Polar region wind driven generator capacity and wind speed parameter optimization method based on data driving

Abstract

The invention relates to a wind driven generator capacity and wind speed parameter optimization method, in particular to a polar region wind driven generator capacity and wind speed parameter optimization method based on data driving, which comprises the following steps of: learning statistical characteristics from polar region historical wind power data by adopting an improved generative adversarial network; and generating a wind power output scene with similar statistical characteristics with the actual data of the polar region. Secondly, according to polar region wind energy characteristics, the expected utilization rate and research and development and operation and maintenance cost of a fan are calculated, and a polar region wind driven generator capacity and wind speed parameter optimization model based on data driving is constructed; and finally, solving the model by using a fast non-dominated sorting algorithm to realize multi-objective optimization, thereby improving the efficiency and precision of parameter optimization. According to the method, a large number of scenes conforming to polar region wind power resource characteristics can be generated through a small amount of data, the authenticity and diversity of scene generation are improved, the method is suitable for fan capacity and wind speed parameter optimization in different environments, and reasonable control over the accuracy and economical efficiency of results is guaranteed.

Description

Technical Field

[0001] The present invention relates to a method for optimizing the capacity and wind speed parameters of a wind turbine generator, and specifically to a method for optimizing the capacity and wind speed parameters of a polar wind turbine generator based on data-driven. Background Art

[0002] At present, all countries are vigorously developing new energy in polar scientific research, and using new energy as the main energy supply method for polar scientific research stations to reduce carbon emissions in polar regions and protect the polar natural environment. Due to the polar day and night phenomenon, photovoltaic power generation alone cannot guarantee the electricity supply of scientific research stations throughout the year, so polar wind power generation has received widespread attention. Since the polar regions are currently under development, it may not be possible to obtain accurate historical wind speeds in new wind turbine areas. It is necessary to generate wind speed data for wind turbine installation locations based on wind speeds in neighboring areas. In addition, there are many extreme weather conditions in the polar regions, and wind speeds often show extreme changes (too large or too small). Although traditional wind speed probability density functions (such as Weibull distribution) are often used for wind turbine parameter estimation, they are difficult to reflect the complex wind speed characteristics of the polar regions and may mislead wind turbine capacity and wind speed parameters.

[0003] The extreme weather environment in the polar regions makes it difficult for existing inland or offshore wind turbines to adapt to the unique wind resource characteristics of the polar regions. The incompatibility of wind turbine parameters not only affects the construction cost of wind turbines, but also directly affects the utilization rate of wind turbines. Therefore, it is urgent to carry out the research and development of special wind turbines suitable for polar environments. In the process of wind turbine research and development, the improvement of capacity and wind speed parameters will significantly increase the research and development costs. Too small a range of cut-in and cut-out wind speeds may lead to too low a utilization rate of wind turbines, while increasing the cut-in and cut-out wind speed range will lead to high research and development costs of wind turbines. In the process of determining wind turbine parameters, it is necessary to consider the matching of wind turbine utilization and parameters so that both wind turbine cost and performance are optimized. The wind turbine parameter design of polar wind farms needs to fully consider the impact of the probability distribution of wind speed on the expected utilization rate of wind turbines, comprehensively consider extreme wind speed conditions and insufficient data, and adopt a more flexible and comprehensive design method based on typical scenarios and probability distribution characteristics of wind speed to ensure the reliability and adaptability of wind turbine capacity and wind speed-related parameter solutions, so as to achieve a balance between research and development operation and maintenance costs and power generation efficiency. Summary of the invention

[0004] The traditional wind speed probability density function is difficult to reflect the complex wind speed characteristics of the polar regions, which may mislead the estimation of wind turbine capacity and wind speed parameters. In view of this problem, the present invention proposes a data-driven polar wind turbine capacity and wind speed parameter optimization method. First, an improved generative adversarial network (GAN) is used to generate typical wind speed scenarios, which overcomes the problem of insufficient historical wind speed data in polar regions, avoids the limitations of traditional single probability distribution models, and can directly generate a large number of scenarios similar to the statistical characteristics of polar wind speed. Secondly, in view of the extreme characteristics of polar wind resources, the expected utilization rate of wind turbines and the R&D and operation and maintenance costs are calculated respectively, and a data-driven comprehensive wind turbine capacity and wind speed parameter optimization model is proposed. Finally, the model is solved using a fast non-dominated sorting algorithm to optimize the efficiency and accuracy of the wind turbine parameter design to ensure that the developed wind turbine achieves optimal performance in extreme environments such as the polar regions.

[0005] The technical solution adopted by the present invention is as follows: a data-driven polar wind turbine capacity and wind speed parameter optimization method, comprising the following steps:

[0006] Step 1: Construct a wind power output scenario generation method based on an improved generative adversarial network. The objective function of the improved generative adversarial network is as follows: The improved generative adversarial network learns the historical wind speed data of the polar adjacent areas for training. After the training is completed, the generator G inputs the noise z and obtains V1=G(z); where: E is the mathematical expectation function under a certain probability distribution, x represents historical data, z represents noise, G(z) represents generated data, x′=εx+(1-ε)G(z), ε~U[0,1], indicating that it obeys the standard normal distribution with a mean of 0 and a variance of 1, λ represents the weight coefficient of the gradient penalty term; ||g||2 represents the 2-norm, P data is the probability distribution of historical data, P Z is the probability distribution of noise z; Represents gradient calculation;

[0007] Step 2: Build a data-driven wind turbine capacity and wind speed parameter optimization model:

[0008] 1) Set the objective function F1, Where F1 is the expected fan utilization factor, P W is the output power, P r is the rated power, V co is the cut-out wind speed; V ci is the cut-in wind speed; T is the total duration of the whole year; f(V) is the wind speed probability function, P(V) is the output characteristic of the wind turbine, and the calculation formula is as follows:

[0009] Where V r is the rated wind speed, and the wind speed V at the observation point is converted to the wind speed at the fan impeller hub: Where: h is the height of the fan impeller hub from the ground; V1 is the wind speed at the wind speed observation point with a ground height of h1, which is generated based on the improved generative adversarial network; b is the exponential parameter;

[0010] 2) Set the objective function F2, where F2 is the R&D and operation and maintenance cost of the wind turbine. Where i is the interest rate; O and M are the unit power operation and maintenance costs; n is the service life, and I is the total R&D cost. In the formula, λ P With λ V are the R&D cost coefficients related to capacity and wind speed, respectively; k is the R&D capacity coefficient; α and β are the exponential growth coefficients related to capacity and wind speed, respectively;

[0011] Step 3: Use the fast non-dominated sorting algorithm to solve the dual-objective model in step 2. After obtaining the Pareto frontier, use the fuzzy membership function to classify the Pareto frontier of the dual-objective model. The calculation formula of the fuzzy membership function is shown as follows: Where: α i,j is the decision vector of the i-th group {P r 、V co 、V ci 、V r} is the optimal solution of the jth objective function; α j max is the maximum value of the jth objective function on the Pareto frontier; α j min is the minimum value of the jth objective function on the Pareto frontier; θ i,j is α i,j Corresponding membership function value; after solving the fuzzy membership function of each group of decision vectors in the Pareto frontier, use Determine the optimal solution, where: N j is the number of objective functions in the Pareto frontier, N j =2; α best，j is the objective function corresponding to the optimal decision vector.

[0012] In the above data-driven polar wind turbine capacity and wind speed parameter optimization method, the constraint condition followed by the objective function in step 2 is: In the formula, are the minimum and maximum values ​​of the rated power of the fan respectively; is the minimum value of historical wind speed data; are the maximum values ​​of historical wind speed data respectively.

[0013] In summary, compared with the prior art, the beneficial effects of the present disclosure are:

[0014] The present invention proposes a data-driven polar wind turbine capacity and wind speed parameter optimization method. First, an improved generative adversarial network (GAN) is used to learn statistical features from polar historical wind power data, and a large number of wind power output scenarios with similar statistical features to polar actual data are directly generated, avoiding the limitation of relying on model assumptions or prior distributions. Secondly, according to the characteristics of polar wind energy, the expected utilization rate of wind turbines and the R&D and operation and maintenance costs are calculated respectively, and a data-driven polar wind turbine capacity and wind speed parameter optimization model is constructed. Finally, the model is solved using a fast non-dominated sorting algorithm to achieve multi-objective optimization, thereby improving the efficiency and accuracy of parameter optimization. This method can generate a large number of scenarios that meet the characteristics of polar wind resources using a small amount of data, improve the authenticity and diversity of scenario generation, and is not only suitable for the optimization of wind turbine capacity and wind speed parameters in different environments, but also ensures the accuracy and economy of the results. Reasonable control. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is an implementation block diagram of the present invention. DETAILED DESCRIPTION

[0016] A data-driven polar wind turbine capacity and wind speed parameter optimization method comprises the following steps:

[0017] Step 1: Construct a wind power output scenario generation method based on an improved generative adversarial network (GAN).

[0018] 1) In order to solve the problem of vanishing gradients in GAN during training, Wasserstein distance is introduced to replace the JS divergence of GAN. The definition of Wasserstein distance is as follows:

[0019]

[0020] Where: P data , P G are the probability distributions of historical data and generated data respectively, sup represents the supremum, θ d is the weight parameter of the neural network of the discriminator D, E is the mathematical expectation function under a certain probability distribution, x represents historical data, z represents noise, G(z) represents generated data, P Z is the probability distribution of noise z.

[0021] 2) Introducing gradient penalty to enhance Lipschitz continuity makes the model converge faster. Its expression is:

[0022]

[0023] Where: x′=εx+(1-ε)G(z), ε~U[0,1], indicating that it obeys the standard normal distribution with mean 0 and variance 1; λ represents the weight coefficient of the gradient penalty term; ||g||2 represents the 2-norm; Represents gradient calculation.

[0024] 3) The goal of GAN is to model the distribution of a given training dataset. The generator maps the noise to the corresponding generated sample G(z). Then, the discriminator D distinguishes the real samples from the generated samples. The objective function of the improved generative adversarial network is as follows:

[0025]

[0026] The improved generative adversarial network learns the historical wind speed data of adjacent areas for training. After the training is completed, the generator G inputs the noise z and obtains G(z)=V1.

[0027] Step 2: Construct a data-driven wind turbine capacity and wind speed parameter optimization model.

[0028] 1) Objective function F1: The expected utilization rate of the wind turbine is the ratio of the expected power generation of the wind turbine in a period of time to the rated power generation in this period of time, which is also the ratio of the expected output power to the rated power:

[0029]

[0030]

[0031] Where F1 is the expected fan utilization factor, P W is the output power, P r is the rated power, E(P W ) is the expected output power, V co is the cut-out wind speed; V ci is the cut-in wind speed; T is the total duration of the whole year; f(V) is the wind speed probability function. The present invention is based on an improved generative adversarial network to learn the historical wind speed data of adjacent areas and generate a large number of scenarios to solve the above formula.

[0032] P(V) is the output characteristic of the wind turbine, and the calculation formula is as follows:

[0033]

[0034] Where V r is the rated wind speed.

[0035] In addition, the wind speed V at the observation point needs to be converted into the wind speed at the fan impeller hub:

[0036] Where: h is the height of the fan impeller hub from the ground; V1 is the wind speed at the wind speed observation point with a ground height of h1, which is generated based on the improved generative adversarial network; b is an exponential parameter, which depends on the atmospheric stability and ground roughness, and its value is 0.5~0.125.

[0037] 2) Objective function F2: Wind turbine cost

[0038] The cost of wind turbines includes R&D costs and operation and maintenance costs:

[0039]

[0040] Where i is the interest rate; O and M are the unit power operation and maintenance costs; and n is the useful life.

[0041] I is the total R&D cost:

[0042] In the formula, λ P With λ V are the R&D cost coefficients related to capacity and wind speed respectively; k is the R&D capability coefficient; α and β are the exponential growth coefficients related to capacity and wind speed respectively.

[0043] 3) Constraints

[0044]

[0045] In the formula, are the minimum and maximum values ​​of the rated power of the fan respectively; is the minimum value of historical wind speed data; are the maximum values ​​of historical wind speed data respectively.

[0046] Step 3: Use a fast non-dominated sorting algorithm (NSGA-Ⅱ) to solve the dual-objective model in step 2. After obtaining the Pareto frontier, use the fuzzy membership function to classify the Pareto frontier of the dual-objective model. The solution with the highest comprehensive membership function value is the optimal solution of the dual-objective model. The calculation formula of the fuzzy membership function is shown as follows:

[0047]

[0048] Where: α i,j is the decision vector of the i-th group {P r 、V co 、V ci 、V r} is the optimal solution of the jth objective function; α j maxis the maximum value of the jth objective function on the Pareto frontier; α j min is the minimum value of the jth objective function on the Pareto frontier; θ i,j is α i,j Corresponding membership function value.

[0049] After solving the membership function of each group of decision vectors in the Pareto frontier, the optimal solution is determined using the following formula:

[0050]

[0051] Where: N j is the number of objective functions in the Pareto frontier, N j =2; α best，j is the objective function corresponding to the optimal decision vector.

Claims

1. A data-driven polar wind turbine capacity and wind speed parameter optimization method, characterized in that: The following steps are involved: Step 1: Construct a wind power output scenario generation method based on an improved generative adversarial network. The objective function of the improved generative adversarial network is as follows: The improved generative adversarial network learns the historical wind speed data of the polar adjacent areas for training. After the training is completed, the generator G inputs the noise z and obtains V1=G(z); where: E is the mathematical expectation function under a certain probability distribution, x represents historical data, z represents noise, G(z) represents generated data, x′=εx+(1-ε)G(z), ε~U[0,1], λ represents the weight coefficient of the gradient penalty term; ||g||2 represents the 2-norm, P data is the probability distribution of historical data, P Z is the probability distribution of noise z; Represents gradient calculation; Step 2: Build a data-driven wind turbine capacity and wind speed parameter optimization model: 1) Set the objective function F1, Where F1 is the expected fan utilization factor, P W is the output power, P r is the rated power, V co is the cut-out wind speed; V ci is the cut-in wind speed; T is the total duration of the whole year; f(V) is the wind speed probability function, P(V) is the output characteristic of the wind turbine, and the calculation formula is as follows: Where V r is the rated wind speed, and the wind speed V at the observation point is converted to the wind speed at the fan impeller hub: Where: h is the height of the fan impeller hub from the ground; V1 is the wind speed at the wind speed observation point with a ground height of h1, which is generated based on the improved generative adversarial network; b is the exponential parameter; 2) Set the objective function F2, where F2 is the R&D and operation and maintenance cost of the wind turbine. Where i is the interest rate; O and M are the unit power operation and maintenance costs; n is the service life, and I is the total R&D cost. In the formula, λ P With λ V are the R&D cost coefficients related to capacity and wind speed, respectively; k is the R&D capacity coefficient; Δ and β are the exponential growth coefficients related to capacity and wind speed, respectively; Step 3: Use the fast non-dominated sorting algorithm to solve the dual-objective model in step 2. After obtaining the Pareto frontier, use the fuzzy membership function to classify the Pareto frontier of the dual-objective model. The calculation formula of the fuzzy membership function is shown as follows: Where: α i,j is the decision vector of the i-th group {P r 、V co 、V ci 、V r } is the optimal solution of the jth objective function; α j max is the maximum value of the jth objective function on the Pareto frontier; α j min is the minimum value of the jth objective function on the Pareto frontier; θ i,j is α i,j Corresponding membership function value; after solving the fuzzy membership function of each group of decision vectors in the Pareto frontier, use Determine the optimal solution, where: N j is the number of objective functions in the Pareto frontier, N j =2; α best，j is the objective function corresponding to the optimal decision vector.

2. The data-driven polar wind turbine capacity and wind speed parameter optimization method according to claim 1, characterized in that: The constraints followed by the objective function in step 2 are In the formula, are the minimum and maximum values ​​of the rated power of the fan respectively; is the minimum value of historical wind speed data; are the maximum values ​​of historical wind speed data respectively.

Images