Hydraulic damping adaptive adjustment method based on wave energy prediction model and electronic equipment

Abstract

The invention discloses a hydraulic damping self-adaptive adjustment method based on a wave energy prediction model and electronic equipment, and relates to the field of hydraulic damping adjustment, and the method realizes prospective intelligent control of a gear rack type PTO system through deep coupling of the wave energy prediction model and a hydraulic damping self-adaptive adjustment mechanism. In a training stage, a target device and a neighborhood device are sequentially alternated, and a general prediction model covering buoys in a preset sea area range is constructed; in the operation stage, a model is constructed based on newest multivariable observation time sequence data to input and output a future wave energy predicted value of the position of a buoy corresponding to a target device, the opening degree of a control valve on a hydraulic oil way is adjusted in advance according to the future wave energy predicted value, and meanwhile feedback correction is conducted in combination with the actual operation state of the device. The limitation that a traditional PTO system only depends on current state feedback is broken through, the predictability and robustness of damping adjustment are remarkably improved, and the energy capture insufficiency under the small wave working condition and the mechanical overload risk under the large wave impact are effectively avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of hydraulic damping adjustment, and in particular to a hydraulic damping self-adaptive adjustment method based on a wave energy prediction model and an electronic device. BACKGROUND

[0002] At present, the wave energy conversion device generally adopts a gear and rack type or a hydraulic type power take-off system (PTO), which drives the rack through the vertical movement of the buoy, and then drives the gear to rotate to drive the generator to generate electricity. However, the existing system mostly adopts fixed damping or only relies on the feedback control strategy of the current state (such as displacement, speed), which is difficult to adapt to the dramatic changes of wave conditions in the ocean environment. In small wave conditions, fixed damping is easy to cause system response delay and low energy capture efficiency; while in large wave impact, insufficient damping may cause high-speed movement of the rack, causing gear meshing impact to intensify, mechanical wear and even structural damage. Although the traditional feedback regulation can respond to the current state, it lacks the ability to predict the future wave situation, and cannot realize the forward-looking adjustment of damping, which still cannot avoid the risk of instantaneous overload.

[0003] In recent years, although some patents have introduced hydraulic or gear and rack PTO structures into the wave energy conversion device, the control logic is still limited to real-time feedback, and has not been fused with the spatio-temporal evolution characteristics of wave energy for predictive regulation. In fact, wave energy has significant randomness, multi-peak, cycle-seasonal mixed characteristics and spatial propagation correlation between buoys, and traditional time series models (such as ARIMA, LSTM) are difficult to fully model such complex dynamics. In the prior art, wave energy prediction and PTO damping control are still in a fragmented state: the prediction model mostly focuses on single-point time series fitting, and it is difficult to describe the spatial propagation characteristics between multiple buoys; and the hydraulic damping adjustment generally relies on lag feedback, and lacks a cooperative mechanism with high-dimensional spatio-temporal prediction models. Especially, there is a lack of a unified modeling framework that can fuse local time series characteristics, spatial correlation between buoys and long-term trend periodicity, and accordingly realize the forward-looking adaptive adjustment of hydraulic damping in the gear and rack PTO system. SUMMARY

[0004] In order to realize the forward-looking adaptive adjustment of hydraulic damping in the gear and rack PTO system, the present application proposes a hydraulic damping self-adaptive adjustment method based on a wave energy prediction model, which is applied to a wave energy conversion device, the device comprising: a buoy; a transmission assembly connected to one end of the buoy; an adjustment assembly comprising a hydraulic cylinder, a piston rod and a hydraulic oil path with a control valve, one end of the piston rod movably extends into the hydraulic cylinder, separating the inner cavity of the hydraulic cylinder into a first oil cavity and a second oil cavity, and the other end is connected to the transmission assembly; the hydraulic oil path communicates the first oil cavity and the second oil cavity, and is controlled by the control valve; the method comprises: The method comprises the following steps: acquiring a plurality of wave energy conversion devices in a preset sea area range; for each of the plurality of wave energy conversion devices, sequentially setting it as a target device and the rest as neighborhood devices; for each setting: acquiring historical multivariate time series data of the positions of the buoys of the target device and the neighborhood devices; performing time alignment processing on each historical multivariate time series data to obtain multivariate observation time series data corresponding to each buoy; synchronously intercepting the multivariate observation time series data corresponding to each buoy through a sliding window, constructing a plurality of training samples based on the intercepted data; constructing training labels corresponding to the training samples; forming a data set based on the training samples and the training labels obtained in each setting; training a preset hybrid neural network model using the data set to obtain a wave energy prediction model; synchronously intercepting the latest multivariate observation time series data of each buoy through a sliding window to obtain a real-time input segment, inputting the real-time input segment into the wave energy prediction model, and outputting a wave energy prediction value of the position of the buoy corresponding to the target device; generating a control signal based on the wave energy prediction value, adjusting the opening of the control valve on the hydraulic oil circuit in the target device according to the control signal, and feedback correcting the control signal based on the actual operating state of the target device.

[0005] Further, the transmission assembly comprises at least one set of intermeshing gears and racks, one end of the rack is connected with the buoy, and the gear is connected with the input end of the generator; The other end of the piston rod is connected with one end of the rack away from the buoy, the two ends of the hydraulic oil circuit are respectively communicated with the first oil cavity and the second oil cavity, and the control valve is arranged on the hydraulic oil circuit, and the control valve is used for controlling the on-off of the hydraulic oil circuit to switch the communication and disconnection state of the first oil cavity and the second oil cavity.

[0006] Further, the time alignment processing is performed on each historical multivariate time series data to obtain multivariate observation time series data corresponding to each buoy; specifically: The historical multivariate time series data of each buoy is statistically processed according to the hourly time period, and is resampled to a common time axis with hourly whole point as the reference to obtain multivariate observation time series data corresponding to each buoy; The multivariate observation time series data comprises the following variables corresponding to each hourly time period: The effective wave height is obtained by zero-crossing analysis on the sea surface height time series in the hourly time period, taking the wave segment between the two consecutive upward crossings of the still water surface as a single wave, taking the difference between the maximum value and the minimum value of each wave as the wave height, and taking the arithmetic mean value of the first third after arranging all the wave heights in descending order; The peak period is obtained by performing spectral analysis on the sea surface height time series within the hourly time period to obtain the energy spectrum, and then taking the period corresponding to the peak of the energy spectrum. The wave direction is obtained by performing directional decomposition on the energy spectrum within the hourly time period to obtain the direction-frequency spectrum, and taking the propagation direction corresponding to the largest energy component. Wind speed, which is obtained by averaging the instantaneous wind speed data over the hourly period; Wind direction, which is obtained by vector averaging the instantaneous wind direction data over the hourly period; Wave energy flux, which is calculated based on the effective wave height and spectral peak period using a preset wave energy flux formula, represents the average wave power transmitted per unit wave peak length within the hourly time period.

[0007] Furthermore, the step of synchronously extracting multivariate observation time-series data corresponding to each buoy through a sliding window, and constructing multiple training samples based on the extracted data, specifically involves: The multivariate observation time series data corresponding to each buoy are synchronously extracted by a sliding window to obtain multiple segments; each segment includes the multivariate observation time series data of each buoy within the time period corresponding to the sliding window. In each segment, the buoy of the target device is set as the target buoy, and the other buoys are set as neighborhood buoys; the multivariate observation time series data corresponding to the target buoy and all neighborhood buoys in the segment are used to construct a training sample; The construction of training labels corresponding to the training samples specifically involves: The sequence of wave energies corresponding to each hour of the target buoy in one or more consecutive hourly time periods is used as the training label for the training sample.

[0008] Furthermore, the latest multivariate observation time-series data of each buoy is synchronously captured through a sliding window to obtain a real-time input segment, which is then input into the wave energy prediction model to output the wave energy prediction value for the location of the target device corresponding to the buoy; specifically: Obtain the latest multivariate observation time series data for each buoy; The target device and neighboring devices are set according to the preset requirements; the buoy corresponding to the target device is set as the target buoy, and the other buoys are set as neighboring buoys. The multivariate observation time series data of the target buoy and each neighboring buoy in the current hour and several consecutive whole hours before are constructed into a real-time input segment, which is input into the wave energy prediction model and outputs the wave energy prediction value of the corresponding target buoy in one or more hours in the future.

[0009] Furthermore, the step of generating a control signal based on the predicted wave energy value specifically includes: Wave energy levels are divided according to the predicted wave energy values ​​and mapped to corresponding hydraulic damping levels; control signals are generated based on the hydraulic damping levels.

[0010] Furthermore, the step of generating a control signal based on the predicted wave energy value specifically includes: Within the time window corresponding to the predicted wave energy value, obtain the spectral peak period and effective wave height corresponding to the buoy's location, as well as the movement speed of the piston rod; Based on the predicted wave energy, the peak period, the effective wave height, and the piston rod's movement speed, the optimal damping coefficient is dynamically calculated using a linear combination model, and a control signal for adjusting the opening of the control valve is generated according to the optimal damping coefficient.

[0011] Furthermore, the hybrid neural network model includes a one-dimensional convolutional neural network module, a graph neural network module, and a segmented temporal prediction module; wherein: The one-dimensional convolutional neural network module is used to extract local features from the multivariate observation time series data of each buoy in the training samples and generate a comprehensive time series feature vector corresponding to each buoy. The graph neural network module is used to fuse the comprehensive temporal feature vector of each buoy with the comprehensive temporal feature vector of other buoys with spatial connection relationships based on the spatial correlation between each buoy, so as to obtain the spatial enhancement features corresponding to each buoy; The segmented time series prediction module is used to select the spatial enhancement features corresponding to the target buoy from the spatial enhancement features corresponding to each buoy, and to perform segmented embedding processing on the time series formed by the spatial enhancement features, that is, to divide it into multiple time segments and map them into embedding vectors, and then apply position encoding to obtain the embedding sequence. The Transformer encoder is used to extract the trend and periodic information in the embedding sequence and output the wave energy prediction sequence for future periods.

[0012] Furthermore, the one-dimensional convolutional neural network module includes: The first channel is used to receive wave energy flux time series data corresponding to each buoy in the training samples, and to extract local features from the wave energy flux time series data of each buoy. The second channel is used to receive time series data of other variables besides wave energy flux for each buoy in the training samples, and to extract local features from the time series data of other variables for each buoy. The feature splicing unit is used to splice the local features extracted from the same buoy through the first channel and the second channel to obtain the spliced ​​features corresponding to each buoy; The fully connected projection unit is used to map the splicing features corresponding to each buoy into a comprehensive temporal feature vector of a unified dimension, thereby obtaining the comprehensive temporal feature vector corresponding to each buoy.

[0013] Furthermore, the actual operating state includes at least one of the following: The vertical displacement or vertical velocity of the buoy; The speed of movement of the piston rod; The rotational speed of the gear; The actual opening degree of the control valve.

[0014] To address the aforementioned problems, another aspect of the present invention provides an electronic device, including: a processor and a memory storing a program, the program including instructions that, when executed by the processor, cause the processor to perform the method described above.

[0015] To address the aforementioned problems, in another aspect of this invention, a non-transitory machine-readable medium storing computer instructions is provided, the computer instructions being used to cause the computer to perform the methods described above.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: (1) This invention achieves forward-looking intelligent control of a rack and pinion PTO system by deeply coupling a wave energy prediction model with a hydraulic damping adaptive adjustment mechanism. Specifically, during the training phase, the target device and neighboring devices are rotated sequentially to construct a general prediction model covering buoys within a preset sea area. During the operation phase, the model input is constructed based on the latest multivariate observation time series data, and the predicted future wave energy value of the buoy corresponding to the target device is output. Based on this, the opening of the control valve on the hydraulic oil line is adjusted in advance, and feedback correction is performed in combination with the actual operating state of the device. This technical solution breaks through the limitation of traditional PTO systems that only rely on current state feedback, significantly improves the predictability and robustness of damping adjustment, effectively avoids insufficient energy capture under small wave conditions and mechanical overload risk under large wave impact, and extends the service life of key components such as gears and racks.

[0017] (2) This invention statistically processes the historical multivariate time series data of each buoy by hourly time intervals and resamples them to a common time axis based on the hourly time, resulting in multivariate observation time series data with uniform time granularity. Among them, the significant wave height is obtained through zero-crossing analysis of the sea surface height time series, the spectral peak period and wave direction are determined based on the energy spectrum and its direction decomposition, respectively, and the wind speed and wind direction are obtained by time averaging and vector averaging of the instantaneous values, respectively. Furthermore, the wave energy flux is calculated based on the significant wave height and the spectral peak period. This processing method ensures that each buoy corresponds to a set of synchronized multivariate observations at the hourly time, thereby providing the hybrid neural network model with input data that is time-aligned across buoys, synchronized with variables, and consistent in physical meaning.

[0018] (3) The one-dimensional convolutional neural network module of the present invention adopts a dual-channel structure: the first channel extracts local features from the wave energy flux time series data of each buoy, and the second channel extracts local features from the time series data of other variables of each buoy besides wave energy flux; then the local features extracted by the two channels of the same buoy are concatenated and mapped into a unified-dimensional comprehensive time series feature vector through a fully connected projection unit. This structure realizes the learning and fusion of differentiated features of key energy indicators (wave energy flux) and other environmental variables, and enhances the model's ability to express local dynamic patterns in multivariate time series.

[0019] (4) In this invention, the graph neural network module fuses the comprehensive temporal feature vector of each buoy with the comprehensive temporal feature vector of other buoys with spatial connections based on the spatial correlation between each buoy, thereby obtaining the spatial enhancement features corresponding to each buoy. This design explicitly introduces the spatial dependency relationship between buoys, so that the feature representation of each buoy not only includes its own historical information, but also integrates the co-evolutionary information of neighboring buoys, thereby improving the model's ability to model the spatial propagation characteristics of waves.

[0020] (5) In this invention, the segmented time-series prediction module selects the spatial enhancement features corresponding to the target buoy from the spatial enhancement features corresponding to each buoy, divides them into multiple time segments and maps them into embedding vectors, applies position encoding, and extracts the trend and periodic information in the sequence through a Transformer encoder to output the wave energy prediction sequence for future periods. This mechanism can effectively capture the evolution law of the time series composed of spatial enhancement features in the time dimension, providing a high-precision prediction basis for the subsequent forward-looking adjustment of hydraulic damping. Attached Figure Description

[0021] Figure 1 This is a flowchart of a hydraulic damping adaptive adjustment method based on a wave energy prediction model, according to an embodiment of the present invention.

[0022] Figure 2This is a cross-sectional view of a wave energy conversion device according to an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of the internal structure of the first and second housings in an embodiment of the present invention.

[0024] Figure 4 This is a structural diagram of the hybrid neural network model in an embodiment of the present invention.

[0025] Figure 5 This is a comparison chart of the prediction results and actual values ​​of the wave energy prediction model in this embodiment of the invention.

[0026] Figure 6 This is a comparison chart of the prediction results and the actual values ​​using the LSTM model in this embodiment of the invention.

[0027] Figure 7 This is a comparison chart of the prediction results and the actual values ​​using the CNN+PatchTST model in this embodiment of the invention.

[0028] Figure 8 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention.

[0029] In all the accompanying drawings, the same reference numerals denote the same technical features, specifically: 100, Buoy; 120, Connecting rod; 200, Transmission assembly; 210, Gear; 220, Rack; 241, Second connecting plate; 300, Generator; 400, Adjustment assembly; 410, Hydraulic cylinder; 411, First oil chamber; 412, Second oil chamber; 420, Piston rod; 421, Rod body; 422, Piston part; 430, Adjustment pipe; 440, Control valve; 900, First housing; 901, First receiving cavity; 910, Second housing; 911, Second receiving cavity; 912, First support base. Detailed Implementation

[0030] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments. Example

[0031] like Figure 2 and Figure 3 As shown, this embodiment provides a wave energy conversion device, comprising: Buoy 100; The transmission assembly 200 includes at least one set of meshing gears 210 and racks 220, one end of rack 220 is connected to buoy 100, and gears 210 are connected to the input end of generator 300. The adjusting assembly 400 includes a hydraulic cylinder 410, a piston rod 420, and a hydraulic oil circuit. One end of the piston rod 420 extends movably into the hydraulic cylinder 410, dividing the inner cavity of the hydraulic cylinder 410 into a first oil chamber 411 and a second oil chamber 412. The other end of the piston rod 420 is connected to the end of the rack 220 away from the float 100. The two ends of the hydraulic oil circuit are respectively connected to the first oil chamber 411 and the second oil chamber 412. A control valve 440 is provided on the hydraulic oil circuit to control the opening and closing of the hydraulic oil circuit, thereby switching the connection and disconnection states of the first oil chamber 411 and the second oil chamber 412. This design constitutes a hydraulic damping system with switchable operating modes: when encountering small waves, the two oil chambers can be connected by opening the control valve 440, allowing the hydraulic oil to flow almost unobstructed, thereby significantly reducing the motion resistance of the piston rod 420 and enabling the float 100 and rack 220 to respond sensitively and efficiently capture small wave energy; when encountering large waves, the two oil chambers are blocked by closing the control valve 440, forcing the hydraulic oil to flow in a restricted channel to generate huge damping, effectively suppressing the excessively fast movement of the rack 220, absorbing impact energy, and providing reliable overload protection for subsequent transmission components and generator 300, taking into account both energy conversion efficiency and system safety.

[0032] like Figures 1 to 3 As shown, the wave energy conversion device provided in this embodiment is mainly used to convert the kinetic and potential energy of ocean waves into electrical energy. The device uses a buoy 100 that moves with the waves to drive a rack 220 in the transmission assembly 200 to perform linear reciprocating motion. The rack 220 further drives the gear 210 meshing with it to rotate, thereby triggering a generator 300 to generate electricity. Simultaneously, an adjustment assembly 400 is used to regulate the damping force during the transmission process to adapt to wave conditions of different intensities. This improves the capture efficiency of low-energy waves while effectively preventing overload damage to the transmission system caused by high-energy waves.

[0033] To provide effective protection for the transmission assembly 200 and the adjustment assembly 400, this embodiment also includes a first housing 900 and a second housing 910, which together form a nested sealed protective structure. The first housing 900 has a first accommodating cavity 901 extending along its length to accommodate the second housing 910 and the adjustment assembly 400. The second housing 910 is horizontally fixed within the first accommodating cavity 901 and has a second accommodating cavity 911 arranged along its length to accommodate and protect the transmission assembly 200 and the generator 300. This design achieves functional zoning and environmental isolation, effectively preventing corrosion of key mechanical and electrical components by seawater, salt spray, and marine organisms. It also enhances the overall structural rigidity and impact resistance, facilitating long-term stable operation of the device in complex marine environments.

[0034] Furthermore, the transmission assembly 200 includes at least one set of meshing gears 210 and racks 220. The rack 220 is arranged vertically, with its upper end connected to the connecting rod 120 of the buoy 100 and its lower end connected to the piston rod 420 in the adjustment assembly 400. The gears 210 are connected to the input end of the generator 300 to convert the linear reciprocating motion of the rack 220 into rotational motion, thereby driving the generator 300 to generate electricity.

[0035] Furthermore, both the gear 210 and the generator 300 are housed within the second accommodating cavity 911. The generator 300 is detachably mounted on the first support 912 within the second accommodating cavity 911, facilitating future maintenance, replacement, or upgrades. The rack 220 is movably mounted on the second housing 910, enabling free reciprocating motion in the vertical direction.

[0036] In order to achieve overload protection of the transmission component 200 and efficient wave energy collection, in this embodiment, an adjustment component 400 is provided at the bottom of the first accommodating cavity 901.

[0037] Furthermore, the adjustment assembly 400 includes a hydraulic cylinder 410, a piston rod 420, and a hydraulic oil circuit. The hydraulic cylinder 410 is vertically fixed at the bottom of the first accommodating cavity 901, with its axis parallel to the direction of movement of the rack 220. The piston rod 420 includes a rod body 421 and a piston portion 422 integrally connected or fixedly assembled. One end of the rod body 421 with the piston portion 422 movably extends into the hydraulic cylinder 410, dividing the inner cavity of the hydraulic cylinder 410 into a first oil chamber 411 above the piston portion 422 and a second oil chamber 412 below the piston portion 422. The other end of the rod body 421 is fixedly connected to the end of the rack 220 away from the buoy 100 via a second connecting plate 241, thereby directly transmitting the vertical reciprocating motion of the rack 220 to the piston rod 420, driving the piston portion 422 to reciprocate synchronously within the hydraulic cylinder 410, and thus applying a controllable hydraulic damping force during the movement of the rack 220.

[0038] The two ends of the hydraulic circuit are connected to the first oil chamber 411 and the second oil chamber 412, respectively. A control valve 440 is provided on the hydraulic circuit to control the opening and closing of the hydraulic circuit, thereby switching the connection and disconnection states of the first oil chamber 411 and the second oil chamber 412. Specifically, when the control valve 440 is open, the two oil chambers are connected, and the hydraulic oil can flow freely between them. The system exhibits low damping characteristics, which is beneficial for sensitive response under small wave conditions. When the control valve 440 is closed, the two oil chambers are isolated, and the movement of the piston rod 420 is limited by the compression resistance of the closed oil. The system exhibits high damping characteristics, effectively suppressing overspeed movement under large wave impacts and achieving overload protection for the transmission component 200.

[0039] Furthermore, the control valve 440 is a multi-stage opening regulating valve. The displacement stroke of the valve core of the control valve 440 is configured to have a fully open position, a fully closed position, and at least one partially open position. This design upgrades damping regulation from a simple "on-off" two-state control to continuous or multi-stage fine control. By controlling the valve core of 440 at different opening positions, the flow cross-sectional area of ​​the hydraulic oil circuit can be changed in stages or approximately continuously, thereby achieving stepped or stepless adjustment of the damping force. This allows the device to match the optimal damping force for waves of different intensities, maximizing energy capture efficiency over a wider range of sea states, while ensuring smoother and more stable system operation.

[0040] Preferably, the control valve 440 is a proportional valve or a servo valve.

[0041] Example 2 To achieve proactive adaptive adjustment of hydraulic damping in a rack and pinion PTO system, such as Figure 1 As shown, this invention proposes a hydraulic damping adaptive adjustment method based on a wave energy prediction model, applied to the wave energy conversion device; the method includes: Multiple wave energy conversion devices within a preset sea area are acquired. For each of these devices, it is sequentially designated as the target device, and the rest as neighboring devices. For each designation: historical multivariate time-series data of the buoy positions of the target device and neighboring devices are acquired; time-aligned data is processed to obtain multivariate observation time-series data with uniform time granularity corresponding to each buoy; the multivariate observation time-series data corresponding to each buoy is synchronously captured using a sliding window, and multiple training samples are constructed based on the captured data; training labels corresponding to the training samples are constructed. Time alignment was performed on the historical multivariate time series data to obtain multivariate observation time series data with uniform time granularity corresponding to each buoy; specifically: The historical multivariate time series data of each buoy were statistically processed by hourly time periods and resampled to a common time axis based on the hour, to obtain the multivariate observation time series data corresponding to each buoy; The multivariate observation time series data includes the following variables corresponding to each hourly time period: Significant wave height is determined by performing zero-crossing analysis on the sea surface height time series within the hour. The wave segment between two consecutive upward crossings of the still water surface is taken as a single wave, and the difference between the maximum and minimum values ​​of each wave is taken as its wave height. The arithmetic mean of the first third of all wave heights is then obtained by sorting all wave heights in descending order. The peak period is obtained by performing spectral analysis on the sea surface height time series within the hourly time period to obtain the energy spectrum, and then taking the period corresponding to the peak of the energy spectrum. The wave direction is obtained by performing directional decomposition on the energy spectrum within the hourly time period to obtain the direction-frequency spectrum, and taking the propagation direction corresponding to the largest energy component. Wind speed, which is obtained by averaging the instantaneous wind speed data over the hourly period; Wind direction, which is obtained by vector averaging the instantaneous wind direction data over the hourly period; Wave energy flux, which is calculated based on the effective wave height and spectral peak period using a preset wave energy flux formula (a standard wave energy flux formula commonly used in the field of marine engineering), represents the average wave power transmitted per unit wave peak length within the hourly time period.

[0042] It should be noted that time alignment processing is performed on the historical multivariate time series data to obtain multivariate observation time series data with uniform time granularity for each buoy. During this process, if the data missing rate of a certain buoy exceeds a first threshold within a preset time window, all data for that buoy within that time window are discarded. If only a few variables are missing, interpolation methods are used to fill in the missing data, specifically including: (1) Time series interpolation: using the buoy's own historical observation data for forward and backward filling; (2) Linear interpolation: suitable for variables with gradual changes, and performs linear estimation based on adjacent effective observation points; (3) Spline interpolation: suitable for variables with smooth trends, continuous completion is achieved through piecewise polynomial fitting; (4) Interpolation based on historical average: The average value of the hour in the same historical period (e.g., the same hour in the past 30 days) is used to fill the gap.

[0043] After the above processing, complete, aligned, and uniform time-granular multivariate observation time series data are obtained for subsequent model training.

[0044] It should be noted that, because this invention adopts a collaborative modeling mechanism based on spatially nearby buoys, even if individual buoys are discarded within a specific time window, the observation data of the remaining buoys can still support the effective prediction of the wave energy of the target buoy, thereby ensuring the robust operation of the system in the actual marine environment.

[0045] In one embodiment of the present invention, each buoy integrates a sensing unit for marine environmental perception, including but not limited to: wave height sensors (such as accelerometers or pressure sensors), anemometers, and auxiliary positioning and attitude measurement devices. These sensors collect raw observation data of the buoy's location in real time, including multi-source signals such as sea surface height time series, instantaneous wind speed, and instantaneous wind direction, and store them locally or transmit them wirelessly to a shore-based / cloud processing unit via the buoy's built-in data acquisition module. Therefore, the historical multivariate time series data of each buoy mentioned in the present invention refers to the raw observation sequence collected and recorded by the sensors mounted on each buoy itself.

[0046] In one embodiment of the present invention, for the historical multivariate time-series data (such as sea surface height, wind speed, wind direction, etc.) collected by each buoy, time windows are first divided according to natural hours (e.g., 09:00–10:00). For the raw observation values ​​within each window, statistical aggregation is performed according to standard marine engineering methods: for example, zero-crossing analysis is performed on the sea surface height time series to calculate the significant wave height for that hour, the arithmetic mean of instantaneous wind speed is taken to obtain the hourly average wind speed, and the vector average of instantaneous wind direction is performed to avoid angular ambiguity. At the same time, the spectral peak period and main wave direction for that hour are obtained through spectral analysis, and the wave energy flux is further calculated by combining the significant wave height and the spectral peak period. After completing the above variable-level statistics, the hourly statistical results of all variables are uniformly assigned to the end time of that hour (e.g., 10:00) as a timestamp, thereby forming a common time axis based on the hour. Ultimately, each buoy corresponds to a time-aligned multivariate observation time series, where each time step contains the synchronous representative value of all variables for that hour, ensuring time consistency across buoys and variables, and providing reliable input for the training and inference of subsequent deep learning models.

[0047] This invention statistically processes the historical multivariate time-series data of each buoy by hourly intervals and resamples it to a common time axis based on the hour, resulting in multivariate observation time-series data with uniform time granularity. Specifically, significant wave height is obtained through zero-crossing analysis of the sea surface height time series; spectral peak period and wave direction are determined based on energy spectrum and its directional decomposition, respectively; wind speed and wind direction are obtained by time averaging and vector averaging of instantaneous values, respectively; and wave energy flux is further calculated based on significant wave height and spectral peak period. This processing method ensures that each buoy corresponds to a set of synchronized multivariate observations at every hour, thus providing hybrid neural network models with cross-buoy time-aligned, variable-synchronized, and physically consistent input data.

[0048] The process involves synchronously extracting multivariate observation time-series data corresponding to each buoy using a sliding window, and constructing multiple training samples based on the extracted data. Specifically: The multivariate observation time series data corresponding to each buoy are synchronously extracted by a sliding window to obtain multiple segments; each segment includes the multivariate observation time series data of each buoy within the time period corresponding to the sliding window. In each segment, the buoy of the target device is set as the target buoy, and the other buoys are set as neighborhood buoys; the multivariate observation time series data corresponding to the target buoy and all neighborhood buoys in the segment are used to construct a training sample; The construction of training labels corresponding to the training samples specifically involves: The sequence of wave energies corresponding to each hour of the target buoy in one or more consecutive hourly time periods is used as the training label for the training sample.

[0049] A dataset is formed based on the training samples and training labels obtained from each setting; A pre-defined hybrid neural network model is trained using the dataset to obtain a wave energy prediction model; The AdamW optimizer was used during model training, with an initial learning rate of 1e-4, a loss function of SmoothL1Loss, and a batch size of 32, to achieve stable and efficient parameter updates.

[0050] The hybrid neural network model includes a one-dimensional convolutional neural network module, a graph neural network module, and a segmented temporal prediction module; like Figure 4 As shown, in the hybrid neural network model: The one-dimensional convolutional neural network module is used to extract local features from the multivariate observation time series data of each buoy in the training samples and generate a comprehensive time series feature vector corresponding to each buoy. The one-dimensional convolutional neural network module includes: The first channel is used to receive the wave energy flux time series data corresponding to each buoy in the training samples, and then sequentially extract local features from the wave energy flux time series data of each buoy through a one-dimensional convolutional layer (with a kernel size of 7), a batch normalization layer, and a ReLU activation function. The second channel is used to receive time-series data of other variables besides wave energy flux corresponding to each buoy in the training samples, and then extracts local features of the other variable time-series data of each buoy through a one-dimensional convolutional layer (with a kernel size of 5), a batch normalization layer, and a ReLU activation function in sequence. The feature splicing unit is used to splice the local features extracted from the same buoy through the first channel and the second channel to obtain the spliced ​​features corresponding to each buoy; The fully connected projection unit is used to map the splicing features corresponding to each buoy into a comprehensive temporal feature vector of a unified dimension, thereby obtaining the comprehensive temporal feature vector corresponding to each buoy.

[0051] The graph neural network module is used to fuse the comprehensive temporal feature vector of each buoy with the comprehensive temporal feature vector of other buoys with spatial connection relationships based on the spatial correlation between each buoy, so as to obtain the spatial enhancement features corresponding to each buoy; It's important to note that although non-target buoys are collectively referred to as "neighborhood buoys" during the sample construction phase, not all neighborhood buoys are directly connected to the target buoy in the graph neural network. The actual connection relationship is determined by preset spatial association rules (such as geographical distance being less than a threshold), and information aggregation only occurs between connected buoys.

[0052] In one specific embodiment of the present invention, the graph neural network module includes a first graph convolutional layer, a second graph convolutional layer, and a nonlinear activation function (such as ReLU) connected in sequence, for modeling the spatial correlation between the buoys.

[0053] Specifically, the comprehensive temporal feature vectors corresponding to each buoy output by the one-dimensional convolutional neural network module are used as the initial node feature inputs of the graph neural network. Each buoy corresponds to a node in the graph, and its node features are the comprehensive temporal feature vectors corresponding to that buoy in the current training samples. The spatial relationships between buoys are represented by a pre-constructed adjacency matrix. For example, a connection threshold can be set based on the geographical distance between buoys; only when the distance is less than this threshold are edge connections established between the corresponding nodes.

[0054] The first graph convolutional layer performs weighted aggregation of the node features of each buoy and the node features of other buoys with spatial connections based on the adjacency matrix, and performs linear transformation through a learnable weight matrix to obtain the first-level intermediate features corresponding to each buoy. The second graph convolutional layer takes the first-level intermediate features as input, and performs information aggregation and linear transformation again based on the same adjacency relationship to obtain the linear fusion features corresponding to each buoy. Subsequently, a nonlinear activation function (such as ReLU) is applied to the linear fusion features, and finally the spatial enhancement features corresponding to each buoy are output.

[0055] The spatial enhancement features not only preserve the original temporal dynamics, but also integrate the coordinated sea state information of multiple neighboring buoys in the local sea area, thereby more comprehensively representing the spatial context of the target buoy's environment and providing high-quality input for the subsequent segmented temporal prediction module.

[0056] The segmented time series prediction module is used to select the spatial enhancement features corresponding to the target buoy from the spatial enhancement features corresponding to each buoy, and to perform segmented embedding processing on the time series formed by the spatial enhancement features, that is, to divide it into multiple time segments and map them into embedding vectors, and then apply position encoding to obtain the embedding sequence. The Transformer encoder is used to extract the trend and periodic information in the embedding sequence and output the wave energy prediction sequence for future periods.

[0057] In one specific embodiment of the present invention, the segmented time series prediction module includes a channel attention calibration module (Squeeze-Excite module), a patch embedding layer, a positional embedding layer, and a Transformer encoder connected in sequence, for modeling the time series composed of the spatial enhancement features of the target buoy, so as to achieve high-precision prediction of wave energy in future periods.

[0058] Specifically, the spatial augmentation features corresponding to the target buoy are first selected from the spatial augmentation features corresponding to each buoy, forming a temporal feature sequence arranged in chronological order; Subsequently, the time-series feature sequence is input into the channel attention calibration module. The importance of each feature channel is dynamically calibrated through the channel attention mechanism, highlighting key feature channels related to wave energy evolution and suppressing irrelevant or redundant channels, thereby enhancing the model's ability to represent effective sea state information. Next, the calibrated feature sequence is fed into the segment embedding layer, which divides it into multiple consecutive time segments and maps each time segment to a low-dimensional embedding vector to form a segment embedding sequence. The length of each time segment is 24 time steps, and a sliding step size of 12 time steps is used when dividing the segment to preserve the local overlap information of the time series. Then, a positional encoding layer is used to add positional encoding to each time segment in the embedded sequence of this segment in order to preserve the temporal sequence relationship of the original sequence, and finally an embedded sequence with positional information is obtained; Finally, the embedded sequence with location information is input into the Transformer encoder, which contains two stacked coding units, each containing a multi-head self-attention mechanism with four attention heads to capture long-term dependencies and multi-scale periodic patterns in the embedded sequence. Each coding unit includes, in sequence: Multi-head self-attention mechanism is used to capture long-term dependencies and multi-scale periodic patterns in embedded sequences; The first residual connection and layer normalization unit is used to add the output of the multi-head self-attention mechanism to its input and then normalize it; A feedforward neural network is used to perform nonlinear transformations on the features output by the first residual connection and the layer normalization unit; The second residual connection and layer normalization unit is used to add the output of the feedforward neural network to its input and then normalize it again; After the above processing, the Transformer encoder outputs a wave energy prediction sequence for future time periods.

[0059] The latest multivariate observation time series data of each buoy are synchronously captured by a sliding window to obtain a real-time input segment. This real-time input segment is then input into the wave energy prediction model, which outputs the wave energy prediction value of the buoy corresponding to the target device. The process involves synchronously capturing the latest multivariate observation time-series data of each buoy through a sliding window to obtain a real-time input segment, which is then input into the wave energy prediction model to output the predicted wave energy value for the location of the target device corresponding to the buoy. Specifically: Obtain the latest multivariate observation time series data for each buoy; The target device and neighboring devices are set according to the preset requirements; the buoy corresponding to the target device is set as the target buoy, and the other buoys are set as neighboring buoys. The multivariate observation time series data of the target buoy and each neighboring buoy in the current hour and several consecutive whole hours before are constructed into a real-time input segment, which is input into the wave energy prediction model and outputs the wave energy prediction value of the corresponding target buoy in one or more hours in the future.

[0060] A control signal is generated based on the predicted wave energy value, and the opening of the control valve on the hydraulic line of the target device is adjusted according to the control signal. At the same time, the control signal is corrected based on the actual operating state of the target device.

[0061] The actual operating state includes at least one of the following: The vertical displacement or vertical velocity of the buoy; The speed of movement of the piston rod; The rotational speed of the gear; The actual opening degree of the control valve.

[0062] The generation of control signals based on the predicted wave energy values ​​includes two implementation methods: One implementation method is as follows: Wave energy levels are divided according to the predicted wave energy values ​​and mapped to corresponding hydraulic damping levels; control signals are generated based on the hydraulic damping levels.

[0063] Specifically, the process of generating control signals based on wave energy prediction values ​​includes: First, classifying wave energy levels according to the predicted wave energy values ​​of the target buoy for the next hour and mapping them to corresponding hydraulic damping levels; specifically, two energy thresholds are preset, namely a first threshold and a second threshold, with the first threshold being less than the second threshold; when the predicted wave energy is lower than the first threshold, it is determined to be a low-wave condition, corresponding to the minimum damping state, at which point the control valve is fully open to improve system sensitivity and ensure effective wave energy capture even in low-energy sea states; when the predicted wave energy is greater than or equal to the first threshold and less than the second threshold, it is determined to be a medium-wave condition, corresponding to the optimal damping state, at which point the control valve is adjusted to a partially open position to make the system operate at the point of maximum energy capture efficiency; when the predicted wave energy is greater than or equal to the second threshold, it is determined to be a high-wave condition, corresponding to the maximum damping state, at which point the control valve is closed or nearly closed to limit the speed of the rack movement, prevent overload of the transmission mechanism, and protect the safety of the mechanical structure. This method utilizes wave energy prediction results to pre-set damping strategies, thereby achieving proactive control of the hydraulic system and avoiding energy loss or equipment damage caused by lag in traditional passive response control.

[0064] Another implementation method is: Within the time window corresponding to the predicted wave energy value, obtain the spectral peak period and effective wave height corresponding to the buoy's location, as well as the movement speed of the piston rod; Based on the predicted wave energy, the peak period, the effective wave height, and the piston rod's movement speed, the optimal damping coefficient is dynamically calculated using a linear combination model, and a control signal for adjusting the opening of the control valve is generated according to the optimal damping coefficient.

[0065] In this embodiment, the calculation of the optimal damping coefficient is as follows: First, within the time window corresponding to the predicted wave energy value, the peak period, significant wave height, and the movement speed of the hydraulic cylinder piston rod at the location of the target buoy are obtained. Then, the spectral peak period, effective wave height and piston rod speed are multiplied by the corresponding working condition correction coefficients, and the three product results are added together to obtain the optimal damping coefficient; The operating condition correction coefficient is adaptively adjusted according to the current sea state and is pre-calibrated through system simulation or measured data. Specifically: When the wave condition is small (e.g., the significant wave height is less than 1.0 meter), the correction coefficient corresponding to the spectral peak period ranges from 0.1 to 0.3. When the wave conditions are moderate (e.g., the significant wave height is 1.0 to 2.5 meters), the correction factor corresponding to the spectral peak period ranges from 0.4 to 0.6. When operating under high wave conditions (e.g., significant wave height greater than 2.5 meters), the correction factor corresponding to the spectral peak period ranges from 1.0 to 1.4. The correction coefficients corresponding to the significant wave height and the piston rod speed are also divided into three intervals—small waves, medium waves, and large waves—based on the sea state level (with significant wave height as the criterion), and corresponding value ranges are set within each interval to achieve dynamic adjustment.

[0066] Finally, based on the calculated optimal damping coefficient, a control signal is generated to adjust the opening of the control valve, thereby achieving forward-looking control of the hydraulic damping force.

[0067] In one specific embodiment of the invention, the wave energy prediction model outputs a wave energy prediction sequence for the target buoy over several future hours (e.g., 1 to 6 hours). However, when generating the control signal required for hydraulic damping adjustment, only the wave energy prediction value corresponding to the most recent hour in the prediction sequence is used as the decision-making basis. This design aims to balance prediction reliability and control real-time performance: on the one hand, the prediction value of the most recent hour is most affected by the current sea state and has a relatively higher confidence level; on the other hand, the hydraulic adjustment system has a certain response delay, and an excessively long prediction window will introduce significant uncertainty. Therefore, the control decision should be based on short-term prediction results to ensure that the damping parameters can be adjusted and applied to the current wave excitation process within an effective time. Therefore, the controller generates the corresponding control signal based on the wave energy prediction value of the most recent hour.

[0068] Figures 5 to 7 In the figures, the horizontal axis "Time (h)" represents time (unit: hours), reflecting continuous time points within the test period; the vertical axis "Wave Energy Flux (kW / m)" represents the predicted wave energy value, i.e., wave energy flux (unit: kilowatts per meter), used to measure the wave power transmitted per unit wave crest width. The True curve (blue solid line) in the figures represents the actual observed value, i.e., the wave energy flux calculated based on actual ocean buoy observation data; the Pred curve (orange dashed line) represents the predicted wave energy value output by the corresponding model. In addition, the performance indicators shown in the figures include: RMSE (Root Mean Square Error): Root mean square error is used to measure the average deviation between predicted and actual observed values. The smaller the value, the higher the prediction accuracy. COR (Correlation Coefficient): The correlation coefficient is used to measure the degree of linear correlation between the predicted sequence and the actual sequence. The value ranges from 0 to 1. The closer it is to 1, the more consistent the trends of the two are.

[0069] Based on this, to further verify the superiority of the wave energy prediction model described in this invention, this embodiment selects a Long Short-Term Memory (LSTM) network model and a CNN+PatchTST model based on convolutional neural networks and segmented Transformers as comparison models, and evaluates their wave energy prediction performance under the same test dataset and evaluation criteria. The results are as follows: Figures 5 to 7 As shown: Figure 5 The wave energy prediction model of this invention is shown in the figure, applied to target buoy 41040. As can be seen from the figure, the predicted curve and the actual curve are highly consistent in terms of overall trend and local peak and trough positions. The RMSE is 3.03 kW / m and the COR is 0.930, indicating that the model has excellent prediction accuracy and dynamic tracking capability.

[0070] Figure 6 The forecast results for target buoy 41040 using the LSTM model are shown. It can be seen that the overall forecast curve is relatively flat, making it difficult to accurately capture the rapid changes in wave energy, especially at wave crests and troughs, where there is a significant deviation. The RMSE is 7.89 kW / m, and the COR is 0.795.

[0071] Figure 7 The forecast results for target buoy 41040 using the CNN+PatchTST model are shown. Although the model can portray the long-term trend well, it still shows a certain lag in the range of drastic wave energy changes because it does not explicitly model the spatial correlation between buoys. The RMSE is 4.19 kW / m and the COR is 0.878.

[0072] In summary, under the same test conditions, the prediction error of the model of this invention is significantly lower than that of the comparative model, and the correlation is higher. This fully demonstrates that the model of this invention, by integrating a one-dimensional convolutional neural network (extracting local features), a graph neural network (modeling spatial correlation between buoys), and a segmented time series prediction module (capturing long-term trends and periodicity), can accurately track the dynamic changes of actual wave energy and has good prediction performance and engineering application value.

[0073] This invention achieves forward-looking intelligent control of a rack and pinion PTO system by deeply coupling a wave energy prediction model with a hydraulic damping adaptive adjustment mechanism. Specifically, during the training phase, the target device and neighboring devices are rotated sequentially to construct a general prediction model covering buoys within a preset sea area. During the operation phase, the model input is constructed based on the latest multivariate observation time-series data, and the output is the predicted future wave energy value for the location of the target device's corresponding buoy. Based on this, the opening of the control valve on the hydraulic circuit is adjusted in advance, and feedback correction is performed based on the actual operating status of the device. This technical solution overcomes the limitation of traditional PTO systems that rely solely on current state feedback, significantly improves the predictability and robustness of damping adjustment, effectively avoids insufficient energy capture under small wave conditions and the risk of mechanical overload under large wave impacts, and extends the service life of key components such as gears and racks.

[0074] This invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method of this invention.

[0075] The aforementioned electronic equipment can be deployed in a shore-based control center, an offshore platform, or inside the wave energy conversion device itself to execute the hydraulic damping adaptive adjustment method described in the embodiments of the present invention, thereby achieving centralized or distributed control of multiple wave energy conversion devices and realizing forward-looking hydraulic damping adjustment based on prediction results.

[0076] The present invention also provides a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform the method of the present invention.

[0077] This invention also provides a computer program product, including a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform the method of this invention.

[0078] refer to Figure 8The present invention will now describe a structural block diagram of an electronic device that can serve as a server or client in embodiments of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0079] like Figure 8 As shown, the electronic device includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. The RAM 403 may also store various programs and data required for the operation of the electronic device. The computing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0080] Multiple components in the electronic device are connected to I / O interface 405, including: input unit 406, output unit 407, storage unit 408, and communication unit 409. Input unit 406 can be any type of device capable of inputting information into the electronic device. Input unit 406 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of the electronic device. Output unit 407 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 408 may include, but is not limited to, disks and optical discs. Communication unit 409 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0081] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, CPUs, graphics processing units (GPUs), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the methods and processes described above. For example, in some embodiments, the method embodiments of the present invention may be implemented as a computer program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM 402 and / or communication unit 409. In some embodiments, the computing unit 401 may be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0082] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0083] In the context of embodiments of the present invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0084] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0085] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0086] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0087] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

Claims

1. A hydraulic damping adaptive adjustment method based on a wave energy prediction model, characterized by, The method is applied to a wave energy conversion device, which comprises a buoy, a transmission assembly connected to the buoy at one end, an adjustment assembly comprising a hydraulic cylinder, a piston rod and a hydraulic oil circuit with a control valve, one end of the piston rod movably extends into the hydraulic cylinder, separating the inner cavity of the hydraulic cylinder into a first oil cavity and a second oil cavity, the other end of the piston rod is connected to the transmission assembly, the hydraulic oil circuit communicates the first oil cavity and the second oil cavity and is controlled by the control valve, the method comprises: Obtain a plurality of wave energy conversion devices in a predetermined sea area, for each of the plurality of wave energy conversion devices, set it as a target device in turn, and the rest as neighborhood devices, and for each setting: obtain the historical multivariate time series data of the positions of the buoys of the target device and the neighborhood devices; time alignment processing is performed on each historical multivariate time series data to obtain the corresponding multivariate observation time series data of each buoy; the multivariate observation time series data corresponding to each buoy is synchronously intercepted by a sliding window, and a plurality of training samples are constructed based on the intercepted data; training labels corresponding to the training samples are constructed; Form a data set based on the training samples and training labels obtained in each setting; Train a pre-set hybrid neural network model using the data set to obtain a wave energy prediction model; Synchronously intercept the latest multivariate observation time series data of each buoy by a sliding window to obtain a real-time input segment, and input the real-time input segment into the wave energy prediction model to output the wave energy prediction value of the position of the buoy corresponding to the target device; Generate a control signal based on the wave energy prediction value, and adjust the opening of the control valve on the hydraulic oil circuit in the target device according to the control signal, while feedback correcting the control signal based on the actual operating state of the target device.

2. The method of claim 1, wherein, The transmission assembly comprises at least one set of intermeshing gears and racks, one end of the rack is connected to the buoy, and the gear is connected to the input end of the generator; The other end of the piston rod is connected to the end of the rack away from the buoy, the two ends of the hydraulic oil circuit are respectively communicated with the first oil cavity and the second oil cavity, and the control valve is arranged on the hydraulic oil circuit, which is used to control the on-off of the hydraulic oil circuit to switch the communication and disconnection state of the first oil cavity and the second oil cavity.

3. The method of claim 1, wherein, The historical multivariate time series data of each buoy is statistically processed by hour time period, and resampled to a common time axis based on hour whole point to obtain the multivariate observation time series data corresponding to each buoy; The multivariate observation time series data includes the following variables corresponding to each hour time period: ​ The effective wave height is obtained by zero-crossing analysis on the sea surface height time series in the hour time period, taking the wave segment between two consecutive upward crossings of the still water surface as a single wave, taking the difference between the maximum value and the minimum value of each wave as the wave height, and taking the arithmetic mean of the first third of all wave heights in descending order; The spectral peak period is obtained by frequency spectrum analysis on the sea surface height time series in the hour time period, obtaining the energy spectrum, and taking the period corresponding to the peak value of the energy spectrum; The wave direction is obtained by direction decomposition on the energy spectrum in the hour time period, obtaining the direction-frequency spectrum, and taking the propagation direction corresponding to the maximum energy component; The wind speed is obtained by time averaging on the instantaneous wind speed data in the hour time period; The wind direction is obtained by vector averaging on the instantaneous wind direction data in the hour time period; The wave energy flux is calculated based on the effective wave height and the spectral peak period by a preset wave energy flux formula, representing the average wave power transmitted per unit wave peak length in the hour time period.

4. The hydraulic damping self-adaptive adjustment method based on wave energy prediction model according to claim 3, characterized in that, The multi-variable observation time series data corresponding to each buoy is synchronously intercepted by a sliding window, and a plurality of training samples are constructed based on the intercepted data, specifically: The multi-variable observation time series data corresponding to each buoy is synchronously intercepted by a sliding window, and a plurality of intercepted segments are obtained; each intercepted segment includes the multi-variable observation time series data of each buoy in the time period corresponding to the sliding window; In each intercepted segment, the buoy of the target device is set as the target buoy, and the remaining buoys are set as the neighborhood buoys; the multi-variable observation time series data corresponding to the target buoy and all neighborhood buoys in the intercepted segment are constructed as a training sample; The training label corresponding to the training sample is constructed, specifically: The sequence of wave energy corresponding to each hour in one or more consecutive hour time periods in the future of the target buoy is taken as the training label corresponding to the training sample.

5. The method of claim 4, wherein, The latest multi-variable observation time series data of each buoy is synchronously intercepted by a sliding window to obtain a real-time input segment, and the real-time input segment is input into the wave energy prediction model to output the wave energy prediction value of the position of the buoy corresponding to the target device; specifically: The latest multi-variable observation time series data of each buoy is obtained; The target device and the neighborhood device are set according to the preset requirements; the buoy corresponding to the target device is set as the target buoy, and the remaining buoys are set as the neighborhood buoys; the multi-variable observation time series data corresponding to each hour of the target buoy and each neighborhood buoy in the current hour and the previous consecutive whole hour are constructed into a real-time input segment, which is input into the wave energy prediction model to output the wave energy prediction value of the corresponding target buoy in one or more hours in the future.

6. The method of claim 1, wherein, The control signal is generated according to the wave energy prediction value, specifically: The wave energy level is divided according to the wave energy prediction value, and is mapped to the corresponding hydraulic damping level; the control signal is generated based on the hydraulic damping level.

7. The method of claim 1, wherein, The control signal is generated according to the wave energy prediction value, specifically: acquire a spectral peak period and an effective wave height corresponding to a position where the buoy is located within a time window corresponding to the wave energy prediction value, and a movement speed of the piston rod; based on the wave energy prediction value, the spectral peak period, the effective wave height, and the movement speed of the piston rod, dynamically calculate an optimal damping coefficient through a linear combination model, and generate a control signal for adjusting the opening degree of the control valve according to the optimal damping coefficient.

8. The method of claim 4, wherein, The mixed neural network model comprises a one-dimensional convolutional neural network module, a graph neural network module, and a segmented time series prediction module; wherein: The one-dimensional convolutional neural network module is used to extract local features of multivariate observation time series data of each buoy in the training sample and generate a comprehensive time series feature vector corresponding to each buoy; The graph neural network module is used to fuse the comprehensive time series feature vector of each buoy with the comprehensive time series feature vectors of other buoys having a spatial connection relationship with it according to the spatial correlation relationship between the buoys, to obtain a spatial enhanced feature corresponding to each buoy; The segmented time series prediction module is used to select a spatial enhanced feature corresponding to a target buoy from the spatial enhanced features corresponding to the buoys, and perform segmented embedding processing on the time series formed by the spatial enhanced feature, i.e., divide it into multiple time segments and map it into an embedding vector, then apply position coding to obtain an embedding sequence, and extract trend and periodicity information in the embedding sequence through a Transformer encoder to output a wave energy prediction sequence in a future period.

9. The method of claim 8, wherein, The one-dimensional convolutional neural network module comprises: a first channel for receiving wave energy flux time series data corresponding to each buoy in the training sample and extracting local features of the wave energy flux time series data of each buoy respectively; a second channel for receiving time series data of other variables except wave energy flux corresponding to each buoy in the training sample and extracting local features of the time series data of other variables of each buoy respectively; a feature splicing unit for splicing the local features extracted by the first channel and the second channel for the same buoy to obtain a spliced feature corresponding to each buoy; a fully connected projection unit for mapping the spliced features corresponding to each buoy into comprehensive time series feature vectors of a unified dimension to obtain a comprehensive time series feature vector corresponding to each buoy.

10. An electronic device comprising: A processor and a memory storing a program, characterized in that the program comprises instructions which, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 9.

Images