A short-term wind power combination prediction method and system under icing conditions

Abstract

A short-term wind power combination prediction method and system under icing conditions, based on historical meteorological data and power data, constructs an icing label data set and establishes an icing identification model; combines multi-source meteorological forecasts to obtain the wind speed of wind turbines in the station and constructs a wind turbine icing growth model; constructs an ice-covered wind turbine CFD simulation model based on different wind speeds, icing degrees and wind turbine types, and through simulation calculations, constructs a wind speed-power correction coefficient table for wind turbines under icing; establishes at least two power prediction models based on different machine learning models, constructs and trains a short-term wind power combination prediction model; calculates the hourly power correction coefficient of the ice-covered station through icing identification and icing growth models, combined with the wind speed-power correction coefficient table for wind turbines under icing, and calculates the power prediction value under icing conditions based on the short-term wind power combination prediction model. The present application can improve the accuracy of power prediction for ice-covered wind turbines and obtain station output more efficiently and accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy power prediction, and in particular relates to a method and system for predicting short-term wind power combination under icing conditions. Background Art

[0002] In recent years, reducing carbon emissions has played a great role in environmental protection and ecological balance. At present, my country's energy-related carbon dioxide emissions are 1 billion tons per year. As the single industry with the largest share of coal consumption and carbon emissions in my country, electricity accounts for more than 40% of carbon emissions. Therefore, vigorously developing new energy power represented by wind and solar energy and promoting the grid connection and consumption of a high proportion of renewable energy will become a top priority for reducing carbon emissions in the power system.

[0003] As new energy represented by photovoltaic and wind power gradually replaces traditional energy, it presents a good development prospect. In recent years, extreme weather has occurred frequently in my country and has shown an increasing and intensifying trend. Affected by extreme weather processes such as cold waves and blizzards, wind farms have low temperatures and high humidity, which are prone to ice disasters and last for a long time. Studies have shown that when the temperature is between -10 and 0°C, the relative humidity is above 70%, and the average wind speed is ≤6 m / s, it is most conducive to the occurrence of freezing. Wind power icing can easily cause large fluctuations in wind power / photovoltaic new energy power generation. There is great uncertainty on both the source and load sides, and it is difficult to predict the power of new energy. The power grid faces the risk of insufficient system defense capabilities under extreme / transitional weather processes, which brings huge challenges to the power balance of the power grid.

[0004] Therefore, there is an urgent need to predict short-term wind power. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the present invention provides a short-term wind power combination prediction method and system under icing conditions, so as to reduce the difficulty of wind power prediction and improve the accuracy of icing wind power prediction.

[0006] The present invention adopts the following technical solution.

[0007] A first aspect of the present invention provides a method for forecasting short-term wind power combination under icing conditions, comprising:

[0008] Based on historical meteorological data and power data, an icing label dataset is constructed and an icing identification model is established;

[0009] Combined with multi-source weather forecasts, the wind speed of wind turbines in the station is obtained and a wind turbine icing growth model is constructed;

[0010] Based on different wind speeds, icing degrees and fan types, a CFD (Computational Fluid Dynamics) simulation model of the icing fan is constructed to obtain the power characteristic curve of the icing fan. Through simulation calculation, a wind speed power correction coefficient table of the fan under icing is constructed.

[0011] Establish at least two power prediction models based on different machine learning models, use IPSO algorithm and / or BO algorithm to optimize model parameters, and build and train a short-term wind power combination prediction model;

[0012] Through icing identification and icing growth model, combined with the wind speed-power correction coefficient table under icing of wind turbines, the hourly power correction coefficient of the icing station is calculated. Based on the hourly power correction coefficient of the icing station, the power prediction value output by the short-term wind power combination prediction model is corrected to obtain the wind power prediction value under icing state.

[0013] Optionally, based on the historical data of the station, the ice cover is calibrated as 1, and the non-ice cover is calibrated as 0. By dividing the training set and the test set, the logistic regression algorithm is used to complete the creation and verification of the ice cover identification model.

[0014] Optionally, the fan icing growth model is as follows:

[0015]

[0016] In the formula,

[0017] is the mass of ice, in kg;

[0018] is the freezing time, in seconds. The freezing start time is obtained through the icing identification model.

[0019] represent the collision, capture, and freezing coefficients of supercooled water droplets,

[0020] is the water drop mass fraction, a dimensionless parameter,

[0021] is the relative speed of the colliding water droplet relative to the fan impeller, in m / s,

[0022] is the effective cross-sectional area of ​​the object that the water droplet collides with, in m 2 .

[0023] Optionally, the relative velocity of the impinging water droplets relative to the fan impeller Calculated using the following formula:

[0024]

[0025] in, is the fan impeller tangential speed, in m / s, is the wind speed of the wind turbine obtained based on the wake model, in m / s.

[0026] Optionally, the wind speed of the wind turbine calculated based on the wake model includes:

[0027] When the current wind turbine is located in a flat terrain, the wind speed of the current wind turbine is calculated according to the first wake model; and / or,

[0028] When there is no wind turbine in the front row of the current wind turbine, the wind speed of the current wind turbine is calculated according to the second wake model; and / or,

[0029] When there are fans in the front row of the current fan, the wind speed of the current fan is calculated according to the third wake model.

[0030] Optionally, the icing power correction coefficient value of the icing station is determined according to the icing power correction coefficient value of each wind turbine in the station according to the following formula:

[0031]

[0032] in,

[0033] is the icing power correction coefficient of the station during the icing period at the i-th moment,

[0034] is the power correction coefficient value of wind turbine k at the i-th moment during the icing period,

[0035] is the rated capacity of fan k,

[0036] n is the total number of wind turbines in the station.

[0037] Optionally, at least two power prediction models are established based on different machine learning models, and constructing and training a short-term wind power combination prediction model includes:

[0038] Dividing the training data set into a plurality of subsets; the data in the training set includes historical meteorological data when there is no ice and historical power data of the wind turbine;

[0039] Use at least two machine learning models for training and prediction for each subset, perform cross-validation on each machine learning model, and stack all cross-validation results to form a prediction result;

[0040] The prediction results of all machine learning models are combined into a new training data set, and the meta-model is trained using the new training data set to obtain a short-term wind power combination prediction model;

[0041] Therein, the machine learning model and / or meta-model is trained with the goal of minimizing the fitness function.

[0042] Optionally, the IPSO algorithm is used to find the optimal parameters during the machine learning model training process, including:

[0043] Randomly generate a population particle, wherein the population particle includes hyperparameters of the machine learning model, and define intervals for each hyperparameter of the population particle;

[0044] Construct the fitness function of the particle and calculate the fitness value corresponding to each particle position;

[0045] Determine the individual extreme value and group extreme value according to the initial particle fitness value, and take the individual extreme value of each particle as its historical best position;

[0046] In each iteration, the particle's own speed and position are updated through individual extremum and global extremum according to the fitness value and inertia weight;

[0047] When the maximum number of iterations is reached or the fitness function value is minimum, the iteration is stopped to obtain the optimal parameters.

[0048] Optionally, construct a fitness function for the particle, including:

[0049] Constructing a fitness function of the particle based on the fitting errors of the validation sample and the training sample, wherein the weights of the fitting errors of the validation sample and the training sample are the same;

[0050] The fitness function is shown as follows:

[0051]

[0052] in,

[0053] , are the predicted value and true value of the jth sample data in the training data set, respectively.

[0054] , are the predicted value and true value of the j-th sample data in the validation dataset, respectively.

[0055] Optionally, the inertia weight is calculated using a dynamic nonlinear method, which is expressed as follows:

[0056]

[0057] in, is the maximum inertia weight, is the minimum inertia weight, t is the current iteration number, is the maximum number of iterations.

[0058] The second aspect of the present invention provides a short-term wind power combination prediction system, including a first building module, a second building module, a third building module, a fourth building module and a calculation module. Among them, the first building module is used to construct an ice label data set based on historical meteorological data and power data, and establish an ice identification model. The second building module is used to combine multi-source meteorological forecasts to obtain the wind speed of the wind turbine in the station and build a wind turbine icing growth model. The third building module is used to build an ice-covered wind turbine CFD simulation model based on different wind speeds, ice coverage and wind turbine types, obtain the power characteristic curve of the ice-covered wind turbine, and construct a typical wind turbine icing wind speed-power correction coefficient table through simulation calculation. The fourth building module is used to establish at least two power prediction models based on different machine learning models, use IPSO and / or BO algorithms to optimize model parameters, and construct and train a short-term wind power combination prediction model. The calculation module is used to calculate the hourly power correction coefficient of the ice-covered station through ice identification and ice growth model, combined with the wind speed-power correction coefficient table under wind turbine icing, and calculate the power prediction value under icing state based on the short-term wind power combination prediction model.

[0059] A third aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the short-term wind power combination prediction method under icing conditions is implemented.

[0060] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned short-term wind power combination prediction method under icing conditions.

[0061] Compared with the prior art, the beneficial effects of the present invention include at least:

[0062] The present invention identifies the icing condition of the wind turbine, and when the identification result is icing, uses a CFD simulation calculation method to calculate the wind speed of the wind turbine, and then combines the wind turbine icing growth model to quantitatively analyze the power characteristics of the icing wind turbine under different icing conditions and different wind speeds to obtain the icing power correction coefficient value. Based on the icing power correction coefficient value, the prediction result output by the power prediction model is corrected, so that the predicted value of the wind power of the station can be obtained more efficiently and accurately, and then the station output can be obtained more efficiently and accurately. The present invention solves the problem of large deviations in wind power prediction caused by power fluctuations due to wind turbine icing caused by bad weather, reduces the difficulty of prediction, and improves the accuracy of power prediction of icing wind turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0064] Figure 1 is a flow chart of a method for predicting wind power provided by an embodiment of the present invention;

[0065] Figure 2 is a schematic diagram of a wind turbine blade icing model provided by an embodiment of the present invention;

[0066] Figure 3 is a schematic diagram of a wind turbine airfoil change with different ice coverage degrees provided by an embodiment of the present invention;

[0067] Figure 4 is a schematic diagram of output power comparison under different ice coverage degrees provided by an embodiment of the present invention;

[0068] Figure 5 It is a schematic diagram of a particle swarm optimization algorithm model parameter optimization process provided by an embodiment of the present invention;

[0069] Figure 6 It is a schematic diagram of a framework of a system for predicting wind power provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0071] According to the length of the prediction time, wind power prediction can generally be divided into ultra-short-term prediction, short-term prediction, medium-term prediction and long-term prediction. The embodiments of the present invention are mainly applied to ultra-short-term prediction and / or short-term prediction.

[0072] Combination Figure 1 As shown, embodiment 1 of the present invention provides a short-term wind power combination prediction method under icing conditions, including the following contents:

[0073] S1. Based on historical meteorological data and power data, an icing label dataset is constructed and an icing identification model is established.

[0074] S1 specifically includes: obtaining the current meteorological data of the station wind turbine; the meteorological data includes one or more of wind speed, temperature, wind direction, air pressure and humidity. Inputting the current meteorological data into the icing identification model to identify whether the wind turbine is covered with ice, wherein the icing identification model is obtained by machine learning training using multiple sets of data, and the multiple sets of data include a first type of data and a second type of data, and each set of data in the first type of data includes: historical meteorological data when the wind turbine is covered with ice and a label indicating that the wind turbine is covered with ice in the historical meteorological data; each set of data in the second type of data includes: historical meteorological data when the wind turbine is not covered with ice and a label indicating that the wind turbine is not covered with ice in the historical meteorological data. Obtain the output information of the icing identification model, wherein the output information includes the icing identification result corresponding to the wind turbine.

[0075] The machine learning in S1 can be a logistic regression algorithm, a decision tree algorithm, a random forest algorithm, etc.

[0076] According to the icing characteristics, wind turbine icing is very likely to occur in a low-temperature and high-humidity environment with a temperature below 0°C, a humidity greater than 70%, and a low wind speed (below 6m / s). In S1, based on the existing historical data of the station, icing is calibrated as 1 and non-icing is calibrated as 0. By dividing the training set and the test set, the logistic regression algorithm is used to complete the creation and verification of the icing identification model.

[0077] S2. Combine multi-source weather forecasts to obtain wind speeds of wind turbines in the station and construct a wind turbine icing growth model.

[0078] The icing growth model of the fan in S2 is as follows:

[0079]

[0080] Where: is the mass of ice, in kg;

[0081] is the freezing time, in seconds. The freezing start time is obtained through the icing identification model;

[0082] represent the collision, capture, and freezing coefficients of supercooled water droplets,

[0083] is the water drop mass fraction, a dimensionless parameter,

[0084] is the relative speed of the colliding water droplet relative to the fan impeller, in m / s,

[0085] is the effective cross-sectional area of ​​the object that the water droplet collides with, in m 2 .

[0086] The effective cross-sectional area of ​​the object hit by the water drop is calculated according to the following formula:

[0087]

[0088] in, D i is the leading edge diameter, in m, and is linearly interpolated at each meter along the blade length, used to calculate the leading edge cross-sectional area.

[0089] The relative velocity of the impinging water droplets with respect to the fan impeller Calculated using the following formula:

[0090]

[0091] in, is the fan impeller tangential speed, in m / s, It is the wind speed of the wind turbine obtained based on the wake model, in m / s.

[0092] Collision coefficient Calculated as follows:

[0093] ,

[0094] Where, K is the inertia parameter, ; is the cylinder diameter, in m (using the blade leading edge diameter); is the water density, in kg / m 3 , take 1000kg / m³, is the air density, in kg / m 3 , the value is 1.225 kg / m3; is the absolute viscosity of air, in kg / m*s, and its value is 1.725·10 −5 kg / (m*s); is the free stream relative velocity in m / s.

[0095] Capture coefficient Calculated as follows:

[0096]

[0097] Freeze coefficient Calculated as follows:

[0098]

[0099] The ratio of the amount of water droplets that freeze on the surface of an object per unit time to the amount of water droplets that remain permanently on the surface of the object. It is related to wind speed, temperature, liquid water content, object diameter and ice coverage duration. In actual application, the free stream relative velocity is approximately equal to the relative velocity of the colliding water droplets relative to the fan impeller.

[0100] Where F is the water flux density, F=η 1 *η 2 * w * v , liquid ratio λ Pick 0.26,

[0101] is the latent heat of freezing of water,

[0102] h is the convective heat transfer coefficient,

[0103] is the viscous heating recovery coefficient ( Can be taken 0.79 ),

[0104] is the free stream relative velocity in m / s

[0105] is the specific heat of air,

[0106] is the temperature of the ice surface (0°C for wet growth),

[0107] is the air temperature,

[0108] ɛ is the molecular mass ratio of dry air and water vapor (the value is 0.622),

[0109] Le is the latent heat of vaporization,

[0110] is the saturated water vapor pressure,

[0111] is the partial pressure of water vapor in the air,

[0112] Air pressure,

[0113] is the specific heat of water,

[0114] is the temperature of the water droplets when they collide,

[0115] σ is the Stefan-Boltzmann constant (5.67 * 10 -8 W / (m 2K 4 ),

[0116] a is the radiation linearization constant (8.1*10 7 K 3 ).

[0117] Combination Figure 2 and 3 As shown in the figure, the ice shape at the blade tip (r=0.95R) is used to approximate the ice shape of the entire blade, ignoring the small protrusions and burr-like structures of natural ice. By selecting appropriate contour points on the upper and lower surfaces of the airfoil, the parametric representation of the outer contour of the ice is achieved. The ice thickness has a one-to-one correspondence with the ice mass, which can be calculated by integration.

[0118] The degree of ice coverage is calculated according to the following formula:

[0119]

[0120] In the formula, is the ice thickness of the wind turbine, is the leaf chord length.

[0121] In this embodiment, ice cover is identified through an ice cover identification model, and the accurate incoming wind speed of a single wind turbine is obtained using high-precision meteorological forecast data and a wake model. The ice growth model is combined to accurately calculate the amount of ice cover, and quantifying the degree of ice cover helps to improve the accuracy of subsequent wind power prediction.

[0122] In S2, the wind speed of the wind turbine calculated based on the wake model includes:

[0123] When the current wind turbine is located in a flat terrain, the wind speed of the current wind turbine is calculated according to the first wake model; and / or,

[0124] When there is no wind turbine in the front row of the current wind turbine, the wind speed of the current wind turbine is calculated according to the second wake model; and / or,

[0125] When there are fans in the front row of the current fan, the wind speed of the current fan is calculated according to the third wake model.

[0126] Specifically, the first wake model is expressed as follows:

[0127]

[0128] The above formula, is the fan speed, is the front row fan speed of the current fan, Take 0.2, is the distance between two wind turbines, is the impeller radius, is the height of the wind turbine impeller, is the surface roughness, which is a constant.

[0129] Specifically, the second wake model is expressed as follows:

[0130]

[0131] in, is the fan speed, is the front row fan speed of the current fan, is the tower height of the wind turbine, is the height difference between the front exhaust fan and the rear exhaust fan, is the coefficient of wind speed variation with height, which can be 1 or 7. is the original height of the front row, is the original wind speed in the front row, is the height after changing the height, is the wind speed after changing altitude.

[0132] Specifically, the third wake model is expressed as follows:

[0133]

[0134] in, is the fan speed, is the front row fan speed of the current fan, Take 0.2, is the distance between two wind turbines, R is the impeller radius, is the tower height of the wind turbine, is the height difference between the front exhaust fan and the rear exhaust fan, is the coefficient of wind speed variation with height, which can be taken as 1 / 7.

[0135] In this embodiment, by adopting different wake models at different wind turbine locations to calculate the wind speed of the wind turbine, a more accurate comprehensive model of the wind farm can be obtained, and the wind speed of the wind turbines at different locations can be obtained, thereby improving the accuracy of subsequent ice coverage calculations.

[0136] S3. Based on different wind speeds, icing degrees and fan types, a CFD simulation model of the icing fan is constructed to obtain the power characteristic curve of the icing fan. Through simulation calculation, a wind speed-power correction coefficient table of a typical fan under icing is constructed.

[0137] Specifically, according to the wind speed and ice coverage degree of the wind turbine, the corresponding ice coverage power correction coefficient value is obtained from a pre-constructed wind speed-power correction coefficient table under typical wind turbine ice coverage.

[0138] In S3, constructing a wind speed-power correction coefficient table under wind turbine icing includes: constructing a CFD simulation model of a wind turbine with different degrees of icing to obtain wind turbine power under different wind speeds;

[0139] Compare the wind turbine power at different wind speeds under different degrees of icing with the wind turbine power without icing to obtain the corresponding icing power correction coefficient value;

[0140] Based on different icing degrees, different wind speeds and corresponding icing power correction coefficient values, a mapping relationship of hourly power correction coefficients for icing stations is constructed to form a wind speed-power correction coefficient table for wind turbines under icing.

[0141] Specifically, clean fans and fan simulation models with different degrees of icing, different wind speeds, and different fan types are constructed, and the power characteristics of icing fans are analyzed using CFD simulation to obtain the fan power at different wind speeds and form an icing fan power characteristic curve, which contains meteorological data such as icing degree and wind speed. Figure 4 As shown, the horizontal axis of the curve is wind speed and the vertical axis is power.

[0142] CFD simulation software calculates torque at different wind speeds And the fan output P, ​​verify the accuracy of the fan simulation model:

[0143]

[0144] Where n is the speed of the fan impeller.

[0145] Power characteristics of fans under icing conditions:

[0146]

[0147] Based on the power characteristic curve of the icing fan, combined with the cubic spline interpolation method, a table of wind speed power correction coefficients for different icing degrees of the fan is constructed, as shown in Table 1, and the icing power correction coefficient value is further obtained. Table 1 is a table of icing-power correction relationship of a fan at a station in the southern region.

[0148] Table 1

[0149] S4. Establish at least two power prediction models based on different machine learning models, use IPSO (Improved Particle Swarm Optimization) algorithm and / or BO (Bayesian Optimization) algorithm to optimize model parameters, and build and train a short-term wind power combination prediction model.

[0150] Specifically, the current meteorological data without ice is input into a pre-trained short-term wind power combination prediction model to predict the site wind power and output the prediction result.

[0151] In S4, at least two power prediction models are established based on different machine learning models, and the short-term wind power combination prediction model is constructed and trained, including:

[0152] Dividing the training data set into a plurality of subsets; the data in the training set includes historical meteorological data when there is no ice and historical power data of the wind turbine;

[0153] Use at least two machine learning models for training and prediction for each subset, perform cross-validation on each machine learning model, and stack all cross-validation results to form a prediction result;

[0154] The prediction results of all machine learning models are combined into a new training data set, and the meta-model is trained using the new training data set to obtain a short-term wind power combination prediction model;

[0155] Therein, the machine learning model and / or meta-model is trained with the goal of minimizing the fitness function.

[0156] In this embodiment, cross-validation can solve small sample problems such as icing, thereby improving the accuracy of power prediction model construction and further improving prediction accuracy.

[0157] The machine learning model can be LSTM (Long Short Term Memory) or LIGHTGBM (Light Gradient Boosting Machine, a distributed gradient boosting tool based on the decision tree algorithm) or KNN (K-Nearest Neighbor) or SVR (Support Vector Regression) or XGBOOST (eXtreme Gradient Boosting) or Random Forest or CATBOOST (Categorical Boosting) or Multi-layer Perception, etc.

[0158] Optionally, combined Figure 5 As shown in the figure, based on the particle swarm optimization algorithm, the optimal parameters are found in the machine learning model training process. The specific steps include:

[0159] Initialization parameters. Specifically include: randomly generating a population particle, the population particle includes the hyperparameters of the short-term wind power prediction model, the hyperparameters include the number of neurons, the number of iterations, etc., and limiting the interval of each parameter of the population particle.

[0160] Construct the fitness function of the particle and calculate the fitness value corresponding to each particle position. Specifically, it includes: determining the individual extreme value and the group extreme value according to the initial particle fitness value, and taking the best position of each particle (i.e., the individual extreme value) as its best historical position.

[0161] In each iteration, the dynamic nonlinear inertia weight is used to update the particle's own speed and position through individual extreme values ​​and global extreme values ​​according to the fitness function value and the inertia weight.

[0162] When the maximum number of iterations is reached or the fitness function value is minimum, stop the iteration to obtain the optimal parameters.

[0163] It can be understood that the inertia parameter represents the ability of the particle to inherit the velocity of the previous iteration, and the optimal value of the particle obtained is the optimal solution of the parameters of the model training.

[0164] Further, constructing the fitness function of the particle includes: constructing the fitness function of the particle based on the fitting errors of the validation sample and the training sample, wherein the weights of the fitting errors of the validation sample and the training sample are the same. The fitness function is shown in the following formula:

[0165]

[0166] in,

[0167] , are the predicted value and true value of the jth sample data in the training data set, respectively.

[0168] , are the predicted value and true value of the j-th sample data in the validation dataset, respectively.

[0169] In this embodiment, by constructing an equally weighted fitness function based on the errors of the validation sample and the training sample, the overfitting and underfitting problems caused by using the training sample error as the fitness value can be reduced.

[0170] Furthermore, the inertia weight of the particle swarm optimization algorithm is expressed as follows:

[0171]

[0172] in, is the maximum inertia weight, is the minimum inertia weight, is the current iteration number, is the maximum number of iterations.

[0173] In this embodiment, a larger w value is conducive to escaping from the local optimum, while a smaller w value is conducive to accelerating convergence. When smaller, Close to ,and The speed of reduction is also slow, which can ensure the global optimization ability of the algorithm; Continuously increasing, decreases nonlinearly, and The reduction rate increases rapidly. This embodiment proposes a dynamic nonlinear weight adjustment strategy to prevent the IPSO algorithm from falling into the local optimum, which can improve the local optimization ability of the algorithm, thereby enhancing the optimization ability and convergence speed by improving the inertia weight, thereby improving the optimization speed.

[0174] Optionally, optimal parameters are found in the machine learning model based on a Bayesian optimization algorithm.

[0175] Specifically, based on the function to be optimized f Construct a probability model and iterate it in turn to obtain the optimal solution for the hyperparameters.

[0176]

[0177] in, is the optimal hyperparameter set, For decision space, f ( x ) is the objective function.

[0178] Furthermore, the cross validation may be a K-fold cross validation, where K may be 5.

[0179] In an optional embodiment, based on a stacking integration algorithm, the prediction results of several prediction methods obtained from different data structures and dimensions are analyzed and integrated based on the prediction effects of the four individual models involved in the combination, so as to analyze and learn the data more accurately and comprehensively, avoiding the concentrated influence of one or more factors on the prediction accuracy.

[0180] Specifically, first divide the data set, for the data set ,in , is the feature vector of the kth sample, p is the number of features contained, is the predicted value corresponding to the kth sample. The data is divided into a training set Strain and a test set Stest.

[0181] Secondly, perform K-fold cross validation to divide the training set Strain into basically equal subsets Strain_1, Strain_2, ..., Strain_n. After using K-fold cross validation on each base learner in the first layer, the output data of each base learner constitutes a new training set ,in .

[0182] Finally, the new training set The meta-learner obtained by summarizing the data as the input data of the second-layer learner in the stacked ensemble algorithm using the second-layer prediction algorithm is the combined prediction model.

[0183] The present invention firstly performs an effective simulation of the clean fan output based on computer fluid dynamics, and the inventor team of the present invention realizes that in open conditions such as icing, the shape and amount of ice accumulated on the fan are difficult to quantify and analyze, the output characteristics of the icing fan are difficult to obtain, and technical difficulties such as superimposed small samples can be solved by constructing a combined prediction model and correcting the output of the combined prediction model.

[0184] Furthermore, the prediction results of the combined prediction model are evaluated, and the combined prediction model is optimized according to the evaluation results.

[0185] Specifically, the prediction results can be evaluated using the root mean square error method or the mean absolute error method.

[0186] S5. Calculate the hourly power correction coefficient of the ice-covered station through ice identification and icing growth model, combined with the wind speed-power correction coefficient table under ice-covered wind turbines. Correct the power forecast value output by the short-term wind power combination forecast model based on the hourly power correction coefficient of the ice-covered station to obtain the wind power forecast value under ice-covered state.

[0187] In S5, the prediction result is corrected based on the ice power correction coefficient value, including:

[0188] Determine the icing power correction coefficient value of the station according to the icing power correction coefficient value of each wind turbine in the station;

[0189] The prediction results are corrected based on the icing power correction coefficient value of the station.

[0190] Specifically, according to the following formula, the icing power correction coefficient value of the station is determined according to the icing power correction coefficient value of each wind turbine in the station:

[0191]

[0192] in,

[0193] is the icing power correction coefficient of the station during the icing period at the i-th moment,

[0194] is the power correction coefficient value of wind turbine k at the i-th moment during the icing period,

[0195] is the rated capacity of fan k,

[0196] n is the total number of wind turbines in the station.

[0197] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0198] Combination Figure 6 As shown, embodiment 2 of the present invention provides a short-term wind power combination prediction system, which runs the short-term wind power combination prediction method under icing conditions as described in embodiment 1, and the system includes:

[0199] The first construction module is used to construct an icing label data set and establish an icing identification model based on historical meteorological data and power data.

[0200] The second construction module is used to combine multi-source meteorological forecasts to obtain the wind speed of the wind turbines in the station and build a wind turbine icing growth model.

[0201] The third construction module is used to construct a CFD simulation model of an ice-covered fan based on different wind speeds, ice coverage degrees and fan types, obtain the power characteristic curve of the ice-covered fan, and construct a wind speed-power correction coefficient table of a typical fan under ice coverage through simulation calculation.

[0202] A fourth construction module is used to establish at least two power prediction models based on different machine learning models, optimize model parameters using IPSO and / or BO algorithms, and construct and train a short-term wind power combination prediction model;

[0203] The calculation module is used to calculate the hourly power correction coefficient of the ice-covered station through ice cover identification and icing growth model, combined with the wind speed-power correction coefficient table under wind turbine ice cover, and correct the power prediction value output by the short-term wind power combination prediction model based on the hourly power correction coefficient of the ice-covered station to obtain the wind power prediction value under the ice cover state.

[0204] The system further comprises: a data acquisition module for acquiring historical meteorological data, historical power data of wind turbines, wind turbine information data, meteorological forecast data and current meteorological data, wherein the wind turbine information data comprises wind turbine type and wind turbine location.

[0205] Embodiment 3 of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the method for predicting the short-term wind power combination under icing conditions described in Embodiment 1 is implemented.

[0206] Embodiment 4 of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for predicting a short-term wind power combination under icing conditions described in Embodiment 1 is implemented.

[0207] Compared with the prior art, the beneficial effects of the present invention include at least:

[0208] The present invention identifies the icing condition of the wind turbine, and when the identification result is icing, uses a CFD simulation calculation method to calculate the wind speed of the wind turbine, and then combines the wind turbine icing growth model to quantitatively analyze the power characteristics of the icing wind turbine under different icing conditions and different wind speeds, and obtains the icing power correction coefficient value. Based on the icing power correction coefficient value, the prediction result output by the power prediction model is corrected, so that the predicted value of the wind power of the station can be obtained more efficiently and accurately, and then the station output can be obtained more efficiently and accurately. The present invention solves the problem of large deviations in wind power prediction caused by power fluctuations due to wind turbine icing caused by bad weather, reduces the difficulty of prediction, and improves the accuracy of power prediction of icing wind turbines.

[0209] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0210] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0211] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0212] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0213] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A short-term wind power combination prediction method under icing conditions, characterized in that: The method comprises: Based on historical meteorological data and power data, an icing label dataset is constructed and an icing identification model is established; Combined with multi-source weather forecasts, the wind speed of wind turbines in the station is obtained and a wind turbine icing growth model is constructed; Based on different wind speeds, icing degrees and fan types, a CFD simulation model of an icing fan is constructed to obtain the power characteristic curve of the icing fan. Through simulation calculation, a wind speed power correction coefficient table of the fan under icing is constructed. At least two power prediction models are established based on different machine learning models, and the model parameters are optimized by using the IPSO algorithm and / or the BO algorithm. The prediction results obtained by stacking the verification results cross-validated by at least two machine learning models are used as new training data sets, and a short-term wind power combination prediction model is obtained by training through a meta-model; Through ice identification and icing growth model, combined with wind speed-power correction coefficient table under wind turbine icing, the hourly power correction coefficient of the icing station is calculated. Based on the hourly power correction coefficient of the icing station, the power forecast value output by the short-term wind power combination forecast model is corrected to obtain the wind power forecast value under icing state. Calculate the icing power correction coefficient of the icing station according to the following formula: in, is the icing power correction coefficient of the station during the icing period at the i-th moment, is the power correction coefficient value of wind turbine k at the i-th moment during the icing period, is the rated capacity of wind turbine k, and n is the total number of wind turbines in the station.

2. The short-term wind power combination prediction method under icing conditions according to claim 1 is characterized by: Based on the historical data of the station, the ice cover is calibrated as 1 and the non-ice cover is calibrated as 0. By dividing the training set and the test set, the logistic regression algorithm is used to complete the creation and verification of the ice cover identification model.

3. The short-term wind power combination prediction method under icing conditions according to claim 1 is characterized by: The fan icing growth model is as follows: In the formula, is the mass of ice, in kg; is the freezing time, in seconds. The freezing start time is obtained through the icing identification model. represent the collision, capture, and freezing coefficients of supercooled water droplets, is the water drop mass fraction, a dimensionless parameter, is the relative speed of the colliding water droplet relative to the fan impeller, in m / s, is the effective cross-sectional area of ​​the object that the water droplet collides with, in m 2 .

4. The short-term wind power combination prediction method under icing conditions according to claim 3 is characterized by: The relative velocity of the impinging water droplets with respect to the fan impeller Calculated using the following formula: in, is the fan impeller tangential speed, in m / s, is the wind speed of the wind turbine obtained based on the wake model, in m / s.

5. The short-term wind power combination prediction method under icing conditions according to claim 4 is characterized by: The wind speed of the wind turbine calculated based on the wake model includes: When the current wind turbine is located in a flat terrain, the wind speed of the current wind turbine is calculated according to the first wake model; and / or, When there is no wind turbine in the front row of the current wind turbine, the wind speed of the current wind turbine is calculated according to the second wake model; and / or, When there are fans in the front row of the current fan, the wind speed of the current fan is calculated according to the third wake model.

6. The short-term wind power combination prediction method under icing conditions according to any one of claims 1 to 5, characterized in that: At least two power prediction models are established based on different machine learning models, and the short-term wind power combination prediction model is constructed and trained, including: Dividing the training data set into a plurality of subsets; the data in the training set includes historical meteorological data when there is no ice and historical power data of the wind turbine; Use at least two machine learning models for training and prediction for each subset, perform cross-validation on each machine learning model, and stack all cross-validation results to form a prediction result; The prediction results of all machine learning models are combined into a new training data set, and the meta-model is trained using the new training data set to obtain a short-term wind power combination prediction model; Therein, the machine learning model and / or meta-model is trained with the goal of minimizing the fitness function.

7. The short-term wind power combination prediction method under icing conditions according to claim 6 is characterized by: The IPSO algorithm is used to find the optimal parameters during the machine learning model training process, including: Randomly generate a population particle, wherein the population particle includes hyperparameters of the machine learning model, and define intervals for each hyperparameter of the population particle; Construct the fitness function of the particle and calculate the fitness value corresponding to each particle position; Determine the individual extreme value and group extreme value according to the initial particle fitness value, and take the individual extreme value of each particle as its historical best position; In each iteration, the particle's own speed and position are updated through individual extremum and global extremum according to the fitness value and inertia weight; When the maximum number of iterations is reached or the fitness function value is minimum, the iteration is stopped to obtain the optimal parameters.

8. The short-term wind power combination prediction method under icing conditions according to claim 7 is characterized by: Construct the fitness function of the particle, including: Constructing a fitness function of the particle based on the fitting errors of the validation sample and the training sample, wherein the weights of the fitting errors of the validation sample and the training sample are the same; The fitness function is shown as follows: in, , are the predicted value and true value of the jth sample data in the training data set, respectively. , are the predicted value and true value of the j-th sample data in the validation dataset, respectively.

9. The short-term wind power combination prediction method under icing conditions according to claim 7 is characterized by: The inertia weight is calculated using a dynamic nonlinear method, expressed as follows: in, is the maximum inertia weight, is the minimum inertia weight, t is the current iteration number, is the maximum number of iterations.

10. A short-term wind power combination prediction system using the short-term wind power combination prediction method under icing conditions according to any one of claims 1 to 9, characterized in that: include: The first construction module is used to construct an ice label data set and establish an ice recognition model based on historical meteorological data and power data; The second construction module is used to combine multi-source meteorological forecasts to obtain the wind speed of the wind turbine in the station and build a wind turbine icing growth model; The third construction module is used to construct a CFD simulation model of an ice-covered fan based on different wind speeds, ice coverage levels and fan types, obtain the power characteristic curve of the ice-covered fan, and construct a wind speed-power correction coefficient table of a typical fan under ice coverage through simulation calculation; A fourth construction module is used to establish at least two power prediction models based on different machine learning models, optimize model parameters using IPSO and / or BO algorithms, and construct and train a short-term wind power combination prediction model; The calculation module is used to calculate the hourly power correction coefficient of the ice-covered station through ice cover identification and icing growth model, combined with the wind speed-power correction coefficient table under wind turbine ice cover, and correct the power prediction value output by the short-term wind power combination prediction model based on the hourly power correction coefficient of the ice-covered station to obtain the wind power prediction value under the ice cover state.

11. An electronic device, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instruction to execute the steps of the short-term wind power combination prediction method under icing conditions according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the short-term wind power combination prediction method under icing conditions described in any one of claims 1 to 9 are implemented.

Images