Control methods, equipment, and media for yaw systems of large wind turbines based on wind forecasting

Abstract

This invention relates to a control method, equipment, and medium for the yaw system of large wind turbines based on wind forecasting. It belongs to the field of wind power generation technology and addresses the problems of low wind-following capability and short lifespan of existing yaw systems. The method includes: processing time-series data of wind speed and direction to extract high-frequency and low-frequency components; establishing a prediction model for the low-frequency components of wind speed and direction; calculating the yaw system adjustment angle using the prediction model; setting the adjustment period; and generating an actual adjustment system based on the yaw system adjustment angle and adjustment period to achieve control of the wind turbine yaw system. This invention is applicable to solving the wind-following problem of wind turbines in the field of wind power generation.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and in particular to the control of yaw systems for large wind turbines. Background Technology

[0002] Wind is uncertain and volatile. Existing active yaw technology judges the direction by the angle between the wind turbine direction and the wind direction. If the angle is greater than a certain degree, the main controller will adjust the yaw. When the angle is adjusted to a certain position and the angle is less than a certain degree, the yaw will stop. This yaw method has a large lag and is not conducive to improving power generation efficiency.

[0003] On the other hand, the yaw system does not distinguish between high-frequency and low-frequency changes in wind direction. Tracking high-frequency changes in wind direction cannot effectively increase the power generation of wind turbines and will greatly increase the working time of the yaw system, reducing its lifespan. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of low wind-following capability and short yaw system life in the prior art, and to provide a control method, equipment and medium for the yaw system of large wind turbines based on wind prediction.

[0005] This invention is achieved through the following technical solution: In one aspect, this invention provides a control method for a large wind turbine yaw system based on wind power prediction, the method comprising:

[0006] The time-series data of wind speed and direction are processed to extract high-frequency and low-frequency components respectively.

[0007] Establish a predictive model for the low-frequency components of wind speed and direction;

[0008] The yaw system adjustment angle is calculated using the prediction model of the low-frequency components of the wind speed and direction.

[0009] Set the adjustment cycle;

[0010] The actual adjustment system is generated based on the yaw system adjustment angle and adjustment cycle, thereby realizing the control of the wind turbine yaw system.

[0011] Further, in step 1, the processing specifically includes:

[0012] Wavelet decomposition was performed on the time series data of wind speed and direction to extract high-frequency and low-frequency components respectively.

[0013] Furthermore, the wavelet decomposition of the time-series data on wind speed and direction to extract high-frequency and low-frequency components specifically includes:

[0014] The components are separated using the following decomposition algorithm:

[0015]

[0016] In the formula, d j,2n i ,d j,2n+1 i h represents the wavelet packet decomposition coefficients. 2k-1 g k-2i represents the low-pass and high-pass filter banks for wavelet packet decomposition; i, j, n, and k are the subscripts of the variables;

[0017] The time-series data was restored using the following reconstruction algorithm:

[0018]

[0019] In the formula, h i-2k * g i-2k * Low-pass and high-pass filters for wavelet packet reconstruction.

[0020] Furthermore, a low-frequency component prediction model for wind speed and direction is established using a long short-term memory network.

[0021] Furthermore, the method of establishing a low-frequency component prediction model for wind speed and direction using a long short-term memory network specifically includes:

[0022] The forget gate uses formula (1) to determine the information to be discarded. Formula (1) is as follows:

[0023] F t =F( S (W) f ·[h t-1 ,x t ]+b f (1);

[0024] The input gate processes the input data using formula (2) and passes it to the subsequent stages. Formula (2) is as follows:

[0025] I t =F (S) (W i ·[h t-1 ,x t ]+b i )

[0026]

[0027] The element state is calculated using formula (3), which is as follows:

[0028]

[0029] The output gate uses formula (4) to pass the new cell state and the new hidden state to the next time step. Formula (4) is as follows:

[0030] O t =F (S) (W o ·[h t-1 ,x t ]+b o )

[0031] h t =O t ·tanh(C t (4);

[0032] In the formula, I t For the value of the input gate, O t F is the value of the output gate. t F is the value of the forget gate. (S) For the Sigmoid function, C represents the state value of the candidate memory cell. t W represents the current state value of the memory cell. i W o W f W c b represents the connection weights between the current time-input data and the previous time-input LSTM unit output to the input gate; i b o b f b c For the bias of each gate and memory unit; h t x is the output of the memory cell at time t; t Let x be the value of variable x at time t.

[0033] Furthermore, the yaw calculation system adjusts the angle, specifically including:

[0034] Calculate the weighted wind direction θ during the adjustment period T. T The calculation formula is as follows:

[0035]

[0036] In the formula, v n θ n Let n be the wind speed and direction at time n;

[0037] The yaw system adjustment angle is:

[0038] θ=θ T -θ0;

[0039] In the formula, θ0 represents the fan orientation at the moment the adjustment system operates.

[0040] Furthermore, the setting of the adjustment period specifically includes:

[0041] The projected daily revenue from electricity generation is represented by α:

[0042]

[0043] In the formula, a n Let E be the electricity price for a time period of n. T，n Let T be the power generation during time n under the adjustment period T, and b be the average cost of a single action of the yaw system.

[0044] The value of T is determined by changing T to make α reach its maximum value.

[0045] Secondly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, it executes the steps of a large wind turbine yaw system control method based on wind prediction as described above.

[0046] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of computer instructions, the plurality of computer instructions being used to cause a computer to execute a large wind turbine yaw system control method based on wind prediction as described above.

[0047] Fourthly, the present invention provides an electronic device, comprising:

[0048] At least one processor; and,

[0049] A memory communicatively connected to the at least one processor; wherein,

[0050] The memory stores instructions that can be executed by the at least one processor, which, when executed, enable the at least one processor to perform a large wind turbine yaw system control method based on wind prediction as described above.

[0051] The beneficial effects of this invention are:

[0052] The control method of the present invention separates the high-frequency and low-frequency wind power for separate prediction, adjusts the yaw system adjustment frequency according to the proportion of low-frequency components, and adjusts the yaw angle according to the prediction results to achieve active control of the yaw system, which can improve power generation and yaw system life.

[0053] Using the control method of this invention, the number of actions of the wind turbine yaw system can be controlled by adjusting the period T. In the example, the number of actions is reduced from 1845 to 1440, and the lifespan is improved.

[0054] To address the issue of low wind-following capability in existing technologies, this invention presents a wind turbine yaw system adjustment method based on predictive control. Calculations using data from a wind farm indicate that this method is expected to increase power generation by 1.59%.

[0055] This invention is applicable to solving the wind turbine's wind resistance problem in the field of wind power generation. Detailed Implementation

[0056] Implementation Method 1: A control method for a large wind turbine yaw system based on wind power prediction, the method comprising:

[0057] The time-series data of wind speed and direction are processed to extract high-frequency and low-frequency components respectively.

[0058] Establish a predictive model for the low-frequency components of wind speed and direction;

[0059] The yaw system adjustment angle is calculated using the prediction model of the low-frequency components of the wind speed and direction.

[0060] Set the adjustment cycle;

[0061] The actual adjustment system is generated based on the yaw system adjustment angle and adjustment period, thereby realizing the control of the wind turbine yaw system. That is, the wind turbine performs a yaw adjustment once every time interval T, and the adjustment amount is θ.

[0062] In this embodiment, in order to establish an accurate wind speed and wind direction prediction model and eliminate the influence of high frequency components, the historical data of wind speed and wind direction are first processed to extract high frequency components and low frequency components respectively.

[0063] Then, the yaw system's adjustment frequency is adjusted according to the proportion of low-frequency components, and the yaw angle is adjusted according to the prediction results to achieve active control of the yaw system, which can improve power generation and yaw system life.

[0064] Implementation Method Two: This implementation method further defines the control method for a large wind turbine yaw system based on wind prediction described in Implementation Method One. In this implementation method, the processing in step 1 is further defined, specifically including:

[0065] In step 1, the processing specifically includes:

[0066] Wavelet decomposition was performed on the time series data of wind speed and direction to extract high-frequency and low-frequency components respectively.

[0067] In this embodiment, wavelet decomposition is used to extract high-frequency components and low-frequency components separately, which can improve the subsequent control effect.

[0068] Implementation method three is a further limitation on the control method for a large wind turbine yaw system based on wind prediction described in implementation method two. In this implementation method, wavelet decomposition is performed on the time-series data of wind speed and direction to extract high-frequency and low-frequency components respectively, which further limits the implementation method. Specifically, it includes:

[0069] The wavelet decomposition of the time-series data of wind speed and direction, extracting high-frequency and low-frequency components respectively, specifically includes:

[0070] The components are separated using the following decomposition algorithm:

[0071]

[0072] In the formula, d j,2n i ,d j,2n+1 i h represents the wavelet packet decomposition coefficients. 2k-1 g k-2i represents the low-pass and high-pass filter banks for wavelet packet decomposition; i, j, n, and k are the subscripts of the variables;

[0073] To make the data more suitable for the subsequent LSTM algorithm, it is restored to time series data using the following reconstruction algorithm:

[0074]

[0075] In the formula, h i-2k * g i-2k * Low-pass and high-pass filters for wavelet packet reconstruction.

[0076] In this embodiment, the low-frequency components in the original wind speed and wind direction time series data can be extracted using the above algorithm and used as the training set for subsequent modeling.

[0077] Implementation Method Four: This implementation method further defines the control method for a large wind turbine yaw system based on wind prediction described in Implementation Method One. In this implementation method, the method for establishing the low-frequency component prediction model for wind speed and direction is further defined, specifically including:

[0078] A low-frequency component prediction model for wind speed and direction is established using a long short-term memory network.

[0079] In this embodiment, a prediction model for wind speed and direction is established based on a very short-term memory network. This model can accurately predict the wind speed and direction for a period of time in the future from time t using data before time t.

[0080] Implementation Method Five: This implementation method further defines the control method for a large wind turbine yaw system based on wind prediction described in Implementation Method Four. In this implementation method, the method of establishing a low-frequency component prediction model for wind speed and direction using a long short-term memory network is further defined, specifically including:

[0081] The method of establishing a low-frequency component prediction model for wind speed and direction using a long short-term memory network specifically includes:

[0082] The forget gate uses formula (1) to determine the information to be discarded. Formula (1) is as follows:

[0083] F t =F( S (W) f ·[h t-1 ,x t ]+b f (1);

[0084] The input gate processes the input data using formula (2) and passes it to the subsequent stages. Formula (2) is as follows:

[0085] I t =F (S) (W i ·[h t-1 ,x t ]+b i )

[0086]

[0087] The element state is calculated using formula (3), which is as follows:

[0088]

[0089] The output gate uses formula (4) to pass the new cell state and the new hidden state to the next time step. Formula (4) is as follows:

[0090] O t =F (S) (W o ·[h t-1 ,x t ]+b o )

[0091] h t =O t ·tanh(C t (4);

[0092] In the formula, I t For the value of the input gate, O t F is the value of the output gate. tF is the value of the forget gate. (S) For the Sigmoid function, C represents the state value of the candidate memory cell. t W represents the current state value of the memory cell. i W o W f W c b represents the connection weights between the current time-input data and the previous time-input LSTM unit output to the input gate; i b o b f b c For the bias of each gate and memory unit; h t x is the output of the memory cell at time t; t Let x be the value of variable x at time t.

[0093] This embodiment provides a specific method and formula for establishing a prediction model of low-frequency components of wind speed and direction based on a very short-term memory network. This model can accurately predict wind speed and direction for a period of time in the future from time t using data before time t.

[0094] Implementation method six is ​​a further limitation on the control method for a large wind turbine yaw system based on wind power prediction described in implementation method one. In this implementation method, the calculation method for the yaw system adjustment angle is further limited, specifically including:

[0095] The calculation of the yaw system adjustment angle specifically includes:

[0096] Calculate the weighted wind direction θ during the adjustment period T. T The calculation formula is as follows:

[0097]

[0098] In the formula, v n θ n Let n be the wind speed and direction at time n;

[0099] The yaw system adjustment angle is:

[0100] θ=θ T -θ0;

[0101] In the formula, θ0 represents the fan orientation at the moment the adjustment system operates.

[0102] In this embodiment, the prediction results obtained from the prediction model based on the low-frequency components of wind speed and wind direction are used to provide a method for calculating the yaw system adjustment angle. This calculation method can achieve effective system adjustment.

[0103] Implementation method seven is a further limitation on the control method for a large wind turbine yaw system based on wind power prediction described in implementation method one. In this implementation method, the method for setting the adjustment period is further limited, specifically including:

[0104] The setting of the adjustment period specifically includes:

[0105] The projected daily revenue from electricity generation is represented by α:

[0106]

[0107] In the formula, a n Let E be the electricity price for a time period of n. T，n Let T be the power generation during time n under the adjustment period T, and b be the average cost of a single action of the yaw system.

[0108] The value of T is determined by changing T to make α reach its maximum value.

[0109] It should be noted that 0-23 refers to 24 hours; 1440 = 24 * 60, which is the number of minutes in a day.

[0110] In this embodiment, reducing the adjustment period T causes the wind turbine's yaw control system and actuators to operate more frequently with changes in wind direction, which will shorten the lifespan of the yaw mechanism and, in severe cases, lead to yaw system failure, thus affecting the safety of the wind turbine generator. Increasing the adjustment period will reduce the yaw system's wind-following ability. Therefore, the adjustment period T should be set by considering all the above factors.

[0111] Implementation method eight is an embodiment of a large wind turbine yaw system control method based on wind prediction as described above, specifically including:

[0112] Step 1: To establish an accurate wind speed and direction prediction model and eliminate the influence of high-frequency components, the historical data of wind speed and direction are first processed. To improve the subsequent control effect, wavelet decomposition is performed on the time series data of wind speed and direction to extract the high-frequency and low-frequency components separately.

[0113] The components are separated using the following decomposition algorithm:

[0114]

[0115] In the formula: d j,2n i ,d j,2n+1 i h represents the wavelet packet decomposition coefficients. 2k-1 g k-2i represents the low-pass and high-pass filter banks for wavelet packet decomposition; i, j, n, and k are the subscripts of the variables.

[0116] To make the data more suitable for the subsequent LSTM algorithm, it is restored to time series data using the following reconstruction algorithm:

[0117]

[0118] Where: h i-2k * g i-2k * Low-pass and high-pass filters for wavelet packet reconstruction.

[0119] The above algorithm can be used to extract the low-frequency components from the original wind speed and direction time series data and use them as the training set for subsequent modeling.

[0120] Step 2: Establish a prediction model for the low-frequency components of wind speed and direction.

[0121] Then, a low-frequency component prediction model for wind speed and direction (two models) is established using a long short-term memory network, specifically including:

[0122] Formula F t =F( S (W) f ·[h t-1 ,x t ]+b f The forget gate determines which information needs to be discarded.

[0123] As an input gate, it processes the input data and passes it to subsequent stages.

[0124] Perform unit state calculations.

[0125] As an output gate, it passes the new cell state and the new hidden state to the next time step.

[0126] In the formula: I t For the value of the input gate, O t F is the value of the output gate. t F is the value of the forget gate. (S) For the Sigmoid function, C represents the state value of the candidate memory cell. t W represents the current state value of the memory cell. i W o W f W c b represents the connection weights between the current time-input data and the previous time-input LSTM unit output to the input gate; i b o b f b cFor the bias of each gate and memory unit; h t x is the output of the memory cell at time t; t Let x be the value of variable x at time t.

[0127] A prediction model for low-frequency components of wind speed and direction is established based on ultra-short-term memory networks. This model can predict wind speed and direction for a period of time in the future, based on data before time t.

[0128] Step 3: Calculate the yaw system adjustment angle.

[0129] Based on the above prediction results, the weighted wind direction θ during the adjustment period T is calculated. T The calculation formula is as follows:

[0130]

[0131] In the formula: v n θ n Let n be the wind speed and wind direction.

[0132] Then the yaw system adjustment angle θ = θ T -θ0.

[0133] θ0 represents the moment the system operates and the direction of the fan.

[0134] Step 4: Adjusting the setting method of period T.

[0135] Reducing the adjustment period T causes the wind turbine's yaw control system and actuators to operate more frequently with changes in wind direction, shortening the yaw mechanism's lifespan and potentially leading to yaw system failure, thus impacting the safety of the wind turbine generator. Increasing the adjustment period reduces the yaw system's ability to follow wind. Therefore, the adjustment period T should be set by considering all these factors. The expected daily power generation revenue is represented by α:

[0136]

[0137] In the formula, a n Let E be the electricity price for a time period of n. T，n Let T be the power generation during the n-time period under the T-regulation cycle, and b be the average cost of a single action of the yaw system.

[0138] The value of T is determined by changing T to make α reach its maximum value.

[0139] Step 5: Based on the yaw adjustment angle obtained in step 3 and the adjustment period obtained in step 4, generate the actual adjustment system to control the wind turbine yaw system. That is, the wind turbine performs a yaw adjustment once every time interval T, and the adjustment amount is θ.

Claims

1. A control method for a large wind turbine yaw system based on wind power prediction, characterized in that, The method includes: The time-series data of wind speed and direction are processed to extract high-frequency and low-frequency components respectively. Establish a predictive model for the low-frequency components of wind speed and direction; The yaw system adjustment angle is calculated using the prediction model of the low-frequency components of the wind speed and direction. Set the adjustment cycle; The actual adjustment system is generated based on the yaw system adjustment angle and adjustment cycle to achieve control of the wind turbine yaw system; A low-frequency component prediction model for wind speed and direction is established using a long short-term memory network; The method of establishing a low-frequency component prediction model for wind speed and direction using a long short-term memory network specifically includes: The forget gate uses formula (1) to determine the information to be discarded. Formula (1) is as follows: （1）； The input gate processes the input data using formula (2) and passes it to the subsequent stages. Formula (2) is as follows: （2）； The unit state is calculated using formula (3), which is as follows: （3）； The output gate uses formula (4) to pass the new cell state and the new hidden state to the next time step. Formula (4) is as follows: （4）； In the formula, I t For the value of the input gate, O t The value of the output gate. F t The value of the forget gate. F (S) For the Sigmoid function, The state value of the candidate memory cell; C t This represents the current state value of the memory cell. W i , W o , W f , W c The connection weights between the current data input and the previous LSTM unit output and the input gate are: b i , b o , b f , b c Bias for each gate and memory unit; h t For memory units in t Output at time x t Let x be the value of variable x at time t; The calculation of the yaw system adjustment angle specifically includes: Calculate the weighted wind direction during the adjustment period T. θ T The calculation formula is as follows: In the formula, v n , θ n Let n be the wind speed and direction at time n; The yaw system adjustment angle is: θ=θ T -θ0; In the formula, θ0 is the fan orientation at the moment the adjustment system is activated; The setting of the adjustment cycle specifically includes: The projected daily revenue from electricity generation is represented by α: In the formula, a n Let n be the electricity price over a time period. E T，n Let T be the power generation during time n under the adjustment period T, and b be the average cost of a single action of the yaw system. The value of T is determined by changing T to make α reach its maximum value.

2. The control method for a large wind turbine yaw system based on wind prediction according to claim 1, characterized in that, In step 1, the processing specifically includes: Wavelet decomposition was performed on the time series data of wind speed and direction to extract high-frequency and low-frequency components respectively.

3. The control method for a large wind turbine yaw system based on wind prediction according to claim 2, characterized in that, The wavelet decomposition of the time-series data of wind speed and direction, extracting high-frequency and low-frequency components respectively, specifically includes: The components are separated using the following decomposition algorithm: In the formula, , These are the wavelet packet decomposition coefficients; , represents the low-pass and high-pass filter banks for wavelet packet decomposition; i, j, n, and k are the subscripts of the variables; The time-series data was restored using the following reconstruction algorithm: In the formula, , Low-pass and high-pass filters for wavelet packet reconstruction.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, The steps of the method according to any one of claims 1 to 3 are performed when the processor runs the computer program stored in the memory.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of computer instructions, which are used to cause a computer to perform the method of any one of claims 1 to 3.

6. An electronic device, characterized in that, include: At least one processor; as well as, 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 to enable the at least one processor to perform the method of any one of claims 1 to 3.

Images