Method and device for processing loss of icing power of fan

Abstract

The invention discloses a fan icing power loss processing method and device. The fan icing power loss processing method comprises the steps that meteorological environment parameters and icing time are obtained; preprocessing the meteorological environment parameters and the icing time to obtain target meteorological environment parameters and target icing time; inputting the target meteorological environment parameters and the target icing time into a pre-trained multi-parameter dynamic prediction model, and predicting to obtain loss probability distribution of the icing fan; wherein the multi-parameter dynamic prediction model is constructed by Gaussian process regression; gaussian process regression is obtained by determining a mean value function and a covariance function; the multi-parameter dynamic prediction model is obtained by training historical meteorological environment parameters and historical icing time in advance; and determining deicing measures according to the loss probability distribution. Therefore, the timeliness and accuracy of fan loss prediction can be effectively improved through the multi-parameter dynamic prediction model, and early warning can be carried out in time according to loss probability distribution.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a wind turbine icing power loss processing method and device. BACKGROUND

[0002] Wind power generation, as a clean and renewable energy source, has been widely used around the world. However, the operation efficiency and power generation capacity of wind turbine generators are affected by various factors, especially environmental factors (such as temperature, wind speed, etc.) and equipment status (such as icing, wear, etc.). Among them, the power loss of the wind turbine is one of the key factors affecting the efficiency of wind power generation. How to accurately predict and evaluate the power loss of the wind turbine is crucial for improving the reliability and economy of the wind power generation system.

[0003] In the existing technology, the linear modeling method is mainly used, that is, the average wind speed data and environmental temperature data are obtained through the SCADA system, the wind speed data is then processed, the temperature data and the wind speed data are time-aligned by using the nearest neighbor interpolation method, the manually input icing level is converted into a fixed power loss coefficient, and finally the final power is calculated according to the power loss coefficient, the temperature data and the wind speed data.

[0004] Since the linear modeling method lacks an accurate early warning mechanism, the de-icing system is often started only after the power has decreased significantly, which causes additional energy waste and equipment wear and tear, and lacks the timeliness of power prediction. SUMMARY

[0005] Based on the above shortcomings of the prior art, the present application provides a wind turbine icing power loss processing method and device to solve the problem of lack of timeliness of power prediction in the prior art.

[0006] In order to achieve the above purpose, the present application provides the following technical solutions:

[0007] The first aspect of the present application provides a wind turbine icing power loss processing method, comprising:

[0008] obtaining meteorological environment parameters and icing time;

[0009] preprocessing the meteorological environment parameters and the icing time to obtain target meteorological environment parameters and target icing time;

[0010] input the target meteorological environment parameter and the target icing time into a pre-trained multi-parameter dynamic prediction model to predict a loss probability distribution of the icing wind turbine; wherein the multi-parameter dynamic prediction model is constructed by a Gaussian process regression; the Gaussian process regression is determined by a mean function and a covariance function; the multi-parameter dynamic prediction model is pre-trained by historical meteorological environment parameters and historical icing times;

[0011] According to the loss probability distribution, determine the de-icing measures.

[0012] Optionally, in the wind turbine icing power loss processing method described above, the training method of the multi-parameter dynamic prediction model comprises:

[0013] Construct a Gaussian process regression based on the mean function and the covariance function, and take the Gaussian process regression as the multi-parameter dynamic prediction model;

[0014] Obtain historical meteorological environment parameters and historical icing times;

[0015] Standardize the historical meteorological environment parameters to obtain standard meteorological environment parameters, and perform logarithmic transformation on the historical icing times to obtain logarithmic icing times;

[0016] Input the standard meteorological environment parameters and the logarithmic icing times into the mean function contained in the multi-parameter dynamic prediction model to obtain a predicted value expectation of the icing wind turbine;

[0017] Input the standard meteorological environment parameters and the logarithmic icing times into the covariance function contained in the multi-parameter dynamic prediction model to obtain the covariance of the icing wind turbine, and integrate the predicted value expectation and the covariance to obtain a sample loss probability distribution of the icing wind turbine;

[0018] Determine whether the sample loss probability distribution meets the requirement of true value;

[0019] If the sample loss probability distribution meets the requirement of true value, determine the multi-parameter dynamic prediction model as a trained multi-parameter dynamic prediction model;

[0020] If the sample loss probability distribution does not meet the requirement of true value, determine an optimized hyperparameter according to the sample loss probability distribution;

[0021] Adjust the multi-parameter dynamic prediction model according to the optimized hyperparameter, and return to execute the inputting of the standard meteorological environment parameters and the logarithmic icing times into the mean function contained in the multi-parameter dynamic prediction model to obtain the predicted value expectation of the icing wind turbine.

[0022] Optionally, in the wind turbine icing power loss processing method, the inputting the standard meteorological environment parameter and the logarithmic icing time into a covariance function included in the multi-parameter dynamic model to obtain the covariance of the icing wind turbine comprises:

[0023] obtaining signal variance, icing saturation time and noise standard deviation;

[0024] calculating a meteorological parameter kernel based on the signal variance and the standard meteorological environment parameter;

[0025] calculating a time kernel based on the icing saturation time and the logarithmic icing time;

[0026] calculating a noise kernel based on the noise standard deviation by using a Kronecker function;

[0027] calculating the covariance of the icing wind turbine based on the meteorological parameter kernel, the time kernel and the noise kernel by using the covariance function included in the multi-parameter dynamic model.

[0028] Optionally, in the wind turbine icing power loss processing method, the determining the optimized hyperparameter based on the sample loss probability distribution comprises:

[0029] constructing a hyperparameter and initializing the hyperparameter to obtain an initial hyperparameter;

[0030] calculating a log marginal likelihood function based on the sample loss probability distribution and the covariance;

[0031] calculating a gradient vector based on the log marginal likelihood function by using an optimization algorithm;

[0032] calculating an optimized hyperparameter based on the initial hyperparameter and the gradient vector;

[0033] determining whether the optimized hyperparameter is less than a preset threshold;

[0034] if the optimized hyperparameter is less than the preset threshold, determining the optimized hyperparameter as the optimized hyperparameter;

[0035] if the optimized hyperparameter is not less than the preset threshold, taking the optimized hyperparameter as the initial hyperparameter and returning to execute the calculation of the optimized hyperparameter based on the initial hyperparameter and the gradient vector.

[0036] Optionally, in the wind turbine icing power loss processing method, the method further comprises:

[0037] obtaining a test group, and performing standardization processing on the test group to obtain a standard test group;

[0038] inputting the standard test set into a pre-trained multi-parameter dynamic prediction model to obtain a test loss probability distribution of the icing fan;

[0039] judging whether the multi-parameter dynamic prediction model meets the accuracy requirement based on the test loss probability distribution and the actual power loss;

[0040] if the multi-parameter dynamic prediction model meets the accuracy requirement, determining that the multi-parameter dynamic prediction model is feasible;

[0041] if the multi-parameter dynamic prediction model does not meet the accuracy requirement, determining that the multi-parameter dynamic prediction model is not feasible.

[0042] Optionally, in the method for processing the power loss of the icing fan, the judging whether the multi-parameter dynamic prediction model meets the accuracy requirement based on the test loss probability distribution and the actual power loss comprises:

[0043] calculating the average deviation percentage and the root mean square error between the test loss probability distribution and the actual power loss;

[0044] judging whether the average deviation percentage and the root mean square error are both less than a preset threshold;

[0045] if the average deviation percentage and the root mean square error are both less than the preset threshold, determining that the multi-parameter dynamic prediction model meets the accuracy requirement;

[0046] if the average deviation percentage and the root mean square error are both not less than the preset threshold, determining that the multi-parameter dynamic prediction model does not meet the accuracy requirement.

[0047] The second aspect of the present application provides a device for processing the power loss of the icing fan, comprising:

[0048] a data acquisition unit configured to acquire meteorological environmental parameters and icing time;

[0049] a preprocessing unit configured to preprocess the meteorological environmental parameters and the icing time to obtain target meteorological environmental parameters and target icing time;

[0050] a prediction unit configured to input the target meteorological environmental parameters and the target icing time into a pre-trained multi-parameter dynamic prediction model to obtain a loss probability distribution of the icing fan; wherein the multi-parameter dynamic prediction model is constructed by a Gaussian process regression; the Gaussian process regression is determined by a mean function and a covariance function; and the multi-parameter dynamic prediction model is trained by historical meteorological environmental parameters and historical icing time in advance;

[0051] A measure determination unit is configured to determine an ice removal measure according to the loss probability distribution.

[0052] Optionally, in the ice-coated fan power loss processing device, further comprising:

[0053] A construction unit is configured to construct a Gaussian process regression based on the mean function and the covariance function, and use the Gaussian process regression as the multi-parameter dynamic prediction model.

[0054] A first acquisition unit is configured to acquire historical meteorological environment parameters and historical icing times.

[0055] A data processing unit is configured to perform standardization processing on the historical meteorological environment parameters to obtain standard meteorological environment parameters, and perform logarithmic transformation processing on the historical icing times to obtain logarithmic icing times.

[0056] A first input unit is configured to input the standard meteorological environment parameters and the logarithmic icing times into the mean function included in the multi-parameter dynamic prediction model to obtain a predicted value expectation of an icing fan.

[0057] A second input unit is configured to input the standard meteorological environment parameters and the logarithmic icing times into the covariance function included in the multi-parameter dynamic prediction model to obtain a covariance of the icing fan, and integrate the predicted value expectation and the covariance to obtain a sample loss probability distribution of the icing fan.

[0058] A requirement judgment unit is configured to judge whether the sample loss probability distribution meets a requirement of a true value.

[0059] A model determination unit is configured to, if the sample loss probability distribution meets the requirement of the true value, determine the multi-parameter dynamic prediction model as a trained multi-parameter dynamic prediction model.

[0060] A parameter determination unit is configured to, if the sample loss probability distribution does not meet the requirement of the true value, determine an optimized hyperparameter according to the sample loss probability distribution.

[0061] An adjustment unit is configured to adjust the multi-parameter dynamic prediction model according to the optimized hyperparameter, and return to perform the inputting of the standard meteorological environment parameters and the logarithmic icing times into the mean function included in the multi-parameter dynamic prediction model to obtain the predicted value expectation of the icing fan.

[0062] Optionally, in the ice-coated fan power loss processing device, the second input unit includes:

[0063] A second acquisition unit is configured to acquire a signal variance, an icing saturation time, and a noise standard deviation.

[0064] a first calculation unit configured to calculate a weather parameter kernel based on the signal variance and the standard weather environment parameter;

[0065] a second calculation unit configured to calculate a time kernel according to the icing saturation time and the logarithmic icing time;

[0066] a third calculation unit configured to calculate a noise kernel based on the noise standard deviation by using a Kronecker function;

[0067] a fourth calculation unit configured to calculate the covariance of the icing fan according to the weather parameter kernel, the time kernel and the noise kernel by using a covariance function of the multi-parameter dynamic inclusion.

[0068] Optionally, in the fan icing power loss processing device, the parameter determination unit comprises:

[0069] an initialization unit configured to construct a hyperparameter and initialize the hyperparameter to obtain an initial hyperparameter;

[0070] a fifth calculation unit configured to calculate a log marginal likelihood function according to the sample loss probability distribution and the covariance;

[0071] a sixth calculation unit configured to calculate a gradient vector by using an optimization algorithm according to the log marginal likelihood function;

[0072] a seventh calculation unit configured to calculate an optimized hyperparameter based on the initial hyperparameter and the gradient vector;

[0073] a threshold judgment unit configured to judge whether the optimized hyperparameter is less than a preset threshold;

[0074] a parameter determination subunit configured to determine the optimized hyperparameter as an optimized hyperparameter if the optimized hyperparameter is less than the preset threshold;

[0075] a unit configured to take the optimized hyperparameter as the initial hyperparameter and return to execute the calculation of the optimized hyperparameter based on the initial hyperparameter and the gradient vector if the optimized hyperparameter is not less than the preset threshold.

[0076] Optionally, in the fan icing power loss processing device, the parameter determination unit comprises:

[0077] a third acquisition unit configured to acquire a test group, and perform standardization processing on the test group to obtain a standard test group;

[0078] a third input unit configured to input the standard test group into a pre-trained multi-parameter dynamic prediction model to predict a test loss probability distribution of the icing fan;

[0079] The first judging unit is configured to judge whether the multi-parameter dynamic prediction model meets the accuracy requirement based on the test loss probability distribution and the actual power loss.

[0080] The first determining unit is configured to determine that the multi-parameter dynamic prediction model is feasible if the multi-parameter dynamic prediction model meets the accuracy requirement.

[0081] The second determining unit is configured to determine that the multi-parameter dynamic prediction model is not feasible if the multi-parameter dynamic prediction model does not meet the accuracy requirement.

[0082] Optionally, in the device for processing the loss of the icing power of the fan, the first judging unit comprises:

[0083] The seventh calculating unit is configured to calculate the average deviation percentage and the root mean square error between the test loss probability distribution and the actual power loss.

[0084] The second judging unit is configured to judge whether the average deviation percentage and the root mean square error are both less than a preset threshold.

[0085] The third determining unit is configured to determine that the multi-parameter dynamic prediction model meets the accuracy requirement if the average deviation percentage and the root mean square error are both less than the preset threshold.

[0086] The fourth determining unit is configured to determine that the multi-parameter dynamic prediction model does not meet the accuracy requirement if the average deviation percentage and the root mean square error are both not less than the preset threshold.

[0087] The method for processing the loss of the icing power of the fan provided in the present application comprises the following steps: obtaining meteorological environment parameters and icing time, pre-processing the meteorological environment parameters and the icing time to obtain target meteorological environment parameters and target icing time, inputting the target meteorological environment parameters and the target icing time into a pre-trained multi-parameter dynamic prediction model to predict a loss probability distribution of the icing fan, wherein the multi-parameter dynamic prediction model is constructed by a Gaussian process regression, the Gaussian process regression is determined by a mean function and a covariance function, the multi-parameter dynamic prediction model is trained by historical meteorological environment parameters and historical icing time in advance, and finally, a de-icing measure is determined according to the loss probability distribution. Thus, the multi-parameter dynamic prediction model can effectively improve the timeliness and accuracy of the loss prediction of the fan, and can also timely perform early warning according to the loss probability distribution. BRIEF DESCRIPTION OF DRAWINGS

[0088] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced. Obviously, the accompanying drawings in the following description only need to be used to explain the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0089] Figure 1 A flowchart of a wind turbine icing power loss processing method provided by an embodiment of the present application is shown in the figure.

[0090] Figure 2 A flowchart of a training method of a multi-parameter dynamic prediction model provided by another embodiment of the present application is shown in the figure.

[0091] Figure 3 A flowchart of a covariance calculation method provided by another embodiment of the present application is shown in the figure.

[0092] Figure 4 A flowchart of a verification method of a multi-parameter dynamic prediction model provided by another embodiment of the present application is shown in the figure.

[0093] Figure 5 A flowchart of another verification method of a multi-parameter dynamic prediction model provided by another embodiment of the present application is shown in the figure.

[0094] Figure 6 A structural diagram of a comparison of a power loss prediction result and a true value and a relative error provided by another embodiment of the present application is shown in the figure.

[0095] Figure 7 A flowchart of an acquisition method of an optimized hyperparameter provided by another embodiment of the present application is shown in the figure.

[0096] Figure 8 A structural diagram of a wind turbine icing power loss processing device provided by another embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0097] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0098] In this application, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitation, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article or equipment including the element.

[0099] The embodiment of the present application provides a loss processing method of icing power of a fan, as shown in the method, the method specifically comprises the following steps: Figure 1 The method specifically comprises the following steps:

[0100] S101, acquiring meteorological environment parameters and icing time.

[0101] The meteorological environment parameters can include wind speed (v), temperature (T), humidity (H) and liquid water content (LWC) and the like.

[0102] Specifically, the meteorological environment parameters and the icing time can be acquired from a meteorological station, so that the subsequent wind farm can monitor meteorological conditions and fan states in real time, and enable deicing measures or adjust the operation mode of the fan according to the predicted loss power, and ensure efficient operation of the fan.

[0103] S102, preprocessing the meteorological environment parameters and the icing time to obtain target meteorological environment parameters and target icing time.

[0104] It can be understood that, in order to improve the quality, accuracy and consistency of data, and provide effective support for subsequent model prediction and decision-making, it is necessary to pre-process the meteorological environment parameters and the icing time. The preprocessing means can include standardization processing means.

[0105] S103, inputting the target meteorological environment parameters and the target icing time into a pre-trained multi-parameter dynamic prediction model to predict a loss probability distribution of an icing fan.

[0106] The multi-parameter dynamic prediction model is obtained by Gaussian process regression, the Gaussian process regression is determined by a mean function and a covariance function, and the multi-parameter dynamic prediction model is trained by using historical meteorological environment parameters and historical icing time in advance.

[0107] Specifically, the multi-parameter dynamic prediction model provided in the embodiments of the present application establishes a nonlinear mapping relationship between wind speed (v), temperature (T), humidity (H), liquid water content (LWC), icing time (Δt) and power loss (Ploss), which is suitable for accurate prediction of power loss and active deicing control under icing conditions of wind farms in cold regions. Therefore, by inputting the target meteorological environmental parameters and the target icing time, the predicted power loss can be obtained.

[0108] Optionally, the embodiments of the present application provide a training method of a multi-parameter dynamic prediction model, as shown in Figure 2 The method comprises the following steps:

[0109] S201, constructing a Gaussian process regression based on a mean function and a covariance function, and taking the Gaussian process regression as a multi-parameter dynamic prediction model.

[0110] It can be understood that the Gaussian process regression (GPR) is a set of an infinite number of random variables, any finite variables of which obey a joint Gaussian distribution. It is completely determined by the mean function m(x) and the covariance function (kernel function) k(x, x'), so the expression of the Gaussian process regression is f(x) ~ gp(m(x), k(x, x')), and the multi-parameter dynamic prediction model is constructed based on the Gaussian process regression.

[0111] S202, obtaining historical meteorological environmental parameters and historical icing time.

[0112] Specifically, the historical meteorological environmental parameters and the historical icing time are selected from the measured data of the icing of a 2WM wind turbine of a certain wind farm, which includes wind speed v (3-15 m / s), temperature T (-10-0℃), humidity H (70-100%), liquid water content LWC (0.05-0.3 g / m³), and icing time Δt (0-6 h), a total of 60 groups of data. Among them, 50 groups of data are obtained as a sample set for training of the multi-parameter dynamic prediction model, and the remaining 10 groups are used for subsequent feasibility determination of the multi-parameter dynamic prediction model. Therefore, the historical meteorological environmental parameters and the historical icing time obtained here are 50 groups of data.

[0113] S203, standardizing the historical meteorological environmental parameters to obtain standard meteorological environmental parameters, and logarithmically transforming the historical icing time to obtain a logarithmic icing time.

[0114] Specifically, the historical meteorological environmental parameters in the sample set are respectively subjected to Z-score standardization, and the expression formula of Z-score standardization is:

[0115]

[0116] where z denotes the normalized value, x denotes the specific meteorological parameter, denotes the mean value, denotes the standard deviation. The normalized meteorological parameter has a mean value of 0 and a standard deviation of 1, which is more likely to meet the requirement of the input scale consistency of the covariance function (such as the RBF kernel) and avoid some features from dominating the calculation of the kernel function due to the large dimension.

[0117] Then, after normalization, the historical icing time is logarithmically transformed, and the specific expression formula is: where the logarithmically transformed Δt (logarithmic icing time) is used as a subsequent input feature and can participate in covariance calculation with the normalized meteorological parameter on a similar scale, thereby improving the effectiveness of the kernel function.

[0118] S204, input the standard meteorological environment parameter and the logarithmic icing time into the mean function contained in the multi-parameter dynamic prediction model to obtain the expected value of the prediction of the icing wind turbine.

[0119] Specifically, after data preprocessing of the historical meteorological environment parameter and the historical icing time, the mean function m(x) needs to be set to 0 to simplify the multi-parameter dynamic prediction model, so that the standard meteorological environment parameter and the logarithmic icing time fluctuate around the zero mean value after normalization, avoiding the complexity of fitting the mean function.

[0120] S205, input the standard meteorological environment parameter and the logarithmic icing time into the covariance function contained in the multi-parameter dynamic prediction model to obtain the covariance of the icing wind turbine, and integrate the expected value of the prediction and the covariance to obtain the sample loss probability distribution of the icing wind turbine.

[0121] It should be noted that the kernel function (covariance function) is the core component of Gaussian process regression, which determines the flexibility, smoothness and generalization ability of the model.

[0122] It should also be noted that the mean function provides the expectation of the prediction value, the variance function quantifies the uncertainty of the prediction value, and the probability distribution integrates these information together to provide a complete statistical description. Therefore, after obtaining the covariance and the expected value of the prediction, they need to be integrated to obtain the probability distribution, that is, the sample loss probability distribution of the icing wind turbine.

[0123] Specifically, the mean function is first obtained, and the specific expression is:

[0124]

[0125] where, is the predicted power loss of the new input point x*, k * is the covariance vector of the new point and all training points , which reflects the similarity between the new point and the historical data, the higher the similarity, the greater the weight of the corresponding historical data, K represents the kernel matrix, which is calculated by , that is, calculated by the covariance function, y represents the power loss observation value of the sample set, represents the noise variance, and I represents the unit matrix.

[0126] Then, based on the mean function and the covariance, the variance function is calculated,

[0127]

[0128] where the output standard deviation is used to construct the confidence interval (95% interval: )

[0129] Finally, the variance function and the mean function are integrated, and the sample loss probability distribution can be obtained.

[0130]

[0131] where, is the predicted power loss, is the variance function.

[0132] Optionally, in another embodiment of the present application, the standard meteorological environment parameters and the logarithmic icing time are input into the covariance function included in the multi-parameter dynamic prediction model in step S205 to obtain a specific implementation of the covariance of the iced wind turbine, as shown in Figure 3 , which specifically includes the following steps:

[0133] S301, obtaining signal variance, icing saturation time and noise standard deviation.

[0134] Specifically, the initial value of the signal variance is set to 4, the icing saturation time is about 5h, and 1 / 3 of the time length is taken as the initial value, and the noise standard deviation , that is, the quantification of sensor error, simulates the actual measurement error.

[0135] S302, calculating the meteorological parameter kernel based on the signal variance and the standard meteorological environment parameters.

[0136] Specifically, the calculation formula of the meteorological parameter kernel is:

[0137]

[0138] where the initial value of the signal variance is assumed =4, l d: meteorological parameter length scale, controlling the sensitivity of each input dimension, the larger the value, the more local the meteorological parameter affects the power loss (small changes may cause the output to fluctuate dramatically), the smaller the value, the more global the meteorological parameter affects (it needs a larger change to significantly affect the output). In the design of the kernel function, the initial value of the length scale is often set to 1 / 4~1 / 2 of the parameter range, which is an empirical assumption, the purpose is to let the kernel function cover a reasonable sensitive interval in the parameter range. Therefore, the initial value corresponding to each meteorological parameter is: l v = 3, l T = 2.5, l H = 15, l lwc = 0.06, l d x i，d represents the dth meteorological parameter of the ith sample (such as v for wind speed, T for temperature, etc.), x j，d represents the dth meteorological parameter of the jth sample.

[0139] S303, calculate the time kernel according to the icing saturation time and the logarithmic icing time.

[0140] Specifically, the calculation formula of the time kernel (Matern 3 / 2) is:

[0141]

[0142] The reason for choosing Matern kernel is that it is more suitable for non-smooth changes in time dimension than RBF kernel. The icing saturation time is about 5h, and 1 / 3 of it is taken as the initial value of the time length, l t = 1.67, which represents the overall fluctuation range of power loss, △t i represents the logarithmic icing time of the ith sample, △t j represents the logarithmic icing time of the jth sample.

[0143] S304, calculate the noise kernel based on the noise standard deviation using the Kronecker function.

[0144] Specifically, the calculation formula of the noise kernel (White Kernel) is:

[0145] knoise=

[0146] Noise standard deviation , quantification of sensor error, simulating actual measurement error. Assuming that the sensor error is 10% of the signal variance, = 0.3. is the Kronecker function, which is 1 when i=j, otherwise it is 0.

[0147] S305, the covariance of the iced wind turbine is calculated according to the covariance function of the multi-parameter dynamic inclusion, the meteorological parameter kernel, the time kernel and the noise kernel.

[0148] It can be understood that the covariance function adopts a composite kernel structure and designs heterogeneous kernels according to the physical characteristics of different parameters, so that after the meteorological parameter kernel, the time kernel and the noise kernel are obtained, the covariance of the iced wind turbine can be calculated, and the specific calculation formula is:

[0149]

[0150] S206, it is judged whether the sample loss probability distribution meets the true value requirement.

[0151] It should be noted that after obtaining the sample loss probability distribution, in order to know whether the sample loss probability distribution predicted by the multi-parameter dynamic prediction model is accurate, it is necessary to further determine the sample loss probability distribution, that is, whether the sample loss probability distribution meets the true value requirement, wherein the true value requirement is a range of the actual value, so if the sample loss probability distribution meets the true value requirement, it means that the sample loss probability distribution predicted by the parameter dynamic prediction model is accurate, so step S207 is executed at this time. If the sample loss probability distribution does not meet the true value requirement, it means that the sample loss probability distribution predicted by the parameter dynamic prediction model is not accurate enough, so step S208 is executed to optimize and train the multi-parameter dynamic prediction model.

[0152] S207, the multi-parameter dynamic prediction model is determined as a trained multi-parameter dynamic prediction model.

[0153] Alternatively, after the multi-parameter dynamic prediction model is determined as a trained multi-parameter dynamic prediction model, it is necessary to verify whether the multi-parameter dynamic prediction model has certain feasibility, so as to avoid problems such as inaccurate prediction and unstable model in actual application. Therefore, in another embodiment of the present application, a verification method of the multi-parameter dynamic prediction model is also provided, as shown in Figure 4 The method comprises the following steps:

[0154] S401, a test group is obtained, and the test group is standardized to obtain a standard test group.

[0155] Specifically, 10 test groups are obtained and standardized, so that the test groups meet the requirement of the multi-parameter dynamic prediction model for the consistency of the characteristic scales of the input, and avoid that some features dominate the calculation of the kernel function due to too large dimension. The 10 test groups all include historical meteorological environmental parameters and historical icing time.

[0156] S402, input the standard test group into the pre-trained multi-parameter dynamic prediction model to obtain a test loss probability distribution of the iced wind turbine.

[0157] It should be noted that the specific implementation of step S402 can be correspondingly referred to steps S204 and S205 in the above method embodiment, which will not be described here.

[0158] S403, judging whether the multi-parameter dynamic prediction model meets the accuracy requirement based on the test loss probability distribution and the actual power loss.

[0159] It can be understood that, in order to know whether the multi-parameter dynamic prediction model has certain feasibility for the dynamic prediction of the iced wind turbine power loss, the multi-parameter dynamic prediction model can be judged whether it meets the accuracy requirement based on the test loss probability distribution and the actual power loss. If the multi-parameter dynamic prediction model meets the accuracy requirement, it means that the multi-parameter dynamic prediction model has certain feasibility for the dynamic prediction of the iced wind turbine power loss, and step S404 is executed. If the multi-parameter dynamic prediction model does not meet the accuracy requirement, it means that the multi-parameter dynamic prediction model does not have certain feasibility for the dynamic prediction of the iced wind turbine power loss, and step S405 is executed.

[0160] Alternatively, in another embodiment of the present application, one specific implementation of step S403 comprises the following steps as shown in the following table: Figure 5

[0161] S501, calculating the average deviation percentage and the root mean square error between the test loss probability distribution and the actual power loss.

[0162] It should be noted that, in order to show that the accuracy of the multi-parameter dynamic prediction model meets the excellent requirement, the model performance indicators MAE and RMSE are selected for calculation. MAE refers to the average absolute error, which refers to the average deviation percentage of the predicted power loss and the actual power loss, and RMSE refers to the root mean square error.

[0163] S502, judging whether the average deviation percentage and the root mean square error are both less than the preset threshold.

[0164] Specifically, a threshold for judging the average deviation percentage and a threshold for judging the root mean square error are preset, so as to know whether the multi-parameter dynamic prediction model meets the accuracy requirement through the judgment of the two thresholds. Therefore, if the average deviation percentage and the root mean square error are both less than the preset threshold, it means that the multi-parameter dynamic prediction model meets the accuracy requirement, and step S503 is executed. If the average deviation percentage and the root mean square error are both not less than the preset threshold, it means that the multi-parameter dynamic prediction model does not meet the accuracy requirement, and step S504 is executed.

[0165] ​Optionally, the threshold of the average deviation percentage is set to 1.60%, and the threshold of the root mean square error is set to 2%, and of course other thresholds can also be set, which can be set according to requirements.

[0166] S503, determine whether the multi-parameter dynamic prediction model meets the accuracy requirement.

[0167] Specifically, the comparison between the test loss probability distribution and the actual power loss and the relative error are as shown in FIG. 6. Figure 6 . Figure 6 It can be seen from FIG. 6 that the predicted value is close to the true value, and the true value is basically within the 95% confidence interval, and the absolute value of the relative error is controlled within 8%. In addition, it is calculated that the average deviation MAE = 1.47% is less than the preset threshold 1.60%, and the root mean square error RMSE = 1.81% is less than the preset threshold 2%, so it can be illustrated that the multi-parameter dynamic prediction model meets the accuracy requirement.

[0168] S504, determine whether the multi-parameter dynamic prediction model meets the accuracy requirement.

[0169] S404, determine whether the multi-parameter dynamic prediction model is feasible.

[0170] It should be noted that when the multi-parameter dynamic prediction model meets the accuracy requirement, it means that the multi-parameter dynamic prediction model can be put into field application for use, so as to know whether to take corresponding deicing measures on the iced wind turbine.

[0171] S405, determine whether the multi-parameter dynamic prediction model is feasible.

[0172] It should be noted that when the multi-parameter dynamic prediction model does not meet the accuracy requirement, the multi-parameter dynamic prediction model needs to be optimized, and the parameters of the multi-parameter dynamic prediction model need to be adjusted, so the next step S208 can be executed.

[0173] S208, determine the optimized hyperparameters according to the sample loss probability distribution.

[0174] It should be emphasized that when the sample loss probability distribution does not meet the requirement of the true value, it means that the prediction performance of the model is not enough, and therefore in the embodiment of the present application, through the dynamic optimization strategy, the hyperparameters in the kernel function are optimized to maximize the log marginal likelihood function, so as to improve the prediction performance of the model.

[0175] Optionally, in another embodiment of the present application, a specific implementation of step S208 includes the following steps as shown in FIG. 7. Figure 7

[0176] S701, construct the hyperparameters and initialize the hyperparameters to obtain the initial hyperparameters. ​

[0177] Specifically, it is necessary to first construct a hyperparameter θ and initialize the hyperparameter θ to provide a suitable starting point for the log-marginal likelihood function, so as to ensure that the training process can converge stably and quickly, thereby improving the prediction accuracy and generalization ability of the model.

[0178] S702: Calculate the log-marginal likelihood function based on the sample loss probability distribution and covariance.

[0179] Specifically, the calculation formula of the logarithmic marginal likelihood function is:

[0180]

[0181] The first term on the right side of the above formula is the data fitting term, which is used to make the model match the field observation data. The second term is the complexity term, which is used to suppress unnecessary parameter fluctuations. y is the power loss observation value of the sample set (sample loss probability distribution), X is the input feature matrix, each row is a sample [v, T, H, LWC, △t], is the hyperparameter set, K is the kernel matrix, and the covariance is given by Calculated, I is the identity matrix and n is the number of samples.

[0182] S703 : Calculate the gradient vector using an optimization algorithm according to the log-marginal likelihood function.

[0183] It can be understood that by using the L-BFGS-B optimization algorithm to perform gradient descent on the log-marginal likelihood function, that is, to optimize it, a gradient vector can be obtained, where each element of the gradient vector is the partial derivative of the objective function with respect to each hyperparameter. The specific calculation formula for the gradient vector is:

[0184]

[0185] in, To find the partial derivative, the partial derivative is used to calculate the gradient of the log-marginal likelihood function with respect to the hyperparameters, so as to optimize the hyperparameters. is the signal variance, l d is the length scale of meteorological parameters, including wind speed, temperature, humidity, and liquid water content, is the noise variance.

[0186] S704: Calculate the optimized hyperparameters based on the initial hyperparameters and the gradient vector.

[0187] It should be noted that the gradient vector is used to guide the optimization process of the hyperparameters. In the gradient descent L-BFGS-B algorithm, the optimization formula for the hyperparameter θ is:

[0188]

[0189] in is the learning rate, which determines the step size of the update of the hyperparameters in each iteration. The L-BFGS-B algorithm determines the step size by approximating the Hessian matrix.

[0190] S705, determine whether the optimized hyperparameters are less than a preset threshold.

[0191] Specifically, in order to know whether the optimized hyperparameters meet the parameters of the optimized multi-parameter dynamic prediction model, a threshold can be used to know, that is, the preset threshold is 1e-6. Therefore, it is determined whether the optimized hyperparameters are less than the preset threshold. If the optimized hyperparameters are less than the preset threshold, it means that the optimized hyperparameters meet the parameters of the optimized multi-parameter dynamic prediction model, and step S706 can be executed at this time. If the optimized hyperparameters are not less than the preset threshold, it means that the optimized hyperparameters do not meet the parameters of the optimized multi-parameter dynamic prediction model, and the optimized hyperparameters need to be further optimized. Therefore, step S707 needs to be executed at this time.

[0192] S706, determine the optimized hyperparameters as the optimized hyperparameters.

[0193] Specifically, when the optimized hyperparameters are less than the preset threshold, the multi-parameter dynamic prediction model can be adjusted based on the optimized hyperparameters, and the next step S209 is executed.

[0194] S707, take the optimized hyperparameters as the initial hyperparameters.

[0195] Specifically, when the optimized hyperparameters are not less than the preset threshold, only the optimized hyperparameters need to be taken as the initial hyperparameters, and step S704 is returned to be executed until the optimized hyperparameters are less than the preset threshold. The purpose of optimizing the hyperparameters can be achieved.

[0196] S209, adjust the multi-parameter dynamic prediction model according to the optimized hyperparameters.

[0197] Specifically, the optimized hyperparameters can refer to the contents shown in Table 1:

[0198] Table 1

[0199] Symbol Optimized value Physical meaning l v ]]> 1.2 Significant effect enhancement when wind speed > 8 m / s l T ]]> 0.8 High sensitivity in the interval of -2℃ ~ 0℃ l H ]]> 12 Obvious effect when humidity is high (> 85%) l lwc ]]> 0.3 Power loss rises sharply when LWC > 0.1 g / m³ l t ]]> 0.53 Icing effect tends to be saturated after 1.6 hours #timg# 5.2 Maximum power loss fluctuation is about 23% #timg# 0.09 Sensor noise is about 3%

[0200] It should be noted that the contents shown in Table 1 are only used for example. Specifically, after adjusting the multi-parameter dynamic prediction model according to the optimized hyperparameters, the model needs to be further trained according to the standard meteorological environment parameters and the log icing time, and step S204 needs to be returned to be executed until the sample loss probability distribution meets the requirements of the true value.

[0201] S104, determine the deicing measures according to the loss probability distribution.​

[0202] Specifically, by monitoring the on-site meteorological environmental parameters and icing time, the power loss of the icing fan is output in real time through a multi-parameter dynamic prediction model, and whether corresponding deicing measures need to be taken is determined accordingly. For example, the deicing measure is to trigger a warning and strengthen the monitoring of the fan condition when the predicted power loss exceeds 10%.

[0203] The method provided by the application comprises the following steps: obtaining meteorological environmental parameters and icing time; preprocessing the meteorological environmental parameters and the icing time to obtain target meteorological environmental parameters and target icing time; inputting the target meteorological environmental parameters and the target icing time into a pre-trained multi-parameter dynamic prediction model to predict a loss probability distribution of an icing fan, wherein the multi-parameter dynamic prediction model is constructed by a Gaussian process regression, the Gaussian process regression is determined by a mean function and a covariance function, the multi-parameter dynamic prediction model is trained by historical meteorological environmental parameters and historical icing time in advance, and finally, deicing measures are determined according to the loss probability distribution. Therefore, the timeliness and accuracy of fan loss prediction can be effectively improved by the multi-parameter dynamic prediction model, and timely warning can be performed according to the loss probability distribution.

[0204] Another embodiment of the application provides a wind turbine icing power loss processing device, as shown in Figure 8 The device comprises the following units:

[0205] A data acquisition unit 801 is configured to acquire meteorological environmental parameters and icing time.

[0206] A preprocessing unit 802 is configured to preprocess the meteorological environmental parameters and the icing time to obtain target meteorological environmental parameters and target icing time.

[0207] A prediction unit 803 is configured to input the target meteorological environmental parameters and the target icing time into a pre-trained multi-parameter dynamic prediction model to predict a loss probability distribution of an icing fan, wherein the multi-parameter dynamic prediction model is constructed by a Gaussian process regression, the Gaussian process regression is determined by a mean function and a covariance function, and the multi-parameter dynamic prediction model is trained by historical meteorological environmental parameters and historical icing time in advance.

[0208] A measure determination unit 804 is configured to determine deicing measures according to the loss probability distribution.

[0209] It should be noted that the specific working processes of the above modules in the embodiments of the application can be correspondingly referred to steps S101-S104 in the above method embodiments, which will not be described here.

[0210] Optionally, another embodiment of the application provides a wind turbine icing power loss processing device, comprising:

[0211] The construction unit is configured to construct a Gaussian process regression based on the mean function and the covariance function, and use the Gaussian process regression as the multi-parameter dynamic prediction model.

[0212] The first acquisition unit is configured to acquire historical meteorological environmental parameters and historical icing times.

[0213] The data processing unit is configured to perform standardization processing on the historical meteorological environmental parameters to obtain standard meteorological environmental parameters, and perform logarithmic transformation processing on the historical icing times to obtain logarithmic icing times.

[0214] The first input unit is configured to input the standard meteorological environmental parameters and the logarithmic icing times into the mean function included in the multi-parameter dynamic prediction model to obtain a predicted value expectation of the icing wind turbine.

[0215] The second input unit is configured to input the standard meteorological environmental parameters and the logarithmic icing times into the covariance function included in the multi-parameter dynamic prediction model to obtain a covariance of the icing wind turbine, and integrate the predicted value expectation and the covariance to obtain a sample loss probability distribution of the icing wind turbine.

[0216] The requirement judgment unit is configured to judge whether the sample loss probability distribution meets a requirement of a true value.

[0217] The model determination unit is configured to determine the multi-parameter dynamic prediction model as a trained multi-parameter dynamic prediction model if the sample loss probability distribution meets the requirement of the true value.

[0218] The parameter determination unit is configured to determine an optimized hyperparameter according to the sample loss probability distribution if the sample loss probability distribution does not meet the requirement of the true value.

[0219] The adjustment unit is configured to adjust the multi-parameter dynamic prediction model according to the optimized hyperparameter, and return to input the standard meteorological environmental parameters and the logarithmic icing times into the mean function included in the multi-parameter dynamic prediction model to obtain the predicted value expectation of the icing wind turbine.

[0220] Optionally, another embodiment of the application provides a wind turbine icing power loss processing device, wherein the second input unit comprises:

[0221] The second acquisition unit is configured to acquire a signal variance, an icing saturation time, and a noise standard deviation.

[0222] The first calculation unit is configured to calculate a meteorological parameter kernel based on the signal variance and the standard meteorological environmental parameters.

[0223] a second calculation unit configured to calculate a time kernel according to the icing saturation time and the logarithmic icing time;

[0224] a third calculation unit configured to calculate a noise kernel based on the noise standard deviation using a Kronecker function;

[0225] a fourth calculation unit configured to calculate a covariance of the icing wind turbine according to the meteorological parameter kernel, the time kernel and the noise kernel using a covariance function of multi-parameter dynamic inclusion.

[0226] Optionally, another embodiment of the application provides a wind turbine icing power loss processing device, and the device comprises a parameter determination unit, which comprises:

[0227] an initialization unit configured to construct a hyperparameter and initialize the hyperparameter to obtain an initial hyperparameter;

[0228] a fifth calculation unit configured to calculate a log marginal likelihood function according to the sample loss probability distribution and the covariance;

[0229] a sixth calculation unit configured to calculate a gradient vector using an optimization algorithm according to the log marginal likelihood function;

[0230] a seventh calculation unit configured to calculate an optimized hyperparameter based on the initial hyperparameter and the gradient vector;

[0231] a threshold determination unit configured to determine whether the optimized hyperparameter is less than a preset threshold;

[0232] a parameter determination subunit configured to determine the optimized hyperparameter as an optimized hyperparameter if the optimized hyperparameter is less than the preset threshold;

[0233] a unit configured to take the optimized hyperparameter as the initial hyperparameter and return to calculate the optimized hyperparameter based on the initial hyperparameter and the gradient vector if the optimized hyperparameter is not less than the preset threshold.

[0234] Optionally, another embodiment of the application provides a wind turbine icing power loss processing device, and the device further comprises:

[0235] a third acquisition unit configured to acquire a test group, and perform standardization processing on the test group to obtain a standard test group;

[0236] a third input unit configured to input the standard test group into a pre-trained multi-parameter dynamic prediction model to predict a test loss probability distribution of the icing wind turbine;

[0237] a first determination unit configured to determine whether the multi-parameter dynamic prediction model meets an accuracy requirement based on the test loss probability distribution and an actual power loss;

[0238] The first determining unit is configured to determine that the multi-parameter dynamic prediction model is feasible if the multi-parameter dynamic prediction model meets the accuracy requirement.

[0239] The second determining unit is configured to determine that the multi-parameter dynamic prediction model is not feasible if the multi-parameter dynamic prediction model does not meet the accuracy requirement.

[0240] Optionally, another embodiment of the application provides a wind turbine icing power loss processing device, and the first judging unit comprises:

[0241] The seventh calculating unit is configured to calculate the average deviation percentage and the root mean square error between the test loss probability distribution and the actual power loss.

[0242] The second judging unit is configured to judge whether the average deviation percentage and the root mean square error are both less than a preset threshold.

[0243] The third determining unit is configured to determine that the multi-parameter dynamic prediction model meets the accuracy requirement if the average deviation percentage and the root mean square error are both less than the preset threshold.

[0244] The fourth determining unit is configured to determine that the multi-parameter dynamic prediction model does not meet the accuracy requirement if the average deviation percentage and the root mean square error are both not less than the preset threshold.

[0245] It should be noted that the specific working processes of the various modules provided by the above embodiments of the application can be correspondingly referred to the corresponding steps in the above method embodiments, which will not be described here.

[0246] It should be further noted that the wind turbine icing power loss processing device provided by the embodiments of the application has the technical effects of any one of the above embodiments, and the embodiments of the application will not be described here.

[0247] The skilled person can further realize that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0248] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for processing wind turbine power loss due to icing, characterized in that: include: Obtain meteorological environment parameters and ice cover time; Preprocessing the meteorological environment parameters and the icing time to obtain target meteorological environment parameters and target icing time; The target meteorological environment parameters and the target icing time are input into a pre-trained multi-parameter dynamic prediction model to predict the loss probability distribution of the icing wind turbine; wherein the multi-parameter dynamic prediction model is constructed by Gaussian process regression; the Gaussian process regression is determined by a mean function and a covariance function; the multi-parameter dynamic prediction model is pre-trained using historical meteorological environment parameters and historical icing time; Deicing measures are determined based on the loss probability distribution.

2. The method according to claim 1, characterized in that The training method of the multi-parameter dynamic prediction model includes: Construct Gaussian process regression based on mean function and covariance function, and use Gaussian process regression as a multi-parameter dynamic prediction model; Obtain historical meteorological environment parameters and historical ice cover time; Standardizing the historical meteorological environment parameters to obtain standard meteorological environment parameters, and performing logarithmic transformation on the historical ice cover time to obtain logarithmic ice cover time; Inputting the standard meteorological environment parameters and the logarithmic icing time into the mean function included in the multi-parameter dynamic prediction model to obtain the expected predicted value of the icing fan; Inputting the standard meteorological environment parameters and the logarithmic icing time into the covariance function included in the multi-parameter dynamic prediction model to obtain the covariance of the icing fan, and integrating the predicted value expectation and the covariance to obtain the sample loss probability distribution of the icing fan; Determining whether the sample loss probability distribution meets the requirements of the true value; If the sample loss probability distribution meets the requirement of the true value, the multi-parameter dynamic prediction model is determined as the trained multi-parameter dynamic prediction model; If the sample loss probability distribution does not meet the requirements of the true value, determining the optimized hyperparameters according to the sample loss probability distribution; The multi-parameter dynamic prediction model is adjusted according to the optimized hyperparameters, and the step of inputting the standard meteorological environment parameters and the logarithmic icing time into the mean function included in the multi-parameter dynamic prediction model is returned to execute to obtain the expected predicted value of the icing fan.

3. The method according to claim 2, characterized in that The step of inputting the standard meteorological environment parameter and the logarithmic icing time into the covariance function of the multi-parameter dynamic inclusion to obtain the covariance of the icing fan comprises: Obtain signal variance, ice saturation time and noise standard deviation; Calculating a meteorological parameter kernel based on the signal variance and the standard meteorological environment parameter; Calculating a time kernel according to the ice saturation time and the logarithmic ice time; Calculating a noise kernel based on the noise standard deviation using a Kronecker function; The covariance of the de-icing fan is calculated according to the meteorological parameter kernel, the time kernel and the noise kernel using the covariance function of the multi-parameter dynamic inclusion.

4. The method according to claim 2, characterized in that Determining the optimized hyperparameters according to the sample loss probability distribution includes: Constructing hyperparameters and initializing the hyperparameters to obtain initial hyperparameters; Calculating a log-marginal likelihood function based on the sample loss probability distribution and the covariance; Calculating a gradient vector using an optimization algorithm according to the log-marginal likelihood function; Calculating optimized hyperparameters based on the initial hyperparameters and the gradient vector; Determine whether the optimized hyperparameter is less than a preset threshold; If the optimized hyperparameter is less than the preset threshold, determining the optimized hyperparameter as the optimized hyperparameter; If the optimized hyperparameter is not less than the preset threshold, the optimized hyperparameter is used as the initial hyperparameter, and the process returns to execute the calculation of the optimized hyperparameter based on the initial hyperparameter and the gradient vector.

5. The method according to claim 2, characterized in that Also includes: Acquire a test group, and perform standardization on the test group to obtain a standard test group; Inputting the standard test group into a pre-trained multi-parameter dynamic prediction model to predict the test loss probability distribution of the ice-covered wind turbine; Based on the test loss probability distribution and the actual power loss, determining whether the multi-parameter dynamic prediction model meets the accuracy requirements; If the multi-parameter dynamic prediction model meets the accuracy requirement, determining that the multi-parameter dynamic prediction model is feasible; If the multi-parameter dynamic prediction model does not meet the accuracy requirement, it is determined that the multi-parameter dynamic prediction model is not feasible.

6. The method according to claim 5, characterized in that The determining, based on the test loss probability distribution and the actual power loss, whether the multi-parameter dynamic prediction model meets the accuracy requirement includes: Calculating an average deviation percentage and a root mean square error between the test loss probability distribution and the actual power loss; Determining whether the average deviation percentage and the root mean square error are both less than a preset threshold; If the average deviation percentage and the root mean square error are both less than a preset threshold, it is determined that the multi-parameter dynamic prediction model meets the accuracy requirement; If both the average deviation percentage and the root mean square error are not less than a preset threshold, it is determined that the multi-parameter dynamic prediction model does not meet the accuracy requirement.

7. A device for processing the power loss of a fan due to icing, characterized in that: include: A data acquisition unit, used to obtain meteorological environment parameters and ice cover time; a preprocessing unit, configured to preprocess the meteorological environment parameters and the icing time to obtain target meteorological environment parameters and target icing time; A prediction unit is configured to input the target meteorological environment parameters and the target icing time into a pre-trained multi-parameter dynamic prediction model to predict a loss probability distribution of an icing wind turbine; wherein the multi-parameter dynamic prediction model is constructed by Gaussian process regression; the Gaussian process regression is determined by a mean function and a covariance function; and the multi-parameter dynamic prediction model is pre-trained using historical meteorological environment parameters and historical icing time; A measure determination unit is used to determine a de-icing measure according to the loss probability distribution.

8. The device according to claim 7, characterized in that Also includes: A construction unit for constructing Gaussian process regression based on mean function and covariance function, and using Gaussian process regression as a multi-parameter dynamic prediction model; The first acquisition unit is used to obtain historical meteorological environment parameters and historical ice cover time; a data processing unit, configured to perform standardization processing on the historical meteorological environment parameters to obtain standard meteorological environment parameters, and perform logarithmic transformation processing on the historical ice cover time to obtain logarithmic ice cover time; A first input unit is configured to input the standard meteorological environment parameter and the logarithmic icing time into a mean function included in the multi-parameter dynamic prediction model to obtain an expected predicted value of the icing fan; a second input unit, configured to input the standard meteorological environment parameter and the logarithmic icing time into a covariance function included in the multi-parameter dynamic prediction model to obtain a covariance of the icing fan, and integrate the predicted value expectation and the covariance to obtain a sample loss probability distribution of the icing fan; A requirement judgment unit, used to judge whether the sample loss probability distribution meets the requirement of the true value; a model determination unit, configured to determine the multi-parameter dynamic prediction model as a trained multi-parameter dynamic prediction model if the sample loss probability distribution meets the requirement of a true value; a parameter determination unit, configured to determine optimized hyperparameters based on the sample loss probability distribution if the sample loss probability distribution does not meet the requirements of the true value; An adjustment unit is used to adjust the multi-parameter dynamic prediction model according to the optimized hyperparameters, and return to execute the step of inputting the standard meteorological environment parameters and the logarithmic icing time into the mean function contained in the multi-parameter dynamic prediction model to obtain the expected predicted value of the icing fan.

9. The device according to claim 8, characterized in that The second input unit includes: The second acquisition unit is used to obtain signal variance, ice saturation time and noise standard deviation; a first calculation unit, configured to calculate a meteorological parameter kernel based on the signal variance and the standard meteorological environment parameter; a second calculation unit, configured to calculate a time kernel according to the ice saturation time and the logarithmic ice time; a third calculation unit, configured to calculate a noise kernel based on the noise standard deviation using a Kronecker function; A fourth calculation unit is configured to calculate the covariance of the ice-covering fan according to the meteorological parameter kernel, the time kernel, and the noise kernel by using the covariance function dynamically included by the multi-parameter.

10. The device according to claim 8, characterized in that The parameter determination unit includes: An initialization unit, used to construct hyperparameters and initialize the hyperparameters to obtain initial hyperparameters; a fifth calculation unit, configured to calculate a log-marginal likelihood function according to the sample loss probability distribution and the covariance; a sixth computing unit, configured to compute a gradient vector using an optimization algorithm according to the log-marginal likelihood function; a seventh computing unit, configured to compute an optimized hyperparameter based on the initial hyperparameter and the gradient vector; A threshold judgment unit, used to judge whether the optimization hyperparameter is less than a preset threshold; a parameter determination subunit, configured to determine the optimized hyperparameter as the optimized hyperparameter if the optimized hyperparameter is less than the preset threshold; As a unit, if the optimized hyperparameter is not less than the preset threshold, the optimized hyperparameter is used as the initial hyperparameter, and the calculation of the optimized hyperparameter based on the initial hyperparameter and the gradient vector is returned to execute.

Images