Short-term prediction and adjustment method of floating wind turbine motion response and mooring tension based on deep learning

Abstract

The short-term prediction and adjustment method of floating wind turbine motion response and mooring tension based on deep learning belongs to the field of floating wind power. In order to solve the problems of offshore wind turbine motion response and mooring tension prediction and adjustment, the key point is that the prediction module selects the prediction model matching the environmental sea conditions in the model training module according to the environmental sea conditions where the target wind turbine is located, and the signal acquisition module collects the motion response, mooring tension and wave height time history, and inputs the matching prediction module through the transmission module. The prediction module predicts the wind turbine platform motion response and mooring tension in the next period of time after the current moment; the control module determines whether the wind turbine platform motion response is within an acceptable range in the next period of time, and whether the mooring tension exceeds the safety critical value or is lower than the set value in the next period of time. When it is determined that at least one of the above two conditions is not met, the control module sends an adjustment command to the mooring adjustment module, which improves the working safety of the offshore wind turbine platform.

Description

Technical Field

[0001] The present invention belongs to the technical field of floating wind power, and relates to a short-term prediction and adjustment method and a safety protection device for the motion response of a floating wind turbine and the motion response of a mooring system based on a deep learning recurrent neural network and a sensor monitoring system. Background Art

[0002] In recent years, my country's offshore wind power industry has developed rapidly, and the application of offshore floating wind turbines has gradually moved to the deep sea. Although offshore wind power has broad development prospects, the environment faced by offshore wind power is more severe than that of onshore wind power. Due to climate, topography and other reasons, my country's southeastern coastal areas are often plagued by typhoons. When extreme weather comes, the motion response and mooring tension of floating wind turbines change dramatically. If the motion response of floating wind turbines can be accurately predicted in advance under extreme sea conditions, and the mooring system is adjusted for safety protection, it can effectively prevent the damage and fracture of the mooring system caused by excessive wind turbine motion response exceeding the safety range. The development of deep learning technology has made it possible to realize this technology. By using recurrent neural network technology, the motion response of offshore floating wind turbines can be effectively predicted in the short term and safety protection adjusted. Summary of the invention

[0003] In order to solve the problem of offshore wind turbine motion response, mooring tension prediction and adjustment, according to the floating wind turbine motion response and mooring tension short-term prediction and adjustment method based on deep learning recorded in the embodiment of the present application, a prediction system is used for prediction, and the prediction system includes a software part and a hardware part. The software part includes a database module, a model training module, a prediction module and a control module, and the hardware part includes a positioning module, a posture module, a tension module, a signal acquisition module, a transmission module and an adjustment module;

[0004] S10. The database module calculates the motion response, mooring tension data and wave height time history of the wind turbine under various sea conditions according to the sea conditions at the location of the wind turbine through the floating wind turbine simulation software SIMA, and forms a target floating wind turbine motion response database, wherein the motion response includes the position and the change angles of roll, pitch and yaw in three directions of the wind turbine;

[0005] S20. The model training module trains the wind turbine motion response and tension prediction model, and the training data used is from the database module. After completing the prediction model training, the model training module saves different wind turbine motion response and tension prediction models obtained under various sea conditions;

[0006] S30. The prediction module selects the prediction model in the model training module that matches the environmental sea conditions according to the environmental sea conditions where the target wind turbine is located, the real-time motion response collected on-site by the positioning module and the attitude module, the real-time mooring tension collected on-site by the tension module, and the wave height time history collected on-site, and the motion response, mooring tension and wave height time history are collected by the signal acquisition module, and input into the matching prediction module through the transmission module, and the prediction module predicts the motion response and mooring tension of the wind turbine platform in a period of time after the current moment;

[0007] S40. The control module determines whether the wind turbine platform motion response and the mooring tension in a period of time after the current moment predicted by the prediction module are within an acceptable range in the period of time after the current moment, and whether the mooring tension exceeds a safety critical value or is lower than a set value in the period of time after the current moment. When it is determined that at least one of the above two conditions is not satisfied, the control module sends an adjustment command to the mooring adjustment module.

[0008] S50. The mooring adjustment module receives the adjustment instruction issued by the control module, and adjusts the anchor chain length of the mooring system according to the predicted value of the wind turbine platform motion response and mooring tension in the next period of time predicted by the prediction module;

[0009] S60. After the length of the anchor chain of the mooring system is adjusted, the tension module collects the mooring tension of the mooring system on site, and the positioning module and the attitude module collect the motion response of the wind turbine after the length of the anchor chain is adjusted, collect the wave height time history on site, input the motion response of the wind turbine after the length of the anchor chain is adjusted, the mooring tension and the wave height time history into the prediction module, and the prediction module predicts the motion response of the wind turbine platform and the mooring tension in a period of time after the current moment;

[0010] S70. Iteratively execute steps S40 to S60 until the control module determines, based on the wind turbine platform motion response and mooring tension in a period of time after the current moment predicted by the prediction module, that the wind turbine platform motion response in the period of time after the current moment is within an acceptable range and the mooring tension in the period of time after the current moment does not exceed the safety critical value.

[0011] According to the method for short-term prediction and adjustment of floating wind turbine motion response and mooring tension based on deep learning described in the embodiment of the present application, in step S40, when it is determined that at least one of the above two conditions is not satisfied, it includes:

[0012] The wind turbine platform motion response is not within the acceptable range during the latter period of time, and the mooring tension exceeds the safety critical value during the latter period of time;

[0013] The wind turbine platform motion response is within an acceptable range during the latter period of time, and the mooring tension is lower than a set value during the latter period of time;

[0014] In step S50, the mooring adjustment module adjusts the anchor chain length of the mooring system according to the predicted value of the wind turbine platform motion response and mooring tension in the next period of time predicted by the prediction module, which specifically includes:

[0015] The wind turbine platform motion response is not within an acceptable range during the latter period of time, and the mooring tension exceeds a safety critical value during the latter period of time, and the mooring adjustment module loosens the anchor chain;

[0016] The motion response of the wind turbine platform is within an acceptable range during the latter period of time, and the mooring tension is lower than a set value during the latter period of time, and the mooring adjustment module tightens the anchor chain.

[0017] According to the method for short-term prediction and adjustment of motion response and mooring tension of a floating wind turbine based on deep learning described in the embodiment of the present application, the positioning module includes a dynamic positioning system installed in the wind turbine, and the dynamic positioning system includes an inertial navigation system and a satellite positioning system, which are used to determine the real-time position of the wind turbine;

[0018] The attitude module includes sensors for measuring the real-time changing angles of the wind turbine's roll, pitch and yaw in three directions.

[0019] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, the attitude module includes a gyroscope.

[0020] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, the tension module includes a tension sensor installed on the mooring system, which is used to collect and monitor the tension of the wind turbine mooring system in real time.

[0021] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method recorded in the embodiments of the present application, the signal acquisition module is used to collect the signal of the real-time position of the wind turbine output by the positioning module, to collect the signal of the real-time change angle of the wind turbine's roll, pitch and bow in three directions output by the attitude module, and to collect the signal of the pulling force on the wind turbine mooring system output by the tension module.

[0022] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, the transmission module transmits the signals collected by the signal acquisition module to the prediction module.

[0023] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, the mooring adjustment module includes an anchor windlass, a traction rope and a load weight.

[0024] According to the method for short-term prediction and adjustment of motion response of a floating wind turbine and mooring tension based on deep learning described in an embodiment of the present application, the load weight is set at the first end of the traction rope, and the second end of the traction rope is connected to the anchor chain of the mooring system through the anchor windlass. The anchor windlass pulls up or down the load weight through the traction rope, so that the anchor chain connected to the traction is relaxed or tightened as the load weight is pulled up or down, and the length of the anchor chain is adjusted to adjust the mooring tension.

[0025] According to the method for short-term prediction and adjustment of the motion response and mooring tension of a floating wind turbine based on deep learning recorded in the embodiments of the present application, a multi-input long short-term memory neural network is used to build the training model, and the multi-input long short-term memory neural network used adds the wave height time history to the training process of the prediction model, and participates in the training with the input motion response and anchor tension to form three-dimensional data, wherein the hyperparameters of the multi-input long short-term memory neural network are as follows: the optimizer is the Adam method, the number of neurons is 128, the BatchSize is 256, the multi-input long short-term memory neural network uses a 3-layer network structure to consider the nonlinear characteristics of the motion response and anchor tension, and has accuracy under the influence of second-order hydrodynamics, and the multi-input long short-term memory neural network controls the transmission of information by a gating equation, and the basic form of the gating equation is as follows:

[0026] G(x)=σ(W•x+b)

[0027] Where w is the weight matrix, b is the bias term, σ is the activation function sigmoid, and the gate output vector is obtained by element-wise multiplication. The gate effect is achieved by controlling the vector. The gate output vector is a real number vector between 0 and 1. When the gate output G(x) is 0, any input information multiplied by G(x) will get 0, indicating that no information can pass. When the gate output G(x) is 1, multiplication will not produce any change, indicating that all information can pass.

[0028] The multi-input long short-term memory neural network is composed of gate equations, including forget gate, input gate, genetic gate, and output gate. The forget gate determines the degree of retention of internal genetic information at time t-1 according to ct-1, and transmits it to the internal genetic information c of the unit at time t; the input gate determines the current unit state according to ht-1, and saves and transmits it; the genetic gate retains part of the information in ht-1 and inherits it to ct; the output gate selects to retain and transmit the output to ht according to ht-1;

[0029] The gating equations are as follows:

[0030] f t =σ(W f ·[h t-1 , x t ]+b f )

[0031] i t =σ(W i ·[h t-1 , x t ]+b i )

[0032] c t =f t ·c t-1 +i c tanh(W C ·[h t-1 , x t ]+b c )

[0033] o t =σ(W o ·[h t-1 , x t ]+b o )

[0034] h t =o t ·tanh(c t )

[0035] The deep learning model adopts the sequence-to-point method in the process of predicting the motion response of the wind turbine and the anchor chain tension, and uses the sliding window method to establish the sequence-to-point relationship between the model input and output. The three-dimensional time series data from t-20 to t is used as an input window X, and the motion response and anchor chain force at t+20 are used as the output point Y. The XY window is then slid backward through time step iteration.

[0036] The beneficial effects of the present invention are as follows:

[0037] (1) The components of the device, including stress strain gauges, sensors, anchor winches, etc., are all mature industrial products and easy to implement.

[0038] (2) It has a good effect in dealing with the time series problem of floating wind turbine motion response and anchor chain tension.

[0039] (3) The wind turbine’s ability to respond to dangerous situations is improved, effectively preventing the occurrence of anchor chain overload and enhancing the safety of the mooring system.

[0040] (4) According to the experimental data recorded in the examples of this application, the method of the present invention has a very high accuracy in predicting motion response and mooring tension, thereby enabling adaptive, real-time, safe and reliable adjustment of the mooring system according to complex environments. In particular, the real-time adjustment is carried out through iterative collection and prediction, and the mooring system sets a safety threshold and a set value. The safety threshold setting ensures that the wind turbine is always within the safe tightening range of the mooring system in a complex environment, and does not cause uncontrollable motion response due to drastic changes in the environment. The set value setting enables the mooring system to relax in time when the environment changes to a friendly trend, so that the mooring system is not always in a tense state, thereby improving the response sensitivity of the mooring system and extending the use time under a better use state. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is an overview of the mooring of a semi-submersible floating wind turbine.

[0042] Figure 2 This is a diagram of the layout of the mooring system adjustment module.

[0043] Figure 3 This is the schematic diagram of the model training module.

[0044] Figure 4 It is a schematic diagram of the signal acquisition-response prediction-mooring adjustment method.

[0045] Figure 5 are prediction example diagrams, where (a) is a prediction example diagram of the platform motion response, and (b) is a prediction example diagram of the platform anchor chain tension response.

[0046] Figure 6 This is a schematic diagram of the internal structure of a multi-input long short-term memory neural network.

[0047] Figure 7 It is a schematic diagram of the sliding window method.

[0048] Among them: 1. Water surface line, 2. Platform side column, 3. Anchor chain, 4. Bottom buoy, 5. Towing rope, 6. Towing channel, 7. Anchor windlass, 8. Load weight. DETAILED DESCRIPTION

[0049] The following is a detailed description of the specific implementation of the present invention in combination with the accompanying drawings and technical solutions. The purpose of the present invention is to propose a method and a safety protection device for short-term prediction of the motion response of a floating wind turbine and the tension of a mooring system, thereby improving the reliability and safety of the wind turbine in complex sea conditions. The method for short-term prediction and adjustment of the motion response of a floating wind turbine and the tension of a mooring system based on deep learning includes:

[0050] The short-term motion response prediction method adopts a prediction system for prediction. The prediction system is mainly composed of software and hardware parts. The software part includes a database module, a model training module, a prediction module, and a control module. The hardware part includes a positioning module, a posture module, a tension module, a signal acquisition module, a transmission module, and an adjustment module.

[0051] The database module is a target floating wind turbine motion response database calculated by the floating wind turbine simulation software SIMA, which takes into account a variety of sea conditions that may be encountered at the wind turbine deployment location.

[0052] The model training module is a prediction model training system built using a multi-input long short-term memory neural network (MI-LSTM). The training data used comes from the database module. After the prediction model training is completed, different wind turbine motion response prediction models obtained under various sea conditions are saved.

[0053] The prediction module selects the prediction model in the model training module that matches the environmental sea conditions according to the environmental sea conditions where the target wind turbine is located, and uses the real-time motion response, mooring tension and wave height time history collected in the signal acquisition module as input factors to predict the motion response and mooring tension of the wind turbine platform after a period of time.

[0054] The control module determines whether the platform is in a dangerous state according to the motion response of the wind turbine after a short time and the mooring tension obtained by the prediction module, that is, whether the motion response is within an acceptable range and whether the mooring tension is about to exceed the safety critical value or be lower than the set value, specifically including that the motion response of the wind turbine platform is not within the acceptable range in the latter period of time, and the mooring tension exceeds the safety critical value in the latter period of time; the motion response of the wind turbine platform is within the acceptable range in the latter period of time, and the mooring tension is lower than the set value in the latter period of time, thereby sending an adjustment command to the mooring adjustment module; if the motion response is within the acceptable range at the prediction time, and the mooring tension will not exceed the safety critical value at the prediction time, then no adjustment command is sent.

[0055] Specifically, the steps include:

[0056] S10. The database module calculates the motion response, mooring tension data and wave height time history of the wind turbine under various sea conditions according to the sea conditions at the location of the wind turbine through the floating wind turbine simulation software SIMA, and forms a target floating wind turbine motion response database, wherein the motion response includes the position and the change angles of roll, pitch and yaw in three directions of the wind turbine;

[0057] S20. The model training module trains the wind turbine motion response and tension prediction model, and the training data used is from the database module. After completing the prediction model training, the model training module saves different wind turbine motion response and tension prediction models obtained under various sea conditions;

[0058] S30. The prediction module selects the prediction model in the model training module that matches the environmental sea conditions according to the environmental sea conditions where the target wind turbine is located, the real-time motion response collected on-site by the positioning module and the attitude module, the real-time mooring tension collected on-site by the tension module, and the wave height time history collected on-site, and the motion response, mooring tension and wave height time history are collected by the signal acquisition module, and input into the matching prediction module through the transmission module, and the prediction module predicts the motion response and mooring tension of the wind turbine platform in a period of time after the current moment;

[0059] S40. The control module determines whether the wind turbine platform motion response and the mooring tension in a period of time after the current moment predicted by the prediction module are within an acceptable range in the period of time after the current moment, and whether the mooring tension exceeds a safety critical value or is lower than a set value in the period of time after the current moment. When it is determined that at least one of the above two conditions is not satisfied, the control module sends an adjustment command to the mooring adjustment module.

[0060] S50. The mooring adjustment module receives the adjustment instruction issued by the control module, and adjusts the anchor chain length of the mooring system according to the predicted value of the wind turbine platform motion response and mooring tension in the next period of time predicted by the prediction module;

[0061] S60. After the length of the anchor chain of the mooring system is adjusted, the tension module collects the mooring tension of the mooring system on site, and the positioning module and the attitude module collect the motion response of the wind turbine after the length of the anchor chain is adjusted, collect the wave height time history on site, input the motion response of the wind turbine after the length of the anchor chain is adjusted, the mooring tension and the wave height time history into the prediction module, and the prediction module predicts the motion response of the wind turbine platform and the mooring tension in a period of time after the current moment;

[0062] S70. Iteratively execute steps S40 to S60 until the control module determines, based on the wind turbine platform motion response and mooring tension in a period of time after the current moment predicted by the prediction module, that the wind turbine platform motion response in the period of time after the current moment is within an acceptable range and the mooring tension in the period of time after the current moment does not exceed the safety critical value.

[0063] According to the method for short-term prediction and adjustment of floating wind turbine motion response and mooring tension based on deep learning described in the embodiment of the present application, in step S40, when it is determined that at least one of the above two conditions is not satisfied, it includes:

[0064] The wind turbine platform motion response is not within the acceptable range during the latter period of time, and the mooring tension exceeds the safety critical value during the latter period of time;

[0065] The wind turbine platform motion response is within an acceptable range during the latter period of time, and the mooring tension is lower than a set value during the latter period of time;

[0066] In step S50, the mooring adjustment module adjusts the anchor chain length of the mooring system according to the predicted value of the wind turbine platform motion response and mooring tension in the next period of time predicted by the prediction module, which specifically includes:

[0067] The wind turbine platform motion response is not within an acceptable range during the latter period of time, and the mooring tension exceeds a safety critical value during the latter period of time, and the mooring adjustment module loosens the anchor chain;

[0068] The motion response of the wind turbine platform is within an acceptable range during the latter period of time, and the mooring tension is lower than a set value during the latter period of time, and the mooring adjustment module tightens the anchor chain.

[0069] According to the method for short-term prediction and adjustment of motion response and mooring tension of a floating wind turbine based on deep learning described in the embodiment of the present application, the positioning module includes a dynamic positioning system installed in the wind turbine, and the dynamic positioning system includes an inertial navigation system and a satellite positioning system, which are used to determine the real-time position of the wind turbine;

[0070] The attitude module includes sensors for measuring the real-time changing angles of the wind turbine's roll, pitch and yaw in three directions.

[0071] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, the attitude module includes a gyroscope.

[0072] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, the tension module includes a tension sensor installed on the mooring system, which is used to collect and monitor the tension of the wind turbine mooring system in real time.

[0073] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method recorded in the embodiments of the present application, the signal acquisition module is used to collect the signal of the real-time position of the wind turbine output by the positioning module, to collect the signal of the real-time change angle of the wind turbine's roll, pitch and bow in three directions output by the attitude module, and to collect the signal of the pulling force on the wind turbine mooring system output by the tension module.

[0074] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, the transmission module transmits the signals collected by the signal acquisition module to the prediction module.

[0075] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, the mooring adjustment module includes an anchor windlass, a traction rope and a load weight.

[0076] According to the method for short-term prediction and adjustment of motion response of a floating wind turbine and mooring tension based on deep learning described in an embodiment of the present application, the load weight is set at the first end of the traction rope, and the second end of the traction rope is connected to the anchor chain of the mooring system through the anchor windlass. The anchor windlass pulls up or down the load weight through the traction rope, so that the anchor chain connected to the traction is relaxed or tightened as the load weight is pulled up or down, and the length of the anchor chain is adjusted to adjust the mooring tension.

[0077] According to the deep learning-based floating wind turbine motion response and mooring tension short-term prediction and adjustment method described in the embodiments of the present application, a multi-input long short-term memory neural network is constructed to obtain the training model.

[0078] The mooring adjustment module receives the adjustment instruction issued by the control module, and adjusts the length of the anchor chain of the mooring system according to the predicted value of the wind turbine platform motion response and the mooring tension in the next period of time predicted by the prediction module. Specifically, the mooring adjustment module is composed of an anchor machine, a traction rope, a load weight, etc. More specifically, the first end of the traction rope is provided with a load weight, and the second end of the traction rope is connected to the anchor chain of the mooring system through the anchor machine. The anchor machine pulls up or lowers the load weight through the traction rope, so that the anchor chain connected to the traction is relaxed or tightened as the load weight is pulled up or lowered, and the length of the anchor chain is adjusted to adjust the mooring tension. Among them, the mooring (anchor chain) length is adjusted according to the predicted value of the wind turbine platform response and the mooring tension given by the prediction module, specifically including that the wind turbine platform motion response is not within the acceptable range in the next period of time, and the mooring tension exceeds the safety critical value in the next period of time, and the mooring adjustment module loosens the anchor chain; the wind turbine platform motion response is within the acceptable range in the next period of time, and the mooring tension is lower than the set value in the next period of time, and the mooring adjustment module tightens the anchor chain.

[0079] After the length of the mooring system anchor chain is adjusted, the tension module collects the mooring tension of the mooring system again on site, and the positioning module and the attitude module collect the motion response of the wind turbine after the adjustment of the anchor chain length again, and the wave height time history is collected again on site, and the motion response of the wind turbine after the adjustment of the anchor chain length, the mooring tension and the wave height time history are input into the prediction module, and the prediction module predicts the wind turbine platform motion response and mooring tension for a period of time after the current moment again;

[0080] The execution is iterated until the control module determines that the wind turbine platform motion response and the mooring tension in a period of time after the current moment predicted by the prediction module are within an acceptable range in the period of time after the current moment, and the mooring tension in the period of time after the current moment does not exceed the safety critical value or is lower than the set value.

[0081] Among them: the positioning module is a dynamic positioning system installed in the wind turbine. The inertial navigation system and satellite positioning system contained in the dynamic positioning system determine the real-time position of the wind turbine through sensors. The attitude module is a sensor that measures the real-time change angles of the wind turbine's roll, pitch, and yaw in three directions, such as a gyroscope. The tension module is a tension sensor installed on the mooring system, which can collect and monitor the tension of the wind turbine mooring system in real time. The signal acquisition module is used to collect signals output by the positioning module, attitude module, and tension module. The transmission module transmits the signals collected by the signal acquisition module to the prediction module.

[0082] Among them, the present invention adopts a multi-input long short-term memory neural network to build the training model. The long short-term memory neural network belongs to a recurrent neural network, and because a memory unit is added to each small unit structure, it can save a part of the timing information inherited by the previous sequence, so it has better performance when facing irregular timing prediction problems. The motion response of the offshore floating wind turbine platform is obviously affected by the sea conditions when it is working. The multi-input long short-term memory neural network used adds the wave height time history to the training process of the prediction model, and the input motion response and anchor chain tension form three-dimensional data to participate in the training. Among them, the hyperparameters of the multi-input long short-term memory neural network are as follows: the optimizer is the Adam method, the number of neurons is 128, and the Batch Size is 256.

[0083] The multi-input long short-term memory neural network uses a three-layer network structure, which fully considers the nonlinear characteristics of motion response and anchor chain tension, and also has high accuracy when considering the influence of second-order hydrodynamics.

[0084] The multi-input long short-term memory neural network uses the gating equation to control the transmission of information. The basic form of the gating equation is as follows:

[0085] G(x)=σ(W·x+b)

[0086] Where w is the weight matrix, b is the bias term, and σ is the activation function sigmoid. The gate output vector is obtained by element-wise multiplication, and the gate effect is achieved by controlling the vector. The gate output vector is a real number vector between 0 and 1. When the gate output G(x) is 0, any input information multiplied by G(x) will result in 0, indicating that no information can pass. When the gate output G(x) is 1, multiplication will not produce any change, indicating that all information can pass.

[0087] The multi-input long short-term memory neural network is composed of gate equations, including forget gate, input gate, genetic gate, and output gate. The forget gate determines the degree of retention of internal genetic information at time t-1 according to ct-1, and transmits it to the internal genetic information c of the unit at time t; the input gate determines the current unit state according to ht-1, and saves and transmits it; the genetic gate retains part of the information in ht-1 to ct; the output gate selects to retain and transmit the output to ht according to ht-1.

[0088] The gating equations are as follows:

[0089] f t =σ(W f ·[h t-1 , x t ]+b f )

[0090] i t =σ(W i ·[h t-1 , x t ]+b i )

[0091] c t =f t ·c t-1 +i c tanh(W C ·[h t-1 , x t ]+b c )

[0092] o t =σ(W o ·[h t-1 , x t ]+b o )

[0093] h t =o t ·tanh(c t )

[0094] The deep learning model uses a sequence-to-point method to predict the motion response of the wind turbine and the anchor chain tension. The sliding window method is used to establish the sequence-to-point relationship between the model input and output. The three-dimensional time series data from t-20 to t is used as an input window X, the motion response and anchor chain force at t+20 are used as the output point Y, and the XY window is slid backward through time step iteration.

[0095] The preferred wind turbine of the present invention is a floating offshore wind turbine, specifically:

[0096] (1) Taking a 5MW Braceless semi-submersible floating wind turbine as an example, considering the possible sea conditions in the sea area where the wind turbine is located, the motion response and mooring tension data of the wind turbine platform under various sea conditions are simulated and calculated in the fully coupled analysis software SIMA. The wave height time history and the motion response and tension at time t are taken as input factors, and the response and tension after t+20s are taken as the target output. MI-LSTM is imported for deep learning model training to obtain the short-term response prediction model of the wind turbine under various sea conditions, and a prediction model library is established.

[0097] (2) In the actual marine environment where the wind turbine is located, based on the real-time six-degree-of-freedom motion response and mooring tension of the wind turbine collected by the positioning module, attitude module, and tension module, a prediction model matching the current sea conditions is selected from the model library. In the prediction module, the short-term predicted motion response and mooring tension after 20 seconds are obtained from the real-time data, and the prediction results are transmitted to the control module.

[0098] (3) The control module compares the six-degree-of-freedom motion response after 20 seconds with the mooring tension and the critical value and the normal set value. If it exceeds the dangerous value, it is judged as dangerous and sends an adjustment command to the mooring adjustment module to loosen the anchor chain; if it is lower than the normal value, it is judged as abnormal and sends an adjustment command to the mooring adjustment module to tighten the anchor chain. For example, in extreme conditions such as typhoons, the longitudinal and transverse swings of the wind turbine platform may deviate greatly. When it is predicted that the longitudinal swing response will be too large after 20 seconds, causing the tension of anchor chains No. 2 and 3 to exceed the critical dangerous value, the control module sends a command to the mooring adjustment module to loosen the anchor chain. After the typhoon weakens, when it is predicted that the tension of anchor chains No. 2 and 3 will be lower than the set value after 20 seconds, the control module sends a command to the mooring adjustment module to tighten the anchor chain.

[0099] (4) The mooring adjustment module performs relevant operations according to the commands sent by the control module. When the command received is to loosen the anchor chain, the adjustment module automatically releases the load weight stored on the wind turbine. The anchor chain force can be used to pull the weight up and down through the anchor windlass to increase the movement distance of the wind turbine, release energy and reduce the force on the anchor chain; when the command received is to tighten the anchor chain, the adjustment module recovers a part of the anchor chain by actively rotating and fixing the anchor windlass. The load weight is fixed and cannot move up and down, and the anchor chain force returns to the normal set value.

[0100] Figure 5 (a) is a diagram showing an example of the prediction of the platform motion response. Figure 5 (b) is a prediction example diagram of the platform anchor chain tension response. It can be seen that the method of the present invention has a very high accuracy in predicting the motion response and mooring tension, thereby enabling the mooring system to be adaptively, real-time, safely and reliably adjusted according to the complex environment. In particular, the real-time adjustment is achieved through iterative collection and prediction. The mooring system sets a safety threshold and a set value. The safety threshold setting ensures that the wind turbine is always within the safe tightening range of the mooring system in a complex environment, and the motion response is not uncontrollable due to drastic changes in the environment. The set value setting enables the mooring system to be released in time when the environment changes to a friendly trend. The mooring system is not always in a taut state, thereby improving the response sensitivity of the mooring system and extending the use time under a better use state.

[0101] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for short-term prediction and adjustment of motion response and mooring tension of floating wind turbines based on deep learning, characterized in that: The prediction system is used for prediction. The prediction system includes a software part and a hardware part. The software part includes a database module, a model training module, a prediction module and a control module. The hardware part includes a positioning module, a posture module, a tension module, a signal acquisition module, a transmission module and an adjustment module. S10. The database module calculates the motion response, mooring tension data and wave height time history of the wind turbine under various sea conditions according to the sea conditions at the location of the wind turbine through the floating wind turbine simulation software SIMA, and forms a target floating wind turbine motion response database, wherein the motion response includes the position and the change angles of roll, pitch and yaw in three directions of the wind turbine; S20. The model training module trains the wind turbine motion response and tension prediction model, and the training data used is from the database module. After completing the prediction model training, the model training module saves different wind turbine motion response and tension prediction models obtained under various sea conditions; S30. The prediction module selects the prediction model in the model training module that matches the environmental sea conditions according to the environmental sea conditions where the target wind turbine is located, the real-time motion response collected on-site by the positioning module and the attitude module, the real-time mooring tension collected on-site by the tension module, and the wave height time history collected on-site, and the motion response, mooring tension and wave height time history are collected by the signal acquisition module, and input into the matching prediction module through the transmission module, and the prediction module predicts the motion response and mooring tension of the wind turbine platform in a period of time after the current moment; S40. The control module determines whether the wind turbine platform motion response and the mooring tension in a period of time after the current moment predicted by the prediction module are within an acceptable range in the period of time after the current moment, and whether the mooring tension exceeds the safety critical value or is lower than the set value in the period of time after the current moment. When it is determined that at least one of the two conditions is not satisfied, the control module sends an adjustment command to the mooring adjustment module. S50. The mooring adjustment module receives the adjustment instruction issued by the control module, and adjusts the anchor chain length of the mooring system according to the predicted value of the wind turbine platform motion response and mooring tension in the next period of time predicted by the prediction module; S60. After the length of the anchor chain of the mooring system is adjusted, the tension module collects the mooring tension of the mooring system on site, and the positioning module and the attitude module collect the motion response of the wind turbine after the length of the anchor chain is adjusted, collect the wave height time history on site, input the motion response of the wind turbine after the length of the anchor chain is adjusted, the mooring tension and the wave height time history into the prediction module, and the prediction module predicts the motion response of the wind turbine platform and the mooring tension in a period of time after the current moment; S70. Iteratively execute steps S40 to S60 until the control module determines that the wind turbine platform motion response and the mooring tension in a period of time after the current moment predicted by the prediction module are within an acceptable range in the period of time after the current moment, and the mooring tension does not exceed the safety critical value in the period of time after the current moment; In step S40, when it is determined that at least one of the two conditions is not satisfied, it includes: The wind turbine platform motion response is not within the acceptable range during the latter period of time, and the mooring tension exceeds the safety critical value during the latter period of time; The wind turbine platform motion response is within an acceptable range during the latter period of time, and the mooring tension is lower than a set value during the latter period of time; In step S50, the mooring adjustment module adjusts the anchor chain length of the mooring system according to the predicted value of the wind turbine platform motion response and mooring tension in the next period of time predicted by the prediction module, which specifically includes: The wind turbine platform motion response is not within an acceptable range during the latter period of time, and the mooring tension exceeds a safety critical value during the latter period of time, and the mooring adjustment module loosens the anchor chain; The wind turbine platform motion response is within an acceptable range during the latter period of time, and the mooring tension is lower than a set value during the latter period of time, and the mooring adjustment module tightens the anchor chain; The positioning module includes a dynamic positioning system installed in the wind turbine, and the dynamic positioning system includes an inertial navigation system and a satellite positioning system, which is used to determine the real-time position of the wind turbine; The attitude module includes sensors for measuring the real-time changing angles of the wind turbine's roll, pitch and yaw in three directions; The attitude module includes a gyroscope; The tension module includes a tension sensor installed on the mooring system, which is used to collect and monitor the tension of the wind turbine mooring system in real time; The signal acquisition module is used to acquire the signal of the real-time position of the wind turbine output by the positioning module, to acquire the signal of the real-time change angle of the wind turbine's roll, pitch and yaw in three directions output by the attitude module, and to acquire the signal of the pulling force on the wind turbine mooring system output by the tension module; The transmission module transmits each signal collected by the signal collection module to the prediction module; The mooring adjustment module includes an anchor windlass, a traction rope and a load weight; The load weight is arranged at the first end of the traction rope, and the second end of the traction rope is connected to the anchor chain of the mooring system through the anchor windlass. The anchor windlass pulls up or down the load weight through the traction rope, so that the anchor chain connected to the traction rope is relaxed or tightened as the load weight is pulled up or down, and the length of the anchor chain is adjusted to adjust the mooring tension; A training model was built using a multi-input long short-term memory neural network. The multi-input long short-term memory neural network added the wave height time history to the training process of the prediction model, and the input motion response and anchor chain tension formed three-dimensional data for training. The hyperparameters of the multi-input long short-term memory neural network are as follows: the optimizer is the Adam method, the number of neurons is 128, the batch size is 256, and the multi-input long short-term memory neural network uses a 3-layer network structure to consider the nonlinear characteristics of motion response and anchor chain tension, and has accuracy under the influence of second-order hydrodynamics. The multi-input long short-term memory neural network uses a gating equation to control the transmission of information. The basic form of the gating equation is as follows: Where w is the weight matrix, b is the bias term, σ is the activation function sigmoid, and the gate output vector is obtained by element-wise multiplication. The gate effect is achieved by controlling the vector. The gate output vector is a real number vector between 0 and 1. When the gate output G(x) is 0, any input information multiplied by G(x) will get 0, indicating that no information can pass. When the gate output G(x) is 1, multiplication will not produce any change, indicating that all information can pass. The deep learning model uses a sequence-to-point method to predict the wind turbine motion response and anchor chain tension, and uses a sliding window method to establish a sequence-to-point relationship between the model input and output. The three-dimensional time series data from t-20 to t is used as an input window X, and the motion response and anchor chain force at t+20 are used as output points Y. The XY window is then slid backward through time step iteration. The mooring adjustment module performs relevant operations according to the commands sent by the control module. When the command received is to loosen the anchor chain, the adjustment module automatically releases the load weight stored on the wind turbine. The anchor chain force can pull the weight up and down through the anchor machine, thereby increasing the movement distance of the wind turbine, releasing energy and reducing the stress on the anchor chain. When the command received is to tighten the anchor chain, the adjustment module recovers part of the anchor chain by actively rotating and fixing the anchor machine. The load weight is fixed and cannot move up and down, and the anchor chain force returns to the normal set value.

Images