A wind turbine grid management system based on wind speed prediction and adaptive control strategy

Abstract

The present invention discloses a fan power grid management system based on wind speed prediction and adaptive control strategy. The present invention relates to the technical field of wind power generation. When predicting the wind speed, the present invention divides wind turbines into several clusters through KDBSCAN, trains wind speed prediction models for each fan in the cluster by using HOT-RNN and combines these models, obtains the thresholds corresponding to each prediction model through an optimization algorithm, introduces a kernel ridge regression model to find the non-linear function relationship between the predicted value and the prediction error, and obtains the total wind speed prediction model. This prediction model considers the non-linear situations of multiple fans and actual wind speed data, making the wind speed prediction more accurate. At the same time, for the moments when the wind speed fluctuates greatly, an adaptive angle of attack increment and an adaptive rotational speed adjustment rate are respectively introduced, so that each fan can convert electrical energy to the greatest extent according to the actual change of the wind speed and ensure the safety of the equipment itself.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and particularly to a fan power grid management system based on wind speed prediction and adaptive control strategy. Background Art

[0002] In the field of wind power generation, the uncertainty of wind speed and the coordination problem between the fan and the power grid have always troubled the development of the industry. Especially in wind speed prediction, traditional prediction methods only predict the wind speed of a single fan of a wind turbine. However, due to the influence of factors such as the position and environment of each fan, using only one for prediction has poor effect. Moreover, the wind speed data itself has extremely strong nonlinearity, and it is also extremely important to select a suitable wind speed prediction model. At the same time, traditional fan control methods are difficult to accurately adapt to wind speed changes, resulting in low fan power generation efficiency and the impact on the stability of the power grid. For example, when the wind speed changes suddenly, the output power of the fan fluctuates violently, easily exceeding the acceptance capacity of the power grid and causing power grid failures; and inaccurate wind speed prediction makes it impossible to reasonably allocate energy storage devices, resulting in energy waste or insufficient power supply. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the present invention provides a fan power grid management system based on wind speed prediction and adaptive control strategy, which solves the problems of inaccurate wind speed prediction of a single fan and the difficulty in coordinating the fan and the power grid.

[0004] To achieve the above object, the present invention is realized through the following technical solutions: A fan power grid management system based on wind speed prediction and adaptive control strategy, comprising:

[0005] A wind speed prediction module, which uses the KDBSCAN density clustering algorithm to divide wind turbines into n clusters, each cluster has Vc (c ∈ [1, n]) similar fans, trains each fan in the n clusters using HOT-RNN, combines the trained Vc similar fans in each cluster, obtains their respective corresponding weights through an optimization algorithm, and at the same time introduces a kernel ridge regression model to determine the relationship between the prediction error and the predicted value, and constructs a final wind speed prediction model;

[0006] An energy storage control module, which continuously monitors the wind speed state through the wind speed prediction module, calculates the wind speed change amount k at adjacent moments. If |k| > d, the energy storage device is activated. If |k| <= d, the energy storage device is not activated, where d is the wind speed change amount threshold;

[0007] A fan blade angle of attack adaptive control module. If |k| > d and k > 0, according to the formula Set an adaptive angle of attack increment, and adjust the angle of attack of the wind turbine blades through the adaptive angle of attack increment. If P(t)>Pg(t), activate the energy storage device for charging, where P(t) is the wind turbine power generation at time t, Pg(t) is the stable power that the power grid can accept at time t, ɑ is the current angle of attack, ɑ max is the maximum value of the blade angle of attack, and m is the proportionality coefficient;

[0008] For the wind turbine speed adaptive control module, if |k|>d and k<0, according to the formula obtain the wind turbine adaptive speed adjustment rate, and adjust the wind turbine speed according to the adaptive speed adjustment rate. If Pr(t)>P(t), activate the energy storage device for discharging, where Pr(t) is the real-time power demand of the power grid, R max is the maximum allowable speed change rate of the wind turbine, N(t) is the current wind turbine speed, and n1 is a constant.

[0009] As a further solution of the present invention, the specific steps for dividing the wind turbine units by using the KDBSCAN density clustering algorithm include:

[0010] Collect multi-dimensional data of the wind turbine units, including wind speed v, power p, longitude lon, latitude lat, and generator speed g, and construct a data set X={x1,x2,...,xm}, where xi=(vi,pi,loni,lati,gi), and m is the number of wind turbines in the wind turbine units;

[0011] Perform Z-score standardization processing on each xi in the data set. If there are missing values, mean filling can be used;

[0012] Determine the core point neighborhood radius eps according to the scale and distribution density of the wind turbine units, and set the minimum number of points min_samples;

[0013] According to the formula calculate the number of points within the eps neighborhood of the point xi. If Neps(xi)>=min_samples, then xi is a core point, where d(xi,xj) is the Euclidean distance between two points, and I(·) is the indicator function, which is 1 if satisfied, otherwise 0;

[0014] Starting from the core point xi, if xj is within the eps neighborhood of xi, then classify xj and xi into the same cluster;

[0015] According to the formula calculate the silhouette coefficient S to evaluate the clustering effect, and adjust the parameters accordingly to optimize the clustering, where ai is the average distance from xi to other points in the same cluster, and bi is the average distance from xi to all points in the nearest neighboring cluster.

[0016] As a further solution of the present invention, the specific method for training the fan using HOT-RNN is as follows:

[0017] Collect a large amount of historical operation data from the monitoring system of the wind turbine, including wind speed, wind direction, power output, humidity, ambient temperature, and air pressure, and construct a multivariate original data set;

[0018] Perform Min-Max normalization and data smoothing operations on the original data;

[0019] Divide the preprocessed data into a training set and a test set according to 80% and 20%, and convert the divided multivariate time series data into a high-order tensor form;

[0020] Determine the network architecture parameters, including the number of hidden layers, the number of neurons in each layer, the tensor order, and the time step;

[0021] Based on the training set data, use the mean square error MSE and the mean absolute error MAE as measures of the deviation between the predicted value and the true value, and use Adam to iteratively update the model parameters;

[0022] Substitute the test set data into the trained prediction model, and calculate MSE, MAE, and the prediction accuracy rate to comprehensively measure the prediction accuracy, stability, and reliability of the model. If the requirements are not met, return to the above steps to continue training the model.

[0023] As a further solution of the present invention, the optimization algorithm is selected from one of the goose flock optimization algorithm, the black-winged kite optimization algorithm, the genetic optimization algorithm, the simulated annealing optimization algorithm, and the particle swarm optimization algorithm.

[0024] As a further solution of the present invention, the specific steps for introducing the kernel ridge regression model to determine the relationship between the prediction error and the predicted value are as follows:

[0025] Collect the predicted values of the Vc group prediction models as A1p, A2p,..., AVcp and the corresponding errors dp, and construct a data set where g is the total amount of predicted data;

[0026] According to the formula Map the input data to a high-dimensional feature space and calculate the kernel matrix K, where xp = (A1 p, A2p,..., AVcp), xq = (A1 q, A2q,..., AVcq), and σ is the kernel bandwidth parameter;

[0027] Let the regression coefficient vector be a = (a0, a1,..., aVc) T , from Take it as the minimization objective function, where λ is the regularization parameter;

[0028] Take the partial derivative of J(a) with respect to a and set it to 0 to determine the regression coefficient a;

[0029] Based on the regression coefficient a, construct the functional relationship f between the prediction error and the predicted value:

[0030]

[0031] As a further solution of the present invention, the constructed final wind speed prediction model is w1A1 + w2A2 +... + wVcAVc + f(A1, A2,..., AVc), where w1, w2,..., wVc are the weights calculated by the optimization algorithm.

[0032] As a further solution of the present invention, according to the formula Adjust the energy storage device for charging, where Ps(t) is the charging power of the energy storage device.

[0033] As a further solution of the present invention, according to the formula Adjust the energy storage device for discharging, where Pd(t) is the discharging power of the energy storage device.

[0034] The present invention provides a fan power grid management system based on wind speed prediction and adaptive control strategy, which has the following beneficial effects compared with the prior art:

[0035] (1) In the present invention, the KDBSCAN density clustering algorithm is used to cluster the fans in the wind turbine group, and the fans with similar operating states can be grouped into one cluster. Wind speed prediction models are constructed for each cluster respectively, focusing on training and optimization of the common characteristics within the cluster, making the model more suitable for the actual working conditions. Compared with the overall unified modeling, it can greatly reduce the influence of wind speed random fluctuations;

[0036] (2) In the present invention, the HOT-RNN is used to predict the wind speed of each fan in each cluster. The wind speed, wind direction, power output, humidity, ambient temperature, and air pressure are constructed in the form of a high-order tensor and input into the model to accurately capture the complex non-linear correlations and dynamic change laws of various factors, mine deep-level information, comprehensively reflect the characteristics of the fan operating environment, and enable the wind speed prediction to fully consider the comprehensive effects of multiple factors, thereby improving the prediction accuracy;

[0037] (3) In the present invention, the kernel ridge regression model is used to determine the relationship between the prediction error and the predicted value. In the face of the change of wind speed fluctuation characteristics, this model can dynamically optimize the prediction value correction strategy to ensure that the wind speed prediction is close to the real situation;

[0038] (4) In the present invention, for the moments when the wind speed fluctuates greatly, an adaptive angle of attack increment and an adaptive rotational speed adjustment rate are respectively introduced, both of which can maximize the conversion of electrical energy by the fan according to the actual change of the wind speed and ensure the safety of the equipment itself. Description of the Drawings

[0039] Figure 1 This is the system principle block diagram of the present invention. Specific implementation manners

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0041] Such as Figure 1 , this application provides a wind turbine grid management system based on wind speed prediction and adaptive control strategy, including:

[0042] A wind speed prediction module divides the wind turbines into n clusters through the KDBSCAN density clustering algorithm, and each cluster has Vc (c ∈ [1, n]) similar wind turbines:

[0043] Collect multi-dimensional data of wind turbines, including wind speed v, power p, longitude lon, latitude lat, and generator speed g, and construct a data set X = {x1, x2,..., xm}, xi = (vi, pi, loni, lati, gi), where m is the number of wind turbines in the wind farm;

[0044] Perform Z-score standardization processing on each xi in the data set, that is where xij is the j-th feature value of the i-th sample, μ j and σ' j are the mean and standard deviation of the j-th feature, making the mean of each feature 0 and the standard deviation 1 to avoid the influence of inconsistent dimensions on the clustering effect;

[0045] If there are missing values, mean filling can be used to ensure the integrity of the data;

[0046] It is initially determined according to the scale of the wind turbines and the density of the distribution of the turbines. For example, eps = D / sqrt(sum(Vc)), D is the geographical range diameter of the wind turbines, sum(Vc) represents the total number of wind turbines in the wind farm, and subsequent optimization is carried out according to the clustering effect. Considering the data noise and the complexity of the expected cluster structure, set the initial minimum number of points, such as min_samples = In(sum(Vc)), and then optimize according to the actual situation;

[0047] According to the formula Calculate the number of points within the eps neighborhood of the point xi. If Neps(xi) >= min_samples, then xi is a core point, where d(xi, xj) is the Euclidean distance between two points, and I(·) is an indicator function that is 1 if the condition is met and 0 otherwise;

[0048] Starting from the core points, construct clusters according to density reachability: If xj is within the eps neighborhood of the core point xi, then xj is density reachable from xi, and points that are mutually density reachable are grouped into one cluster;

[0049] According to the formula Calculate the silhouette coefficient S to evaluate the clustering effect. The mean of the silhouette coefficients reflects the overall clustering quality. A value close to 1 indicates good clustering, and a value close to -1 indicates poor clustering. Adjust the parameters accordingly to optimize the clustering, where ai is the average distance from xi to other points in the same cluster, and bi is the average distance from xi to all points in the nearest neighboring cluster.

[0050] Use HOT-RNN to train each wind turbine in n clusters, specifically including:

[0051] Collect various operating data of multiple wind turbines from the monitoring system of the wind farm, including variables such as wind speed, wind direction, power, temperature, and air pressure, as well as the corresponding timestamp information. These data constitute the original multivariate time series dataset, denoted as X = {(x1t, x2t,..., xqt), t ∈ [1, T]}, where q represents the number of variables, T represents the total length of the time series, and xit represents the observed value of the i-th variable at time step t;

[0052] For each variable i, Normalize it to the interval [0, 1] and use the moving average method to smooth the normalized data to reduce noise interference in the data. Usually, the smoothing window is set to r, and the formula for the smoothed data is For example, if the window size r = 5 is selected, for the smoothing calculation of the wind speed data at time step t = 10, the normalized wind speed values from the 6th to 10th time steps are added and averaged as the smoothed wind speed value. This helps the model better capture the trend and periodic characteristics of the data, rather than being misled by short-term noise fluctuations;

[0053] Divide the preprocessed data into a training set and a test set in a ratio of 8:2. Let the training set data be Xtrain = {(x1t, x2t,..., xqt), t ∈ [1, Ttrain]}, and the test set data be Xtest = {(x1t, x2t,..., xqt), t ∈ [Train + 1, T]}, where Ttrain = 0.8T;

[0054] Convert the divided multivariate time series data into a high-order tensor form. The selection of the tensor order needs to be determined according to the multivariate and multi-time-step characteristics of the data. It is necessary to make full use of the structural information of the data so that the model can learn the interaction relationships between variables and between different time steps. At the same time, it is also necessary to adjust the time step according to the periodicity of the data and the prediction requirements. For example, if it is found that the wind speed shows a relatively stable periodic characteristic within 24 time steps (assuming each time step is 1 hour), the time step can be set to 24;

[0055] Taking a third-order tensor as an example, if the time step is set to m1, the data can be organized into a tensor with a shape of (N, m1, q). N is the number of samples (for the training set, N = Train / m1; for the test set, N = (T - Ttrain) / m1). For example, for a data set with three variables of wind speed, wind direction, and power and a time step of 10, if there are a total of 1000 time steps of training data, then for the training set, N = (1000 * 0.8) / 10 = 80, and the shape of the tensor is (80, 10, 3). In this tensor, the first dimension represents different sample sequences, the second dimension represents the time step, and the third dimension represents the variable;

[0056] Determine the appropriate number of hidden layers according to the complexity of the data and the requirements of the model's expressive ability. For wind speed prediction when the surrounding environment is stable, 2 - 3 hidden layers can be set. For prediction when the surrounding environment is complex and the wind speed fluctuates greatly, 3 - 5 hidden layers can be set;

[0057] The number of neurons in each layer also needs to be adjusted according to the data characteristics. For the case where the input data dimension is relatively high (many variables or long time steps), the number of neurons can be appropriately increased;

[0058] Based on the training set data, using the mean squared error MSE and mean absolute error MAE as measures of the deviation between the predicted value and the true value, the Adam optimizer will automatically adjust the learning rate of each parameter according to the gradient change situation of different parameters. For parameters with large gradient changes, the learning rate is appropriately reduced, and for parameters with small gradient changes, a relatively large learning rate is maintained, thereby improving the efficiency and stability of model training;

[0059] Load the training set data into the model in small batches for training. If the batch size is b, then in each training epoch, N / b batch training needs to be performed. For each small batch, training is carried out according to the above steps. After traversing all batches of data, one round of training is completed;

[0060] Substitute the test set into the trained prediction model, and calculate evaluation metrics such as MSE, MAE, and prediction accuracy in the same way as the training process. If the performance of the model on the test set fails to meet the preset requirements, return to the above steps to continue training the model, which may involve the following operations, such as increasing or decreasing the number of hidden layers, adjusting the number of neurons, changing the tensor order or time step, or adjusting data preprocessing, such as modifying the normalization method, adjusting the window size of data smoothing, or adjusting the weights of the loss function or optimizer parameters, until the prediction effect of the prediction model on the test set reaches the preset standard.

[0061] Combine the Vc similar wind turbines trained in each cluster, and obtain their corresponding weights through algorithms such as the goose flock optimization algorithm, the black-winged kite optimization algorithm, and the genetic optimization algorithm:

[0062] For the selected optimization algorithm, determine its required specific parameters. For example, for the goose flock and the black-winged kite, it is necessary to set the population size N1, the maximum number of iterations Tmax, etc. For the genetic algorithm, it is also necessary to set the crossover probability Pc and the mutation probability Pm. At the same time, determine the range of the weights, set it as [wmin, wmax], and randomly initialize the weight combinations of the Vc models in each cluster within this range to form an initial population. Each individual can be expressed as Li = (wi1, wi2,..., wiVc) (i ∈ [1, N1]);

[0063] For each individual Li in the current population, substitute it into the wind speed prediction model wi1A1 + wi2A2 +... + wiVcAVc + f(A1, A2,..., AVc), combine the known historical wind speed data, calculate the predicted wind speed value of this model, and calculate the fitness value F(Li) according to the error between the predicted wind speed value and the actual wind speed value. Define it as F(Li) = 1 / (1 + MSE(Li)), so that the larger the fitness value, the better the prediction effect of the weight combination;

[0064] For algorithms such as the genetic algorithm that also include selection operations, select the individuals in the current population according to the fitness value. Common selection algorithms include roulette wheel selection, tournament selection, etc. Through the selection operation, retain the excellent weight combination individuals;

[0065] Determine the optimal individual in the population, such as the leading goose of the goose flock and the guiding individual of the black-winged kite. Its position represents the optimal solution of the current weight. Then, according to the specific update rules of the algorithm, let other individuals approach the optimal individual or explore new positions around it. The updated individual positions need to be processed at the boundaries to ensure that each weight value is still within the range of [wmin, wmax];

[0066] If it is a genetic algorithm, it is also necessary to perform a crossover operation on the selected individuals. Select two individuals (parent generations) with probability Pc, exchange some genes (weight values) in their chromosomes (weight combinations), and generate new individuals (offspring). In addition, operations such as mutation can be performed on the offspring to randomly change some gene values (weight values) in the chromosome to enhance the diversity of the population and avoid falling into local optima;

[0067] Check whether the termination condition is satisfied. Common termination conditions include reaching the maximum number of iterations Tmax, the population fitness converging (i.e., the change in fitness values for several consecutive generations is less than the set threshold), or finding a weight combination that meets specific accuracy requirements. If the termination condition is satisfied, stop the algorithm iteration and output the weight combination Li corresponding to the optimal individual in the current population as the optimal result. If not, continue the iterative optimization.

[0068] At the same time, introduce a kernel ridge regression model to determine the relationship between the prediction error and the predicted value, and construct the final wind speed prediction model:

[0069] Collect the predicted values A1p, A2p,..., AVcp of the prediction model in the Vc group and the corresponding errors dp, and construct a data set where g is the total amount of predicted data;

[0070] By introducing a kernel function, convert the low-dimensional predicted value into a high-dimensional feature representation to enhance the nonlinear fitting ability of the model. The Gaussian kernel function is selected here. According to the formula Map the input data to the high-dimensional feature space and calculate the kernel matrix K. K is positive definite and has a dimension of g×g. Among them, xp = (A1 p, A2p,..., AVcp), xq = (A1 q, A2q,..., AVcq), and σ is the kernel bandwidth parameter;

[0071] Let the regression coefficient vector be a = (a0, a1,..., aVc) T , from as the minimization objective function, where λ is the regularization parameter;

[0072] Take the partial derivative of J(a) with respect to a and set it to 0 to obtain the linear variance group as:

[0073] where Q is a g-dimensional all-1 vector, E is a g×g identity matrix, and d = (d1, d2,..., dg) T , solve the equation to obtain the regression coefficient a;

[0074] Based on the regression coefficient a, construct the functional relationship f between the prediction error and the predicted value:

[0075]

[0076] The final wind speed prediction model is \(w_1A_1 + w_2A_2+\cdots+w_{Vc}A_{Vc}+f(A_1,A_2,\cdots,A_{Vc})\).

[0077] The energy storage control module continuously monitors the wind speed status through the wind speed prediction module, calculates the wind speed change amount \(k=(v(t_2)-v(t_1)) / (t_2 - t_1)\) at adjacent moments. If \(|k|>d\), it means that the wind speed fluctuates violently at this time, and the energy storage device needs to be activated to regulate the balance of power transmission between the wind turbine and the power grid. If \(|k|\leq d\), it means that the wind speed changes smoothly at this time, and the energy storage device does not need to be activated. Here, \(v(t_1)\) and \(v(t_2)\) are the wind speeds at times \(t_1\) and \(t_2\), and \(d\) is the threshold of the wind speed change amount.

[0078] For the wind turbine blade pitch angle adaptive control module, if \(|k|>d\) and \(k > 0\), the blade pitch angle of the wind turbine is adjusted according to the adaptive pitch angle increment:

[0079] Let the maximum blade pitch angle be \(\alpha\) max , which is usually determined by the aerodynamic design of the wind turbine blades, material strength, and the tolerance of the overall mechanical structure. For most common horizontal axis wind turbines, this value is generally in the range of 20 - 25 degrees;

[0080] Under the adaptive control strategy, the pitch angle increment \(\Delta\alpha\) is not only related to the wind speed change amount \(k\), but also needs to consider the distance between the current pitch angle \(\alpha\) and the maximum value \(\alpha\) max . Specifically, the adaptive pitch angle increment satisfies:

[0081] where \(m\) is a proportionality coefficient fitted according to the aerodynamic characteristics of the wind turbine and the power generation efficiency curve;

[0082] For example, a certain wind turbine unit has a 2.5 MW wind turbine, the maximum pitch angle \(\alpha\) max \(= 22^{\circ}\), the current blade pitch angle \(\alpha = 8^{\circ}\), \(m = 3\), and the wind speed change amount \(k = 2\). The calculated pitch angle increment according to the formula is \(\Delta\alpha=3\times2 = 6^{\circ}\). Then the new pitch angle is \(\alpha+\Delta\alpha = 14^{\circ}\), which is within the range of the maximum pitch angle. Therefore, the pitch angle can be increased by 6° normally, and the wind turbine impeller can capture wind energy more efficiently to improve the power generation efficiency, meeting the performance optimization requirements of the current wind speed rising condition. On the contrary, in an extreme weather condition, affected by the wind speed fluctuation, the current pitch angle of the wind turbine has reached \(20^{\circ}\), and the wind speed changes suddenly, \(k = 1.5\). Calculated, \(\Delta\alpha=3\times1.5 = 4.5^{\circ}\). If the formula is followed at this time, the pitch angle will exceed the maximum pitch angle. So the actual \(\Delta\alpha=22 - 20 = 2^{\circ}\), only increasing the pitch angle by 2°, avoiding the risk of stall and structural damage of the wind turbine due to excessive increase of the pitch angle, ensuring the dynamic adjustment of the wind turbine within the safe and efficient operation boundary in the scenario of rapid wind speed change, and maintaining the balance between power generation stability and equipment reliability;

[0083] With the sudden increase in power generation, in order to maintain the stability of the power grid, the excess electricity needs to be properly disposed of. The energy storage device accurately absorbs the excess electric energy according to the real-time power difference. Let the power generation of the wind turbine at time t be P(t), the stable power that the power grid can accept be Pg(t), and the charging power of the energy storage device Ps(t) follows:

[0084] Ensure the stable power injected into the power grid and avoid the instability of the power grid frequency and voltage caused by the sudden increase in wind power;

[0085] For the wind turbine speed adaptive control module, if |k|>d and k<0, adjust the wind turbine speed according to the adaptive speed adjustment rate:

[0086] When the wind speed decreases rapidly, setting the adaptive speed adjustment rate can not only protect the mechanical components of the wind turbine but also take into account the power generation efficiency. The key here is to dynamically generate a reasonable speed adjustment instruction based on the real-time state of the wind turbine and the change in wind speed to ensure the smooth transition of the wind turbine through the wind speed mutation period;

[0087] Let the maximum allowable speed change rate that the mechanical components of the wind turbine can withstand be Rmax, the current wind turbine speed be N(t), and the constant based on the mechanical inertia of the wind turbine and the performance of the speed control system be n. The specific adaptive speed adjustment rate satisfies:

[0088]

[0089] When the speed adjustment rate does not exceed Rmax and the adjusted speed will not drop to a negative value, adjust at a rate adapted to the wind speed change so that the wind turbine can closely follow the wind speed and flexibly decelerate;

[0090] When the speed adjustment rate exceeds Rmax but the adjusted speed is still not negative, to protect the equipment components, force the adjustment rate to be set to Rmax to ensure that there is no mechanical damage caused by excessive deceleration impact;

[0091] When the speed becomes negative, it means that the large decrease in wind speed may cause the wind turbine to stop naturally. At this time, the current speed should be maintained, waiting for the wind speed to rise or further instructions later to avoid abnormal braking of the wind turbine;

[0092] At the same time, turn on the discharge mode of the energy storage device to fill the instantaneous power gap of the power grid and maintain the power supply and demand balance. The discharge power Pd(t) is dynamically adjusted according to the real-time power deficit of the power grid. The real-time power demand of the power grid is Pr(t), and the discharge power Pd(t) of the energy storage device follows:

[0093] Ensure the continuous and stable power supply of the key loads of the power grid and always ensure the safe and efficient operation of the wind power equipment.

[0094] Some of the data in the above formula are numerically calculated after removing their dimensions, and the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0095] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A wind turbine power grid management system based on wind speed prediction and adaptive control strategy, characterized in that, Including: A wind speed prediction module that uses the KDBSCAN density clustering algorithm to divide wind turbines into n clusters. Each cluster has Vc (c ∈ [1, n]) similar wind turbines. The HOT-RNN is used to train each wind turbine in the n clusters, and the trained Vc similar wind turbines in each cluster are combined. The respective weights are obtained through an optimization algorithm, and at the same time, a kernel ridge regression model is introduced to determine the relationship between the prediction error and the predicted value, and a final wind speed prediction model is constructed. A energy storage control module that continuously monitors the wind speed state through the wind speed prediction module, calculates the wind speed change k at adjacent moments. If |k| > d, the energy storage device is activated; if |k| ≤ d, the energy storage device is not activated, where d is the wind speed change threshold. The fan angle of attack adaptive control module, if |k|>d and k>0, according to the formula sets the adaptive angle of attack increment, and adjusts the fan blade angle of attack through the adaptive angle of attack increment. If P(t)>Pg(t), the energy storage device is activated for charging. Wherein, P(t) is the fan power generation power at time t, Pg(t) is the stable power that the power grid can accept at time t, ɑ is the current angle of attack, ɑ max is the maximum value of the blade angle of attack, and m is the proportionality coefficient; The fan speed adaptive control module, if |k|>d and k<0, according to the formula obtains the fan adaptive speed adjustment rate, and adjusts the fan speed according to the adaptive speed adjustment rate. If Pr(t)>P(t), the energy storage device is activated to discharge. Among them, Pr(t) is the real-time power demand of the power grid, R max is the maximum allowable speed change rate of the fan, N(t) is the current fan speed, and n1 is a constant.

2. The wind turbine grid management system based on wind speed prediction and adaptive control strategy according to claim 1, wherein The specific steps for dividing wind turbines using the KDBSCAN density clustering algorithm include: Collect multi-dimensional data of wind turbines, including wind speed v, power p, longitude lon, latitude lat, and generator speed g, and construct a data set X = {x1, x2,..., xm}, where xi = (vi, pi, loni, lati, gi), and m is the number of wind turbines in the wind farm. Perform Z-score normalization on each xi in the data set. If there are missing values, use mean filling. Determine the core point neighborhood radius eps according to the scale and distribution density of the wind turbines, and set the minimum number of points min_samples. According to the formula Calculate the number of points within the eps neighborhood of point xi. If Neps(xi) >= min_samples, then xi is a core point, where d(xi, xj) is the Euclidean distance between two points, and I(·) is the indicator function, which is 1 if satisfied and 0 otherwise; Starting from the core point xi, if xj is within the eps neighborhood of xi, then xj and xi are grouped into one cluster. According to the formula Calculate the silhouette coefficient S to evaluate the clustering effect, and adjust the parameters to optimize the clustering accordingly. Among them, ai is the average distance from xi to other points in the same cluster, and bi is the average distance from xi to all points in the nearest neighboring cluster.

3. A fan power grid management system based on wind speed prediction and adaptive control strategy according to claim 1, characterized in that, The specific method for training wind turbines using HOT-RNN is: Collect a large amount of historical operation data from the monitoring system of wind turbines, including wind speed, wind direction, power output, humidity, ambient temperature, and air pressure, and construct a multi-variable original data set. Perform Min-Max normalization and data smoothing operations on the original data. Divide the preprocessed data into a training set and a test set according to 80% and 20%, and convert the divided multi-variable time series data into a high-order tensor form. Determine the network architecture parameters, including the number of hidden layers, the number of neurons in each layer, the tensor order, and the time step. Based on the training set data, use the mean square error MSE and the mean absolute error MAE as measures of the deviation between the predicted value and the true value, and use Adam to iteratively update the model parameters. Substitute the test set data into the trained prediction model, and calculate MSE, MAE, and the prediction accuracy rate to comprehensively measure the prediction accuracy, stability, and reliability of the model. If the requirements are not met, return to the above steps to continue training the model.

4. A wind turbine power grid management system based on wind speed prediction and adaptive control strategy according to claim 1, wherein the optimization algorithm selects one of the goose flock optimization algorithm, the black-winged kite optimization algorithm, the genetic optimization algorithm, the simulated annealing optimization algorithm, and the particle swarm optimization algorithm.

5. A wind turbine power grid management system based on wind speed prediction and adaptive control strategy according to claim 1, characterized in that, The specific steps for introducing a kernel ridge regression model to determine the relationship between the prediction error and the predicted value are: Collect the predicted values A1p, A2p,..., AVcp of the prediction model for the Vc group and the corresponding errors dp to construct a data set where g is the total amount of predicted data; According to the formula map the input data to a high-dimensional feature space and calculate the kernel matrix K, where xp = (A1 p, A2p,..., AVcp), xq = (A1 q, A2q,..., AVcq), and σ is the kernel bandwidth parameter, where p and q are sample indices; Let the regression coefficient vector be a = (a0, a1,..., aVc) T , and use as the minimized objective function, where λ is the regularization parameter; Take the partial derivative of J(a) with respect to a and set it to 0 to determine the regression coefficient a. Based on the regression coefficient a, construct a functional relationship f between the prediction error and the predicted value:

6. The fan power grid management system based on wind speed prediction and adaptive control strategy according to claim 1, characterized in that The finally constructed final wind speed prediction model is \(w_1A_1 + w_2A_2+\cdots+w_{V_c}A_{V_c}+f(A_1,A_2,\cdots,A_{V_c})\), where \(w_1, w_2,\cdots,w_{V_c}\) are the weights calculated by the optimization algorithm.

7. A fan power grid management system based on wind speed prediction and adaptive control strategy according to claim 1, adjusts the energy storage device for charging according to the formula wherein, \(P_s(t)\) is the charging power of the energy storage device.

8. A wind turbine grid management system based on wind speed prediction and adaptive control strategy according to claim 1, characterized in that, Adjust the energy storage device to discharge according to the formula where Pd(t) is the discharge power of the energy storage device.

Images