A wind energy resource endowment assessment method, system, device and medium

Abstract

The present invention discloses a wind energy resource endowment assessment method, system, device, and medium, relating to the field of artificial intelligence intelligent meteorological technology, including: obtaining back-calculated data, reanalyzed wind speed data, and terrain data of the area to be assessed in the target assessment year; constructing a hierarchical feature fusion model and a feature differentiation processing model based on the preprocessed back-calculated data to obtain refined wind energy resource feature data; constructing a terrain wind energy generator based on the preprocessed reanalyzed wind speed data and terrain data to optimize wind energy resource assessment and obtain a simulated wind energy resource image; and inputting the refined wind energy resource feature data and the simulated wind energy resource image into a discriminator to obtain a refined wind energy resource assessment result. The present invention utilizes deep learning and refined feature capture technology to achieve a refined assessment of wind energy resource endowment, providing a scientific basis for the rational utilization and optimal allocation of wind energy resources.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence intelligent meteorological technology, and in particular to a wind energy resource endowment assessment method, system, equipment and medium. Background Art

[0002] In recent years, with the rapid increase in computing power and the rapid advancement of deep learning technology, artificial intelligence has made significant progress in fields such as perception, assessment, and prediction. Although traditional statistical methods (such as the Weibull probability distribution), wind power density assessment indicators, and spatial analysis methods based on geographic information systems (GIS) can reveal a certain degree of wind energy resource endowment in a region, these methods have significant limitations: They are overly simplistic and lack the ability to comprehensively consider multiple influencing factors; they also fail to fully leverage the advantages of deep learning technology in complex pattern recognition and high-dimensional data processing, resulting in a lack of precision and detail in regional wind energy resource assessments. Therefore, a model that can integrate multi-feature data and accurately capture the distribution patterns of wind energy resources is urgently needed to fill the gap in the refined assessment of regional wind energy resource endowment and provide strong support for the precise positioning of wind power generation. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a wind energy resource endowment assessment method, system, device and medium to solve the limitations of existing wind energy resource assessment methods.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides a method for evaluating wind energy resource endowment, comprising:

[0007] Obtain back-calculated data, reanalyzed wind speed data, and terrain data for the area to be assessed in the target assessment year;

[0008] Based on the pre-processed back-calculated data, a hierarchical feature fusion model and a feature differentiation processing model are constructed. Dilated convolution is used to capture multi-scale features, an attention mechanism is used to adjust weights, and convolution and feature transformation are performed after setting thresholds to divide features, thereby obtaining refined feature data of wind energy resources.

[0009] Based on the preprocessed reanalysis wind speed data and terrain data, a terrain wind energy generator is constructed to optimize the wind energy resource assessment and obtain a simulated wind energy resource image;

[0010] The refined feature data of wind energy resources and the simulated wind energy resource images are input into the discriminator to obtain the refined assessment results of wind energy resources.

[0011] As a preferred embodiment of the wind energy resource endowment assessment method of the present invention, wherein:

[0012] The back-calculation data after pre-processing includes the following steps:

[0013] The back-calculated data includes wind speed data, wind power density data, wind shear data and turbulence intensity data;

[0014] Normalize the wind speed data and wind power density data in the back-calculated data;

[0015] Perform conditional normalization on the wind shear data in the back-calculated data;

[0016] Perform denormalization on the turbulence intensity data in the back-calculated data;

[0017] The normalized wind speed data and wind power density data in the back-calculated data are stacked in the channel dimension to obtain the wind energy intensity characteristics;

[0018] The conditionally normalized wind shear data and the denormalized turbulence intensity data in the back-calculated data are stacked in the channel dimension to obtain the stability characteristics.

[0019] As a preferred embodiment of the wind energy resource endowment assessment method of the present invention, wherein:

[0020] The hierarchical feature fusion model includes the following steps:

[0021] The wind energy intensity characteristics and stability characteristics are spliced in the time dimension to obtain the original combined feature map;

[0022] Apply different dilation rates to the combined feature map to perform dilated convolution operations on the spatial dimension to obtain multi-scale spatial feature maps;

[0023] Perform dilated convolution operations on the spatial feature map in the time dimension to capture the changing characteristics of the time series and obtain the temporal feature map;

[0024] Combine the spatial feature map and the temporal feature map with the original combined feature map through element-wise multiplication to obtain a fused feature map;

[0025] Based on the fusion feature map, the attention mechanism is guided to adjust the weights of the intensity feature and the stability feature to obtain the attention feature map;

[0026] The acquired spatial feature map, temporal feature map and attention feature map are integrated to obtain the comprehensive wind energy resource characteristics;

[0027] The dilated convolution operation on the spatial dimension by applying different dilation rates to the combined feature map is expressed as:

[0028] ,

[0029] in, express The feature map of the t-th time step, and Indicates that the current convolution operation is The position in and Represents the position index of the convolution kernel, and The value range is from -1 to 1. Indicates the position at the current expansion rate The surrounding receptive field; represents the weight in the convolution kernel, represents the bias term of the convolution operation, Indicates that the dilated convolution operation is at position The output eigenvalue at represents the spatial expansion rate;

[0030] The dilated convolution operation on the spatial feature map in the time dimension Expressed as:

[0031] ,

[0032] in, Indicates location At time step The above features, represents the feature map obtained by spatial dilation convolution, Represents the index on the time dimension, The value range is from -1 to 1; represents the temporal convolution kernel weight, represents the bias term of the convolution operation, represents the time dilation rate.

[0033] As a preferred embodiment of the wind energy resource endowment assessment method of the present invention, wherein:

[0034] The feature differentiation processing model includes the following steps:

[0035] The average pooling operation is used to calculate the average values of wind energy intensity and stability characteristics in the time dimension, and the average values are used as the feature weight threshold.

[0036] According to the feature weight threshold, the wind energy intensity feature is divided into relatively strong features and relatively weak features, and the wind energy stability feature is divided into relatively stable features and relatively unstable features;

[0037] Merge relatively strong features with relatively stable features to generate enriched features, and apply convolution operation to obtain enriched feature maps;

[0038] Merge relatively weak features with relatively unstable features and input them into the feature conversion unit to obtain a rich semantic feature map;

[0039] The enriched feature map and the rich semantic feature map are merged to obtain the differential wind energy resource characteristics.

[0040] As a preferred embodiment of the wind energy resource endowment assessment method of the present invention, wherein:

[0041] The pre-processed reanalysis of wind speed data and terrain data comprises the following steps:

[0042] The reanalysis wind speed data and topographic data were resampled to the same spatial resolution as the back-calculated data;

[0043] Reanalysis wind power density data are calculated using wind speed data;

[0044] The reanalysis wind power density data and terrain data are normalized respectively to obtain normalized reanalysis wind power density data and terrain data.

[0045] As a preferred embodiment of the wind energy resource endowment assessment method of the present invention, wherein:

[0046] The construction of the terrain wind energy generator comprises the following steps:

[0047] The normalized reanalysis wind power density data P and terrain data are merged D and input into the terrain wind energy generator;

[0048] The terrain wind energy generator includes an encoder, a bottleneck layer and a decoder;

[0049] Based on the combined input data, the encoder uses convolutional layers with instance normalization and activation functions to extract key features of the terrain. ;

[0050] Key terrain features extracted based on encoder ,The bottleneck layer uses dilated convolution to capture multi-scale spatial dependencies and generate high-level feature representation B;

[0051] Based on the high-level feature representation B provided by the bottleneck layer, the decoder recovers the spatial dimension of the image through transposed convolution operations;

[0052] The decoder uses skip connections to transmit the key terrain features extracted by the encoder Combined with the normalized reanalysis wind power density data P, a simulated wind energy resource image is generated. .

[0053] As a preferred embodiment of the wind energy resource endowment assessment method of the present invention, wherein:

[0054] The step of inputting the refined wind energy resource feature data and the simulated wind energy resource image into the discriminator comprises the following steps:

[0055] Integrate comprehensive wind energy resource characteristics and differentiated wind energy resource characteristics to form refined wind energy resource characteristic data;

[0056] The accuracy of the refined wind energy resource characteristic data in restoring local meteorological characteristics is evaluated using normalized reanalysis wind power density data.

[0057] The output of the generator is corrected by distinguishing the refined characteristic data of wind energy resources from the simulated wind energy resource image.

[0058] In a second aspect, the present invention provides a wind energy resource endowment assessment system, comprising:

[0059] Data acquisition and preprocessing module, used to obtain back-calculated data, reanalyzed wind speed data and terrain data of the area to be assessed in the target assessment year;

[0060] The hierarchical feature fusion and differentiation processing module is used to build a hierarchical feature fusion model and a feature differentiation processing model based on the preprocessed back-calculation data. It captures multi-scale features through dilated convolution, adjusts weights through the attention mechanism, and sets thresholds to divide features before performing convolution and feature transformation to obtain refined feature data of wind energy resources.

[0061] The terrain wind energy generator module is used to build a terrain wind energy generator based on the preprocessed reanalyzed wind speed data and terrain data, optimize the wind energy resource assessment, and obtain the simulated wind energy resource image;

[0062] The discriminator module is used to input the refined feature data of wind energy resources and the simulated wind energy resource image into the discriminator to obtain the refined evaluation results of wind energy resources.

[0063] In a third aspect, the present invention provides an electronic device, comprising:

[0064] Memory, used to store programs;

[0065] The processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the wind energy resource endowment assessment method.

[0066] In a fourth aspect, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the steps of implementing the wind energy resource endowment assessment method are implemented.

[0067] The present invention has the following beneficial effects: By designing hierarchical feature fusion and feature differentiation processing modules, the present invention maximizes the preservation of wind energy resource intensity and stability characteristics, reduces redundant information, and thus improves the accuracy of wind energy resource assessment and achieves higher spatial resolution. Furthermore, the terrain wind energy generator module combines terrain data to generate wind energy resource images that are more consistent with local meteorological characteristics. The discriminator module further optimizes the generated results, significantly improving the reliability and accuracy of the assessment model. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0069] Figure 1 A schematic diagram of a basic flow chart of a wind energy resource endowment assessment method provided by one embodiment of the present invention;

[0070] Figure 2 This is an overall structural diagram of a network model of a wind energy resource endowment assessment method provided by one embodiment of the present invention;

[0071] Figure 3 A diagram showing the structure of a hierarchical feature fusion module for a wind energy resource endowment assessment method provided by one embodiment of the present invention;

[0072] Figure 4 A structural diagram of a feature differentiation processing module of a wind energy resource endowment assessment method provided by one embodiment of the present invention;

[0073] Figure 5 A structural diagram of a feature conversion unit of a wind energy resource endowment assessment method provided by one embodiment of the present invention;

[0074] Figure 6 A structural diagram of a terrain wind energy generator module of a wind energy resource endowment assessment method provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0075] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0076] Example 1, reference Figures 1 to 6 , as one embodiment of the present invention, provides a wind energy resource endowment assessment method, comprising:

[0077] S1: Obtain back-calculated data, reanalyzed wind speed data, and terrain data for the area to be assessed in the target assessment year;

[0078] In an embodiment of the present invention, the back-calculated data is the operational Weather Research and Forecasting model (WRF) back-calculated data, and the reanalysis wind speed data is the European Centre for Medium-Range Weather Forecasts Reanalysis Data Version 5 (ERA5) reanalysis wind speed data, wherein the WRF back-calculated data is generated by the operational Weather Research and Forecasting model (WRF), and these data are obtained by numerical simulation based on historical meteorological observations (such as global weather station, satellite and radar data) and reanalysis data (such as ERA5 and the National Centers for Environmental Prediction NCEP of the United States).

[0079] In an embodiment of the present invention, WRF back-calculated data, ERA5 reanalysis wind speed data, and terrain data have the following characteristics: the spatial resolution of WRF back-calculated data is 1 kilometer, and the temporal resolution is 1 hour, including four elements such as wind speed, wind power density, wind shear, and turbulence intensity. In order to match the analysis requirements, the 1-hour data will be aggregated and converted into daily average data. The spatial resolution of ERA5 reanalysis wind speed data is 25 kilometers, and the temporal resolution is also 1 hour. The 1-hour data also needs to be aggregated into daily average data. The spatial resolution of the terrain data is 30 meters, covering five elements such as altitude, slope, slope curvature, slope direction, and flow direction.

[0080] S2: Based on the preprocessed back-calculated data, a hierarchical feature fusion model and a feature differentiation processing model are constructed. Dilated convolution is used to capture multi-scale features, an attention mechanism is used to adjust weights, and convolution and feature transformation are performed after setting thresholds to divide features, thereby obtaining refined feature data of wind energy resources.

[0081] In this embodiment of the present invention, traditional statistical analysis methods are used to study the long-term variations in wind energy resources within the region to be assessed, and the results are used as prior information. Next, the wind speed data (normalized), wind power density data (normalized), wind shear data (conditionally normalized), and turbulence intensity data (denormalized) from the WRF back-calculated data are processed. The processed data are stacked two by two along the channel dimension to obtain wind energy intensity and stability characteristics, which serve as inputs to the hierarchical feature fusion module and the feature differentiation processing module.

[0082] In an embodiment of the present invention, conditional normalization is specifically as follows: setting multiple wind shear absolute value ranges, and estimating the probability distribution of historical wind shear data in each range based on the Weibull probability distribution, and performing a visual analysis of the probability distribution of wind energy resources in different ranges. By comparing the resource differences within each range, the optimal wind shear absolute value range that can significantly reflect the wind energy resource potential of the grid in the area to be evaluated is determined, and it is determined as the optimal range. Based on the optimal range, the wind shear value within the time period of the wind energy resource endowment to be evaluated is normalized, that is, the wind shear value within the range is close to 1 after normalization, and the wind shear value away from the range will gradually decrease according to the exponential decay rule.

[0083] In an embodiment of the present invention, a hierarchical feature fusion module and a feature differentiation processing module are constructed to optimize the fusion of multi-feature information. The hierarchical feature fusion module is used to enhance the spatiotemporal correlation between features. The feature differentiation processing and aggregation module is used for feature reconstruction and conversion to reduce the redundant features generated in the feature fusion process and retain more information about the small target grid. The outputs of the two modules are integrated as the input of the discriminator module.

[0084] In an embodiment of the present invention, a hierarchical feature fusion module (HFFM) such as Figure 3 The specific process is as follows:

[0085] After concatenating the wind intensity and stability features, dilated convolution is used to capture the changing patterns of wind energy resources at different scales. The output of the convolution operation is nonlinearly processed using a Sigmoid activation function. The activated result is used to weight the feature map, highlighting the most relevant features. Finally, through hierarchical feature fusion, the initial intensity features, stability features, and features processed by dilated convolution are combined, retaining the original information while performing deep feature learning, thereby more accurately capturing the changing patterns of wind energy resources. The specific operation of HFFM is as follows:

[0086] Wind energy intensity characteristics and stability characteristics , each time step Feature map Represents different feature states, is the height of the feature map, Is the width of the feature map. Feature map is obtained after feature concatenation , the specific formula is:

[0087] ,

[0088] Among them, concat is a splicing operation used to connect two or more tensors in a specified dimension. The concatenated feature map Perform dilated convolution. The specific operations are as follows:

[0089] For any position in the feature map A 3×3 convolution kernel and a dilation rate r are applied to perform dilated convolution operations to extract multi-scale spatial features. The output can be expressed as:

[0090] ,

[0091] in, yes The feature map of the t-th time step, and Is the current convolution operation in The position in and Is the position index of the convolution kernel, traversing from -1 to 1 respectively; Indicates the position at the current expansion rate The surrounding receptive field; is the weight in the convolution kernel; is the bias term of the convolution operation, Is the dilated convolution operation at position The output eigenvalue at the location, the entire convolution process is achieved by continuously At each time step, each position The local area is convolved and finally the spatial feature map is obtained by splicing . Represents the spatial expansion rate, for different expansion rates: when = 1, the convolution output is ;when = 2, the convolution output is ;when = 3, the convolution output is .

[0092] Perform an expansion convolution operation on the time dimension of the feature map after spatial convolution , for any position in the feature map The expansion rate is calculated at each time step The convolution of , the output is expressed as:

[0093] ,

[0094] in, It's location At time step The above features, here It is the feature map obtained by spatial dilation convolution, is the index on the time dimension, from -1 to 1; is the temporal convolution kernel weight; Is the bias term of the convolution operation. The whole convolution process is to continuously Each position Perform convolution operations with different receptive fields at each time step, and finally splice to obtain the temporal feature map . Represents the time expansion rate, for different expansion rates: when = 1, the convolution output is ;when = 2, the convolution output is ;when = 3, the convolution output is The output of spatial and temporal dilated convolutions is combined with Perform element-level multiplication to obtain fused features , allowing the model to "enhance" or "adjust" important features at specific locations and time steps.

[0095] Using attention mechanism to fuse features 、 and , aims to utilize high-level features Unguided attention mechanism to more efficiently adjust wind energy intensity features and stability characteristics By fusing weights, we can automatically focus on more important features or areas and ignore less relevant parts, thus obtaining more representative wind energy resource endowment characteristics.

[0096] In an embodiment of the present invention, a Feature Differentiation Processing Module (FDPM) such as Figure 4 The specific process is as follows:

[0097] The average pooling operation is used to calculate the average values of the wind energy intensity feature and the stability feature in the time dimension respectively, and the average value is used as the feature weight threshold to capture the global information of each spatial position at all time steps. According to the feature weight threshold, the wind energy intensity feature is divided into relatively strong features and relatively weak features, and the wind energy stability feature is divided into relatively stable features and relatively unstable features. Then, the relatively strong features are merged with the relatively stable features to generate enriched features, and a convolution operation is applied to extract more detailed information; the relatively weak features are merged with the relatively unstable features and input into the feature conversion unit to generate a feature map with richer semantic information using fewer computing resources. Finally, the features are merged to obtain more detailed wind energy resource endowment features after FDPM processing. The specific operations of FDPM are as follows:

[0098] Wind energy intensity characteristics and stability characteristics , each time step Feature map Represents different feature states, is the height of the feature map, is the width of the feature map. The average pooling operation is used to calculate the wind energy intensity feature and stability characteristics The average value in the time dimension is used to obtain the feature weight threshold. This operation captures the global information of each spatial position at all time steps. The specific formula is:

[0099] ,

[0100] ,

[0101] in, yes No. The feature map of time steps, is the weight threshold of wind energy intensity feature, yes No. The feature map of time steps, is the weight threshold of the wind energy stability feature. According to the calculated feature weight threshold, the wind energy intensity feature Divide into relatively strong characteristics and relatively weak characteristics; divide the wind energy stability characteristics Divided into relatively stable characteristics and relatively unstable characteristics, the specific formula is:

[0102] ,

[0103] ,

[0104] ,

[0105] ,

[0106] in, is a relatively strong feature, It is a relatively weak feature. It is a relatively stable feature. It is a relatively unstable feature. Since the features have differences in time steps after threshold classification, the relatively strong features and relatively stable features are added element by element at the same time step, while the features at different moments are not added. The features generated after this processing are enriched features. Through the convolution operation, more detailed information is further extracted to enhance the expressive power of features. The convolution operation has an impact on the time step. At each position of the feature map:

[0107] ,

[0108] in, and Is the current convolution operation in the enrichment feature The position at each time step; and is the position index of the convolution kernel, traversing 1 to 3 respectively; Represents the convolution operation on the position The spatial convolution range at ; is the weight in the convolution kernel; is the bias term of the convolution operation; is the convolution output Position in the feature map The value at .

[0109] The relatively weak features and relatively unstable features are added element by element at the same time step, while the features at different moments are not added. The features generated after this process are Input Feature Transformation Unit (FTU) to obtain ,This unit generates feature maps with richer semantic information using less computational resources.

[0110] Enrichment features The output result after convolution and Adding each element together to obtain more detailed wind energy resource endowment characteristics after FDPM processing .

[0111] In the embodiment of the present invention, Figure 5 As shown, the specific operations of FTU are as follows:

[0112] Extraction using depthwise separable convolution The local characteristics of , the specific formula is:

[0113] ,

[0114] in, Represents half the size of the depthwise convolution kernel (for example, 3×3 convolution corresponds to ); Represents the depth convolution kernel (usually of size )’s weight; express of Position in the moment feature graph The value at express of Position in the moment feature graph The output after depth convolution. Then use 1×1 convolution to all time channels of depth convolution output Perform linear combination to achieve information fusion between channels:

[0115] ,

[0116] in, Represents the weight matrix of point-by-point convolution, responsible for Input channels are converted to output channels; express middle Position in the moment feature graph The eigenvalue at ; Indicates location The fusion result of all time channels at .

[0117] The attention weights are generated by pooling layer dimensionality reduction, 3×3 convolution operation to extract features and normalization layer standardization for adaptive control. The importance of local features, the specific formula is:

[0118] ,

[0119] in, Represents a pooling operation; Represents the convolution kernel weight; Represents the convolution operation; and denote the normalized mean and standard deviation respectively; and represents the normalized trainable parameters; It is a decimal that prevents the denominator from being zero; Represents the sigmoid activation function; As attention weights for feature enhancement. and Perform element-wise multiplication to enhance the feature expression of important areas, and the output is .

[0120] S3: Based on the preprocessed reanalysis wind speed data and terrain data, a terrain wind energy generator is constructed to optimize the wind energy resource assessment and obtain a simulated wind energy resource image;

[0121] In this embodiment of the present invention, a terrain wind energy generator module is constructed, using an architecture inspired by U-Net and optimized for wind resource assessment. This generator module replaces the noise input in traditional cGANs with terrain data and combines it with wind power density data to generate simulated wind resource images.

[0122] In this embodiment of the present invention, the terrain wind generator module adopts an architecture inspired by U-Net, modified for the task of wind resource refinement. The encoder uses convolutional layers with instance normalization and LeakyReLU activations to compress spatial dimensions while increasing feature depth. The bottleneck layer uses dilated convolutions to capture multi-scale spatial dependencies, which is crucial for understanding wind patterns of varying intensities within the assessed area. The decoder uses transposed convolutions and skip connections to preserve spatial details while improving the quality of wind resource assessment.

[0123] In the embodiment of the present invention, Figure 6 As shown, the specific operations of the terrain wind energy generator module are as follows:

[0124] Combine the terrain data and wind power density data to form a new input. The specific formula is:

[0125] ,

[0126] = ,

[0127] in, Represents terrain data, which is the main input of the generator. The generator extracts features from the terrain data through the encoder and then generates output related to wind power density; Represents the normalized reanalysis wind power density data, representing the wind energy density at each location at different time steps, and is used to conditionally constrain the generator to generate a matching wind energy image. The convolution layer uses different filters and stride length Extract input data of The spatial characteristics of the channel and the LeakyReLU activation function are used to introduce nonlinear characteristics; Represents padding in convolution operation; Represents the bias term in the convolutional layer of the encoder; output is a high-dimensional representation of terrain features and will serve as the input to the bottleneck layer. The bottleneck layer uses dilated convolution to capture spatial dependencies at different scales, which is crucial for understanding the spatial pattern of wind energy distribution. The specific formula is:

[0128] ,

[0129] in, is the convolution kernel size, is the expansion rate, represents the bias term in the convolution layer of the bottleneck layer, the dilated convolution ( ) can enable the generator to capture the spatial characteristics of wind energy in a larger range The decoder gradually restores the spatial resolution and combines the terrain features extracted by the encoder with the wind power density data through skip connections. The specific formula is:

[0130] ,

[0131] ,

[0132] in, represents the bias term in the deconvolution layer of the decoder, the convolution transpose ( ) operation restores the spatial dimension of the image through deconvolution operation to generate a wind energy resource image The skip connection combines the topographic features from the encoder and wind power density data Combined with the decoder input, it helps the terrain wind energy generator to better consider the relationship between terrain and wind energy resources when generating wind energy resource maps, thereby generating a wind energy resource map that can accurately reflect the spatial distribution of wind energy potential and closely integrate the actual terrain characteristics. .

[0133] S4: Input the refined feature data of wind energy resources and the simulated wind energy resource image into the discriminator to obtain the refined assessment results of wind energy resources.

[0134] In this embodiment of the present invention, a discriminator module is constructed to receive the refined wind resource feature data from S2, the reanalyzed wind power density data from S3, and the simulated wind resource image. Serving as a "quality controller" in the RWEA-MFFC model, the discriminator module not only uses wind power density data to verify the effectiveness of the refined wind resource feature data in restoring local meteorological characteristics but also distinguishes between the refined wind resource feature data and the generated data, correcting the generator module output to gradually approximate the actual wind resource characteristics, thereby improving the reliability and accuracy of the assessment results.

[0135] In an embodiment of the present invention, a model training and validation process is also described, including:

[0136] The years of wind energy resource endowment to be assessed (such as 2023 or 2005-2009) are selected, and the WRF back-calculated data, ERA5 reanalysis wind speed data and terrain data during the years are used as model input data to build the training data set of the RWEA-MFFC model.

[0137] Based on the constructed training dataset, the RWEA-MFFC model loss is calculated, the network parameters are optimized, and a network model that conforms to the RWEA-MFFC architecture is obtained.

[0138] The grids with higher wind energy resource potential in the refined wind energy resource assessment results output by the RWEA-MFFC model are compared with the site data of meteorological observation stations in the area to be assessed to verify the model's ability to restore the wind energy resource endowment of local grids.

[0139] It should be noted that the present invention discloses a refined wind energy resource endowment assessment based on multi-feature fusion and cGAN (RWEA-MFFC) method, which can not only verify the consistency of the multi-index wind energy resource assessment results based on WRF back-calculated data and ERA5 data in the spatiotemporal distribution of wind energy resources, but also restore the local wind energy resource characteristics lost due to low resolution of ERA5 data, thereby realizing a refined assessment of the wind energy resource endowment in the area to be assessed.

[0140] Example 2, which is an embodiment of the present invention, provides a method for evaluating wind energy resource endowment. In order to verify its beneficial effects, scientific demonstration is carried out through specific implementation methods and implementation effects.

[0141] The specific details of this embodiment are as follows:

[0142] The experimental results presented in this example rely on a study area between 24.51° and 29.29° north latitude and 103.49° and 110.12° east longitude, which roughly covers the main part of a certain province in China and contains complex topographic features and diverse climatic conditions. The selection of this area relies on its unique geographical and meteorological characteristics, making it an ideal place for studying wind energy resource endowment assessment models. The experimental results show that the method of the present invention performs back-calculation and comprehensive assessment of historical wind energy resource data in this area with high accuracy and reliability.

[0143] This embodiment uses data from 2000 to 2004, 2006 to 2010, 2012 to 2016, and 2018 to 2022 as training sets, and data from 2005, 2011, 2017, and 2023 as test sets.

[0144] In this embodiment, analysis of the existing ERA5 wind power density data found that the data lacked refinement in spatial distribution and failed to effectively distinguish between wind energy resource-rich and barren areas. In addition, when compared with the actual site data of a provincial meteorological observation station, it was found that the ERA5 data had a significant underestimate in the estimation of wind energy resources and failed to accurately reflect the real differences in wind energy resources within a province. In contrast, the refined wind energy resource endowment assessment method based on multi-feature fusion and conditional generative adversarial network proposed in the present invention can better fit the site data and show a significant similar trend with the ERA5 data in the spatiotemporal distribution of wind energy resource endowment. At the same time, it is verified that the method of the present invention can restore the local wind energy resource feature loss caused by the low resolution of ERA5.

[0145] To further validate the advantages of the proposed method in assessing the wind energy resource endowment of a specific province, a module ablation experiment was conducted. The experimental results showed that the wind resource distribution map generated by the traditional generative adversarial network method was similar to the ERA5 data, but failed to fully reflect the spatial differences in the province's wind energy resource endowment. The wind resource distribution map generated by adding HFFM to the generative adversarial network method, while able to reveal certain regional differences in wind energy resources, still failed to accurately fit the site data. The wind resource distribution map generated by adding FDPM to the generative adversarial network method, while effectively recovering the underestimation in the ERA5 data by learning wind energy stability characteristics, failed to accurately represent regional differences in wind energy resource endowment. The wind resource distribution map generated by combining HFFM and FDPM was significantly better than using either module alone. Further, replacing the generator of the generative adversarial network with the terrain wind energy generator module resulted in a wind resource distribution map that better fitted the site data and showed significant similarity to the ERA5 data in the spatiotemporal distribution of wind energy resource endowment.

[0146] Example 3: This example provides a wind energy resource endowment assessment system, including:

[0147] Data acquisition and preprocessing module, used to obtain back-calculated data, reanalyzed wind speed data and terrain data of the area to be assessed in the target assessment year;

[0148] The hierarchical feature fusion and differentiation processing module is used to build a hierarchical feature fusion model and a feature differentiation processing model based on the pre-processed back-calculation data to obtain refined feature data of wind energy resources;

[0149] The terrain wind energy generator module is used to build a hierarchical feature fusion model and a feature differentiation processing model based on preprocessed back-calculation data. It captures multi-scale features through dilated convolution, adjusts weights through an attention mechanism, and sets thresholds to divide features before performing convolution and feature transformation to obtain refined feature data of wind energy resources.

[0150] The discriminator module is used to input the refined feature data of wind energy resources and the simulated wind energy resource image into the discriminator to obtain the refined evaluation results of wind energy resources.

[0151] This embodiment further provides an electronic device applicable to a method for evaluating wind energy resource endowment, including:

[0152] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a wind energy resource endowment assessment method as proposed in the above embodiment.

[0153] This embodiment also provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, a method for evaluating wind energy resource endowment as proposed in the above embodiment is implemented.

[0154] The storage medium proposed in this embodiment and the method for implementing a wind energy resource endowment assessment proposed in the above embodiment belong to the same inventive concept. Technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0155] Through the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general-purpose hardware. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0156] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A wind energy resource endowment assessment method, characterized in that: include: Obtain back-calculated data, reanalyzed wind speed data, and terrain data for the area to be assessed in the target assessment year; Based on the pre-processed back-calculated data, a hierarchical feature fusion model and a feature differentiation processing model are constructed. Dilated convolution is used to capture multi-scale features, an attention mechanism is used to adjust weights, and convolution and feature transformation are performed after setting thresholds to divide features, thereby obtaining refined feature data of wind energy resources. Based on the preprocessed reanalysis wind speed data and terrain data, a terrain wind energy generator is constructed to optimize the wind energy resource assessment and obtain a simulated wind energy resource image; The refined feature data of wind energy resources and the simulated wind energy resource images are input into the discriminator to obtain the refined assessment results of wind energy resources.

2. A wind energy resource endowment assessment method according to claim 1, characterized in that: The back-calculation data after pre-processing includes the following steps: The back-calculated data includes wind speed data, wind power density data, wind shear data and turbulence intensity data; Normalize the wind speed data and wind power density data in the back-calculated data; Perform conditional normalization on the wind shear data in the back-calculated data; Perform denormalization on the turbulence intensity data in the back-calculated data; The normalized wind speed data and wind power density data in the back-calculated data are stacked in the channel dimension to obtain the wind energy intensity characteristics; The conditionally normalized wind shear data and the denormalized turbulence intensity data in the back-calculated data are stacked in the channel dimension to obtain the stability characteristics.

3. A wind energy resource endowment assessment method according to claim 2, characterized in that: The hierarchical feature fusion model includes the following steps: The wind energy intensity characteristics and stability characteristics are spliced in the time dimension to obtain the original combined feature map X combined ; Apply different dilation rates to the combined feature map to perform dilated convolution operations on the spatial dimension to obtain multi-scale spatial feature maps; Perform dilated convolution operations on the spatial feature map in the time dimension to capture the changing characteristics of the time series and obtain the temporal feature map; Combine the spatial feature map and the temporal feature map with the original combined feature map through element-wise multiplication to obtain a fused feature map; Based on the fusion feature map, the attention mechanism is guided to adjust the weights of the intensity feature and the stability feature to obtain the attention feature map; The acquired spatial feature map, temporal feature map and attention feature map are integrated to obtain the comprehensive wind energy resource characteristics; The dilated convolution operation on the spatial dimension by applying different dilation rates to the combined feature map is expressed as: Among them, X t Represents X combined The feature map of the t-th time step, i and j represent the current convolution operation in X t The position in the convolution kernel is m and n, and the value range of m and n is from -1 to 1. t (i+m·r,j+n·r) represents the receptive field around position (i,j) at the current expansion rate; K(m+1,n+1) represents the weight in the convolution kernel, b spatial represents the bias term of the convolution operation, represents the output eigenvalue of the dilated convolution operation at position (i, j), and r represents the spatial dilation rate; The dilated convolution operation on the spatial feature map in the time dimension Expressed as: in, represents the feature of position (i, j) at time step td·p, X r represents the feature map obtained by spatial dilation convolution, p represents the index in the time dimension, and the value range of p is from -1 to 1; K(p+1) represents the time convolution kernel weight, b temporal represents the bias term of the convolution operation, and d represents the time dilation rate.

4. A wind energy resource endowment assessment method according to claim 3, characterized in that: The feature differentiation processing model includes the following steps: The average pooling operation is used to calculate the average values of wind energy intensity and stability characteristics in the time dimension, and the average values are used as the feature weight threshold. According to the feature weight threshold, the wind energy intensity feature is divided into relatively strong features and relatively weak features, and the wind energy stability feature is divided into relatively stable features and relatively unstable features; Merge relatively strong features with relatively stable features to generate enriched features, and apply convolution operation to obtain enriched feature maps; Merge relatively weak features with relatively unstable features and input them into the feature conversion unit to obtain a rich semantic feature map; The enriched feature map and the rich semantic feature map are merged to obtain the differential wind energy resource characteristics.

5. A wind energy resource endowment assessment method according to claim 4, characterized in that: The pre-processed reanalysis of wind speed data and terrain data comprises the following steps: The reanalysis wind speed data and topographic data were resampled to the same spatial resolution as the back-calculated data; Reanalysis wind power density data are calculated using wind speed data; The reanalysis wind power density data and terrain data are normalized respectively to obtain normalized reanalysis wind power density data and terrain data.

6. A wind energy resource endowment assessment method according to claim 5, characterized in that: The construction of the terrain wind energy generator comprises the following steps: The normalized reanalysis wind power density data P and terrain data are merged D and input into the terrain wind energy generator; The terrain wind energy generator includes an encoder, a bottleneck layer and a decoder; Based on the merged input data, the encoder uses a convolutional layer with instance normalization and activation function to extract the key features of the terrain E i ; Based on the terrain key features E extracted by the encoder i ,The bottleneck layer uses dilated convolution to capture multi-scale spatial dependencies and generate high-level feature representation B; Based on the high-level feature representation B provided by the bottleneck layer, the decoder recovers the spatial dimension of the image through transposed convolution operations; The decoder uses skip connections to transmit the terrain key features E extracted by the encoder i Combined with the normalized reanalysis wind power density data P, a simulated wind energy resource image G is generated. skip .

7. A wind energy resource endowment assessment method according to claim 6, characterized in that: The step of inputting the refined wind energy resource feature data and the simulated wind energy resource image into the discriminator comprises the following steps: Integrate comprehensive wind energy resource characteristics and differentiated wind energy resource characteristics to form refined wind energy resource characteristic data; The accuracy of the refined wind energy resource characteristic data in restoring local meteorological characteristics is evaluated using normalized reanalysis wind power density data. The output of the generator is corrected by distinguishing the refined feature data of wind energy resources from the simulated wind energy resource image.

8. A wind energy resource endowment assessment system, applying a wind energy resource endowment assessment method according to any one of claims 1 to 7, characterized in that: include: Data acquisition and preprocessing module, used to obtain back-calculated data, reanalyzed wind speed data and terrain data of the area to be assessed in the target assessment year; The hierarchical feature fusion and differentiation processing module is used to build a hierarchical feature fusion model and a feature differentiation processing model based on the preprocessed back-calculation data. It captures multi-scale features through dilated convolution, adjusts weights through the attention mechanism, and sets thresholds to divide features before performing convolution and feature transformation to obtain refined feature data of wind energy resources. The terrain wind energy generator module is used to build a terrain wind energy generator based on the preprocessed reanalyzed wind speed data and terrain data, optimize the wind energy resource assessment, and obtain the simulated wind energy resource image; The discriminator module is used to input the refined feature data of wind energy resources and the simulated wind energy resource image into the discriminator to obtain the refined evaluation results of wind energy resources.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a wind energy resource endowment assessment method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a wind energy resource endowment assessment method according to any one of claims 1 to 7 are implemented.

Images