Wind turbine generator axial thrust prediction and reduction control method and equipment

Abstract

The invention discloses a wind turbine generator axial thrust prediction and reduction control method and equipment, and aims to solve the problems that a complete and effective model is difficult to establish and real-time control measurement is difficult to carry out on the axial thrust of a wind turbine generator by adopting a machine learning method and utilizing the advantage that a BP neural network can approach a nonlinear relation. The real-time rotating speed, the pitch angle and the wind speed of the unit serve as input, the real-time thrust of the unit is predicted, meanwhile, a sparrow optimization algorithm based on chaotic mapping is adopted for optimizing the weight and the threshold value of the BP neural network, and the accuracy of approximation of the BP neural network to the nonlinear model of the thrust of the unit is improved; according to the feedback control strategy, the axial thrust of the unit near the rated wind speed can be rapidly and effectively reduced, the axial thrust is kept within the safety limit value, and the loss of electric power is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power control, and in particular relates to a method and device for predicting and reducing the axial thrust of a wind turbine generator set. Background Art

[0002] With the development of larger and more intelligent wind turbines, while intelligent control of wind turbines is being pursued, greater attention is being paid to load variations during operation. However, while fatigue loads are currently receiving greater attention, with corresponding load reduction controls implemented, less attention is being paid to peak loads. Axial thrust in wind turbines is one of the most common peak loads. When a wind turbine operates near rated wind speed, the axial thrust of the turbine increases rapidly, causing turbine components to experience greater thrust than designed, and more seriously, creating safety issues.

[0003] With the expansion of wind turbine installations, wind turbines are now moving towards larger and more intelligent units. Blade length and hub height have increased significantly. Currently, the world's longest blade is 123 meters long, and the proportion of wind turbines with rotor diameters greater than 150 meters is increasing year by year. Increased blade length increases the area swept by the rotor, increasing the axial thrust of the nacelle. This results in greater peak loads on the wind turbine, severely impacting the service life of turbine components and increasing operational risks. Therefore, given the trend toward larger wind turbines, suppressing peak loads and limiting axial thrust are crucial and urgent.

[0004] Doubly-fed variable-speed, variable-pitch wind turbines are still the mainstream wind turbine model. Given the blade shape and structural design of variable-speed, variable-pitch wind turbines, when wind speeds are below the rated speed, the turbine's axial thrust continues to increase as the wind speed rises. When the wind speed reaches near the rated speed, the turbine's axial thrust rapidly rises. After reaching its peak, the turbine begins to change pitch, limiting wind energy absorption and reducing axial thrust. Furthermore, as wind speed rises, the pitch angle increases, reducing axial thrust. Traditional thrust reduction control methods typically use a lookup table, using the turbine speed to set a conservative pitch angle to limit peak thrust. However, as wind turbines become larger and their inertia increases, the speed may not have fully changed by the time the wind speed reaches near the rated speed. This can lead to untimely control, causing axial thrust to exceed safety limits. Furthermore, the overly conservative speed-pitch angle lookup table design wastes a significant amount of wind energy during thrust reduction, resulting in power loss. To solve the above problems, some experts use the table lookup method to perform peak-shaving control on the unit, which achieves the effect of thrust reduction by giving a pitch angle before the rated speed and superimposing the pitch control command and then transmitting it to the actuator; some experts also designed an envelope protection control algorithm, which performs real-time detection of various states of the wind turbine by detailed modeling of the output itself, so as to ensure that the operation of the unit is always kept within a safe range, thereby ensuring that the thrust does not exceed the limit value; or by modeling the thrust of the wind turbine, the relevant state space expression is obtained, and on this basis, the model prediction is adopted Control is used to estimate the thrust of the unit in the next state and to give the next pitch instruction angle to achieve the control effect of thrust reduction; some technicians use the Kalman filter algorithm to estimate the wind speed of the wind turbine, and then find the minimum pitch angle within the safety limit of thrust under the current wind speed based on the three-dimensional surface diagram of the tip speed ratio, pitch angle, and unit thrust coefficient, and use it as the pitch control instruction; or based on the standard pitch PID controller, the minimum pitch angle limit curve is obtained by linear interpolation of two pre-set pitch angles to limit the generation of thrust spikes, thereby achieving the control effect of thrust reduction. Among the above methods, traditional table lookup control has the problems of untimely and inaccurate control, and other control methods rely too much on accurate models, making it difficult to achieve axial thrust reduction control of wind turbines and achieve as fast and error-free control as possible. Summary of the Invention

[0005] The present invention aims to reduce the hidden dangers caused by the axial peak thrust of wind turbines through thrust prediction and thrust reduction control strategies based on optimized neural networks.

[0006] The first object of the present invention is to provide the following technical solution: a method for predicting and reducing the axial thrust of a wind turbine generator set, comprising:

[0007] Obtain historical operating data of wind turbines during stable operation, and construct a training dataset for wind turbine axial thrust prediction based on the historical operating data;

[0008] Based on the optimized BP neural network, a wind turbine axial thrust prediction model is constructed and trained using a training data set.

[0009] Inputting the real-time collected wind turbine operation data into the trained wind turbine axial thrust prediction model, and outputting the result as the wind turbine axial thrust prediction result;

[0010] According to the prediction result of the axial thrust of the wind turbine generator set, the operating data of the wind turbine generator set is adjusted to achieve the reduction control of the axial thrust of the wind turbine generator set.

[0011] Furthermore, historical operating data of the wind turbine during stable operation is obtained, and based on the historical operating data, a training data set for wind turbine axial thrust prediction is constructed, including:

[0012] The torque control of the wind turbine adopts the standard torque control of the wind turbine, and the pitch control adopts the second-order active disturbance rejection controller. The operation data generated by the preset time interval of the stable operation of the wind turbine is obtained as the historical operation data;

[0013] The wind speed, generator speed, pitch angle and axial thrust in the historical operation data are selected as the training data set for wind turbine axial thrust prediction.

[0014] Furthermore, based on the optimized BP neural network, a wind turbine axial thrust prediction model is constructed, including:

[0015] Based on the Tent chaos map, the initial population position of the sparrow optimization algorithm is generated to optimize the sparrow optimization algorithm;

[0016] Based on the optimized sparrow optimization algorithm, the initial weights and thresholds of the BP neural network are optimized and calculated, and the overall mean square error of the training data set is used as the fitness function to determine the most appropriate initial weights and thresholds of the BP neural network.

[0017] The BP neural network optimized by the sparrow optimization algorithm is used as the wind turbine axial thrust prediction model.

[0018] Furthermore, the wind turbine axial thrust prediction model consists of a three-layer structure: input layer, hidden layer, and output layer. After training the nodes of different hidden layers, the axial thrust of the turbine is used as the output, and the wind speed, generator speed, and pitch angle, which have a strong relationship with the axial thrust, are used as the three inputs of the network input layer.

[0019] Among them, the hidden layer input and output expressions are:

[0020]

[0021] Where: net i (2) (k) is the input of the hidden layer, O i (2) (k) is the output of the hidden layer, f(x) is the Sigmoid activation function, ω ij (2) is the weight coefficient corresponding to the hidden layer; k is the variable number, and x refers to the input variable of the current layer;

[0022] Output layer input and output expressions:

[0023]

[0024] Where: net l (3) (k) is the input of the output layer, O l (3) (k) is the output of the output layer, g(x) is the Sigmoid activation function, ω li (3) is the weight coefficient corresponding to the output layer.

[0025] Furthermore, the Tent chaos mapping formula is expressed as:

[0026]

[0027] Among them, a is the main parameter of Tent chaos mapping, and its value is between 0 and 1; X n The initial population position for the sparrow optimization algorithm.

[0028] Furthermore, based on the optimized sparrow optimization algorithm, the initial weights and thresholds of the BP neural network are optimized and calculated, and the overall mean square error of the training data set is used as the fitness function to determine the most appropriate initial weights and thresholds of the BP neural network, including:

[0029] Initialize the position and fitness of the sparrow population, determine the initial values ​​of the maximum number of algorithm iterations N, the number of individuals in the population n, the number of discoverers PD, the number of danger-sensing individuals SD, the safety value ST, and the warning value R2;

[0030] Calculate the individual fitness of the initial population and find the initial optimal individual position;

[0031] The finder performs foraging behavior to find food locations for the population, that is, to find locations with higher fitness. Its location is updated as follows:

[0032]

[0033] Where Q is a random number that obeys the normal distribution, L is the row unit vector, and a is a random number between [0, 1].

[0034] The joiner forages for food based on the food location found by the discoverer. The joiner's location is updated as follows:

[0035]

[0036] Among them, X worst is the position of the sparrow with the lowest fitness, A + is a row vector containing two elements: 1 and -1; t is the current iteration number, X p The best position for the discoverer to occupy;

[0037] The early warning system performs anti-predator actions, continuously monitoring the surrounding environment, and issues a warning when a predator is detected, and updates the population location as follows:

[0038]

[0039] Among them, β is a random number that obeys the normal distribution and is used to control the step size of the updated position; K is a random number whose value range is between -1 and 1, f i is the individual fitness value; ∈ is a very small number used to prevent f i =f worst ;

[0040] Update the best individual and the best fitness that appear in the record, repeat the iteration until the number of iterations reaches the maximum value, select the best individual and the best fitness as the output.

[0041] Furthermore, according to the prediction result of the axial thrust of the wind turbine generator set, the operating data of the wind turbine generator set is adjusted to achieve the reduction control of the axial thrust of the wind turbine generator set, including:

[0042] Construct a wind turbine axial thrust reduction controller and set the thrust limit of the axial thrust;

[0043] The predicted result of the axial thrust of the wind turbine is compared with the thrust limit. When the predicted result of the axial thrust of the wind turbine is greater than the thrust limit, the axial thrust reduction controller of the wind turbine is used to provide pitch angle compensation to the pitch control link of the wind turbine to reduce the axial thrust of the wind turbine and realize the reduction control of the axial thrust of the wind turbine.

[0044] A second object of the present invention is to provide a wind turbine axial thrust prediction and reduction control device, comprising:

[0045] A data acquisition module is used to obtain historical operating data of the wind turbine when it is in stable operation, and to construct a training data set for wind turbine axial thrust prediction based on the historical operating data;

[0046] A model building module is used to build a wind turbine axial thrust prediction model based on an optimized BP neural network and train the wind turbine axial thrust prediction model using a training data set;

[0047] A prediction module is used to input the real-time collected wind turbine operation data into the trained wind turbine axial thrust prediction model and output the result as the wind turbine axial thrust prediction result;

[0048] The axial thrust reduction control module is used to adjust the wind turbine operating data according to the axial thrust prediction results of the wind turbine, so as to achieve axial thrust reduction control of the wind turbine.

[0049] The third object of the present invention is to provide an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute each step in the method of the aforementioned technical solution.

[0050] A fourth object of the present invention is to provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute each step in the method according to the aforementioned technical solution.

[0051] Compared with the prior art, the advantages of the present invention are:

[0052] The present invention addresses the problems of difficulty in establishing a complete and effective model and performing real-time control measurement of the axial thrust of a wind turbine generator set. By adopting a machine learning method and taking advantage of the BP neural network's ability to approximate nonlinear relationships, the real-time thrust of the unit is predicted using the unit's real-time rotational speed, pitch angle, and wind speed as inputs. At the same time, a sparrow optimization algorithm based on chaos mapping is used to optimize the BP neural network's weights and thresholds, thereby improving the accuracy of the BP neural network's approximation of the unit's thrust nonlinear model. The feedback control strategy of the present invention can quickly and effectively reduce the unit's axial thrust near the rated wind speed, keeping it within safety limits, and reducing the loss of electric power. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic flow chart of a method for predicting and reducing the axial thrust of a wind turbine provided by the present invention;

[0054] Figure 2 A schematic diagram of a curve of wind energy utilization coefficient in a method for predicting and reducing axial thrust of a wind turbine provided by the present invention;

[0055] Figure 3A schematic diagram of the Tent chaos map scatter points in a method for predicting and reducing the axial thrust of a wind turbine provided by the present invention;

[0056] Figure 4 A schematic diagram of the structure of a BP neural network in a method for predicting and reducing the axial thrust of a wind turbine provided by the present invention;

[0057] Figure 5 A schematic diagram of the improved BP neural network process in a wind turbine axial thrust prediction and reduction control method provided by the present invention;

[0058] Figure 6 A schematic diagram of a curve showing how the fitness of the sparrow optimization algorithm changes with the number of iterations in a method for predicting and reducing the axial thrust of a wind turbine provided by the present invention;

[0059] Figure 7 A schematic diagram of a comparison curve between the predicted value and the expected value of the improved model test set in a method for predicting and reducing the axial thrust of a wind turbine provided by the present invention;

[0060] Figure 8 A schematic diagram of an error comparison curve of an improved model test set in a wind turbine axial thrust prediction and reduction control method provided by the present invention;

[0061] Figure 9 A schematic diagram of a wind speed-thrust curve in a method for predicting and reducing the axial thrust of a wind turbine provided by the present invention;

[0062] Figure 10 A schematic flow chart of a wind speed-thrust curve thrust reduction control strategy in a method for predicting and reducing the axial thrust of a wind turbine provided by the present invention;

[0063] Figure 11 A schematic structural diagram of a wind turbine axial thrust prediction and reduction control device provided by the present invention;

[0064] Figure 12 A schematic structural diagram of a non-transitory computer-readable storage medium storing computer instructions provided by the present invention. DETAILED DESCRIPTION

[0065] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 paying any creative work.

[0066] like Figure 1A method for predicting and reducing the axial thrust of a wind turbine generator system is shown, comprising:

[0067] S110: Acquire historical operating data of the wind turbine generator set during stable operation, and construct a training data set for wind turbine generator set axial thrust prediction based on the historical operating data.

[0068] The wind turbine involved in the embodiment of the present invention is a variable speed and variable pitch wind turbine, and its operation control is divided into two stages: torque-speed control below the rated wind speed and pitch-speed control above the rated wind speed.

[0069] The formula for wind turbines to absorb wind energy is:

[0070]

[0071] Among them, P is the wind energy absorbed by the unit, ρ is the air density, R is the impeller radius, v is the wind speed, C P is the wind energy utilization coefficient, which is a function of the tip speed ratio λ and the pitch angle β. Figure 2 A wind energy utilization coefficient curve diagram of a wind turbine model is selected for the present invention.

[0072] The tip speed ratio is defined as the ratio of the linear velocity of the blade tip to the wind speed, and the formula is expressed as:

[0073]

[0074] Where ω is the impeller speed.

[0075] At rated wind speed, wind turbines need to absorb as much wind energy as possible in the non-transition phase to improve the wind energy utilization coefficient. The maximum power tracking mode is adopted, and its maximum power tracking gain coefficient is as follows:

[0076]

[0077] Among them, K opt is the maximum power tracking gain coefficient, and T is the electromagnetic torque.

[0078] Above the rated wind speed, the wind turbine adopts variable pitch control, which limits the wind energy absorbed by the turbine by adjusting the pitch angle, while keeping the generator running smoothly at the rated speed. In this invention, the torque control adopts the standard torque control of the wind turbine, and the pitch control adopts the second-order active disturbance rejection controller for control, and on this basis, the thrust reduction of the wind turbine is studied. The types of wind turbine operating parameters are as follows:

[0079] As shown in Table 1.

[0080]

[0081] Table 1 Wind turbine operating parameters

[0082] S120: Based on the optimized BP neural network, a wind turbine axial thrust prediction model is constructed, and the wind turbine axial thrust prediction model is trained using a training data set.

[0083] With a sufficient number of hidden layers, a BP neural network only requires a three-layer network structure to approximate any nonlinear function, and has strong generalization capabilities. However, neural network training takes a long time, and the gradient descent method easily causes the BP neural network to fall into a local optimum during training, preventing optimal results. The present invention optimizes the BP neural network using a sparrow optimization algorithm, selects the initial weights and thresholds of the BP neural network using the sparrow optimization algorithm, and uses the overall mean square error of the training data set as a fitness function to determine the optimal initial weights and thresholds. At the same time, a tent chaotic map is used to generate a large number of initial sparrow individuals to improve the quality and population diversity of the initial sparrow individuals and enhance their nonlinear characteristics. Sparrow individuals with the best fitness are selected from these individuals as the initial population to ensure the accuracy of model approximation.

[0084] The process of building a wind turbine axial thrust prediction model includes the following steps:

[0085] S121: Based on the tent chaotic map, an initial population position of the sparrow optimization algorithm is generated to optimize the sparrow optimization algorithm.

[0086] While the sparrow optimization algorithm offers good stability, high accuracy, and requires few parameter adjustments, it relies on the initial population position. Randomly selecting a position can result in poor optimization results and fail to meet requirements. This paper employs a tent chaotic map to generate the initial population position for the sparrow optimization algorithm, enhancing its search capabilities and adaptability.

[0087] The mathematical expression of Tent chaos map is as follows:

[0088]

[0089] Among them, a is the main parameter of Tent chaos mapping, and its value is between 0 and 1; X n is the initial population position of the sparrow optimization algorithm. In the present invention, the value of a is selected as 0.7, and the Tent chaos map scatter plot is as follows: Figure 3 shown.

[0090] S122: Based on the optimized sparrow optimization algorithm, the initial weights and thresholds of the BP neural network are optimized and calculated, and the overall mean square error of the training data set is used as the fitness function to determine the most appropriate initial weights and thresholds of the BP neural network.

[0091] The sparrow optimization algorithm is a novel intelligent optimization algorithm developed based on observations of sparrows' foraging and anti-predator behaviors. A sparrow population consists of three types of individuals: finders determine food locations and seek out locations with high fitness; joiners forage based on food locations provided by finders; and early warning agents monitor the surrounding environment and issue warnings upon detecting predators. The basic idea behind the sparrow optimization algorithm is to divide the search space into several blocks and then search each block individually until the optimal solution is found or all subspaces have been exhausted. This algorithm exhibits strong convergence and local search capabilities, and can, to a certain extent, overcome the tendency of BP neural networks to become trapped in local optimal solutions.

[0092] The basic optimization process of the sparrow optimization algorithm includes:

[0093] 1. Initialize the location and fitness of the sparrow population, determine the initial values ​​of the algorithm, such as the maximum number of iterations N, the number of individuals in the population n, the number of discoverers PD, the number of danger-sensing individuals SD, the safety value ST, and the warning value R2;

[0094] 2. Calculate the individual fitness of the initial population and find the initial optimal individual position;

[0095] 3. The finder performs foraging behavior to find food locations for the population, that is, to find locations with higher fitness. Its location is updated as follows:

[0096]

[0097] Where Q is a random number that obeys the normal distribution, L is the row unit vector, and a is a random number between [0, 1].

[0098] 4. The joiner forages for food based on the food location found by the discoverer. The joiner's location is updated as follows:

[0099]

[0100] Among them, X worst is the position of the sparrow with the lowest fitness, A + is a row vector (containing only 1 and -1 elements);

[0101] 5. The early warning device performs anti-predator actions, continuously monitoring the surrounding environment. When a predator is detected, it issues a warning and updates the population location as follows:

[0102]

[0103] Among them, β is a random number (obeying normal distribution) used to control the step size of the updated position; K is a random number (its value range is between -1 and 1), f i is the individual fitness value; ∈ is a very small number (used to prevent fi =f worst );

[0104] 6. Update the best individual and the best fitness that appear in the record;

[0105] 7. Repeat iterations 3-6 until the number of iterations reaches the maximum value, select the best individual and the best fitness as the output.

[0106] S123: Using the BP neural network optimized by the sparrow optimization algorithm as a wind turbine axial thrust prediction model.

[0107] Neural networks, without requiring knowledge of the specific mathematical relationship between input and output, can derive the expected output value based on the input data through self-training and rules. Therefore, they are often used for data regression prediction to approximate unknown nonlinear functional relationships. The BP neural network is a multi-layer feedforward network that uses backpropagation of errors. It uses gradient descent, with the square of the network error as the objective function, and uses gradient search to minimize the objective function to ensure the accuracy of the expected value.

[0108] The BP neural network structure is divided into three layers, namely the input layer, the hidden layer, and the output layer. After training different hidden layer nodes, this paper finally adopts the 3-11-1 structure, with the axial thrust of the unit as the output, and the wind speed, generator speed, and pitch angle that have a strong relationship with the axial thrust as the three inputs of the network input layer. The BP neural network structure is shown in the figure below. Figure 4 shown.

[0109] In the present invention, the wind turbine axial thrust prediction model includes a three-layer structure, namely an input layer, a hidden layer, and an output layer. After training the nodes of different hidden layers, the axial thrust of the turbine is used as the output, and the wind speed, generator speed, and pitch angle, which have a strong relationship with the axial thrust, are used as the three inputs of the network input layer.

[0110] Among them, the hidden layer input and output expressions are:

[0111]

[0112] Where: net i (2) (k) is the input of the hidden layer, O i (2) (k) is the output of the hidden layer, f(x) is the Sigmoid activation function, ω ij (2) is the weight coefficient corresponding to the hidden layer; k is the variable number, and x refers to the input variable of the current layer;

[0113] Output layer input and output expressions:

[0114]

[0115] Where: net l (3) (k) is the input of the output layer, O l (3) (k) is the output of the output layer, g(x) is the Sigmoid activation function, ω li (3) is the weight coefficient corresponding to the output layer.

[0116] S130: Inputting the wind turbine operating data collected in real time into the trained wind turbine axial thrust prediction model, and outputting the result as the axial thrust prediction result of the wind turbine.

[0117] When the number of hidden layers is sufficient, a BP neural network only requires a three-layer network structure to approximate any nonlinear function, and has strong generalization capabilities. However, neural network training takes a long time, and the gradient descent method easily causes the BP neural network to fall into a local optimum during training, preventing it from obtaining the best results. The present invention optimizes the BP neural network using the sparrow algorithm, selects the initial weights and thresholds of the BP neural network using the sparrow algorithm, and uses the overall mean square error of the training set and test set as the fitness function to determine the optimal initial weights and thresholds. At the same time, this paper uses the tent chaotic map to generate a large number of initial sparrow individuals to improve the quality and population diversity of the initial sparrow individuals and enhance their nonlinear characteristics. The sparrow individuals with the best fitness are selected as the initial population to ensure the accuracy of the model approximation.

[0118] For the BP neural network, the input layer, hidden layer, and output layer nodes are selected to determine the BP neural network topology. Then, the chaotic map is used to generate multiple initial individuals of the sparrow population. The individuals with the best fitness are selected as the initial population. Based on the sparrow optimization algorithm, the discoverers, joiners, and early warnings are replaced. The optimal initial weights and thresholds of the BP neural network are selected. Finally, the neural network is trained based on the selected optimal initial values ​​to obtain the final trained neural network model. The overall flow chart is as follows: Figure 5 shown.

[0119] The BP neural network output is the unit's axial thrust. Its input selection criterion is a strong correlation with the unit's axial thrust. This paper selects the unit's wind speed, pitch angle, and generator speed as inputs. The number of hidden layer nodes is selected as follows. Based on the standard BP neural network, different hidden layer nodes are trained, and the optimal number of hidden layer nodes is selected using the mean square error of the training set as the metric. The mean square error metrics for different nodes are shown in Table 2. After comparing the mean square errors, the present invention selects 11 hidden layer nodes.

[0120]

[0121] Among them, N hidden is the number of hidden layer nodes, m and n are the number of input layer and output layer nodes respectively, and a is a random integer between 1 and 10.

[0122]

[0123] Table 2 Mean square error of training set under different hidden layer nodes

[0124] The initial weight threshold of the BP neural network is optimized by the sparrow optimization algorithm to improve the accuracy of the nonlinear approximation of the BP neural network. The parameter selection of the sparrow optimization algorithm is shown in Table 3.

[0125]

[0126] Table 3 Initial parameters of the sparrow optimization algorithm

[0127] The curve of the fitness of the sparrow optimization algorithm changing with the number of iterations is as follows: Figure 6 As shown in the figure, the comparison curves of the BP neural network after the optimization of chaos mapping and sparrow optimization algorithm and the standard BP neural network for the test set samples are as follows: Figure 7 As shown, the error comparison curve is as follows Figure 8 shown.

[0128] The comparison between the BP neural network regression prediction and the optimized indicators is shown in Table 4.

[0129]

[0130] Table 4 BP neural network training indicators

[0131] Depend on Figure 6 、 7 Comparing the curves with the training indicators in Table 4 shows that the optimized BP neural network can more accurately approximate the expected value in regression prediction, with a smaller prediction error. The prediction model of the present invention has better accuracy and can more accurately obtain the current unit axial thrust, allowing for more accurate control later.

[0132] S140: Adjusting the wind turbine operating data according to the wind turbine axial thrust prediction result to achieve reduction control of the wind turbine axial thrust.

[0133] The steps include:

[0134] S141: Construct a wind turbine axial thrust reduction controller and set the thrust limit of the axial thrust.

[0135] The axial thrust of a wind turbine can be expressed as follows:

[0136]

[0137] Where: C t is the thrust coefficient of the wind turbine, which is a function of the tip speed ratio λ and the pitch angle β.

[0138] The present invention conducts control research on the 5MW nonlinear wind turbine model provided by FAST, and obtains the wind speed-thrust variation curve by operating the wind turbine in steady state under different wind speeds, such as Figure 9 shown.

[0139] The thrust trend can be seen from the wind speed-thrust curve. Below the rated wind speed, the thrust increases continuously with the wind speed. At the rated wind speed of 11.4m / s, it reaches its maximum value, producing a thrust peak of 0.728MN. Subsequently, as the wind speed increases, the pitch angle begins to change, and the thrust value decreases rapidly. After the wind speed increases, the wind turbine has a large inertia and the speed cannot change in time, resulting in a lag effect. The standard pitch angle control link uses speed error for control, so it will not act before the rated wind speed exceeds the thrust limit. This paper adopts a thrust feedback controller. The thrust can respond to changes in the current wind speed more quickly than the speed, and the thrust can be controlled more accurately and timely, so that it can meet the requirements of a pitch angle compensation angle before the rated wind speed, so that the pitch actuator can act in advance to achieve the control target.

[0140] It can be concluded from formula (11) that the axial thrust of the unit is related to the tip speed ratio, wind speed, and pitch angle during the operation of the unit. Therefore, this paper selects the unit generator speed, wind speed, and pitch angle as the input of the BP neural network model approximation based on their related variables, so as to obtain the axial thrust of the unit for subsequent control. The structure of the unit's thrust reduction control strategy is as follows: Figure 10 shown.

[0141] S142: Compare the predicted result of the axial thrust of the wind turbine generator set with the thrust limit. When the predicted result of the axial thrust of the wind turbine generator set is greater than the thrust limit, provide pitch angle compensation to the pitch control link of the wind turbine generator set through the axial thrust reduction controller of the wind turbine generator set to reduce the axial thrust of the wind turbine generator set and achieve axial thrust reduction control of the wind turbine generator set.

[0142] The thrust reduction controller is independent of the wind turbine's torque control and pitch control, performing completely independent control. Therefore, there's no need to design a corresponding controller switching strategy. However, the PID controller has an integral phase, which can lead to integral saturation. When the wind speed drops from above the rated wind speed to below the rated wind speed, the pitch command of the pitch control phase will drop rapidly. However, due to integral saturation, the thrust reduction control phase will continue to accumulate integrals, resulting in an inability to compensate for the corresponding pitch angle command in a timely manner, causing the thrust to exceed the expected value. Similarly, when the thrust value is less than the limit below the rated wind speed, the integral effect will continue to accumulate, resulting in an inability to effectively generate the corresponding control command in a timely manner when the wind speed rises to near the rated wind speed. Therefore, this article analyzes the pitch angle generated by the pitch control phase and sets upper and lower limits on the integral effect of the thrust reduction controller to ensure that pitch angle compensation can be made in a timely and effective manner to achieve the desired control effect.

[0143] like Figure 11 As shown, the present invention provides a wind turbine axial thrust prediction and reduction control device 800, comprising:

[0144] The data acquisition module 810 is used to acquire historical operating data of the wind turbine when it is in stable operation, and to construct a training data set for wind turbine axial thrust prediction based on the historical operating data;

[0145] The model building module 820 is used to build a wind turbine axial thrust prediction model based on the optimized BP neural network, and train the wind turbine axial thrust prediction model using a training data set;

[0146] The prediction module 830 is used to input the real-time collected wind turbine operation data into the trained wind turbine axial thrust prediction model and output the result as the wind turbine axial thrust prediction result;

[0147] The axial thrust reduction control module 840 is used to adjust the wind turbine operating data according to the wind turbine axial thrust prediction result to achieve axial thrust reduction control of the wind turbine.

[0148] In order to implement the embodiment, the present invention also proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute each step in the method of the aforementioned technical solution.

[0149] like Figure 12As shown, the non-transitory computer-readable storage medium 900 includes a memory 910 of instructions and an interface 930, and the instructions can be executed by a processor 920 to complete the method. Alternatively, the storage medium can be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0150] In order to implement the embodiments, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method according to the embodiments of the present invention is implemented.

[0151] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0152] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0153] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0154] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.

[0155] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the embodiments described, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0156] Those skilled in the art will understand that all or part of the steps of the method for implementing the embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0157] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0158] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it is understood that the embodiments are exemplary and are not to be construed as limiting the present invention. Those skilled in the art may make changes, modifications, substitutions, and variations to the embodiments within the scope of the present invention.

Claims

1. A method for predicting and reducing the axial thrust of a wind turbine generator set, characterized in that: include: Acquire historical operating data of the wind turbine generator set during stable operation, and construct a training data set for wind turbine generator set axial thrust prediction based on the historical operating data; Based on the optimized BP neural network, a wind turbine axial thrust prediction model is constructed, and the wind turbine axial thrust prediction model is trained using the training data set; Inputting the wind turbine operating data collected in real time into the trained wind turbine axial thrust prediction model, and outputting the result as the axial thrust prediction result of the wind turbine; According to the prediction result of the axial thrust of the wind turbine generator set, the operating data of the wind turbine generator set is adjusted to achieve reduction control of the axial thrust of the wind turbine generator set.

2. A method for predicting and reducing the axial thrust of a wind turbine according to claim 1, characterized in that: Acquiring historical operating data of the wind turbine during stable operation, and constructing a training data set for wind turbine axial thrust prediction based on the historical operating data, including: The torque control of the wind turbine generator set adopts the standard torque control of the wind turbine generator set, and the pitch control adopts the second-order active disturbance rejection controller, and the operation data generated by the wind turbine generator set during the preset time interval of stable operation is obtained as the historical operation data; The wind speed, generator speed, pitch angle and axial thrust in the historical operation data are selected as a training data set for predicting the axial thrust of the wind turbine generator set.

3. The method for predicting and reducing the axial thrust of a wind turbine according to claim 1, wherein: Based on the optimized BP neural network, a wind turbine axial thrust prediction model is constructed, including: Based on the Tent chaos map, the initial population position of the sparrow optimization algorithm is generated to improve the nonlinear characteristics and population diversity of the sparrow optimization algorithm; Based on the optimized sparrow optimization algorithm, the initial weights and thresholds of the BP neural network are optimized and calculated, and the overall mean square error of the training data set is used as the fitness function to determine the most appropriate initial weights and thresholds of the BP neural network; The BP neural network optimized by the sparrow optimization algorithm is used as the axial thrust prediction model of the wind turbine.

4. A method for predicting and reducing the axial thrust of a wind turbine according to claim 3, characterized in that: The wind turbine axial thrust prediction model includes a three-layer structure, namely an input layer, a hidden layer, and an output layer. The nodes of different hidden layers are trained, with the axial thrust of the turbine as the output, and the wind speed, generator speed, and pitch angle, which have a strong relationship with the axial thrust, as the three inputs of the network input layer. Among them, the hidden layer input and output expressions are: Where: net i (2) (k) is the input of the hidden layer, O i (2) (k) is the output of the hidden layer, f(x) is the Sigmoid activation function, ω ij (2) is the weight coefficient corresponding to the hidden layer; k is the variable number, and x refers to the input variable of the current layer; The output layer input and output expressions: Where: net l (3) (k) is the input of the output layer, O l (3) (k) is the output of the output layer, g(x) is the Sigmoid activation function, ω li (3) is the weight coefficient corresponding to the output layer.

5. The method for predicting and reducing the axial thrust of a wind turbine generator set according to claim 3, characterized in that: The Tent chaos mapping formula is expressed as: Among them, a is the main parameter of Tent chaos mapping, and its value is between 0 and 1; X n The initial population position for the sparrow optimization algorithm.

6. A method for predicting and reducing the axial thrust of a wind turbine according to claim 5, characterized in that: Based on the optimized sparrow optimization algorithm, the initial weights and thresholds of the BP neural network are optimized and calculated, and the overall mean square error of the training data set is used as the fitness function to determine the most appropriate initial weights and thresholds of the BP neural network, including: Initialize the position and fitness of the sparrow population, determine the maximum number of algorithm iterations N, the number of individuals in the population n, the number of discoverers PD, the number of danger-sensing individuals SD, the safety value ST, and the warning value R2; Calculate the individual fitness of the initial population and find the initial optimal individual position; The finder performs foraging behavior to find food locations for the population, that is, to find locations with higher fitness. Its location is updated as follows: Where Q is a random number that obeys the normal distribution, L is the row unit vector, and a is a random number between [0, 1]. The joiner forages for food based on the food location found by the discoverer. The joiner's location is updated as follows: Among them, X worst is the position of the sparrow individual with the lowest fitness, A + is a row vector containing two elements: 1 and -1. t is the current iteration number. X p The best position for the discoverer to occupy; The early warning system performs anti-predator actions, continuously monitoring the surrounding environment, and issues a warning when a predator is detected, and updates the population location as follows: Among them, X best is the individual position of the sparrow with the best fitness, β is a random number that obeys the normal distribution and is used to control the step size of the updated position; K is a random number whose value range is between -1 and 1, f i is the individual fitness value; ∈ is a very small number used to prevent f i =f worst ; Update the best individual and the best fitness that appear in the record, repeat the iteration until the number of iterations reaches the maximum value, select the best individual and the best fitness as the output.

7. A method for predicting and reducing the axial thrust of a wind turbine according to claim 1, characterized in that: Adjusting the wind turbine operating data according to the wind turbine axial thrust prediction result to achieve reduction control of the wind turbine axial thrust, including: Constructing a wind turbine axial thrust reduction controller and setting a thrust limit of the axial thrust; The predicted result of the axial thrust of the wind turbine generator set is compared with the thrust limit. When the predicted result of the axial thrust of the wind turbine generator set is greater than the thrust limit, the axial thrust reduction controller of the wind turbine generator set provides pitch angle compensation to the pitch control link of the wind turbine generator set to reduce the axial thrust of the wind turbine generator set and realize the reduction control of the axial thrust of the wind turbine generator set.

8. A wind turbine axial thrust prediction and reduction control device, characterized in that: include: A data acquisition module is used to acquire historical operating data of the wind turbine when it is in stable operation, and to construct a training data set for predicting the axial thrust of the wind turbine based on the historical operating data; A model building module is used to build a wind turbine axial thrust prediction model based on an optimized BP neural network, and train the wind turbine axial thrust prediction model using the training data set; A prediction module, configured to input the real-time collected wind turbine operation data into the trained wind turbine axial thrust prediction model, and output the result as the axial thrust prediction result of the wind turbine; The axial thrust reduction control module is used to adjust the operating data of the wind turbine generator set according to the axial thrust prediction result of the wind turbine generator set to achieve axial thrust reduction control of the wind turbine generator set.

9. An electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform each step in the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to enable the computer to execute each step of the method according to any one of claims 1 to 7.

Images