each element in these two matrices represents the element of the NWP trend item sequence of the multi-day wind speed time series data in the two sets of time series combinations. and the elements of the obs trend term sequence of the time series data of the measured wind speed for multiple days The Manhattan distance between the two signals is calculated based on the derivative information. Since the trend change of non-stationary signals needs to be combined with derivative information, the derivative is expanded on the basis of Manhattan distance calculation to include the gradient and curvature of the signal into the distance calculation. The calculation formula is as follows:

[0083] ,

[0084] in, is the distance matrix, and They are the elements of the NWP trend item sequence of multi-day wind speed time series data. and the element j of the trend term sequence of the obs time series data of the measured wind speed for multiple days, , , are weight coefficients, respectively , , The Pearson coefficient is normalized, , , They are and The Manhattan distance, first derivative distance, and second derivative distance between them. , They are , The first-order difference of , . , They are , The second-order difference of , .

[0085] , , The calculation process is:

[0086] remember , , The Pearson correlation coefficient is , , , the calculation formula is:

[0087] ,

[0088] ,

[0089] ,

[0090] in, , , , , , .

[0091] The Pearson correlation coefficient is normalized to obtain , , :

[0092] ,

[0093] ,

[0094] ,

[0095] Step 403: Calculate the cumulative distance using dynamic programming. Define the cumulative distance matrix and , which represents the NWP of multi-day wind speed time series data elements to the time series data obs of the wind speed measured over multiple days The minimum cumulative distance of elements. The recursive formula of dynamic programming is:

[0096] ,

[0097] in, That is, it is the minimum cumulative distance of each element in the two sets of time series. If the NWP trend item sequence of multi-day wind speed time series data and the obs trend item sequence of multi-day measured wind speed time series data are used as the x-axis and y-axis to construct the coordinate system, Indicates from Arrive on the left. Indicates from Arrived above, Indicates from The upper left corner arrives diagonally.

[0098] The initialization condition is , and Can be initialized according to boundary conditions.

[0099] Step 404: Find the optimal path. Start backtracking and find the optimal path with the smallest cumulative distance. The sum of the cumulative distances on the optimal path is the dynamic time warping DTW distance of the two sequences.

[0100] According to the optimal path, the upper and lower envelope data of the NWP trend item of the multi-day wind speed time series data are corrected to gradually approach the upper and lower envelope data of the obs trend item of the multi-day measured wind speed time series data, and finally the new upper and lower envelopes are obtained, that is, and ,in is the date serial number, It is the serial number of the time point for measuring wind speed every day.

[0101] Step 5: Dynamic fitting inversion. The time series data of each date is inverted by using the dynamic adaptive graph attention fitting algorithm for the new upper and lower envelopes to obtain the time series data , as the corrected multi-day wind speed time series data NWP, is the inverted time series data set.

[0102] The dynamic adaptive graph attention fitting algorithm can handle the dynamic alignment problem of non-stationary signals very well, improving the ability to capture non-stationary features while ensuring robustness. The specific steps are as follows:

[0103] Step 501: Build a dynamic graph. Model the local features of the time series (such as amplitude, gradient, curvature) as a graph structure, with nodes representing time points and edge weights reflecting local similarities. This is divided into two steps:

[0104] Step 5011: Extract node features. and Multidimensional features are extracted at each time point, including amplitude, first-order derivative, second-order derivative and local entropy.

[0105] Step 5012: Calculate edge weights. and The adjacency matrix is ​​dynamically constructed based on feature similarity. The formula is:

[0106] ,

[0107] ,

[0108] Where exp is the natural exponential function, , and , are the elements in the data set of the new upper envelope respectively , and the elements in the data set of the new lower envelope , , , They are , The information entropy at each time point can dynamically adjust the weights of key areas of non-stationary signals (such as mutation points) in the dynamic fitting algorithm to give them higher weights. is a parameter that controls the similarity decay rate, is a parameter that balances temporal proximity, usually derived from empirical values ​​(e.g. ) and gradually adjust. You can also choose an appropriate method to determine the specific task and data characteristics. and If further optimization is needed, cross-validation or Bayesian optimization can be combined for automatic parameter adjustment.

[0109] Information entropy is calculated as follows:

[0110] ,

[0111] in, It is a window Middle The probability of each category is calculated with a 4-hour time interval as a window in the time series. The value range is .

[0112] Step 502: Enhance the graph attention feature. Use the multi-head graph attention network (GAT) that aggregates neighborhood information through the attention mechanism and strengthens the key node representation. The importance of different time points is dynamically adjusted through the attention weights to focus on the mutation area and key phase. The expression is as follows:

[0113] ,

[0114] in, is the number of attention heads, For the The attention weight of each head, For the The independent parameter matrix of each head, For the concatenation feature operation, the final output dimension is .

[0115] Step 503: multi-scale dynamic alignment. This is divided into two steps: capturing global trends at a coarse granularity and optimizing local alignment at a fine granularity to avoid noise interference, specifically including:

[0116] Step 5031, coarse-grained alignment. The obs data of the time series of wind speed measured over multiple days is used as the original signal for wavelet downsampling to obtain a low-frequency approximate signal. The improved dynamic time warping (DTW) is applied to the graph attention feature space to generate a global alignment path.

[0117] Step 5032: Fine-grained correction. In the high-frequency subband, fine-tuning is performed using a local adaptive window (the window size is dynamically adjusted by the signal gradient) based on the global path constraint. A derivative consistency loss is introduced to ensure that the alignment path remains smooth in amplitude and first-order / second-order derivatives.

[0118] Step 504: Online incremental update: Using a sliding window mechanism, only the graph structure and attention weights of the most recent 96 time points are maintained for the real-time signal.

[0119] This method is applied to specific engineering applications. The multi-day wind speed time series data NWP and multi-day measured wind speed time series data obs of a wind farm in Rudong, Jiangsu Province from April 2020 to January 2021 are tested with two sets of data from April and January of the following year. The specific implementation example is as follows:

[0120] 1. Verify the effectiveness of the optimized dynamic time warping DTW algorithm proposed in the present invention;

[0121] In order to verify the effectiveness of the optimized dynamic time warping DTW algorithm proposed in this paper, this experiment selected the multi-day wind speed time series data NWP and the multi-day measured wind speed time series data obs in April and January of the following year, and after performing empirical mode decomposition EMD processing, the traditional dynamic time warping DTW algorithm and the optimized dynamic time warping DTW algorithm were used to conduct experiments on the trend term of the multi-day wind speed time series data NWP approaching the trend term of the multi-day measured wind speed time series data obs. The envelope matching results for January are shown in Figure 2. Figure 2 As shown, the envelope matching results for April are as follows Figure 3 As shown. By comparing the approximation results of the two algorithms, it can be found that the traditional dynamic time warping DTW algorithm has the problem of relying on amplitude differences, and the effect of approximating the upper and lower envelopes of obs is not good. The optimized dynamic time warping DTW algorithm takes into account the trend changes of non-stationary signals and needs to combine derivative information, and incorporates the gradient and curvature of the signal into the distance calculation. While calculating the Manhattan distance, the first-order derivative distance and the second-order derivative distance are calculated, which effectively solves the problem. Compared with the upper and lower envelopes generated by the traditional DTW algorithm, they are closer to the upper and lower envelopes of obs.

[0122] Second, verify the effectiveness of the wind speed correction model proposed in the present invention;

[0123] In order to verify the effectiveness of the wind speed correction model proposed in this invention, the wind speed correction results are compared with the multi-day measured wind speed time series data obs and multi-day wind speed time series data NWP. The envelope matching results in January are as follows: Figure 4 As shown, the envelope matching results for April are as follows Figure 5 As shown. By analyzing the curve graph, it can be seen that the wind speed correction method based on upper and lower envelope approximation proposed in the present invention has a relatively good wind speed correction effect in the data tests in April and January of the following year, and has a good reflection of the wind speed characteristics of each time period. It can convert NWP data into wind speed correction data that is closer to obs data and more in line with the actual real wind speed value, that is, the time series data obs of multi-day measured wind speed. It can be seen that the method proposed in the present invention has excellent and stable performance.

[0124] The present invention provides a wind speed correction method based on upper and lower envelope approximation. There are many methods and ways to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. A wind speed correction method based on upper and lower envelope approximation, characterized in that: The following steps are involved: Step 1: Collect data: Select the multi-day wind speed time series data of wind field numerical weather forecast data NWP{X dt ,d=1,2,3...D,t=1,2,3...N} and the time series data of wind speed measured over multiple days obs{Y dt ,d=1,2,3...D,t=1,2,3...N}, where X dt is the NWP time series data set of multiple-day wind speed time series data, Y dt is the obs time series data set of wind speed measured over multiple days; d is the date serial number, t is the serial number of the daily wind speed measurement time point; D is the date number, N is the total number of daily wind speed measurement time points; Step 2: Extract the trend item of the multi-day wind speed time series data NWP: Perform EMD processing on the multi-day wind speed time series data NWP to obtain the trend item of each day's wind speed data, and extract the upper envelope of the trend item of the multi-day wind speed time series data NWP and lower envelope They are the upper and lower envelope data sets of the NWP trend item of multi-day wind speed time series data; Step 3, extract the trend item of the time series data obs of the multi-day measured wind speed: perform EMD processing on the time series data obs of the multi-day measured wind speed to obtain the trend item of the wind speed data of each day, and use the same method as step 2 to extract the upper envelope of the trend item of the time series data obs of the multi-day measured wind speed and lower envelope They are the upper and lower envelope data sets of the obs trend item of the time series data of the wind speed measured over multiple days; Step 4, dynamic time warping: correspond the upper envelope of the NWP trend item of the multi-day wind speed time series data to the upper envelope of the obs trend item of the multi-day measured wind speed time series data, correspond the lower envelope of the NWP trend item of the multi-day wind speed time series data to the lower envelope of the obs trend item of the multi-day measured wind speed time series data, and process it through the optimized dynamic time warping DTW method, so that the upper and lower envelopes of the NWP trend item of the multi-day wind speed time series data are close to the upper and lower envelopes of the obs trend item of the multi-day measured wind speed time series data, and obtain a new upper envelope and lower envelope are the data sets for obtaining the new upper and lower envelopes respectively; Step 5, dynamic fitting inversion: The time series data of each date are inverted by using the dynamic adaptive graph attention fitting algorithm for the new upper and lower envelopes to obtain the time series data {Z dt ,d=1,2,3...D,t=1,2,3...N}, as the corrected multi-day wind speed time series data NWP, Z dt is the time series data set after inversion; Step 4 includes: Step 4-1, define two sets of time series, namely: Upper envelope time series combination: and Lower envelope time series combination: and The length of the time series is D×N, and D×N is denoted as the parameter M; Step 4-2, construct the distance matrix: calculate the Manhattan distance between each pair of points in each time series to form two M×M distance matrices D 0 [i,j] and D 1 [i,j], Distance Matrix D 0 Each element in [i,j] represents the Manhattan distance between element i of the NWP trend item sequence of multi-day wind speed time series data in the upper envelope time series combination and element j of the obs trend item sequence of multi-day measured wind speed time series data; Distance Matrix D 1 Each element in [i,j] represents the Manhattan distance between element i of the NWP trend term sequence of multi-day wind speed time series data in the lower envelope time series combination and element j of the obs trend term sequence of multi-day measured wind speed time series data; On the basis of Manhattan distance calculation, the derivative is extended to include the gradient and curvature of the signal into the distance calculation. The calculation formula is: D[i,j]=α|x i -y j |+β|x' i -y' j |+γ|x” i -y” j |, Where D[i,j] is the distance matrix, x i and j are the element i of the NWP trend item sequence of the multi-day wind speed time series data and the element j of the obs trend item sequence of the multi-day measured wind speed time series data, α, β, γ are weight coefficients, |x i -y j |、|x' i -y' j |、|x” i -y" j |x respectively i and j Manhattan distance, first-order derivative distance and second-order derivative distance between; x' i ,y' j x i The first difference of j The first difference of , x' i =x i+1 -x i , y' i =y i+1 -y i ;x” i ,y” j x i The second-order difference of j The second difference of i =x' i+1 -x' i ,y" i =y' i+1 -y' i ; The calculation process of α, β, and γ is: Note [x i ,y j ]、[x' i ,y' j ], [x” i ,y” j The Pearson correlation coefficients of ] are α1, β1, and γ1, respectively, and the calculation formula is: Among them, the intermediate parameters Intermediate parameters Intermediate parameters Intermediate parameters Intermediate parameters Intermediate parameters Normalize the Pearson correlation coefficient to get α, β, and γ: Step 4-3, dynamic programming to calculate the cumulative distance: define the cumulative distance matrix DTW 0 [i,j] and DTW 1 [i,j],DTW 0 [i,j] represents the minimum cumulative distance between the first i elements of the NWP trend item sequence of the multi-day wind speed time series data in the upper envelope time series combination and the first j elements of the obs trend item sequence of the multi-day measured wind speed time series data, DTW 1 [i,j] represents the minimum cumulative distance between the first i elements of the NWP trend item sequence of the multi-day wind speed time series data in the lower envelope time series combination and the first j elements of the obs trend item sequence of the multi-day measured wind speed time series data; The recursive formula for dynamic programming is: DTW[i,j]=D[i,j]+min(DTW[i-1,j],DTW[i,j-1],DTW[i-1,j-1]), Where DTW[i,j] is the minimum cumulative distance of each element in the two sets of time series. If the NWP trend item sequence of multi-day wind speed time series data and the obs trend item sequence of multi-day measured wind speed time series data are used as the x-axis and y-axis to construct the coordinate system, then DTW[i-1,j] means arriving from the left side of D[i,j], DTW[i,j-1] means arriving from the top of D[i,j], and DTW[i-1,j-1] means arriving from the upper left corner diagonal of D[i,j]; The initialization condition is DTW[0,0]=D[0,0]; Step 4-4, find the optimal path: start backtracking from the lower right corner of the cumulative distance matrix DTW[M,M] to find the optimal path with the smallest cumulative distance. The sum of the cumulative distances on the optimal path is the dynamic time warping DTW distance between the NWP trend item sequence of the multi-day wind speed time series data and the obs trend item sequence of the multi-day measured wind speed time series data; According to the optimal path, the upper and lower envelope data of the NWP trend item of the multi-day wind speed time series data are corrected, so that the upper and lower envelope data of the NWP trend item of the multi-day wind speed time series data gradually approach the upper and lower envelope data of the obs trend item of the multi-day measured wind speed time series data, and finally a new upper envelope is obtained. and lower envelope 2. The method according to claim 1, characterized in that: Step 5 includes: Step 5-1, construct a dynamic graph: model the local features of the time series data obs of the multi-day measured wind speed into a graph structure, where the nodes represent time points and the edge weights reflect local similarities; Step 5-2, enhance the graph attention feature: use the multi-head graph attention network GAT to aggregate neighborhood information through the attention mechanism, strengthen the key node representation, and dynamically adjust the importance of different time points through the attention weight. The expression is: Where K is the number of attention heads, is the attention weight of the kth head, W k is the independent parameter matrix of the kth head, ‖ is the concatenation feature operation, and the final output dimension is K×d'; Step 5-3, perform multi-scale dynamic alignment; Step 5-4, online incremental update.

3. The method according to claim 2, characterized in that Step 5-1 includes: Step 5-1-1, extract node features: for two time series and Extract multidimensional features at each time point, including amplitude, first-order derivative, second-order derivative and local entropy; Step 5-1-2, calculate edge weights: construct time series based on feature similarity dynamics The adjacency matrix of and The adjacency matrix of The formula is: Where exp is the natural exponential function, are the elements i and j in the data set of the new upper envelope, are the elements i and j in the data set of the new lower envelope, θ 0 ,θ 1 The time series Information entropy and time series at each time point The information entropy at each time point; σ is the parameter that controls the similarity decay rate, and λ is the parameter that balances temporal proximity; The information entropy θ is calculated as: Among them, p i (x) is the probability of the kth category in window x, and the value range of k is [1,6D].

4. The method according to claim 3, characterized in that Step 5-3 includes: Step 5-3-1, coarse-grained alignment: The multi-day measured wind speed time series data obs is used as the original signal for wavelet downsampling to obtain a low-frequency approximate signal, and the improved dynamic time warping DTW method is applied to the graph attention feature space to generate a global alignment path; Step 5-3-2, fine-grained correction: In the high-frequency sub-band, fine-tune using a local adaptive window based on the global path constraint.

5. The method according to claim 4, characterized in that Step 5-4 includes: using a sliding window mechanism to maintain only the graph structure and attention weights of the most recent Y1 time points for the real-time signal.

6. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores program codes, and when the program codes are executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 5.

7. A storage medium, characterized in that: A computer program or instruction is stored, and when the computer program or instruction is run on a computer, the steps of the method according to any one of claims 1 to 5 are executed.

Images