Floating type wind turbine generator ballast water control method based on wind power medium and long term prediction

Abstract

A floating type wind turbine generator ballast water control method based on wind power medium and long term prediction comprises the steps that firstly, based on historical meteorological data, long-term wind regime distribution and probability distribution of a wind power plant are obtained through a data processing and clustering algorithm; and predicting the wind condition by using historical weather and external forecast data and a random forest algorithm, and optimizing a ballast water control strategy and generating a wind power time sequence in combination with wave flow load calculation and unit attitude and load analysis. Measuring the position of the impeller of the unit in real time, determining an operation strategy by means of a neural network, and optimizing the attitude and load of the unit by using a particle swarm algorithm; a ballast water capacity range can be determined according to unit parameters and control system performance, a state space model is established, an optimization problem is solved, and a strategy is corrected. And finally, a ballast water system is adjusted through deep learning and feedback control, a real-time optimization model is constructed for cooperative control, and the unit stability and the power generation efficiency are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of ballast water control of floating wind turbines, and in particular relates to a ballast water control method of floating wind turbines based on medium- and long-term wind power prediction. Background Art

[0002] Ballast water control is crucial in the operation of floating wind turbines. It is not only related to the stability of the unit in a complex marine environment, but also has a significant impact on the power generation efficiency. Traditional ballast water control methods often lack accurate prediction and effective use of wind conditions, and it is difficult to achieve optimal control under variable marine environments and different wind conditions. For example, some early methods were based only on simple meteorological data or fixed control strategies, without in-depth consideration of the complexity of the long-term wind distribution of wind farms, as well as the dynamic interaction between wind conditions and wave and current loads, unit attitude and loads, resulting in ballast water control unable to adapt to environmental changes in a timely manner, and unit stability and power generation efficiency were difficult to reach ideal levels.

[0003] In recent years, with the rapid development of data processing technology and machine learning algorithms, new opportunities have been provided for the optimization of ballast water control of floating wind turbines. Wind condition prediction technology based on big data analysis can more accurately grasp the changing laws of wind speed and wind direction in wind farms, thereby providing a basis for adjusting ballast water control strategies in advance. For example, by using historical meteorological data for in-depth analysis and statistics, detailed distribution and probability distribution of long-term wind conditions in wind farms can be obtained, laying the foundation for subsequent accurate prediction and control. Machine learning algorithms such as random forest algorithms have shown strong capabilities in wind condition prediction, and can integrate multiple meteorological factors to make more reliable predictions of future wind condition distribution. At the same time, accurate calculation and analysis of wave and current loads have become a key link in optimizing ballast water control. By converting the wave model from the time domain to the frequency domain through mathematical tools such as Fourier transform, combined with the equation of motion, the wave load can be calculated more accurately, and the calculation results of the current load can be combined to fully grasp the load distribution of the sea conditions. On this basis, considering the relationship between wind power and wave and current loads, the control strategy of the unit attitude and load can be further optimized. For example, when the wind power is large, reasonable adjustment of the unit attitude and ballast water distribution can better balance the force, improve power generation efficiency and ensure safe and stable operation of the unit.

[0004] Based on the above analysis, the existing technology still has the following shortcomings: when analyzing the wind conditions of wind farms, it often only relies on simple short-term data or a single meteorological indicator, lacks a comprehensive and in-depth analysis of the long-term wind condition distribution, and cannot accurately obtain the probability distribution characteristics of wind conditions. This makes the subsequent formulation of ballast water control strategies based on wind conditions lack a solid basis and is prone to deviations. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a ballast water control method for a floating wind turbine based on medium- and long-term wind power forecasts. By utilizing a large amount of historical meteorological data, through data preprocessing, sample point analysis, and clustering algorithms, a comprehensive and accurate long-term wind condition distribution model for the wind farm is constructed, and the probability distribution of wind conditions is accurately calculated, providing a reliable data basis for subsequent wind condition forecasts and ballast water control strategy optimization.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: A method for controlling ballast water of a floating wind turbine based on medium- and long-term wind power prediction, comprising the following steps: Step 1: Based on historical meteorological data, analyze and count the wind speed and wind direction of the wind farm, obtain the distribution of long-term wind conditions in the wind farm, and calculate the probability distribution of wind conditions; Step 2: Use historical meteorological data and external meteorological forecast services to predict future wind distribution using the random forest algorithm; combine the predicted wind distribution and the probability distribution of sea conditions, consider the impact of unit attitude and load on control system performance, establish a function and optimize it using the gradient descent method to determine the ballast water control strategy, and finally generate a wind power time series; Step 3: Real-time measurement of the unit's impeller position, combined with wind conditions and control system performance, using neural networks to determine the unit's operating strategy; combined with the ballast water control strategy results of step 2, based on the optimization objective function and related constraints, the particle swarm optimization algorithm is used to optimize the unit's attitude and load to achieve maximum utilization of wave and current energy; Step 4: Determine the reasonable range of ballast water capacity based on the unit operating parameters and control system performance, establish a state space model and solve the optimization problem in combination with the wave and current sea conditions and real-time operating conditions, and formulate a ballast water control strategy; at the same time, modify the control strategy based on the deviation between the actual wave and current sea conditions and the predicted value; Step 5: Monitor the operating status of the unit in real time, and adjust the ballast water system parameters using deep learning models and feedback control methods; based on the measured wave and current parameters and control system performance, establish a real-time optimization model to solve the unit attitude and load adjustment and apply it to control. At the same time, use the feedback control mechanism to dynamically adjust the ballast water system and control strategy to achieve coordinated control. Preferably, the sub-steps of step 1 are: Step 1.1: Collect wind speed and direction data of the wind farm, construct the corresponding measurement data sequence, and specify the range of wind direction values; Step 1.2: Normalize the wind speed and wind direction data respectively. The wind speed is reduced to the interval of 0 to 1 according to its own maximum value range, and the wind direction is mapped to the interval of 0 to 1 according to the ratio of 360°, forming a normalized data sequence and combining it into a sample space; Step 1.3, calculate the Euclidean distance between sample points according to a specific formula to measure the similarity of wind speed and wind direction combinations at different times; Step 1.4: Use the K-Means clustering algorithm to divide the sample points into different clusters based on distance. After multiple iterations until the cluster center changes to the standard, determine the wind condition distribution, and calculate the proportion of sample points in each cluster to obtain the wind condition probability distribution.

[0007] Preferably, the method of step 1.1 is: Assume that the wind speed measurement data sequence is ,in Indicates The wind speed value measured at each moment; the wind direction measurement data sequence is ,in Indicates The wind direction angle value measured at each moment; The measured data of wind speed and wind direction are normalized to facilitate subsequent analysis and calculation; for wind speed data, the following normalization formula is used: ; in: :Indicates the The wind speed value measured at a moment is is the serial number of the measurement time, the value range is from To the total measurement time; : After normalization The wind speed value at a certain moment has a range that depends on the normalized calculation result. arrive between (depending on the range of the wind speed data itself and the way it is normalized); : represents the wind speed measurement data sequence The minimum value among , that is, the minimum value of wind speed among all measurement moments; : represents the wind speed measurement data sequence The maximum value among them, that is, the maximum value of wind speed among all measurement moments; Get the normalized wind speed data series ; : After normalization The wind speed value at a certain moment has a range that depends on the normalized calculation result. arrive between (depending on the range of the wind speed data itself and the way it is normalized); : represents the wind speed measurement data sequence The minimum value among , that is, the minimum value of wind speed among all measurement moments; : represents the wind speed measurement data sequence The maximum value among them, that is, the maximum value of wind speed among all measurement moments; Get the normalized wind speed data series .

[0008] Preferably, the method of step 1.2 is: Calculate the Euclidean distance between sample points , used to measure the similarity between different sample points; the Euclidean distance calculation formula is as follows: ; in: :Indicates the The sample points and The Euclidean distance between two sample points is used to measure the distance between the two sample points in space. The smaller the distance, the more similar the wind speed and wind direction combinations of the two sample points are. and :They are and The normalized wind speed value at the moment, here and is a different measurement time sequence number, ranging from To the total measurement time; and :They are and The wind direction angle value after normalization at each moment is also and is a different measurement time sequence number, ranging from to the total measurement time.

[0009] Preferably, the method of step 1.3 is: The sample points according to The values ​​of are divided into different clusters as the cluster centers of the classification results to obtain the distribution of long-term wind conditions in the wind farm; Clustering algorithm, its objective function is: ; in: :yes The objective function value of the clustering algorithm is used to measure the effect of clustering; the goal is to continuously adjust the cluster center so that The value of is the smallest, that is, the sum of the squares of the distances from each sample point to the cluster center to which it belongs is minimized, so as to achieve reasonable clustering division; : The pre-set number of clusters, which indicates the number of clusters into which the sample points are to be divided, that is, the number of categories of different wind conditions that are expected to be obtained, which is determined by the user according to the specific situation and analysis requirements; : No. cluster, that is, it contains the A set of sample points of similar wind conditions, The value range is from arrive , indicating different clustering categories; : No. sample points, which consist of normalized wind speed and wind direction data, namely , is the serial number of the sample point, ranging from to the total number of sample points; : No. The center of a cluster, whose initial value can be randomly selected, is a two-dimensional vector, which represents the normalized combination of wind speed and wind direction values ​​corresponding to the cluster center. In the clustering process, it is continuously updated through iteration so that the objective function Minimum; In each iteration, the cluster centers are updated The formula is: ; in: :Indicates the The updated center value of a cluster is a two-dimensional vector, which is also composed of normalized wind speed and wind direction values; : No. clusters, which has the same meaning as in the above objective function , that is, it contains the A collection of sample points of similar wind conditions; : No. sample points, which has the same meaning as in the above objective function , which is the sample point composed of normalized wind speed and wind direction data; :Indicates the Cluster The number of sample points in is used to calculate the average value of the sample points in the cluster to update the cluster center; After multiple iterations, when the change of the cluster center is less than the preset threshold, the clustering algorithm is considered to have converged. Each cluster represents the different distribution of long-term wind conditions in the wind farm.

[0010] Preferably, the method of step 1.4 is: For each wind condition pattern obtained by clustering, the proportion of the number of sample points in the cluster to the total number of sample points is counted to obtain the probability distribution of the wind condition; assuming that the total number of wind condition classifications is Species, respectively , statistics belong to The number of sample points for wind-like conditions is , then the wind conditions Probability , the probability distribution of wind conditions can be expressed as ,in: : represents the total number of wind condition classifications, that is, the number of different wind condition patterns obtained by the clustering algorithm; :Indicates the wind mode, The value range is from arrive , used to distinguish different wind condition categories; :Indicates that it belongs to The number of sample points of each wind mode (cluster) is obtained by counting the sample points in each cluster; : represents the total number of sample points, that is, the total number of sample points corresponding to all measurement moments, which is used to calculate the probability of each wind condition mode; :Indicates that the wind condition belongs to The probability of a class is obtained by calculating the ratio of the number of sample points belonging to this class of wind conditions to the total number of sample points. Its value range is arrive and the sum of the probabilities of all wind condition categories is ; : Indicates wind conditions The probability distribution function is a set of all wind condition category probabilities, which is used to describe the overall probability distribution of wind conditions in a wind farm.

[0011] Preferably, the method of step 2 is: (1) Based on historical meteorological data and external meteorological forecast services, a machine learning model is used to predict wind speed and direction to obtain the distribution of wind conditions in the future: Collect historical weather data, including wind speed ,wind direction ,temperature ,humidity , air pressure Relevant meteorological data; Assume that the wind speed measurement data sequence is ,in Indicates The wind speed value measured at each moment; the wind direction measurement data sequence is ,in Indicates The wind direction angle value measured at each moment; the temperature measurement data sequence is ,in Indicates The temperature value measured at each moment; the humidity measurement data sequence is ,in Indicates The humidity value measured at each moment; the pressure measurement data sequence is ,in Indicates The air pressure value measured at a moment; Divide historical data into training set and test set, and preprocess the training set data. The preprocessing methods include data cleaning and standardization. The normalization formula is as follows: For wind speed : ,in Represents the original wind speed value, is the mean of the wind speed training set data, is the standard deviation, is the normalized wind speed value; For wind direction : ,in Indicates the original wind direction angle value, is the mean of the wind direction training set data, is the standard deviation, is the normalized wind direction value; For temperature : ,in represents the original temperature value, is the mean of the temperature training set data, is the standard deviation, is the standardized temperature value; For humidity : ,in Represents the original humidity value, is the mean of the humidity training set data, is the standard deviation, is the standardized humidity value; For air pressure : ,in At the original pressure value, is the mean of the air pressure training set data, is the standard deviation, is the normalized air pressure value; The random forest algorithm is used to predict wind conditions. Assume that the training set data is: , in is the feature vector, is the true value of the corresponding wind speed or wind direction; When building a decision tree, from the training set Random draw with replacement samples, forming a training subset , then Perform feature selection and node splitting until the stopping condition is met; for prediction, input the test set samples into the constructed random forest, each decision tree gives a prediction result, and the final prediction result is the average or majority vote of all decision tree prediction results; the performance of the model can be evaluated using the mean absolute error: ,in is the number of test set samples, is the true value, is the predicted value; (2) Combined with the wave and current loads of the wind farm, the impact of the unit attitude and load on the control system performance is comprehensively considered to optimize the ballast water control strategy: Calculate the wave and current loads based on the predicted wind distribution and historical sea condition data: Assume that during the calculation of wave and current load, the sequence of wave height changes with time is: ,in Indicates The wave height value measured or calculated at each moment; the wave period sequence is ,in Indicates The wave period corresponding to the moment; the ocean current velocity sequence is ,in Indicates The current velocity values ​​measured at each location; the current direction sequence is ,in Indicates The current direction value measured at each moment; Using Fourier transform, the wave model is converted from time domain to frequency domain, and we get: , in is the variation of wave height with time, For the The amplitude of the frequency component, is the angular frequency, is the phase, is the time that represents the motion caused by the wave period; Applying the time domain and frequency domain conversion relationship to the equation of motion and using the Fourier transform, we get: ,in For quality, is the damping coefficient, is the stiffness coefficient, is the displacement, For external force, is the amplitude of the wave, is the propagation constant, and It is the conversion result between time domain and frequency domain; The wave response of the blades and the unit is calculated, and the wave load is expressed as: ,in: : represents the effective wave period; : is the zero-order moment of the wave energy spectrum; : characteristic energy parameters related to wave energy spectrum; Through the wave load and flow loads The calculation results are statistically analyzed to obtain the load distribution of the sea conditions; at the same time, based on the predicted wind speed , combined with the relevant equipment parameters of the wind farm, calculate the wind power; Analyze the relationship between wind power and wave and current loads. The magnitude of wind power will affect the operating status and stress of the unit, and thus affect the bearing and response of wave and current loads. (3) Based on the predicted wind distribution, combined with the unit attitude and load, the magnitude and direction of the wave and current load are calculated; based on the predicted wind speed, unit attitude and load at different times, the corresponding wind characteristic parameters are comprehensively calculated to generate a wind time series; Assume that the wind condition time series is ,in Indicates The combination of wind condition characteristic parameters at each moment reflects the comprehensive wind condition at that moment; combined with the calculation results of wind power, the relationship between wind condition characteristic parameters and wind power at different moments is analyzed.

[0012] Preferably, the method of step 3 is: Real-time measurement of the impeller position of the unit, based on the blade angle and pitch angle , determine the unit's operating strategy: Combine wind conditions with control system performance, that is, wind speed ,wind direction The performance of the control system is studied by using a neural network model to find out the stable operating range of the unit; a multi-layer perceptron neural network is used, and its structure includes an input layer, a hidden layer, and an output layer; The input layers include: wind speed ,wind direction , and the unit's current blade angle and pitch angle ; The neural network is trained through a large amount of historical data, and the weights and thresholds of the neural network are optimized so that it can accurately predict the stable operating range of the unit. Suppose the input vector of the neural network is , the output is , indicating whether the unit is in a stable operating range; based on a large number of known input And the corresponding output , adjust the connection weights inside the neural network and threshold , so that the predicted output is as close to the true output as possible; According to the input-output relationship of the sample data, the corresponding relationship between the impeller position and the unit operation is determined; by analyzing the model, the optimal impeller position under different working conditions is determined; Based on the close relationship between wind power and unit attitude and load, the control strategy of the unit is further optimized; the optimization objective function considering wind power is established: , At the same time, the stability constraints of the unit are met; the constraints include: inclination range , load limit .

[0013] Preferably, the method of step 4 is: (1) Develop ballast water control strategies based on the unit’s operating parameters and control system performance: Based on the maximum and minimum weight distribution of the unit, the reasonable range of ballast water capacity is obtained; The ballast water system is optimized by combining wave and current conditions, real-time operating conditions and wind power. The model predictive control algorithm is used to first establish a discrete time state space model of the ballast water system, and then, based on the prediction time domain and control time domain , at each sampling moment , solve the following optimization problem: ; in To control the sequence, To predict the output, is the reference trajectory, and is a weight matrix used to balance control performance and control energy; the above optimization problem is solved by a quadratic programming solver to obtain the optimal control sequence at the current moment ,Pick As the control input at the current moment, it is applied to the ballast water system; (2) Modify the control strategy based on actual wave and current conditions: The ballast water control strategy is modified according to the deviation between the actual measured wave and current sea condition parameters and the predicted values; the correction coefficient, wind direction and other parameters are continuously optimized through real-time monitoring and comprehensive analysis of wind power, wave and current sea conditions and unit operating status. , crew attitude, ballast water tank level And make predictions.

[0014] 10. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 1, characterized in that: the method of step 5 is: (1) Adjusting the ballast water system based on deep learning model and feedback control: The long short-term memory network is used to predict the future operation status with the historical operation status data sequence as input; the status data sequence includes wind speed, wind direction, unit attitude, and ballast water tank level; According to the prediction results and control strategies, the ballast water system parameters are adjusted in advance, including: adjusting the ballast water tank level balance load changes according to the prediction of wind speed increase; (2) Optimize crew attitude, load and coordinated control: A real-time optimization model is constructed based on the measured wave and current parameters and the performance of the control system, and the functional relationship related to each parameter is determined by comprehensively considering the wave and current energy utilization efficiency, wind power utilization efficiency and load fluctuation; Solve the optimization problem to obtain the unit attitude and load adjustment, transmit it to the control system, and achieve precise adjustment by adjusting the blade angle, cabin direction and ballast water distribution, which is applied to the unit control while considering the impact on wind power utilization efficiency; A feedback control optimization objective function is constructed, and the expected and actual unit power and ballast water system status differences are comprehensively considered to obtain and apply the ballast water control quantity and system adjustment value. The wind power change trend is analyzed to ensure that the adjustment strategy is conducive to its stable improvement and achieve coordinated control. The present invention can achieve the following beneficial effects: 1. Utilize a large amount of historical meteorological data, through data preprocessing, sample point analysis and clustering algorithm, to build a comprehensive and accurate long-term wind distribution model for wind farms, accurately calculate the probability distribution of wind conditions, and provide a reliable data basis for subsequent wind condition forecasts and ballast water control strategy optimization.

[0015] 2. Comprehensively consider the various characteristic parameters of waves and currents, use advanced mathematical transformations (such as Fourier transform) and physical models to accurately calculate wave and current loads, and accurately analyze the load distribution of sea conditions, thereby providing accurate force basis for the formulation of unit attitude and ballast water control strategy.

[0016] 3. A ballast water control strategy that can automatically adjust according to the real-time operating conditions of the unit, changes in wind conditions, and dynamic changes in wave and current conditions. Adopting machine learning (such as random forest algorithm) and intelligent optimization algorithm (such as particle swarm optimization algorithm), combined with real-time monitoring data, the control strategy is adaptively optimized to ensure that the unit can operate stably and efficiently under different operating conditions.

[0017] 4. Establish a multi-factor collaborative optimization model to organically combine factors such as wind conditions, wave and current conditions, unit attitude and load, and achieve overall collaborative optimization when formulating ballast water control strategies. At the same time, use deep learning models (such as long short-term memory networks) to accurately predict the future operating status of the unit, adjust the control strategy in advance, and improve the unit's ability to respond to complex environmental changes.

[0018] 5. By using wind condition prediction and accurate wave and current load calculation, combined with an adaptive ballast water control strategy, the forces on the unit can be effectively balanced under various wind and sea conditions, reducing unstable phenomena such as shaking and tilting of the unit, greatly improving the stability of the unit, and reducing the risk of equipment damage caused by unstable operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0020] The preferred solution is Figure 1 As shown, a method for controlling ballast water of a floating wind turbine based on medium- and long-term wind power prediction includes the following steps: Step 1: Based on historical meteorological data, analyze and count the wind speed and wind direction of the wind farm, obtain the distribution of long-term wind conditions in the wind farm, and calculate the probability distribution of wind conditions.

[0021] 1. Data preparation and normalization: Assume that the wind speed measurement data sequence is ,in Indicates The wind speed value measured at each moment; the wind direction measurement data sequence is ,in Indicates The wind direction angle value measured at the moment (specify the value range of the wind direction angle, for example, north is , the angle increases gradually in the clockwise direction, and the value range is arrive ).

[0022] The wind speed and wind direction measurement data are normalized to facilitate subsequent analysis and calculation. For wind speed data, the following normalization formula is used: ; in: :Indicates the The wind speed value measured at a moment is is the serial number of the measurement time, the value range is from to the total measurement time.

[0023] : After normalization The wind speed value at a certain moment has a range that depends on the normalized calculation result. arrive (depending on the range of the wind speed data itself and how it is normalized).

[0024] : represents the wind speed measurement data sequence The minimum value of , that is, the minimum value of wind speed in all measurement moments.

[0025] : represents the wind speed measurement data sequence The maximum value among them is the maximum value of wind speed among all measuring moments.

[0026] Get the normalized wind speed data series .

[0027] For wind direction data, since it is an angle value, the following normalization method can be used (mapping the angle value to the [0, 1] interval): ; in: :Indicates the The wind direction angle value measured at the moment is also is the serial number of the measurement time, the value range is from To the total measurement time, the angle value range is arrive .

[0028] : After normalization The wind direction angle value at a moment is in the range of arrive The original wind direction angle value is The ratio is normalized.

[0029] Get the normalized wind direction data sequence .

[0030] At this time, the normalized wind speed and wind direction data are combined into the sample space ,in Represents a sample point.

[0031] 2. Calculate the Euclidean distance between sample points: Calculate the Euclidean distance between sample points , which is used to measure the similarity between different sample points (i.e., wind speed and wind direction combinations at different times). The Euclidean distance calculation formula is as follows: ; in: :Indicates the The sample points and The Euclidean distance between sample points is used to measure the distance between the two sample points (corresponding to the wind speed and wind direction combinations at different times) in space. The smaller the distance, the more similar the wind speed and wind direction combinations of the two sample points are.

[0032] and :They are and The normalized wind speed value at the moment, here and is a different measurement time sequence number, ranging from to the total measurement time.

[0033] and :They are and The wind direction angle value after normalization at each moment is also and is a different measurement time sequence number, ranging from to the total measurement time.

[0034] 3. Divide sample points based on clustering algorithm to obtain wind distribution: The sample points according to The values ​​of are divided into different clusters as the cluster centers of the classification results to obtain the distribution of long-term wind conditions in the wind farm. Clustering algorithm, its objective function is: ; in: :yes The objective function value of the clustering algorithm is used to measure the effect of clustering. The goal is to continuously adjust the cluster center so that The value of is the smallest, that is, the sum of the squares of the distances from each sample point to the cluster center to which it belongs is minimized, so as to achieve reasonable clustering division.

[0035] : The pre-set number of clusters indicates the number of clusters into which the sample points are to be divided, that is, the number of categories of different wind conditions that are expected to be obtained, which is determined by the user based on the specific situation and analysis requirements.

[0036] : No. cluster, that is, it contains the A set of sample points of similar wind conditions, The value range is from arrive , indicating different clustering categories.

[0037] : No. sample points, which consist of normalized wind speed and wind direction data, namely , is the serial number of the sample point, ranging from to the total number of sample points.

[0038] : No. The center of a cluster, whose initial value can be randomly selected, is a two-dimensional vector, which represents the normalized combination of wind speed and wind direction values ​​corresponding to the cluster center. In the clustering process, it is continuously updated through iteration so that the objective function Minimum.

[0039] In each iteration, the cluster centers are updated The formula is: ; in: :Indicates the The updated center value of a cluster is a two-dimensional vector, which is also composed of normalized wind speed and wind direction values.

[0040] : No. clusters, which has the same meaning as in the above objective function , that is, it contains the A collection of sample points of similar wind conditions.

[0041] : No. sample points, which has the same meaning as in the above objective function , which is the sample point composed of normalized wind speed and wind direction data.

[0042] :Indicates the Cluster The number of sample points in is used to calculate the average value of the sample points in the cluster to update the cluster center.

[0043] After multiple iterations, when the change in the cluster center is less than a preset threshold (such as 0.001), the clustering algorithm is considered to have converged. The clusters represent different distributions of long-term wind conditions in wind farms, and each cluster may correspond to a typical wind condition pattern, such as a cluster of light wind and stable wind direction, a cluster of strong wind and changeable wind direction, etc.

[0044] 4. Calculate the probability distribution of wind conditions: For each wind condition pattern (cluster) obtained by clustering, the proportion of the number of sample points in the cluster to the total number of sample points is counted to obtain the probability distribution of the wind condition. Species, respectively , statistics belong to The number of sample points for wind-like conditions is , then the wind conditions Probability , the probability distribution of wind conditions can be expressed as ,in: : Represents the total number of wind condition classifications, that is, the number of different wind condition patterns obtained by the clustering algorithm.

[0045] :Indicates the wind mode, The value range is from arrive , used to distinguish different wind condition categories.

[0046] :Indicates that it belongs to The number of sample points of a wind condition mode (cluster) is obtained by counting the sample points in each cluster.

[0047] : represents the total number of sample points, that is, the total number of sample points corresponding to all measurement moments, which is used to calculate the probability of each wind condition mode.

[0048] :Indicates that the wind condition belongs to The probability of a class is obtained by calculating the ratio of the number of sample points belonging to this class of wind conditions to the total number of sample points. Its value range is arrive and the sum of the probabilities of all wind condition categories is .

[0049] : Indicates wind conditions The probability distribution function is a set of all wind condition category probabilities, which is used to describe the overall probability distribution of wind conditions in a wind farm.

[0050] Step 2: Using historical meteorological data and external meteorological forecast services, the random forest algorithm is used to predict future wind distribution. Combined with the predicted wind distribution and the probability distribution of sea conditions, the influence of unit attitude and load on the control system performance is considered. By establishing a function and optimizing it with the gradient descent method, the ballast water control strategy is determined, and finally a wind power time series is generated.

[0051] The probability of the sea state refers to the possibility that the marine environment (including waves, currents and other factors) is in a certain state or within a certain range. The present invention is mainly obtained by statistically analyzing the calculation results of wave loads and flow loads. For example, the probability of the sea state can represent the probability of the wave height being in a certain interval, the flow velocity being in a certain range, and their combination.

[0052] 1. Based on historical meteorological data and external meteorological forecast services, a machine learning model is used to predict wind speed and direction to obtain the distribution of wind conditions in the future: Collect historical weather data, including wind speed ,wind direction ,temperature ,humidity , air pressure And other relevant meteorological data (consider the influence weight of these data on wind speed and direction prediction, and establish a multivariate prediction model).

[0053] Assume that the wind speed measurement data sequence is ,in Indicates The wind speed value measured at each moment; the wind direction measurement data sequence is ,in Indicates The wind direction angle value measured at the moment (specify the value range of the wind direction angle, for example, north is , the angle increases gradually in the clockwise direction, and the value range is arrive ); The temperature measurement data sequence is ,in Indicates The temperature value measured at each moment; the humidity measurement data sequence is ,in Indicates The humidity value measured at each moment; the pressure measurement data sequence is ,in Indicates The air pressure value measured at a moment.

[0054] Divide historical data into training set and test set, and divide them according to a certain ratio (such as 70% training set and 30% test set).

[0055] The training set data is preprocessed, including data cleaning (removing outliers), standardization (making the data mean 0 and variance 1), etc. The standardization formula is as follows: For wind speed : ,in Represents the original wind speed value, is the mean of the wind speed training set data, is the standard deviation, is the normalized wind speed value.

[0056] For wind direction : ,in Indicates the original wind direction angle value, is the mean of the wind direction training set data, is the standard deviation, is the normalized wind direction value.

[0057] For temperature : ,in represents the original temperature value, is the mean of the temperature training set data, is the standard deviation, is the normalized temperature value.

[0058] For humidity : ,in Represents the original humidity value, is the mean of the humidity training set data, is the standard deviation, is the standardized humidity value.

[0059] For air pressure : ,in At the original pressure value, is the mean of the air pressure training set data, is the standard deviation, is the normalized air pressure value.

[0060] The same preprocessing is performed on the test set data.

[0061] The random forest algorithm is used for wind forecasting. The random forest is an ensemble learning algorithm composed of multiple decision trees. For the construction of decision trees, information gain or Gini index can be used as the feature selection standard. Assume that the training set data is ,in is the characteristic vector (including wind speed, wind direction, temperature, etc.), is the true value of the corresponding wind speed or wind direction. When building a decision tree, from the training set Random draw with replacement samples, forming a training subset , then Perform feature selection and node splitting until the stopping condition is met (such as the tree depth reaches the preset value or the number of node samples is less than the threshold). For prediction, the test set samples are input into the constructed random forest. Each decision tree gives a prediction result, and the final prediction result is the average or majority vote of all decision tree prediction results. The mean absolute error (MAE) can be used to evaluate the performance of the model: ,in is the number of test set samples, is the true value, is the predicted value.

[0062] 2. Combined with the wave and current loads of the wind farm, the impact of the unit attitude and load on the control system performance is comprehensively considered to optimize the ballast water control strategy: According to the predicted wind distribution (including wind speed and direction) and historical sea condition data, wave and current loads are calculated (the calculation method is the same as the improved wave and current load calculation in step 1).

[0063] Assume that during the calculation of wave and current load, the sequence of wave height changes with time is: ,in Indicates The wave height value measured or calculated at each moment; the wave period sequence is ,in Indicates The wave period corresponding to the moment; the ocean current velocity sequence is ,in Indicates The current velocity values ​​measured at each location; the current direction sequence is ,in Indicates The current direction value measured at the moment (specify the value range of the current direction angle, for example, north is , the angle increases gradually in the clockwise direction, and the value range is arrive ).

[0064] Using Fourier transform, the wave model is converted from time domain to frequency domain, and we get: , in is the variation of wave height with time, For the The amplitude of the frequency component, is the angular frequency, is the phase, is the time that represents the motion caused by the wave period.

[0065] Applying the time domain and frequency domain conversion relationship to the equation of motion and using the Fourier transform, we get: ,in is the mass (the mass here refers to the mass of objects related to the action of waves or currents, such as the foundation structure of a floating wind turbine or the mass of a part of the structure affected by waves and currents), is the damping coefficient (a coefficient indicating the damping effect on motion under the action of wave flow), is the stiffness coefficient (a coefficient reflecting the ability of the structure to resist deformation), is the displacement (referring to the displacement of an object under the action of waves and currents, such as the displacement of the foundation of a floating wind turbine), is the external force (in wave load calculation, the external force can be expressed as a wave-related force, such as , The amplitude of the wave , propagation constant etc., the specific relationship can be determined according to the wave theory). is the amplitude of the wave, is the propagation constant, and It is the conversion result between time domain and frequency domain.

[0066] The wave response of the blades and the unit is calculated, and the wave load can be expressed as: ,in: : Represents the effective wave period (the effective wave period here is a characteristic period related to wave energy, etc.).

[0067] : is the zero-order moment of the wave energy spectrum (it is a parameter used to characterize specific properties of the wave energy spectrum when calculating wave loads).

[0068] : Characteristic energy parameters related to wave energy spectrum.

[0069] The calculation of flow load can be based on relevant theories of fluid mechanics, such as ,in: : is the fluid density (referring to the density of seawater).

[0070] : is the drag coefficient (a coefficient related to the shape of the object, surface roughness, etc., which affects the force of the ocean current on the object).

[0071] : is the force-bearing area (referring to the force-bearing area of ​​the floating wind turbine in the direction of the ocean current).

[0072] : is the flow velocity (i.e. the ocean current speed).

[0073] By wave load ( ) and flow loads ( ) are statistically analyzed to obtain the load distribution of the sea conditions.

[0074] At the same time, based on the predicted wind speed , combined with relevant equipment parameters of wind farms (such as wind turbine blade radius , air density , Power coefficient etc.), the wind power can be calculated.

[0075] Wind power The calculation formula is .

[0076] Analyzing the relationship between wind power and wave and current loads, the magnitude of wind power will affect the operating state and stress of the unit, and thus affect the bearing and response to wave and current loads. For example, higher wind power may cause the unit speed to increase, thereby changing the unit's posture in the wave and current and the load distribution it receives. Moreover, under different wind power levels, the unit's response characteristics to wave and current loads may also be different.

[0077] Consider the crew attitude (such as inclination , direction angle ) and loads (such as aerodynamic loads , wave load , Flow Load ) on the performance of the control system, such as establishing a control system performance evaluation function , where aerodynamic load With wind speed , blade angle , pitch angle etc., we can establish relationships such as ( is the aerodynamic coefficient, reflecting the aerodynamic characteristics, is the blade force bearing area).

[0078] Use gradient descent to derive the function To optimize, first calculate the function Regarding each variable ( etc.), and the learning rate is , in each iteration, the variables are updated according to the following formula:

[0079] Repeat the above iterative process until the function The value of converges or reaches the preset number of iterations. The variable value obtained at this time is the optimized combination of parameters such as the unit attitude and load, based on which the ballast water control strategy is determined. For example, the adjustment amount and direction of the ballast water are determined according to the optimized load conditions to balance the overall force of the unit and improve stability and power generation efficiency.

[0080] 3. Combine the above results to generate a wind condition time series: Based on the predicted wind distribution (wind speed and direction), combined with the unit attitude (such as inclination angle , direction angle ) and load ( , , ), calculate the size and direction of wave and current loads. The relationship between wind conditions and wave and current loads can be established through relevant physical models. For example, changes in wind conditions will cause changes in wave height, period, current speed and direction, which in turn affect wave and current loads. According to the predicted wind speed, unit attitude and load information at different times, the corresponding wind condition characteristic parameters (such as equivalent wind speed, wind direction change range, etc.) are comprehensively calculated to generate a wind condition time series. Suppose the wind condition time series is ,in Indicates The combination of wind condition characteristic parameters at a certain moment reflects the comprehensive wind condition at that moment.

[0081] Furthermore, combined with the calculation results of wind power, the relationship between wind characteristic parameters and wind power at different times is analyzed. For example, under certain wind conditions, although the wind speed is high, the wind power may not be at the optimal level due to unstable wind direction or other factors. Through the joint analysis of wind condition time series and corresponding wind power data, the operating status and energy acquisition efficiency of the unit at different times can be more accurately evaluated, providing a more comprehensive basis for subsequent optimization control.

[0082] Step 3: Measure the impeller position of the unit in real time, combine the wind conditions and control system performance, and use the neural network to determine the unit operation strategy; then combine the ballast water control strategy results of step 2, based on the optimization objective function and related constraints, use the particle swarm optimization algorithm to optimize the unit attitude and load to achieve maximum utilization of wave and current energy.

[0083] 1. Measure the impeller position of the unit in real time and determine the unit's operating strategy based on the blade angle and pitch angle: Combined with wind conditions (wind speed ,wind direction ) and control system performance, and use the neural network model to learn and find the stable operating range of the unit. For example, a multi-layer perceptron (MLP) neural network is used, whose structure includes an input layer (input wind condition information such as wind speed ,wind direction etc., as well as the unit's current blade angle and pitch angle ), hidden layer (multiple hidden layers can be set, and the number of neurons in each layer is determined according to the actual situation) and output layer (output stable operation status of the unit or adjustment suggestions). The neural network is trained through a large amount of historical data, and the weights and thresholds of the neural network are optimized so that it can accurately predict the stable operation range of the unit. Suppose the input vector of the neural network is , the output is , indicating whether the unit is in a stable operating range (such as Indicates stability, Indicates instability). The training process is based on a large number of known input And the corresponding output , adjust the connection weights inside the neural network and threshold , making the predicted output as close to the true output as possible.

[0084] According to the input-output relationship of the sample data, the corresponding relationship between the impeller position and the unit operation is determined. For example, the relationship between the impeller position and the unit power generation is established. , stability and other performance indicators ,in is the unit speed. By analyzing the model, the optimal impeller position under different working conditions is determined. The value range is , The value range is , within this range, by changing and The value of ,θ, , calculate the corresponding , find Maximum and meets stability requirements and The combination is the parameter setting corresponding to the optimal impeller position.

[0085] In this process, combined with the calculation of wind power (based on real-time wind speed and other parameters), the wind power output under different impeller positions is analyzed. Since wind power is closely related to factors such as wind speed and blade angle, optimizing wind power output by adjusting the impeller position is an important means to improve the power generation efficiency of the unit. For example, when the impeller position is at a certain angle, the wind power may reach the maximum value, and this optimal position may change with wind conditions.

[0086] For example, when the impeller position is at a certain angle, the wind power may reach the maximum value, and this optimal position may change with the change of wind conditions. Therefore, it is necessary to monitor the wind conditions in real time and dynamically adjust the impeller position according to the change of wind power. At the same time, considering the relationship between wind power and other loads (such as wave loads), the wind power and other loads (such as wind power) may be adjusted. , Flow Load ) because the change of wind power will affect the overall stress condition of the unit, thereby changing the response characteristics of the unit to wave and current loads.

[0087] Based on the close relationship between wind power and unit attitude and load, the control strategy of the unit is further optimized. For example, the optimization objective function considering wind power is established. , At the same time, the stability constraints of the unit (such as the inclination range , load limit When the particle swarm optimization algorithm is used to solve the optimization problem, wind power is taken as an important evaluation index into the fitness function calculation of the particle. Assume that the position vector of the particle represents the attitude and load variables of the unit (such as ), the velocity vector represents the rate of change of the variable. In each iteration, the particle moves according to its historical optimal position , the historical optimal position of the group gbest And its own speed to update the position. Step 4: Determine the reasonable range of ballast water capacity based on the unit operating parameters and control system performance, establish a state space model and solve the optimization problem in combination with wave and current sea conditions and real-time operating conditions, and formulate a ballast water control strategy; at the same time, modify the control strategy based on the deviation between the actual wave and current sea conditions and the predicted values.

[0088] 1. Develop ballast water control strategy based on the unit’s operating parameters and control system performance: Based on the maximum and minimum weight distribution of the unit, a reasonable range of ballast water capacity is obtained.

[0089] Combined with wave and current conditions (wave loads , Flow Load ), real-time operating conditions (unit attitude, load, etc.) and wind power to optimize the ballast water system. Using the model predictive control (MPC) algorithm, we first establish a discrete time state space model of the ballast water system, and then, based on the prediction time domain and control time domain , at each sampling moment , solve the following optimization problem: in To control the sequence, For forecast output (such as predicted crew attitude and ballast tank level, etc.), is the reference trajectory (such as the desired crew attitude and ballast tank level, etc.), and is a weight matrix used to balance control performance and control energy. The above optimization problem is solved by a specific quadratic programming solver to obtain the optimal control sequence at the current moment ,Pick As the control input at the current moment, it is applied to the ballast water system. For example, if the solution is If (water pump flow) is a positive value, it means that water needs to be added to the ballast tank; if it is a negative value, it means that water needs to be discharged, and the valve opening Adjustments are also made accordingly to control the flow rate.

[0090] During the optimization process, the impact of wind power on ballast water control is considered. For example, when wind power is high and has an upward trend, in order to better utilize wind energy and maintain the stability of the unit, it may be necessary to adjust the ballast water distribution so that the unit posture is more conducive to the capture and conversion of wind power. At the same time, according to the relationship between wind power and wave and current loads, the ballast water capacity is reasonably adjusted to balance the stress of the unit under the combined action of wind, waves and currents, to avoid affecting the life of the unit due to excessive stress or reducing the power generation efficiency due to uneven stress.

[0091] 2. Modify the control strategy based on actual wave and current conditions: According to the actual measured wave and current sea condition parameters (such as wave height , flow rate ) and the predicted value, and correct the ballast water control strategy. For example, establish a correction factor ,in is the wave height deviation, The correction factor is applied to the calculation of the ballast water control quantity for the flow rate deviation, such as the water pump flow control quantity ( is the flow control quantity before correction). Assume that the actual wave height measured is , the predicted wave height is ,but Similarly, let the actual flow rate measured be , the predicted flow rate is ,but By adjusting the correction coefficient, the adjustment amount and speed of ballast water can be more carefully controlled to ensure that the unit can operate stably in complex environments and make full use of wind energy resources.

[0092] Through real-time monitoring and comprehensive analysis of wind power, wave and current conditions, and unit operation status, the correction coefficient, wind direction, and wind speed are continuously optimized. , crew attitude, ballast water tank level The input of the LSTM network is the historical running status data sequence. Suppose the input sequence is ,in Indicates The running status data vector at each moment is output as the prediction of the future running status. . Adjust the parameters of the ballast water system according to the prediction results and the set control strategy. For example, according to the change trend of the operating status predicted by the LSTM network, adjust the parameters of the ballast water system in advance so that it can better adapt to the upcoming changes in operating conditions. If it is predicted that the wind speed will increase, which may cause changes in the force on the unit, the ballast water tank level can be adjusted in advance to balance the upcoming increase in load.

[0093] Step 5: Monitor the operating status of the unit in real time, and adjust the ballast water system parameters using deep learning models and feedback control methods; based on the measured wave and current parameters and control system performance, establish a real-time optimization model to solve the unit attitude and load adjustment and apply it to control. At the same time, use the feedback control mechanism to dynamically adjust the ballast water system and control strategy to achieve coordinated control.

[0094] 1. Monitor the operating status of the unit in real time and adjust the ballast water system according to the generated control strategy: Using deep learning models (such as long short-term memory networks (LSTMs)) and feedback control methods to dynamically adjust the ballast water system. For example, using LSTM to monitor the operating status of the unit (including wind speed ,wind direction , Crew attitude , , Ballast water tank level The input of the LSTM network is the historical running status data sequence. Suppose the input sequence is ,in Indicates The running status data vector at each moment is output as the prediction of the future running status. . Adjust the parameters of the ballast water system according to the prediction results and the set control strategy. For example, according to the change trend of the operating status predicted by the LSTM network, adjust the parameters of the ballast water system in advance so that it can better adapt to the upcoming changes in operating conditions. If it is predicted that the wind speed will increase, which may cause changes in the force on the unit, the ballast water tank level can be adjusted in advance to balance the upcoming increase in load.

[0095] According to the measured wave and current parameters (wave height , flow rate ) and control system performance, optimize the unit attitude and load. The wind power utilization efficiency can be calculated based on the ratio of actual wind power to theoretical maximum wind power. According to the real-time measured wave and flow parameters and the unit operating status, the optimization problem is solved to obtain the optimal unit attitude and load adjustment at the current moment. The obtained unit attitude and load adjustment are transmitted to the unit control system in real time, and the unit attitude and load are accurately adjusted by adjusting the blade angle, cabin direction, and ballast water distribution, and applied to the control of the unit. For example, if it is calculated that the load fluctuation should be reduced at present, the aerodynamic load can be changed by adjusting the blade angle, or the ballast water distribution can be adjusted to balance the wave load and flow load, while considering the impact on the wind power utilization efficiency to ensure that the effective utilization of wind power is not reduced in the process of optimizing load fluctuations.

[0096] 2. Dynamically adjust the ballast water system and control strategy through sensors and feedback control mechanisms: Considering the real-time operating conditions of the unit (wind speed ,wind direction , Crew attitude , , Load , , etc.) and offshore environment (wave height , flow rate etc.), adjust the unit operation strategy by optimizing the objective function, and obtain the ballast water control quantity ;By optimizing the objective function, the ballast water system adjustment value is calculated , to achieve coordinated control of the ballast water system and the unit. For example, establish the optimization objective function of feedback control: , in and They are The expected unit power at any given moment (considering the combined power of wind power and other energy sources) and the actual unit power, and They are The expected ballast water system state (such as liquid level, water volume, etc.) and the actual ballast water system state at all times are obtained by solving the optimization objective function. Minimum ballast water control and ballast water system adjustment values , and applied to system control. For example, if Lower than and Higher than , it may be necessary to reduce the amount of ballast water to adjust the unit attitude, increase the power generation capacity, and make the unit operation closer to the expected state. At the same time, analyze the changing trend of wind power during this adjustment process to ensure that the adjustment strategy is conducive to the continuous and stable improvement of wind power.

[0097] In the present invention, the wave and current sea conditions refer to various data information related to waves and currents, including wave and current data and sea condition data. The wave and current data include wave height data, wave period data, wave direction data, and wave spectrum data; the sea condition data include current speed data and current direction data; specifically include: (1) Wave-related data: Wave height data: This is the key data describing the size of waves, such as the sequence of wave height changes over time at different times. It can be obtained through on-site measurement equipment (such as buoys, wave radars, etc.) and is an important indicator of wave energy.

[0098] Wave period data: includes wave period information, such as effective wave period. Wave period is related to wave frequency, which is very important for understanding the motion characteristics and energy distribution of waves. Frequency domain period information can be obtained from time domain wave data through Fourier transform and other methods.

[0099] Wave direction data: The propagation direction of waves is crucial for analyzing their impact on floating wind turbines, because the load conditions (such as the size and direction of wave loads) borne by the turbines will be different under the action of waves in different directions.

[0100] Wave spectrum data: such as the zero-order moment of the wave energy spectrum and other related data, which can describe the characteristics of waves from an energy perspective and is used to calculate the forces exerted by waves on structures.

[0101] (2) Ocean current related data Current velocity data: This is one of the most important data of ocean currents, such as flow velocity. The magnitude and direction of the current velocity will affect the position offset and stress of the floating wind turbine. It can be measured by equipment such as the Acoustic Doppler Current Profiler (ADCP).

[0102] Ocean current direction data: Similar to wave direction, ocean current direction determines the direction of its force on the unit, which is critical for accurate calculation of flow loads, etc. (3) Wave-current interaction data In the actual marine environment, there is interaction between waves and currents. For example, the surface velocity distribution data under the combined action of waves and currents will change the flow field structure around the floating wind turbine, thereby affecting the load, motion response and many other performances of the unit. These interaction data are also part of the wave and current related data, which is essential for accurately analyzing the behavior of the unit in a complex marine environment.

[0103] Wind conditions include many factors, and wind power is a quantitative reflection of one of them. Wind conditions are a comprehensive concept, which includes wind speed, wind direction, wind stability (such as whether the wind speed changes frequently, whether the wind direction is stable, etc.), wind turbulence intensity and other factors. Wind power is mainly related to wind speed, but is also affected by factors such as air density and wind turbine blade parameters. It can be said that wind power is a quantitative description of wind conditions from an energy perspective. For example, under the same wind direction and stability conditions, the greater the wind speed, the greater the wind power.

[0104] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for controlling ballast water of a floating wind turbine based on medium- and long-term wind power prediction, characterized in that The following steps are involved: Step 1: Based on historical meteorological data, analyze and count the wind speed and wind direction of the wind farm, obtain the distribution of long-term wind conditions in the wind farm, and calculate the probability distribution of wind conditions; Step 2: Use historical meteorological data and external meteorological forecast services to predict future wind distribution using the random forest algorithm; combine the predicted wind distribution and the probability distribution of sea conditions, consider the impact of unit attitude and load on control system performance, establish a function and optimize it using the gradient descent method to determine the ballast water control strategy, and finally generate a wind power time series; Step 3: Real-time measurement of the unit's impeller position, combined with wind conditions and control system performance, using neural networks to determine the unit's operating strategy; combined with the ballast water control strategy results of step 2, based on the optimization objective function and related constraints, the particle swarm optimization algorithm is used to optimize the unit's attitude and load to achieve maximum utilization of wave and current energy; Step 4: Determine the reasonable range of ballast water capacity based on the unit operating parameters and control system performance, establish a state space model and solve the optimization problem in combination with the wave and current sea conditions and real-time operating conditions, and formulate a ballast water control strategy; at the same time, modify the control strategy based on the deviation between the actual wave and current sea conditions and the predicted value; Step 5: Monitor the operating status of the unit in real time, and adjust the ballast water system parameters using deep learning models and feedback control methods; based on the measured wave and current parameters and control system performance, establish a real-time optimization model to solve the unit attitude and load adjustment and apply it to control. At the same time, use the feedback control mechanism to dynamically adjust the ballast water system and control strategy to achieve coordinated control.

2. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 1 is characterized in that: The sub-steps of step 1 are: Step 1.1: Collect wind speed and direction data of the wind farm, construct the corresponding measurement data sequence, and specify the range of wind direction values; Step 1.2: Normalize the wind speed and wind direction data respectively. The wind speed is reduced to the interval of 0 to 1 according to its own maximum value range, and the wind direction is mapped to the interval of 0 to 1 according to the ratio of 360°, forming a normalized data sequence and combining it into a sample space; Step 1.3, calculate the Euclidean distance between sample points according to the formula to measure the similarity of wind speed and wind direction combinations at different times; Step 1.4: Use the K-Means clustering algorithm to divide the sample points into different clusters based on distance. After multiple iterations until the cluster center changes to the standard, determine the wind condition distribution, and calculate the proportion of sample points in each cluster to obtain the wind condition probability distribution.

3. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 2 is characterized in that: The method of step 1.1 is: Assume that the wind speed measurement data sequence is ,in Indicates The wind speed value measured at each moment; the wind direction measurement data sequence is ,in Indicates The wind direction angle value measured at each moment; The measured data of wind speed and wind direction are normalized to facilitate subsequent analysis and calculation; for wind speed data, the following normalization formula is used: ； in: :Indicates the The wind speed value measured at a moment is is the serial number of the measurement time, the value range is from To the total measurement time; : After normalization The wind speed value at a moment; : represents the wind speed measurement data sequence The minimum value among , that is, the minimum value of wind speed among all measurement moments; : represents the wind speed measurement data sequence The maximum value among them, that is, the maximum value of wind speed among all measurement moments; Get the normalized wind speed data sequence ; : After normalization The wind speed value at a moment; : represents the wind speed measurement data sequence The minimum value among , that is, the minimum value of wind speed among all measurement moments; : represents the wind speed measurement data sequence The maximum value among them, that is, the maximum value of wind speed among all measurement moments; Get the normalized wind speed data sequence .

4. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 3 is characterized in that: The method for step 1.2 is: Calculate the Euclidean distance between sample points , used to measure the similarity between different sample points; The Euclidean distance calculation formula is as follows: ； in: :Indicates the The sample points and The Euclidean distance between two sample points is used to measure the distance between the two sample points in space. The smaller the distance, the more similar the wind speed and wind direction combinations of the two sample points are. and :They are and The normalized wind speed value at the moment, here and is a different measurement time sequence number, ranging from To the total measurement time; and :They are and The wind direction angle value after normalization at each moment is also and is a different measurement time sequence number, ranging from to the total measurement time.

5. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 4 is characterized in that: The method for step 1.3 is: The sample points according to The values ​​of are divided into different clusters as the cluster centers of the classification results to obtain the distribution of long-term wind conditions in the wind farm; Clustering algorithm, its objective function is: ； in: :yes The objective function value of the clustering algorithm is used to measure the effect of clustering; the goal is to continuously adjust the cluster center so that The value of is the smallest, that is, the sum of the squares of the distances from each sample point to the cluster center to which it belongs is minimized, so as to achieve reasonable clustering division; : The pre-set number of clusters, which indicates the number of clusters into which the sample points are to be divided, that is, the number of categories of different wind conditions that are expected to be obtained, which is determined by the user according to the specific situation and analysis requirements; : No. cluster, that is, it contains the A set of sample points of similar wind conditions, The value range is from arrive , indicating different clustering categories; : No. sample points, which consist of normalized wind speed and wind direction data, namely , is the serial number of the sample point, ranging from to the total number of sample points; : No. The center of a cluster, whose initial value can be randomly selected, is a two-dimensional vector, which represents the normalized combination of wind speed and wind direction values ​​corresponding to the cluster center. In the clustering process, it is continuously updated through iteration so that the objective function Minimum; In each iteration, the cluster centers are updated The formula is: ； in: :Indicates the The updated center value of a cluster is a two-dimensional vector, which is also composed of normalized wind speed and wind direction values; : No. clusters, which has the same meaning as in the above objective function , that is, it contains the A collection of sample points of similar wind conditions; : No. sample points, which has the same meaning as in the above objective function , which is the sample point composed of normalized wind speed and wind direction data; :Indicates the Cluster The number of sample points in is used to calculate the average value of the sample points in the cluster to update the cluster center; After multiple iterations, when the change of the cluster center is less than the preset threshold, the clustering algorithm is considered to have converged. Each cluster represents the different distribution of long-term wind conditions in the wind farm.

6. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 5 is characterized in that: The method for step 1.4 is: For each wind condition pattern obtained by clustering, the proportion of the number of sample points in the cluster to the total number of sample points is counted to obtain the probability distribution of the wind condition; assuming that the total number of wind condition classifications is Species, respectively , statistics belong to The number of sample points for wind-like conditions is , then the wind conditions Probability , the probability distribution of wind conditions can be expressed as ,in: : represents the total number of wind condition classifications, that is, the number of different wind condition patterns obtained by the clustering algorithm; :Indicates the wind mode, The value range is from arrive , used to distinguish different wind condition categories; :Indicates that it belongs to The number of sample points of a wind condition pattern, that is, the number of sample points of a cluster, is obtained by counting the sample points in each cluster; : represents the total number of sample points, that is, the total number of sample points corresponding to all measurement moments, which is used to calculate the probability of each wind condition mode; :Indicates that the wind condition belongs to The probability of a class is obtained by calculating the ratio of the number of sample points belonging to this class of wind conditions to the total number of sample points. Its value range is arrive and the sum of the probabilities of all wind condition categories is ; : Indicates wind conditions The probability distribution function is a set of all wind condition category probabilities, which is used to describe the overall probability distribution of wind conditions in a wind farm.

7. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 1, characterized in that: The method for step 2 is: (1) Based on historical meteorological data and external meteorological forecast services, a machine learning model is used to predict wind speed and direction to obtain the distribution of wind conditions in the future: Collect historical weather data, including wind speed ,wind direction ,temperature ,humidity , air pressure Relevant meteorological data; Assume that the wind speed measurement data sequence is ,in Indicates The wind speed value measured at each moment; the wind direction measurement data sequence is ,in Indicates The wind direction angle value measured at each moment; the temperature measurement data sequence is ,in Indicates The temperature value measured at each moment; the humidity measurement data sequence is ,in Indicates The humidity value measured at each moment; the pressure measurement data sequence is ,in Indicates The air pressure value measured at a moment; Divide historical data into training set and test set, and preprocess the training set data. The preprocessing methods include data cleaning and standardization. The normalization formula is as follows: For wind speed : ,in Represents the original wind speed value, is the mean of the wind speed training set data, is the standard deviation, is the normalized wind speed value; For wind direction : ,in Indicates the original wind direction angle value, is the mean of the wind direction training set data, is the standard deviation, is the normalized wind direction value; For temperature : ,in represents the original temperature value, is the mean of the temperature training set data, is the standard deviation, is the standardized temperature value; For humidity : ,in Represents the original humidity value, is the mean of the humidity training set data, is the standard deviation, is the standardized humidity value; For air pressure : ,in At the original pressure value, is the mean of the air pressure training set data, is the standard deviation, is the normalized air pressure value; The random forest algorithm is used to predict wind conditions. Assume that the training set data is: ， in is the feature vector, is the true value of the corresponding wind speed or wind direction; When building a decision tree, from the training set Random draw with replacement samples, forming a training subset , then Perform feature selection and node splitting until the stopping condition is met; for prediction, input the test set samples into the constructed random forest, each decision tree gives a prediction result, and the final prediction result is the average or majority vote of all decision tree prediction results; the performance of the model can be evaluated using the mean absolute error: ,in is the number of test set samples, is the true value, is the predicted value; (2) Combined with the wave and current loads of the wind farm, the impact of the unit attitude and load on the control system performance is comprehensively considered to optimize the ballast water control strategy: Calculate the wave and current loads based on the predicted wind distribution and historical sea condition data: Assume that during the calculation of wave and current load, the sequence of wave height changes with time is: ,in Indicates The wave height value measured or calculated at each moment; the wave period sequence is ,in Indicates The wave period corresponding to the moment; the ocean current velocity sequence is ,in Indicates The current velocity values ​​measured at each location; the current direction sequence is ,in Indicates The current direction value measured at each moment; Using Fourier transform, the wave model is converted from time domain to frequency domain, and we get: ， in is the variation of wave height with time, For the The amplitude of the frequency component, is the angular frequency, is the phase, is the time that represents the motion caused by the wave period; Applying the time domain and frequency domain conversion relationship to the equation of motion and using the Fourier transform, we get: ,in For quality, is the damping coefficient, is the stiffness coefficient, is the displacement, For external force, is the amplitude of the wave, is the propagation constant, and It is the conversion result between time domain and frequency domain; The wave response of the blades and the unit is calculated, and the wave load is expressed as: ,in: : represents the effective wave period; : is the zero-order moment of the wave energy spectrum; : characteristic energy parameters related to wave energy spectrum; Through the wave load and flow loads The calculation results are statistically analyzed to obtain the load distribution of the sea conditions; at the same time, based on the predicted wind speed , combined with the relevant equipment parameters of the wind farm, calculate the wind power; Analyze the relationship between wind power and wave and current loads. The magnitude of wind power will affect the operating status and stress of the unit, and thus affect the bearing and response of wave and current loads. (3) Based on the predicted wind distribution, combined with the unit attitude and load, the magnitude and direction of the wave and current load are calculated; based on the predicted wind speed, unit attitude and load at different times, the corresponding wind characteristic parameters are comprehensively calculated to generate a wind time series; Assume that the wind condition time series is ,in Indicates The combination of wind condition characteristic parameters at each moment reflects the comprehensive wind condition at that moment; combined with the calculation results of wind power, the relationship between wind condition characteristic parameters and wind power at different moments is analyzed.

8. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 1, characterized in that: The method for step 3 is: Real-time measurement of the impeller position of the unit, based on the blade angle and pitch angle , determine the unit's operating strategy: Combine wind conditions with control system performance, that is, wind speed ,wind direction With the control system performance, use the neural network model to learn and find out the stable operating range of the unit; A multi-layer perceptron neural network is used, and its structure includes an input layer, a hidden layer, and an output layer; The input layers include: wind speed ,wind direction , and the unit's current blade angle and pitch angle ; The neural network is trained through a large amount of historical data, and the weights and thresholds of the neural network are optimized so that it can accurately predict the stable operating range of the unit. Suppose the input vector of the neural network is , the output is , indicating whether the unit is in a stable operating range; based on a large number of known input And the corresponding output , adjust the connection weights inside the neural network and threshold , so that the predicted output is as close to the true output as possible; According to the input-output relationship of the sample data, the corresponding relationship between the impeller position and the unit operation is determined; by analyzing the model, the optimal impeller position under different working conditions is determined; Based on the close relationship between wind power and unit attitude and load, the control strategy of the unit is further optimized; the optimization objective function considering wind power is established: ， At the same time, the stability constraints of the unit are met; the constraints include: inclination range , load limit .

9. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 1, characterized in that: The method for step 4 is: (1) Develop ballast water control strategies based on the unit’s operating parameters and control system performance: Based on the maximum and minimum weight distribution of the unit, the reasonable range of ballast water capacity is obtained; The ballast water system is optimized by combining wave and current conditions, real-time operating conditions and wind power. The model predictive control algorithm is used to first establish a discrete time state space model of the ballast water system, and then, based on the prediction time domain and control time domain , at each sampling moment , solve the following optimization problem: ； in To control the sequence, To predict the output, is the reference trajectory, and is the weight matrix, which is used to balance the control performance and control energy. The above optimization problem is solved by the quadratic programming solver to obtain the optimal control sequence at the current moment. ,Pick As the control input at the current moment, it is applied to the ballast water system; (2) Modify the control strategy based on actual wave and current conditions: The ballast water control strategy is modified according to the deviation between the actual measured wave and current sea condition parameters and the predicted values; the correction coefficient, wind direction and other parameters are continuously optimized through real-time monitoring and comprehensive analysis of wind power, wave and current sea conditions and unit operation status. , crew attitude, ballast water tank level And make predictions.

10. The method for controlling ballast water of a floating wind turbine generator system based on medium- and long-term wind power prediction according to claim 1, characterized in that: The method for step 5 is: (1) Adjusting the ballast water system based on deep learning model and feedback control: The long short-term memory network is used to predict the future operation status with the historical operation status data sequence as input; the status data sequence includes wind speed, wind direction, unit attitude, and ballast water tank level; According to the prediction results and control strategies, the ballast water system parameters are adjusted in advance, including: adjusting the ballast water tank level balance load changes according to the prediction of wind speed increase; (2) Optimize crew attitude, load and coordinated control: A real-time optimization model is constructed based on the measured wave and current parameters and the performance of the control system, and the functional relationship related to each parameter is determined by comprehensively considering the wave and current energy utilization efficiency, wind power utilization efficiency and load fluctuation; Solve the optimization problem to obtain the unit attitude and load adjustment, transmit it to the control system, and achieve precise adjustment by adjusting the blade angle, cabin direction and ballast water distribution, which is applied to the unit control while considering the impact on wind power utilization efficiency; A feedback control optimization objective function is constructed, and the expected and actual unit power and ballast water system status differences are comprehensively considered to obtain and apply the ballast water control quantity and system adjustment value. The wind power change trend is analyzed to ensure that the adjustment strategy is conducive to its stable improvement and achieve coordinated control.

Images