An active control method for a structure in a wave tank for point-wise wave damping
By constructing a machine learning model, the active motion of the structure generates radiation waves to eliminate waves in the flume, solving the problems of reflected wave influence and breakwater applicability in two-dimensional cross-section flume tests, and achieving a high-precision and highly adaptable wave-damping effect.
Patent Information
- Application Number
- CN202510109387.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-23
- Publication Date
- 2026-08-25
- Estimated Expiration
- 2045-01-23
AI Technical Summary
In existing technologies, incident waves generated in two-dimensional cross-sectional flume tests produce reflected waves when propagating within the flume, affecting the accuracy of test results. Fixed breakwaters have high construction and maintenance costs and are not suitable for deep seas, while floating breakwaters have poor wave-damping effects.
Machine learning methods are used to construct a parameter mapping relationship between the submersion depth of a structure, the amplitude of motion, the distance to the target point, and the wave height. The active motion of the structure is used to generate radiation waves to eliminate waves at the point to be eliminated, and time prediction is combined to optimize control decisions.
It achieves high-precision wave elimination at any point within a two-dimensional cross-section flume, making it suitable for efficient wave dissipation in long-term breakwater structures, adapting to water level changes, and applicable to deep-sea environments.
Smart Images
Figure CN120046482B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wave dissipation and disaster reduction, and in particular relates to a method for active control of fixed-point wave dissipation of structures based on machine learning. Background Technology
[0002] Oceans cover seven-tenths of the Earth's surface and contain abundant resources. With increasing societal demands for resources, the development and utilization of marine resources has gradually become a key area. Currently, the design of offshore structures often employs two-dimensional cross-sectional flume tests to assess the reliability of the structure and the feasibility of the selected design.
[0003] One of the limitations is the size of the water tank. When the incident wave generated at the wave-generating end of the water tank propagates inside the water tank, even if wave-damping facilities / equipment are arranged at the end, it will inevitably generate reflected waves. This phenomenon will affect the results of long-term high-fidelity tests.
[0004] On the other hand, breakwaters are currently mainly divided into two categories: fixed breakwaters and floating breakwaters. Fixed breakwaters have high construction and maintenance costs, are not suitable for deep-sea areas, and are difficult to adapt to the impact of water level changes. Floating breakwaters are suitable for deep-sea areas, but because they drift with the current, their wave-damping effect is poor. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention proposes an active control method for structures used in wave-damping at fixed points within wave tanks. This method utilizes machine learning to construct a parameter mapping relationship between the structure's immersion depth, motion amplitude, target point distance, and target point wave height. Combining this with the advantages of time prediction, the method predicts the structure's motion amplitude and controls its movement based on the predicted amplitude to generate the desired radiated waves, thereby eliminating waves at the target point.
[0006] The technical solution adopted in this invention is as follows:
[0007] I. An active control method for structures used for fixed-point wave damping in wave tanks
[0008] The active control method for the structure includes the following steps:
[0009] S1) Obtain the temporal immersion depth and temporal motion amplitude of the feature points of the structure, obtain the temporal radiation wave height of the first side at different measurement points on the first side of the structure, and the temporal radiation wave height of the second side at different measurement points on the second side opposite to the first side, and use the obtained data to construct a dataset.
[0010] In step S1), the immersion depth at the start of each time step, the motion amplitude within the time step, the distance between each measuring point, and the radiation wave height at each measuring point are constructed into a set of sample data for each time step.
[0011] S2) Build a machine learning model and train it using the dataset obtained in step S1).
[0012] In step S2), the machine learning model maps the following relationship:
[0013] ① The relationship between the distance between measuring points, the amplitude of motion within the time step, the immersion depth at the start of the time step, and the radiation wave height at the measuring point on the first side of the structure (i.e., the radiation wave height on the first side).
[0014] ② The relationship between the distance between measuring points, the motion amplitude within each time step, the immersion depth at the start of the time step, and the radiation wave height at the measuring point on the second side of the structure (i.e., the radiation wave height on the second side).
[0015] S3) Obtain the distance between the point to be eliminated and the feature point of the structure as the distance between the point to be eliminated. Obtain the initial immersion depth of the feature point of the structure and the initial wave height of the point to be eliminated. Use the trained machine learning model to predict the motion amplitude of the structure in the first time step based on the collected distance between the point to be eliminated, the initial immersion depth of the feature point of the structure and the initial wave height of the point to be eliminated. Control the structure to start moving based on the predicted motion amplitude.
[0016] Step S3) specifically involves:
[0017] S3.1) An incident wave is generated in the wave tank, and the incident wave is incident from the first side; the incident wave generates an incident wave and a reflected wave on the first side of the structure, and the reflected wave and the incident wave propagate in opposite directions; the incident wave generates a transmitted wave on the second side of the structure, and the transmitted wave and the incident wave propagate in the same direction.
[0018] S3.2) Obtain the distance between the point to be eliminated and the structure, as well as the initial immersion depth of the feature points of the structure. Use the distance between the point to be eliminated and the structure as the distance between the point to be eliminated, and collect the wave height of the point to be eliminated at the initial moment as the initial wave height.
[0019] In specific implementation, the process of obtaining the initial wave height in step S3.2) can be replaced by the following steps: determine whether the point to be eliminated is located on the first or second side of the structure. If the point to be eliminated is located on the first side of the structure, use the Goda two-point method to separate the incident wave height and the reflected wave height in the wave field, and set the sum of the incident wave height and the reflected wave height at the initial moment as the initial wave height. If the point to be eliminated is located on the second side of the structure, collect the transmitted wave height at the point to be eliminated, and set the transmitted wave height at the initial moment as the initial wave height.
[0020] S4) During the wave dissipation process, the wave height of the point to be dissipated and the immersion depth of the structural feature points are collected in real time. The real-time wave height of the point to be dissipated and the real-time immersion depth of the structural feature points are obtained. The trained machine learning model is used to predict the motion amplitude of the structure based on the real-time wave height of the point to be dissipated, the real-time immersion depth of the structural feature points and the distance of the point to be dissipated. The motion of the structure is controlled according to the predicted motion amplitude, so as to dissipate the wave at the point to be dissipated.
[0021] Specifically, in step S4), a trained machine learning model is used to predict the amplitude of the structure's motion in the next time step based on the distance to the point to be eliminated, the immersion depth at the start of the current time step, and the wave height of the point to be eliminated at the end of the current time step. The movement of the structure in the next time step is then controlled based on the amplitude of the structure's motion in the next time step.
[0022] Further, in steps S3) and S4), a pre-trained wave height prediction model can be used to process the initial wave height of the point to be eliminated and the wave height of the point to be eliminated at the end of the current time step to obtain the predicted wave height considering the time delay. Then, the predicted wave height considering the time delay is input into the trained machine learning model. Specifically, the wave height prediction model is a prediction model about the relationship between wave height and time.
[0023] Furthermore, the active control method for the structure may also include the following process: accumulating the time-series data collected during the wave dissipation process in step S4), constructing a dataset based on the time-series data, and using the dataset to optimize the trained machine learning model.
[0024] II. An active control system for fixed-point wave damping in a wave tank
[0025] The active control system for the structure includes:
[0026] A structure placed in a body of water to dissipate waves at a point to be dissipated by generating radiated waves.
[0027] An electric motor is connected to the structure for transmission and is used to drive the structure to move.
[0028] Several wave height meters are arranged at the point to be cleared to collect the initial wave height and real-time wave height of the point. The initial wave height is the wave height, transmitted wave height, or incident wave height and reflected wave height separated using the Goda two-point method at the initial moment at the point to be cleared.
[0029] The training module is used to construct a dataset through theoretical solutions, experimental measurements, or numerical simulations. This dataset is then used to train a machine learning model. The trained model predicts and outputs the real-time motion amplitude of the structure based on the real-time wave height at the point to be mitigated, the real-time immersion depth of the structure's feature points, and the distance to the point to be mitigated, all received from the active wave mitigation control module. Specifically, when the training module constructs the dataset through experimental measurements, the active control system for the structure also includes several wave height meters arranged at various measuring points along the propagation direction of the radiated waves. These wave height meters are used to collect the time-series radiated wave heights on the first and second sides. The training module includes a data acquisition unit and a machine learning unit. The input of the data acquisition unit is communicatively connected to the output of the wave height meters arranged at each measuring point and to the motor. The output of the data acquisition unit is communicatively connected to the input of the machine learning unit, and the machine learning unit is coupled to the active wave mitigation control module.
[0030] The active wave suppression control module is coupled with the training module, and its output is also connected to the control terminal of the motor. It is used to obtain the real-time wave height of the point to be suppressed and the real-time immersion depth of the structural feature point. The real-time wave height of the point to be suppressed, the real-time immersion depth of the structural feature point, and the distance of the point to be suppressed are input into the trained machine learning model. The real-time motion amplitude of the structure is received from the trained machine learning model, and the movement of the structure is controlled by controlling the motor based on the real-time motion amplitude of the structure.
[0031] The beneficial effects of this invention are as follows:
[0032] 1. This invention actively controls the amplitude of the structure's motion in the next time step, thereby eliminating waves at any point in a two-dimensional cross-section water tank.
[0033] 2. The method of the present invention can achieve long-term, high-precision wave generation for the wavefront point to be eliminated of a structure.
[0034] 3. The method of the present invention can eliminate waves at the wave-after point of a structure, and as an actively controlled breakwater structure, it protects the facilities at the target point.
[0035] 4. The method of the present invention is applicable to application scenarios that consider the impact of time processing delay on control during the implementation of the method, making the control decision more convenient for practical application in engineering applications such as control decision for efficient wave dissipation of breakwaters. Attached Figure Description
[0036] Figure 1 This is a schematic diagram of the active control method and system for structures in this invention;
[0037] Figure 2 This is a schematic diagram of the training module in this invention;
[0038] Figure 3 This is a schematic diagram of the active wave damping control module in this invention;
[0039] Figure 4 This is a schematic diagram of the movement of the structure in this invention. Detailed Implementation
[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.
[0041] The first aspect of this invention provides an active control method for structures used in fixed-point wave damping within wave tanks. The active control method for structures of this invention includes the following steps:
[0042] S1) Obtain the temporal immersion depth and temporal motion amplitude of the feature points of the structure during the motion process, obtain the temporal radiation wave height of the first side at different measuring points on the first side of the structure, and the temporal radiation wave height of the second side at different measuring points on the second side opposite to the first side, obtain the distance between each measuring point and the structure as their respective measuring point distances, and construct a dataset using the obtained measuring point distances and temporal data.
[0043] Here, the motion amplitude represents the displacement distance within one time step. For example... Figure 4 As shown, for sway and heave, the amplitude of motion represents the displacement of the object; for roll, the amplitude of motion represents the sway angle of the object.
[0044] In practice, feature points of the structure are selected according to usage requirements. For example, for regular or irregular objects, their edges or center of gravity can be selected as feature points of the structure to facilitate measurement.
[0045] The acquisition methods for the aforementioned time-series data include, but are not limited to, theoretical solutions, experimental measurements, or numerical simulations. The specific process for acquiring time-series data using experimental measurements is as follows: First, a motor is connected to the structure via a transmission system, and the structure is placed in the water within a wave tank. The initial immersion depth of the structure's characteristic points is measured. Then, combined with the structure's preset motion form, preset motion mode, preset motion process, and initial immersion depth, the time-series immersion depth and time-series motion amplitude of the structure's characteristic points during the motion process are obtained. Subsequently, several wave height meters are installed at different preset measuring points on the first and second sides of the structure. The distance between each measuring point and the structure is used as the corresponding measuring point distance. The structure is driven to move by the motor, and the wave height meters collect time-series radiation wave height data at their respective measuring points. The time-series radiation wave height data collected by each wave height meter on the first side is taken as the first-side time-series radiation wave height, and the time-series radiation wave height data collected by each wave height meter on the second side is taken as the second-side time-series radiation wave height. In practice, the number and location of wave height meters or wave acquisition points for collecting training data should be selected according to needs. It is important to note that all wave height meters should be arranged on the same straight line, which is parallel to the line connecting the two sides of the structure and the direction of radiated wave propagation.
[0046] In step S1), the process of constructing a dataset using the acquired measurement point distances and time-series data is as follows: For each time step, the immersion depth at the start of the time step, the motion amplitude within the time step, the measurement point distance, and the radiated wave height at each measurement point are used to construct a set of sample data corresponding to each measurement point. The sample data corresponding to all measurement points at each time step are then combined to form a dataset. Specifically, the structure completes one motion within each time step. The immersion depth at the start of the time step refers to the immersion depth of the structure's feature points underwater before the start of this motion. The radiated wave height at the measurement point is the wave height data collected at the measurement point within a preset wave response period after the end of this motion.
[0047] Furthermore, each set of sample data can also include the length of the time step.
[0048] S2) Build a machine learning model and train it using the dataset obtained in step S1).
[0049] Machine learning models map the following relationships:
[0050] ① For the measuring points on the first side of the structure, the machine learning model maps the relationship between the distance to the measuring points, the motion amplitude within each time step, the immersion depth at the start of the time step, and the radiation wave height at the measuring points on the first side of the structure (i.e., the radiation wave height on the first side); this mapping relationship is set according to the following formula:
[0051] a l :xb ,L,d-η l
[0052] In the formula, a l Let x represent the mapping relationship on the first side. b Let L represent the motion amplitude within a time step, d represent the distance between measuring points, and η represent the immersion depth at the start of the time step. l This indicates the radiation wave height on the first side of the structure.
[0053] ② For the measuring points on the second side of the structure, the machine learning model maps the relationship between the distance to the measuring points, the motion amplitude within each time step, the immersion depth at the start of the time step, and the radiation wave height at the measuring points on the second side of the structure (i.e., the radiation wave height on the second side). This mapping relationship is set according to the following formula:
[0054] a r :x b ,L,d-η r
[0055] In the formula, a r Let x represent the mapping relationship on the second side. b Let L represent the motion amplitude within a time step, d represent the distance between measuring points, and η represent the immersion depth at the start of the time step. r It indicates the radiation wave height on the second side of the structure.
[0056] The machine learning model includes an input layer, hidden layers, and an output layer. In the machine learning model, the parameters of the input layer should be determined by the parameters that can be represented and collected in step S1). As an optional implementation of the present invention, in each set of sample data, the distance between the same measuring point, the immersion depth at the start of the same time step, and the motion amplitude within the same time step are used as inputs, and the radiation wave height of the same measuring point within the same time step is used as the output. As another optional implementation of the present invention, in each set of sample data, the distance between the same measuring point, the immersion depth at the start of the same time step, and the radiation wave height of the same measuring point within the same time step are used as inputs, and the motion amplitude within the same time step is used as the output. In the machine learning model, the number of hidden layers and the number of parameters should be selected under optimal efficiency conditions to ensure that a suitable learning architecture enables fast and accurate analysis and prediction.
[0057] S3) Obtain the distance between the point to be eliminated and the structure as the distance between the point to be eliminated. Obtain the initial immersion depth of the feature point of the structure and the initial wave height at the point to be eliminated. Use the trained machine learning model to predict the motion amplitude of the structure in the first time step based on the collected distance between the point to be eliminated, the initial immersion depth of the feature point of the structure and the initial wave height of the point to be eliminated. Control the structure to start moving based on the predicted motion amplitude, and wave elimination begins.
[0058] Step S3) specifically refers to:
[0059] S3.1) Generate an incident wave in the wave tank, i.e., the wave to be eliminated, so that the incident wave enters from the first side of the structure and propagates to the second side.
[0060] In this invention, the propagation direction of the radiated wave in step S1) is parallel to the propagation direction of the incident wave (wave to be eliminated) in step S3). A coordinate system is established with the propagation direction of the incident wave as the x-axis, the horizontal direction perpendicular to the propagation direction of the incident wave as the y-axis, and the vertical direction as the z-axis. Therefore, the first side and the second side in step S1) represent the negative and positive directions of the x-axis, respectively. The distance between the measuring points in step S1) and the distance between the points to be eliminated in step S3) are relative distances on the x-axis, and the y-axis coordinate is not considered. The water depth, the shape of the structure, and the position within the wave tank remain consistent with those in step 1).
[0061] S3.2) Measure the distance between the point to be eliminated and the structure, as well as the initial immersion depth of the characteristic points of the structure. Use the distance between the point to be eliminated and the structure as the distance between the point to be eliminated. Keep the structure stationary and use a wave height meter to collect the wave height at the initial moment of the point to be eliminated as the initial wave height.
[0062] Furthermore, the process of collecting the wave height at the initial moment of the point to be eliminated in step S3.2) can be replaced by the following steps: determine whether the point to be eliminated is located on the first or second side of the structure; if the point to be eliminated is located on the first side of the structure, use the Goda two-point method to separate the incident wave height and the reflected wave height in the wave field, and set the sum of the incident wave height and the reflected wave height at the initial moment of the point to be eliminated as the initial wave height; if the point to be eliminated is located on the second side of the structure, use a wave height meter to collect the transmitted wave height at the point to be eliminated, and set the transmitted wave height at the initial moment as the initial wave height.
[0063] S4) During the wave dissipation process, the wave height of the point to be dissipated and the immersion depth of the structural feature points are collected in real time. The real-time wave height of the point to be dissipated and the real-time immersion depth of the structural feature points are obtained. The trained machine learning model is used to predict the real-time motion amplitude of the structure based on the real-time wave height of the point to be dissipated, the real-time immersion depth of the structural feature points and the distance of the point to be dissipated. The movement of the structure is controlled according to the real-time motion amplitude of the structure, so as to dissipate the wave at the point to be dissipated.
[0064] In step S4), the process of using a trained machine learning model to predict the real-time motion amplitude of the structure based on the real-time wave height of the point to be eliminated, the real-time immersion depth of the structure's feature points, and the distance to the point to be eliminated is as follows: the trained machine learning model is used to predict the motion amplitude of the structure in the next time step based on the distance to the point to be eliminated, the immersion depth of the structure at the beginning of the current time step, and the real-time wave height of the point to be eliminated at the end of the current time step, and the motion of the structure in the next time step is controlled based on the motion amplitude of the structure in the next time step.
[0065] Furthermore, the active control method for this structure also includes step S5):
[0066] S5) Accumulate the time series data collected during the wave attenuation process in step S4). When the number of data points reaches a preset number, use the time series data collected in step S4) to construct a dataset containing all relevant variables and use the dataset to optimize the trained machine learning model.
[0067] In the time-series data collected in step S4), each data point includes the distance to the point to be eliminated, the motion amplitude within the same time step, the immersion depth at the start of the same time step, and the radiation wave height at the end of the same time step. The radiation wave height at the end of each time step is the difference between the initial wave height and the wave height at the end of the time step.
[0068] In steps S1) and S3) to S4), the irregularly shaped structures should maintain a consistent external shape. The external shape of the structures includes, but is not limited to, horizontal plates, vertical plates, and wedges, with the specific shape selected according to actual needs. The motion of the irregularly shaped structures should also remain consistent; that is, the fulcrum for translational motion or rotational motion should be consistent. Motion can be single-degree-of-freedom or multi-degree-of-freedom coupled, with the specific motion form selected according to actual needs. The motion process of the irregularly shaped structures should also remain consistent, including but not limited to sinusoidal motion, where the amplitude of the object's motion can be constantly monitored based on the motion curve.
[0069] Furthermore, when the model speed and motor response speed are slow and the delay time is large, the impact of the delay time needs to be considered. To solve this problem, the method of this invention introduces a delay time and compensates for it to achieve the requirements of closed-loop control. The delay time is the time interval from reading the wave height at the target point to the final movement of the motor control board. The specific process is as follows:
[0070] After step S2) is completed, after controlling the structure to move one step, the time-series curve of the radiated wave changing with time is collected at each measuring point on both sides of the structure. Each time-series curve is used as a set of training data. The wave height and time relationship prediction model is trained by Deep Reinforcement Learning (DRL) to obtain a trained wave height prediction model.
[0071] The trained wave height prediction model is used to predict the wave height of waves, especially irregular waves, after a time delay. The input of the trained wave height prediction model is the wave height value collected at the current time t and the time delay Δt. The output is the predicted wave height value after the time delay, i.e., at time t+Δt.
[0072] In step S3), the initial wave height of the point to be eliminated is processed using a pre-trained wave height prediction model to obtain the predicted initial wave height after a delay time Δt. The predicted initial wave height is then input into a trained machine learning model. The trained machine learning model is used to predict the motion amplitude of the structure in the first time step based on the distance to the point to be eliminated, the initial immersion depth of the structure's feature points, and the predicted initial wave height of the point to be eliminated.
[0073] In step S4), a pre-trained wave height prediction model is used to process the wave height of the point to be eliminated at the end of the current time step, and the predicted wave height after considering the time delay is obtained after a delay time Δt. The predicted wave height after considering the time delay is then input into a trained machine learning model. The trained machine learning model is used to predict the motion amplitude of the structure in the next time step based on the distance of the point to be eliminated, the immersion depth at the start of the current time step, and the predicted wave height after considering the time delay at the end of the current time step.
[0074] A second aspect of the present invention provides an active control system for fixed-point wave damping within a wave tank. The active control system for the structure of the present invention includes:
[0075] A structure, placed in a body of water, is used to generate radiating waves within a wave tank through its own motion, and to utilize these radiating waves to dampen waves at the point to be damped. Specifically, the structure can be a horizontal plate, a vertical plate, or a wedge. The structure's motion mode is at least one of vertical translation, horizontal translation, and rotational motion; the structure's motion form is single-degree-of-freedom or multi-degree-of-freedom coupled. When the structure's motion form is single-degree-of-freedom, one of the above motion modes is used; when the structure's motion form is multi-degree-of-freedom coupled, a combination of at least two of the above motion modes is used. The specific motion mode and motion form can be selected according to actual needs. The structure's motion curve is sinusoidal. Specifically, the wave tank can be a two-dimensional numerical or physical tank.
[0076] The motor, with its output end connected to the structure for transmission, is used to drive the structure's movement.
[0077] The training module is used to construct a dataset through theoretical solutions, experimental measurements, or numerical simulations, train a machine learning model using the dataset, and use the trained machine learning model to predict the real-time motion amplitude of the structure based on the real-time wave height of the wave-dissipating point, the real-time immersion depth of the structure's feature points, and the distance to the wave-dissipating point, and output the real-time motion amplitude of the structure.
[0078] Several wave height meters are arranged at the point to be cleared to collect the initial and real-time wave heights. The initial wave height is the wave height at the initial moment, the transmitted wave height, or the incident and reflected wave heights separated using the Goda two-point method. When the training module constructs a dataset through experimental measurements, it also includes several wave height meters arranged at various measuring points on the first and second sides of the structure along the propagation direction of the radiated wave. These are used to collect the time-series radiated wave heights on the first and second sides. The output terminals of these wave height meters at each measuring point are all communicatively connected to the training module to transmit the collected time-series wave heights to the training module.
[0079] The active wave suppression control module, coupled with the training module, enables rapid information transmission, thereby achieving real-time control and correction of the structure's motion. Its input terminal is also connected to the output terminal of a wave height meter located at the wave suppression point, and the output terminal is connected to the control terminal of the motor. The active wave suppression control module acquires the real-time wave height at the wave suppression point and the real-time immersion depth of the structure's feature points. It inputs these data—including the real-time wave height, immersion depth, and distance—into a trained machine learning model. The model receives the real-time motion amplitude of the structure and controls the structure's motion by controlling the motor based on these amplitudes.
[0080] like Figure 2 As shown, when the training module constructs a dataset through experimental measurements, it includes a data acquisition unit and a machine learning unit. The input of the data acquisition unit is connected to the output of the wave height meter located at each measuring point and to the motor, respectively. The output of the data acquisition unit is connected to the input of the machine learning unit, which is coupled to the active wave suppression control module. The coupling between the machine learning unit and the active wave suppression control module means that the machine learning unit receives time-series data collected during the wave suppression process from the active wave suppression control module and outputs the predicted amplitude of the structure's motion in the next time step to the active wave suppression control module.
[0081] In the training module, the data acquisition unit is used to receive the first-side time-series radiation wave height from the wave height meter arranged on the first side, receive the second-side time-series radiation wave height from the wave height meter arranged on the second side, obtain the time-series motion amplitude and time-series immersion depth of the structure from the motor, and output the first-side time-series radiation wave height, the second-side time-series radiation wave height, the time-series motion amplitude and time-series immersion depth of the structure to the machine learning unit.
[0082] In the training module, the machine learning unit is used to construct a dataset using data received from the data acquisition unit or the active wave suppression control module, and to train or optimize the machine learning model within the machine learning unit using the dataset; and to use the trained machine learning model to predict the amplitude of the structure's motion in the next time step based on the distance to the wave suppression point received from the active wave suppression control module, the immersion depth at the start of the current time step, and the real-time wave height of the wave suppression point at the end of the current time step, and output the prediction to the active wave suppression control module.
[0083] The technical principles and effects of the present invention will be further explained below with reference to specific embodiments:
[0084] like Figure 1 As shown in Module 1, in this embodiment, a horizontal plate is used as the structure, and the distance between the center of the structure and the water surface is the initial immersion depth of the structure. The structure's motion mode is vertical translational motion, and the displacement in the vertical direction within each time step is the motion amplitude x within that time step. b A motor is positioned above the structure and is connected to the structure via a transmission mechanism to drive its movement. This embodiment does not consider time delay.
[0085] like Figure 2 As shown, the motion of the structure is sinusoidal, represented as:
[0086] x b =A×sin(ω) b ×t)
[0087] In the formula, x b ω represents the amplitude of the structure's motion within one time step, A represents the maximum distance the structure deviates from its equilibrium position, and ω represents the amplitude of the structure's motion within one time step. b The angular frequency of the structure within one time step is represented by t, where t represents the length of one time step.
[0088] In this embodiment, the first side and the second side of the structure respectively correspond to Figure 1 The left and right sides of the structure. After the structure begins to move, the first lateral radiation wave η is generated on the left side. l A second side radiation wave η is generated on the right side. rWhen the structure is shaped like a 30° wedge, a 50° wedge, or an arc, the curves of the first-side time-series radiation wave heights collected at point W3 by the wave height meter are as follows: Figure 1 The training curve is shown in the figure.
[0089] Figure 1 Module 1 in the document is the training module. For example... Figure 1 and Figure 2 As shown, the training module includes a data acquisition unit (top) and a machine learning unit (bottom). The input of the data acquisition unit is connected to the outputs of wave height meters (W1, W2, W3) installed at different preset measurement points on the left side of the structure, wave height meters (W4, W5, W6) installed at different preset measurement points on the right side of the structure, and the motor. The data acquisition unit receives the first-side time-series radiated wave height (η) from the wave height meters (W1, W2, W3) located on the left side. l1 ~η l3 The second-side time-series radiated wave height (η) is received from the wave height meters (W4, W5, W6) located on the right side. r1 ~η r3 The data acquisition unit receives the temporal motion amplitude and temporal immersion depth of the structure from the motor. The data acquisition unit also receives the distance between each measuring point (L1~L6). The data acquisition unit then converts the distance between the measuring points (L1~L6) and the temporal radiation height (η) on the first side into a single data set. l1 ~η l3 ), second-side time-series radiation wave height (η) r1 ~η r3 The temporal motion amplitude and temporal immersion depth of the structure are output to the machine learning unit.
[0090] The machine learning unit uses the received data to construct a dataset, uses the dataset to train or optimize the machine learning model, and then obtains the submersion depth of the structure, the amplitude of motion, and the distance to the target point and the wave height η at the target point. t Parameter mapping relationship between them:
[0091] a l :x b ,L,d-η l
[0092] a r :x b ,L,d-η r
[0093] In the formula, aal represents the mapping relationship on the left side, and ai r The right-hand side represents the mapping relationship, where xb represents the motion amplitude within one time step, L represents the distance to the target point, d represents the immersion depth at the start of the time step, and η... l η represents the radiation wave height on the left side of the structure. rThis indicates the radiation wave height on the right side of the structure.
[0094] Figure 1 Module 2 is the active wave damping control module, which will be discussed below. Figure 3 The principle of the active wave suppression control module controlling the movement of the structure based on the real-time wave height of the point to be suppressed is further explained.
[0095] In this embodiment, the incident wave η is generated from the left side of the structure. i The incident wave propagates from left to right (i.e., from the first side to the second side). The incident wave generates an incident wave η on the left side (wavefront) of the structure. i and reflected wave η rr A transmitted wave η is generated on the right side (behind the wave). t .
[0096] When the point to be mitigated is located on the left side (first side) of the structure, before wave mitigation, two wave height meters are placed on the left side of the structure, and the incident wave and reflected wave at the point to be mitigated are separated using the Goda two-point method. During wave mitigation, the wave height at the left-side point to be mitigated is:
[0097] η t1 =η i -η rr -η l
[0098] In the formula, η t1 η represents the wave height at the point to be eliminated on the left. i η represents the incident wave height at the point to be eliminated on the left. rr η represents the height of the reflected wave at the point to be eliminated on the left. l This indicates the height of the radiated wave at the point to be eliminated on the left.
[0099] To eliminate waves at the left-hand point, the wave height η at that point is obtained. t1 =0, at this time:
[0100] η l =η i -η rr
[0101] In the formula, η l η represents the height of the radiated wave at the point to be eliminated on the left. i η represents the incident wave height at the point to be eliminated on the left. rr This indicates the height of the reflected wave at the point to be eliminated on the left.
[0102] When the point to be mitigated is located on the right side (second side) of the structure, before wave mitigation, a wave height meter is placed at the point to be mitigated to obtain the transmitted wave at that point. During wave mitigation, the wave height at the right-side point to be mitigated is:
[0103] ηt2 =η t +η r
[0104] In the formula, η t2 η represents the wave height at the point to be eliminated on the right. t η represents the height of the transmitted wave at the point to be eliminated. r This indicates the height of the radiated wave at the point to be eliminated on the right.
[0105] To eliminate the wave at the point to be eliminated on the right, the wave height η at the point to be eliminated on the right is obtained. t2 =0, at this time:
[0106] η r =-η t
[0107] In the formula, η t η represents the height of the transmitted wave at the point to be eliminated on the right. r This indicates the height of the radiated wave at the point to be eliminated on the right.
[0108] Taking wave suppression at the left-side point as an example, the output results of wave suppression without using structures, wave suppression using a horizontal plate with a fixed movement process, and active wave suppression using the system of this invention are as follows: Figure 1 As shown in the output results, it can be seen that the system of the present invention can achieve a more accurate active wave elimination effect during long-term wave elimination.
[0109] The above specific embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.
Claims
1. An active control method for structures used for fixed-point wave damping in wave tanks, characterized in that: The active control method for the structure includes the following steps: S1) Obtain the temporal immersion depth and temporal motion amplitude of the feature points of the structure, obtain the temporal radiation wave height of the first side at different measurement points on the first side of the structure, and the temporal radiation wave height of the second side at different measurement points on the second side opposite to the first side, and construct a dataset; In step S1), the immersion depth at the start of each time step, the motion amplitude within the time step, the distance between each measuring point, and the radiation wave height at each measuring point are constructed into a set of sample data for each time step. S2) Construct a machine learning model and train it using the dataset obtained in step S1). The time-series curves of the radiated wave change over time are collected at each measuring point on both sides of the structure. Each time-series curve is used as a set of training data. The wave height prediction model is trained by deep reinforcement learning to obtain a trained wave height prediction model. The wave height prediction model is a prediction model of the relationship between wave height and time. S3) Obtain the distance between the point to be eliminated and the feature point of the structure as the distance between the point to be eliminated. Use the trained machine learning model to predict the motion amplitude of the structure in the first time step based on the distance between the point to be eliminated, the initial immersion depth of the feature point of the structure and the initial wave height of the point to be eliminated. Control the structure to start moving based on the predicted motion amplitude. S4) During the wave dissipation process, a trained machine learning model is used to predict the motion amplitude of the structure based on the real-time wave height of the point to be dissipated, the real-time immersion depth of the structure's feature points, and the distance to the point to be dissipated. The motion of the structure is controlled based on the predicted motion amplitude, thereby dissipating the wave at the point to be dissipated. In step S4), a trained machine learning model is used to predict the motion amplitude of the structure in the next time step based on the distance to the point to be eliminated, the immersion depth at the start of the current time step, and the wave height of the point to be eliminated at the end of the current time step. The motion of the structure in the next time step is controlled based on the motion amplitude of the structure in the next time step. In step S3), the initial wave height of the point to be eliminated is processed using a pre-trained wave height prediction model to obtain the predicted initial wave height after a delay time. The predicted initial wave height is then input into a trained machine learning model. The trained machine learning model is used to predict the motion amplitude of the structure in the first time step based on the distance to the point to be eliminated, the initial immersion depth of the structure's feature points, and the predicted initial wave height of the point to be eliminated. In step S4), a pre-trained wave height prediction model is used to process the wave height of the point to be eliminated at the end of the current time step, and the predicted wave height after considering the time delay is obtained. Then, the predicted wave height after considering the time delay is input into the trained machine learning model. The trained machine learning model is used to predict the motion amplitude of the structure in the next time step based on the distance of the point to be eliminated, the immersion depth at the beginning of the current time step, and the predicted wave height after considering the time delay at the end of the current time step.
2. The active control method for structures used for fixed-point wave attenuation in wave tanks according to claim 1, characterized in that: Step S3) specifically involves: S3.1) Generate an incident wave in the wave tank, so that the incident wave enters from the first side; S3.2) Obtain the distance between the point to be eliminated and the structure, as well as the initial immersion depth of the feature points of the structure. Use the distance between the point to be eliminated and the structure as the distance between the point to be eliminated, and collect the wave height of the point to be eliminated at the initial moment as the initial wave height.
3. The active control method for structures used for fixed-point wave attenuation in wave tanks according to claim 2, characterized in that: In step S3.2), the process of obtaining the initial wave height includes the following steps: determining whether the point to be eliminated is located on the first or second side of the structure; if the point to be eliminated is located on the first side of the structure, the incident wave height and the reflected wave height are separated using the Goda two-point method, and the sum of the incident wave height and the reflected wave height at the initial moment is set as the initial wave height; if the point to be eliminated is located on the second side of the structure, the transmitted wave height at the point to be eliminated is collected, and the transmitted wave height at the initial moment is set as the initial wave height.
4. The active control method for structures used for fixed-point wave attenuation in wave tanks according to claim 1, characterized in that: The active control method for structures further includes: constructing a dataset based on the time-series data collected during the wave attenuation process in step S4), and using the dataset to optimize the trained machine learning model.
5. An active control system for a structure applied to the active control method for a structure as described in any one of claims 1 to 4, characterized in that: The active control system for the structure includes: A structure placed in a body of water to dissipate waves at a point to be dissipated by generating radiated waves. An electric motor, connected to the structure for transmission, is used to drive the structure's movement; Several wave height meters are arranged at the point where wave suppression is to be carried out, to collect the initial wave height and real-time wave height at the point where wave suppression is to be carried out; The training module is used to construct a dataset through theoretical solutions, experimental measurements, or numerical simulations, train a machine learning model using the dataset, and use the trained machine learning model to predict and output the real-time motion amplitude of the structure. The active wave suppression control module is coupled with the training module, and its output is also connected to the control terminal of the motor. It is used to obtain the real-time wave height of the point to be suppressed and the real-time immersion depth of the structural feature point. The real-time wave height of the point to be suppressed, the real-time immersion depth of the structural feature point, and the distance of the point to be suppressed are input into the trained machine learning model. The real-time motion amplitude of the structure is received from the trained machine learning model, and the movement of the structure is controlled by controlling the motor based on the real-time motion amplitude of the structure.
6. The active control system for fixed-point wave damping structures in wave tanks according to claim 5, characterized in that: When the training module constructs a dataset through experimental measurements, the active control system of the structure also includes several wave height meters arranged at various measuring points along the propagation direction of the radiated wave; the training module includes a data acquisition unit and a machine learning unit; the input end of the data acquisition unit is communicatively connected to the output end of the wave height meter arranged at each measuring point and the motor, the output end of the data acquisition unit is communicatively connected to the input end of the machine learning unit, and the machine learning unit is coupled to the active wave suppression control module.