ERA5 downscaling model-based deep and far sea wind field generating capacity evaluation method

Abstract

The invention provides a deep and far sea wind field generating capacity evaluation method based on an ERA5 downscaling model, and belongs to the field of deep and far sea offshore wind power, and the method comprises the following steps: obtaining the original wind field data of ERA5 at a preset height from the sea level in the longitude and latitude range of a target sea area in a preset time range, and carrying out the region cutting and coordinate alignment, preprocessing the original wind field data of the ERA5; training an ERA5 downscaling model through a super-resolution generative adversarial network deep learning model framework; performing super-resolution processing on the original wind field data by using an ERA5 downscaling model to generate high-resolution wind field data; arranging the position of the virtual anemometer tower according to the deep and far sea wind power project, correcting the wind field data of the virtual anemometer tower by using the measured data, and calculating the wind resource and the generating capacity of the deep and far sea wind field. According to the method, the ERA5 low-resolution wind field data is downscaled through the super-resolution generative adversarial network, the original 0.25-degree resolution is improved to 0.0625 degrees, the wind energy resource assessment specification is met, and the early-stage investment and the time period of an offshore wind plant project are reduced.

Description

Technical Field

[0001] This invention belongs to the field of deep-sea offshore wind power, specifically involving a method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model. Background Technology

[0002] ERA5 is the fifth-generation global climate reanalysis dataset released by the European Centre for Medium-Range Weather Forecasts (ECMWF). It provides time-based global climate data with a spatial resolution of 0.25° and an actual distance scale of 25–30 km. It exhibits good continuity and regularity and is widely used for early-stage wind resource and theoretical power generation assessments of offshore wind power projects. According to the "Code for Design of Wind Farms" (GB 51096-2015), the control radius for wind energy resource assessment at a site should ideally be 10 km for the measured data from each wind measurement tower. However, because the 25–30 km spatial resolution of ERA5 is greater than the 10 km requirement in the code, directly using the ERA5 dataset for wind energy resource assessment and power generation performance prediction analysis has several shortcomings.

[0003] Traditional wind resource assessment methods primarily rely on conventional wind measurement data. However, as offshore wind power projects expand from nearshore to deep-sea areas, they often face challenges such as distance from land and a lack of meteorological data. Furthermore, the construction of offshore wind measurement towers is costly, requiring significant upfront investment and a long timeframe, making it difficult to meet the timeliness and cost-effectiveness requirements of deep-sea offshore wind farm projects in the early stages. Therefore, how to utilize deep learning frameworks to generate effective wind resource data at virtual wind measurement tower locations through computer simulation and data analysis, thereby addressing the lack of meteorological data and the low spatial resolution of raw ERA5 wind field data for deep-sea offshore wind power projects, has become an urgent practical problem.

[0004] Existing technologies utilize various methods for downscaling using deep learning. While deep learning has achieved significant results in image processing, further research is needed to apply generative adversarial networks and super-resolution techniques within deep learning frameworks to downscaling regular ERA5 wind field data, treating it as image data. Most current research is limited to local optimization techniques such as downscaling, wind resource assessment, or power generation calculation, failing to effectively establish a complete digital workflow from low-resolution ERA5 wind field data to high-resolution wind field reconstruction, then to wind resource and power generation assessment, and finally to visualization and interaction of the results. Summary of the Invention

[0005] To address the problems in the existing technologies, this invention proposes a method for assessing the power generation of deep-sea wind farms based on the ERA5 downscaling model. By using a super-resolution generative adversarial network enhanced by a deep learning model framework, the low-resolution ERA5 wind farm data is downscaled, and a virtual wind measurement tower is established to provide wind resource and power generation assessment results for deep-sea wind farms, thereby reducing the initial investment and time cycle of offshore wind farm projects.

[0006] The technical solution of the present invention is as follows: In a first aspect, this invention proposes a method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model, comprising the following steps: The system acquires raw ERA5 wind field data at a preset altitude above sea level within a preset time range and latitude and longitude range of the target sea area, performs regional cropping and coordinate alignment, and preprocesses the raw ERA5 wind field data. Select a super-resolution generative adversarial network deep learning model framework, which is configured with a residual module, an upsampling module, and a fidelity constraint module. Deep learning models are trained using a super-resolution generative adversarial network deep learning model framework. An adaptive learning rate optimizer is used to suppress overfitting during training, and a Dropout layer is added to the deep learning model. The original wind field data was super-resolution processed using the ERA5 downscaling model to generate high-resolution wind field data for the target sea area, and fidelity constraints were applied to the high-resolution wind field data. Based on the location of virtual wind measurement towers deployed in high-resolution wind fields in deep-sea wind power projects, the measured wind field data at a preset height above sea level along the coast is spatiotemporally matched and corrected with the virtual wind measurement tower data. Using the corrected high-resolution virtual wind measurement tower wind field data, the wind resources and wind power generation of deep-sea wind farms are estimated.

[0007] Furthermore, the deep learning model training combines learning rate warm-up, cosine decay strategy and dynamic adjustment strategy based on validation loss. The learning rate warm-up is the process of increasing the learning rate from 0 to the initial preset learning rate for a previous preset percentage of training rounds. The cosine decay strategy involves decreasing the learning rate according to a cosine function curve after the learning rate reaches the initial learning rate, eventually decreasing it to 1% of the initial value. The dynamic adjustment strategy based on validation loss is to multiply the learning rate by a coefficient factor=0.95 when the validation loss does not decrease for a preset number of consecutive training iterations.

[0008] Furthermore, the original wind field data is super-resolution processed using the ERA5 downscaling model to generate high-resolution wind field data for the target sea area. Specifically: The raw wind field data of ERA5 is input into the deep learning model, and the raw wind field data of ERA5 is magnified to the target high resolution size through nearest neighbor upsampling to generate the basic high resolution skeleton. High-resolution wind field data are stitched together to generate a residual map. The residual map is then processed using a predefined block mask, retaining only the residuals within the block mask. The residual map after block masking is added to the basic high-resolution skeleton to generate high-resolution wind field data.

[0009] Furthermore, the step of estimating the wind resources and wind farm power generation of the deep-sea wind farm based on the corrected high-resolution virtual meteorological tower wind field data includes the following steps: Turbulence intensity is calculated based on the corrected high-resolution virtual wind field data, and wind speed ranges and wind speed frequencies of the high-resolution virtual wind field data are statistically analyzed. The optimal shape parameters are obtained by fitting the wind speed frequency distribution to a Weibull curve. k and dimensional parameters c ; Calculate the pre-installed wheel hub height, the maximum gust wind speed that occurs once every 50 years. v 50,gust ,according to v 50,gust Select the fan model based on the turbulence intensity; The power generation of the deep-sea wind farm is calculated based on the corrected high-resolution virtual wind measurement tower wind field data, wind turbine model, and wind turbine operating time range.

[0010] Furthermore, the expression for fitting the wind speed frequency distribution using a Weibull curve... for: ; in, v Wind speed; k For shape parameters; c These are dimensional parameters.

[0011] Furthermore, the estimated pre-installed wheel hub height has a 50-year return period for the maximum wind speed. v 50,gust Includes the following steps: The annual maximum wind speed sequence at the virtual wind measurement tower over the past 30 years was statistically analyzed. The maximum wind speed with a 50-year return period was calculated using Gumbel type I and Pearson type III frequency distribution functions. Based on the wind shear index, the maximum wind speed with a 50-year return period at the height of the pre-installed wind turbine hub was estimated. v 50 ; Calculate the gust coefficient C It can be expressed as a formula as follows: ; in, n For frequency parameters; T Time interval; Calculate the maximum gust wind speed with a 50-year return period. v 50,gust It can be expressed as a formula as follows: ; in, σ 50 This represents the standard deviation of the annual maximum wind speed sequence over the past 30 years.

[0012] Furthermore, the wind farm's power generation E th The calculation formula is: ; in, m This refers to the number of generator units in the wind farm. v 1 represents the cut-in wind speed of the wind turbine generator; v 2 represents the cut-out wind speed of the wind turbine generator set; p j ( v ) is the first j The typhoon wind turbine generator is operating at a wind speed of v Power generation at that time; For the first j The Weibull distribution obtained by fitting the wind speed probability distribution of the typhoon wind turbine generator.

[0013] Furthermore, the spatial resolution of the high-resolution wind field data is 0.0625°, and the actual distance scale is 7km.

[0014] Furthermore, the fidelity constraint on the high-resolution wind field data specifically includes: High-resolution wind field data is downsampled to restore low-resolution wind field data, and compared with the original wind field data to see if the wind field data at the same latitude and longitude are consistent. If they are inconsistent, the high-resolution wind field data is corrected to ensure that the downsampled low-resolution wind field data is consistent with the original wind field data.

[0015] Secondly, the present invention proposes a computer-readable storage medium storing computer-executable instructions for executing the above-described method for assessing the power generation of deep-sea wind farms based on the ERA5 downscaling model.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: By using a super-resolution generative adversarial network enhanced by a deep learning model framework, the low-resolution 0.25° ERA5 wind field data is downscaled, and the high-resolution 0.0625° wind field data is reconstructed through computer simulation and data analysis. This effectively solves the problem of lack of meteorological data for deep-sea offshore wind power projects, overcomes the disadvantage of low spatial resolution of the original ERA5 wind field data, and meets the requirements of existing technical specifications. By constructing a comprehensive meteorological database for offshore wind farms and a professional algorithm model for assessing wind resources and power generation, a unified data management and sharing platform is established, effectively improving the consistency and repeatability of data processing results. By combining Geographic Information System (GIS) technology with an interactive visualization platform, multi-dimensional data analysis and visualization are achieved, comprehensively and intuitively reflecting the assessment results of wind resources and power generation of offshore wind farms, and improving the intuitiveness and comprehensibility of the assessment results. The reconstructed high-resolution wind field data addresses the lack of meteorological data and insufficient spatial resolution in the early stages of deep-sea offshore wind power projects. By replacing traditional physical wind measurement towers with virtual ones, the project can achieve wind resource and power generation assessment results, reduce early investment and time cycle, and meet the timeliness and economic requirements of early-stage deep-sea offshore wind power projects. A digital technology process has been established that covers the entire process from high-resolution wind farm reconstruction to wind resource and power generation assessment, and then to result visualization and interaction, which systematically improves the efficiency of wind resource and power generation assessment for deep-sea offshore wind power projects. Attached Figure Description

[0017] Figure 1 A flowchart of a method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model; Figure 2 The training and validation loss curves, the loss curve, and the learning rate variation curve with the number of iterations are used to train the model. Figure 3 The results show the comparison of wind speed and direction between high-resolution wind field data and raw wind field data. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: This embodiment provides a method for obtaining high-resolution wind field data based on the ERA5 downscaling model, such as... Figure 1 As shown, it includes the following steps: We acquired the raw u and v wind field data of ERA5 at a height of 100m above sea level within the latitude and longitude range of the target sea area over the past 30 years, and performed regional cropping and coordinate alignment. In the ERA5 wind field data, the u wind field is the zonal wind component, corresponding to the east-west direction; the v wind field is the meridional wind component, corresponding to the north-south direction. The raw u and v wind field data of ERA5 are preprocessed, including data normalization, noise filtering and missing value processing; data normalization adopts the maximum and minimum value normalization method to normalize the raw data to the [0, 1] interval; noise filtering adopts the median filter with a filter window size of 5×5 to remove white noise in the data; missing value processing adopts the linear interpolation method to fill the missing data. The Super-Resolution Generative Adversarial Network (ESRGAN) deep learning model framework was selected to build a deep learning network containing a residual module, an upsampling module, and a fidelity constraint module to generate super-resolution results. The residual module uses 5 residual layers to capture detailed information in the data. The upsampling module includes 2 sub-layers: the first sub-layer uses bicubic interpolation upsampling, and the second sub-layer uses convolutional upsampling. The fidelity constraint module includes a discriminator to ensure the authenticity of the generated results. Deep learning model training, specifically: 75% of the preprocessed data was used as the training set and 25% as the validation set; the initial learning rate was set to 1×10⁻⁶. -3 Training is performed by combining a learning rate warmup, a cosine decay strategy, and a dynamic adjustment strategy based on validation loss. For the first 30% of training epochs, the learning rate smoothly increases non-linearly from 0 to the initial preset learning rate, set to 1×10⁻⁶. -3 To avoid instability in the early stages of training, the learning rate decays according to a cosine function curve after warm-up. The validation loss is monitored; if it does not decrease for five consecutive training iterations, the learning rate is multiplied by a factor of 0.95 to promote stable model convergence, eventually decreasing to 1% of the initial value. The AdamW optimizer is used to suppress overfitting, with a weight decay factor of 10. -4 It balances adaptive momentum adjustment and regularization; during training, intermediate super-resolution results are generated, and these results are downsampled to restore the initial resolution. The results are then compared with the original resolution, and the mean squared error (MSE) is calculated. When the standard deviation of the loss is less than 10 for five consecutive training rounds... -6 When the model has converged sufficiently, the training monitoring settings include setting the gradient accumulation step count to 1, performing gradient clipping after each training step, setting the maximum allowable gradient norm (max_norm=0.5) to prevent gradient explosion, evaluating model performance, and monitoring for overfitting risk. An early stopping threshold is set based on the validation set loss; if the validation loss falls below the preset early stopping threshold (MIN_DELTA=10) for more than 10 consecutive rounds, the model is considered to have converged sufficiently. -6Training is terminated early, and the best model is saved. A Dropout layer is added to the deep learning model to dynamically and randomly mask the ratio of neurons, initially set to 0.3, and then reduced to a final value of 0.1 during training to improve generalization ability. The training and validation loss curves, loss curves, and learning rate variation curves are shown below. Figure 2 As shown, the training and validation loss curves drop sharply in the early stages of training, indicating that the model quickly learns effective features from the data. Later, the loss decreases more gradually and stabilizes, indicating that the training process is normal and converges well. The loss values ​​on the loss curves use a logarithmic scale, and both curves show stable convergence in the later stages. The learning rate is gradually increased from 0 to its initial value at the beginning of training, and then gradually decreased using a cosine decay and dynamic adjustment strategy based on the validation loss, until the model smoothly converges to the optimal solution.

[0020] The trained model is used to perform super-resolution processing on the original u and v wind field data to generate high-resolution hourly wind field data for the target sea area throughout the year. Specifically: Low-resolution u and v wind field component data are input into the model, and the input data is magnified to the target high-resolution size through nearest neighbor upsampling to generate a basic high-resolution skeleton (u). bone v bone High-resolution wind field data is stitched together, processed through convolution, residual blocks, an upsampling module, and a final convolutional layer to generate a residual map. A predefined block mask is used to process the residual map, retaining only the residuals within the mask blocks and removing boundary effects. The residual map after block mask processing (u) res v res ) and (u bone v bone Adding these together yields high-resolution wind field data (u). hr v hr This enables ERA5 data downscaling and high-resolution wind field reconstruction.

[0021] Compare the u and v wind field components at the same latitude and longitude in the low-resolution wind field data recovered by downsampling from the original ERA5 u and v wind field data and the high-resolution wind field data to verify the model fidelity constraint effect. The fidelity constraint means that while the model improves the resolution, it ensures that the output of the high-resolution wind field data still matches the original u and v wind field data of ERA5 after being degraded back to the low-resolution wind field data. This ensures that the model can recover details and maintain the authenticity of the wind field when the magnification scale=4.

[0022] High-resolution wind field data were saved in NetCDF format with a time step of 1 hour; the spatial resolution of the high-resolution wind field data was 0.0625°, and the actual distance scale was 7km. Hourly wind speed and direction data measured at coastal meteorological stations were used to locally validate the high-resolution wind field data. Correlation coefficients were calculated and significance tests were performed. The correlation coefficients included relative error, mean absolute error, and root mean square error. A linear regression model between the measured wind field data and the high-resolution wind field data was established, and the high-resolution wind field data was corrected based on the measured wind field data.

[0023] The comparison results of wind speed and direction between high-resolution wind field data and raw wind field data are as follows: Figure 3 As shown, in the low-resolution ERA5 raw wind field data, the spatial distribution of wind speed and direction is locally blurred and the outline boundaries are unclear. In contrast, the high-resolution wind field data not only retains the characteristics of the original ERA5 data, but also makes the outline boundaries clearer, which meets the design specifications for wind energy resource assessment.

[0024] Example 2: This embodiment provides a method for assessing the power generation of deep-sea wind farms based on the ERA5 downscaling model. It involves deploying virtual wind measurement towers in high-resolution wind farms for deep-sea wind power projects and using an interactive geographic information system (GIS) visualization platform to assess wind resources and power generation. Figure 1 As shown, the specific steps include: By matching measured wind speed data from existing offshore wind measuring towers at a height of 100m with wind speed data from virtual wind measuring towers in the deep sea area in terms of time and space, the correlation coefficient between the two is calculated and significance is tested. The correlation coefficient includes relative error, mean absolute error and root mean square error. A linear regression model of measured data and virtual wind measuring tower wind speed data is established, and the virtual wind measuring tower wind speed data is corrected based on the measured data.

[0025] Wind speed intervals are defined as 1 m / s, such as the interval between 7.6 and 8.5 m / s for wind speeds of 8 m / s. The wind speed intervals and frequencies at the virtual wind measurement tower are statistically analyzed, and a histogram of wind speed frequency distribution is plotted.

[0026] The wind speed frequency distribution histogram is fitted using a Weibull function. The shape and size parameters of the Weibull probability density function conforming to the wind speed frequency distribution are solved using the maximum likelihood estimation method, and the optimal shape parameters are selected. k and dimensional parameters c Plot the Weibull probability density function curve. Expressed as a formula: ; in, v Wind speed; e It is a natural constant; The maximum likelihood estimation method is specifically as follows: For n independent observations v1, v2, ..., v n Likelihood functionL ( k , c Expressed as a formula: ; in, v i For the first i Each wind speed was an independent observation. Indicates in the parameter k , c Down v i Distribution; Pick L The logarithm yields the log-likelihood function ln L Expressed as a formula: ; The solution is found by minimizing the negative log-likelihood function, expressed as follows: ; Draw wind rose diagrams at the virtual anemometer tower, including wind direction rose diagram, wind energy rose diagram, wind direction frequency rose diagram, average wind speed rose diagram, maximum wind speed rose diagram, wind energy frequency rose diagram, and average wind energy rose diagram; Calculate the daily, monthly, and annual values ​​of wind speed and wind power density at the virtual meteorological tower, and plot the results as a line graph. Calculate the frequency distribution of wind speed and wind power density at the virtual wind measurement tower, and plot the frequency distribution of wind speed and wind power density. The annual maximum wind speed sequence at the virtual wind measurement tower over the past 30 years was statistically analyzed. The maximum wind speed with a 50-year return period was calculated using Gumbel type I and Pearson type III frequency distribution functions. Based on the wind shear index, the maximum wind speed with a 50-year return period at the height of the pre-installed wind turbine hub was estimated. v 50 ; The specific maximum gust wind speed for the 50-year return period, used to estimate the height of the pre-installed wind turbine hub, is as follows: Calculate the gust coefficient C It can be expressed as a formula as follows: ; in, n This is a frequency parameter, in Hz. T Time interval, in seconds; Calculate the maximum gust wind speed with a 50-year return period. v 50,gust It can be expressed as a formula as follows: ; in, σ 50This represents the standard deviation of wind speeds for the annual maximum wind speed sequence over the past 30 years. Calculate the turbulence intensity and turbulence level at the virtual wind measurement tower, and select the wind turbine model based on the 50-year return period maximum gust wind speed at the height of the pre-installed wind turbine hub; The return period refers to the average interval between hydrological and meteorological events calculated through data observation and mathematical statistics methods. It is the average number of years between the occurrence of a hydrological and meteorological event of a certain magnitude.

[0027] Based on the corrected high-resolution virtual meteorological tower wind field data, wind turbine model and wind turbine operating time range, calculate the wind farm power generation assessment data at the monthly or annual scale of the deep-sea wind farm, including total power generation hours, effective power generation hours and theoretical power generation per unit; Ignoring wake effects, the theoretical annual power generation of the wind farm E th The calculation formula is as follows: ; in, m This refers to the number of generator units in the wind farm. v 1 represents the cut-in wind speed of the wind turbine generator; v 2 represents the cut-out wind speed of the wind turbine generator set; p j ( v ) is the first j The typhoon wind turbine generator is operating at a wind speed of v Power generation at that time; For the first j The Weibull distribution obtained by fitting the wind speed probability distribution of a typhoon turbine generator set; The interactive GIS visualization platform enables the analysis and visualization of wind farm power generation assessment data. The visualization results include Weibull distribution curves, wind rose diagrams, wind speed and wind power density frequency distribution maps, daily wind speed and wind power time series diagrams, monthly wind speed and wind power time series diagrams, annual wind speed and wind power time series diagrams, monthly theoretical power generation, and annual theoretical power generation, which intuitively reflect the wind resources and power generation assessment results of deep-sea offshore wind farms.

[0028] Example 3 A computer-readable storage medium is a non-volatile memory storing computer-executable instructions. The computer-executable instructions are used to execute a method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model. For specific methods, please refer to the methods described in Embodiment 1 and Embodiment 2. For the sake of brevity, they will not be repeated here.

Claims

1. A method for assessing the power generation of deep-sea wind farms based on the ERA5 downscaling model, characterized in that, Includes the following steps: The system acquires raw ERA5 wind field data at a preset altitude above sea level within a preset time range and latitude and longitude range of the target sea area, performs regional cropping and coordinate alignment, and preprocesses the raw ERA5 wind field data. Select a super-resolution generative adversarial network deep learning model framework, which is configured with a residual module, an upsampling module, and a fidelity constraint module. Deep learning models are trained using a super-resolution generative adversarial network deep learning model framework. An adaptive learning rate optimizer is used to suppress overfitting during training, and a Dropout layer is added to the deep learning model. The original wind field data was super-resolution processed using the ERA5 downscaling model to generate high-resolution wind field data for the target sea area, and fidelity constraints were applied to the high-resolution wind field data. Based on the location of virtual wind measurement towers deployed in high-resolution wind fields in deep-sea wind power projects, the measured wind field data at a preset height above sea level along the coast is spatiotemporally matched and corrected with the virtual wind measurement tower data. Using the corrected high-resolution virtual wind measurement tower wind field data, the wind resources and wind power generation of deep-sea wind farms are estimated.

2. The method for assessing the power generation of deep-sea wind farms based on the ERA5 downscaling model according to claim 1, characterized in that, The deep learning model is trained by combining learning rate warm-up, cosine decay strategy and dynamic adjustment strategy based on validation loss. The learning rate warm-up is the process of increasing the learning rate from 0 to the initial preset learning rate for a previous preset percentage of training rounds. The cosine decay strategy involves decreasing the learning rate according to a cosine function curve after the learning rate reaches the initial learning rate, eventually decreasing it to 1% of the initial value. The dynamic adjustment strategy based on validation loss is to multiply the learning rate by a coefficient factor=0.95 when the validation loss does not decrease for a preset number of consecutive training iterations.

3. The method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model according to claim 1, characterized in that, The original wind field data is super-resolution processed using the ERA5 downscaling model to generate high-resolution wind field data for the target sea area. Specifically: The original wind field data of ERA5 is input into the deep learning model, and the original wind field data of ERA5 is magnified to the target high resolution size through nearest neighbor upsampling to generate the basic high resolution skeleton. High-resolution wind field data are stitched together to generate a residual map. The residual map is then processed using a predefined block mask, retaining only the residuals within the block mask. The residual map after block masking is added to the basic high-resolution skeleton to generate high-resolution wind field data.

4. The method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model according to claim 1, characterized in that, The process of estimating wind resources and wind power generation of deep-sea wind farms based on corrected high-resolution virtual meteorological tower wind field data includes the following steps: Turbulence intensity is calculated based on the corrected high-resolution virtual wind field data, and wind speed ranges and wind speed frequencies of the high-resolution virtual wind field data are statistically analyzed. The optimal shape parameters are obtained by fitting the wind speed frequency distribution to a Weibull curve. k and dimensional parameters c ; Calculate the pre-installed wheel hub height, the maximum gust wind speed that occurs once every 50 years. v 50,gust ,according to v 50,gust Select the fan model based on the turbulence intensity; The power generation of the deep-sea wind farm is calculated based on the corrected high-resolution virtual wind measurement tower wind field data, wind turbine model, and wind turbine operating time range.

5. The method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model according to claim 4, characterized in that, The expression for fitting the wind speed frequency distribution using a Weibull curve. for: ； in, v Wind speed; k For shape parameters; c These are dimensional parameters.

6. The method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model according to claim 4, characterized in that, The estimated pre-installed wheel hub height has a 50-year return period for the maximum gust wind speed. v 50,gust Includes the following steps: The annual maximum wind speed sequence at the virtual wind measurement tower over the past 30 years was statistically analyzed. The maximum wind speed with a 50-year return period was calculated using Gumbel type I and Pearson type III frequency distribution functions. Based on the wind shear index, the maximum wind speed with a 50-year return period at the height of the pre-installed wind turbine hub was estimated. v 50 ; Calculate the gust coefficient C It can be expressed as a formula as follows: ； in, n For frequency parameters; T Time interval; Calculate the maximum gust speed with a 50-year return period. v 50,gust It can be expressed as a formula as follows: ； in, σ 50 This represents the standard deviation of the annual maximum wind speed sequence over the past 30 years.

7. The method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model according to claim 4, characterized in that, The wind farm's power generation E th The calculation formula is: ； in, m This refers to the number of generator units in the wind farm. v 1 represents the cut-in wind speed of the wind turbine generator; v 2 represents the cut-out wind speed of the wind turbine generator set; p j ( v ) is the first j The typhoon wind turbine generator is operating at a wind speed of v Power generation at that time; For the first j The Weibull distribution obtained by fitting the wind speed probability distribution of the typhoon wind turbine generator.

8. The method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model according to claim 1, characterized in that, The spatial resolution of the high-resolution wind field data is 0.0625°, and the actual distance scale is 7km.

9. The method for evaluating the power generation of deep-sea wind farms based on the ERA5 downscaling model according to claim 1, characterized in that, The fidelity constraint applied to the high-resolution wind field data specifically includes: High-resolution wind field data is downsampled to restore low-resolution wind field data, and compared with the original wind field data to see if the wind field data at the same latitude and longitude are consistent. If they are inconsistent, the high-resolution wind field data is corrected to ensure that the downsampled low-resolution wind field data is consistent with the original wind field data.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for performing a method for assessing the power generation of deep-sea wind farms based on the ERA5 downscaling model as described in any one of claims 1 to 9.

Images