A smart monitoring and optimization system and method for operating parameters of marine energy devices

By combining a two-layer time-scale decoupling architecture with fuzzy adaptive inertial weights, the problem of global optimization and real-time response of marine energy devices in a strongly stochastic marine environment is solved, thereby improving power generation efficiency and reducing equipment losses, and enhancing the real-time performance and robustness of the system.

CN122469640BActive Publication Date: 2026-11-17TIANJIN PORT ENG INST LTD OF CCCC FIRST HARBOR ENG +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610915714.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-06-24
Publication Date
2026-11-17
Estimated Expiration
2046-06-24

AI Technical Summary

Technical Problem

Existing technologies cannot effectively solve the problem of unifying global optimization capability and real-time response speed for marine energy devices in highly random marine environments, resulting in low power generation efficiency and high equipment losses.

Method used

A two-layer time-scale decoupled architecture is adopted, which combines fuzzy adaptive inertial weights, surrogate model-assisted evaluation and health index-driven multi-objective decision-making. Through data perception and edge preprocessing layer, slow layer optimization module, fast layer mapping module and intelligent control execution layer, real-time optimized control of marine energy devices is achieved.

Benefits of technology

It significantly improves power generation efficiency, reduces equipment losses, and enhances the system's engineering reliability and environmental adaptability, while meeting engineering control precision requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122469640B_ABST
    Figure CN122469640B_ABST
Patent Text Reader

Abstract

This invention discloses an intelligent monitoring and optimization system and method for the operating parameters of marine energy devices, belonging to the field of marine renewable energy power generation control technology. The system includes: a data sensing and edge preprocessing layer for collecting operational data and environmental wave data and extracting wave statistical features and real-time wave features; a dual-layer time-scale optimization engine that performs population evolution optimization based on wave statistical features to generate a mapping model in a first preset cycle, and directly calculates control commands based on the real-time wave features according to the mapping model in each control cycle; an intelligent control execution layer for adjusting the controllable operating parameters of the device; and an equipment health assessment and online learning module for calculating the equipment health index and calibrating the proxy model online. This invention balances global optimization capabilities with real-time control response speed, significantly improving power generation efficiency and reducing equipment losses, and is applicable to marine energy power generation equipment such as oscillating water column wave energy devices and tidal current turbines.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of marine renewable energy power generation control technology, specifically to an intelligent monitoring and optimization system and method for the operating parameters of marine energy devices. Background Technology

[0002] Ocean energy (wave energy, tidal energy) is characterized by high energy density and strong predictability. Oscillating water column (OWC) wave energy devices and tidal turbines are currently promising technologies for commercialization. The energy conversion efficiency of these devices is highly dependent on the dynamic matching between operating parameters (such as turbine speed, pitch angle, yaw angle, etc.) and real-time sea conditions (wave height, period, current velocity).

[0003] At the optimization algorithm level, metaheuristic algorithms such as Particle Swarm Optimization (PSO), Differential Evolution (DE), and their hybrid variants have demonstrated powerful global optimization capabilities in multiple engineering fields. The FAHSPSO-DE algorithm proposed by Naderi et al. in 2021, by fusing self-learning PSO with Differential Evolution and introducing fuzzy logic to adjust inertia weights online, has achieved significant results in quasi-steady-state optimization problems such as optimal power flow in power systems. This algorithm reveals an effective hybrid optimization architecture—namely, balancing global exploration and local development through fuzzy adaptive balancing and enhancing population diversity through multi-strategy collaboration—an architecture with certain cross-domain reference value.

[0004] However, applying this hybrid optimization architecture to the control of marine energy device operating parameters requires substantial improvements and adaptations to address the unique characteristics of the marine environment. (1) Difference in time scale. Wave conditions have second-level fluctuation characteristics, while the population iteration mechanism of traditional metaheuristic algorithms requires tens to hundreds of objective function evaluations to converge. If the population evolution is directly placed in the control closed loop, it cannot meet the real-time requirement of second-level control cycle; (2) Heterogeneity of optimization objectives and constraints. The optimal power flow of the power system takes the power generation cost or emissions as the optimization objective, and the constraints are power balance and line power flow limits; while the ocean energy device takes the maximization of energy capture efficiency and the minimization of equipment loss as dual objectives, and the constraints include turbine speed restriction (avoiding resonance), actuator speed limit, air chamber aerodynamic nonlinearity, etc., and there is no direct correspondence between the two. (3) Differences in computational cost of objective function evaluation. Power flow calculation is based on mature analytical models and has low computational cost; however, the power generation evaluation of marine energy devices relies on high-fidelity aero-hydraulic coupling models, which have a huge computational load for a single evaluation and cannot be directly called in the optimization loop; (4) The randomness and unpredictability of environmental inputs. The randomness of waves is much stronger than that of power grid load fluctuations, requiring higher robustness of optimization results; (5) Equipment characteristic drift during long-term operation. During long-term operation, the characteristics of marine energy devices will change slowly due to scaling, wear, biofouling, etc., requiring the optimization system to have online self-adaptive capabilities.

[0005] Therefore, there is a need for an operational parameter optimization method that is specifically designed for the physical characteristics and operating environment of marine energy devices, taking into account both global optimization capabilities and real-time response speed. Summary of the Invention

[0006] To overcome the aforementioned shortcomings, this invention provides an intelligent monitoring and optimization system for the operating parameters of marine energy devices. Specifically, it is an intelligent optimization system and method for the operating parameters of marine energy power generation equipment such as oscillating water column (OWC) wave energy power generation devices and horizontal axis tidal current turbines. This invention achieves real-time decision-making of control parameters in the face of highly stochastic marine environments by employing dual-layer time-scale decoupling, fuzzy adaptive inertial weighting, surrogate model-assisted evaluation, and health index-driven multi-objective decision-making. This significantly improves power generation efficiency and reduces equipment losses, and is applicable to marine energy power generation equipment such as oscillating water column wave energy devices and tidal current turbines.

[0007] In a first aspect, the present invention provides an intelligent monitoring and optimization system for the operating parameters of a marine energy device, comprising: The data sensing and edge preprocessing layer is used to collect real-time operational data and environmental wave data from marine energy devices, and to preprocess the data to extract wave statistical features and real-time wave features. The dual-layer timescale optimization engine includes: a slow-layer optimization module, configured to be triggered at a first preset period, performing population evolution optimization based on the wave statistical characteristics and the current state of the ocean energy device to generate an optimal control parameter mapping model that reflects the mapping relationship between wave characteristics and control parameters; and a fast-layer mapping module, configured to directly calculate and output control commands based on the real-time wave characteristics within each control period, according to the optimal control parameter mapping model. The intelligent control execution layer is used to receive control commands output by the fast layer mapping module and adjust the controllable operating parameters of the marine energy device; the controllable operating parameters include one or more of the following: turbine speed setpoint, air chamber valve opening, generator torque, propeller pitch angle and yaw angle; The equipment health assessment and online learning module is used to continuously monitor the vibration characteristics, temperature change trends, and power generation efficiency degradation rate of key components, calculate the equipment health index, and collect data pairs between control commands and actual power generation effects for online calibration of the proxy model.

[0008] Preferably, the operating data includes: spindle vibration acceleration, turbine speed, rotor torque, bearing temperature, generator voltage and current; the environmental data includes: wave height, tide level and current velocity; the wave statistical characteristics include the following quantities calculated within a sliding time window: mean significant wave height, mean spectral peak period, mean wave energy flux and standard deviation of significant wave height; the real-time wave characteristics include: real-time significant wave height, real-time spectral peak period and real-time wave energy flux.

[0009] Preferably, the slow-layer optimization module adopts an optimization algorithm that combines particle swarm optimization and differential evolution, and includes a fuzzy adaptive mechanism based on wave change trends; the fuzzy adaptive mechanism uses wave height deviation, wave energy flux change rate and efficiency change as input variables, and dynamically adjusts the inertia weight of particle swarm optimization through fuzzy logic rules. The slow-layer optimization module further includes a robust boundary speed restriction constraint processing mechanism. This mechanism, based on the measurement accuracy of the speed sensor, extends the boundary of the original speed restriction zone outward by a safety margin to form an effective restriction zone, and applies a fitness penalty proportional to the intrusion depth to individuals whose control parameters fall into the effective restriction zone.

[0010] Preferably, the fast layer mapping module uses radial basis function interpolation to construct the mapping relationship and includes a sliding window node management mechanism; the sliding window node management mechanism is used to limit the upper limit of the number of nodes used in radial basis function interpolation, and when the addition of a new node causes the number of nodes to exceed the upper limit, nodes with similar sea conditions are clustered and merged.

[0011] Preferably, the slow layer optimization module further includes a surrogate model-assisted fitness evaluation mechanism, that is, to quickly evaluate the fitness of individuals in the population through a pre-trained neural network surrogate model; the device health assessment and online learning module is also used to perform online calibration of the output layer of the surrogate model using the recursive least squares method, and to trigger global retraining when the cumulative prediction error exceeds a preset threshold.

[0012] Preferably, the equipment health assessment and online learning module is further used to dynamically adjust the efficiency preference weights based on the calculated equipment health index; the slow-layer optimization module dynamically weights the two objectives of maximizing power generation efficiency and minimizing equipment losses according to the dynamically adjusted efficiency preference weights, so as to construct a single-objective optimization function.

[0013] Preferably, the equipment loss is quantified by an equipment loss assessment value L, the calculation formula of which is: ; in, The effective value of the spindle vibration acceleration. The rotor torque variation rate, The temperature rise of the bearing during the evaluation period. For vibration limits, To allow the maximum rate of torque change, To allow the maximum temperature rise, , , These are the weighting coefficients, and .

[0014] Preferably, the dual-layer timescale optimization engine further includes a cold start mechanism, which is configured to: when the system starts for the first time or the optimal control parameter mapping model is invalid, control the fast layer mapping module to output control commands using preset conservative control parameters; after the slow layer optimization module completes the first population evolution and generates an effective mapping model, smoothly switch the source of control commands for the fast layer mapping module to the calculation results based on the mapping model.

[0015] Preferably, the fast layer mapping module further includes an extrapolation protection mechanism and an output limiting and smoothing mechanism; the extrapolation protection mechanism is used to backtrack to the control parameter value of the nearest neighbor node when the real-time wave feature falls outside the effective range of the mapping model; the output limiting and smoothing mechanism is used to impose a rate of change limit on the control commands output by the fast layer mapping module.

[0016] Secondly, the present invention provides a method for intelligent monitoring and optimization of operating parameters of a marine energy device based on the system described in the first aspect, applied to the system described in the first aspect, the method comprising the following steps: S1: Real-time acquisition of operational data and environmental wave data of marine energy devices through the data sensing and edge preprocessing layer, and extraction of wave statistical features and real-time wave features; S2: Triggered by the slow layer optimization module at a first preset period, population evolution optimization is performed based on the wave statistical characteristics and the current state of the ocean energy device to generate an optimal control parameter mapping model that reflects the mapping relationship between wave characteristics and control parameters; S3: In each control cycle, the fast layer mapping module directly calculates and outputs control commands based on the optimal control parameter mapping model and the real-time wave characteristics, without the need for online iterative optimization; S4: The intelligent control execution layer adjusts the controllable operating parameters of the marine energy device according to the control commands; S5: The equipment health assessment and online learning module continuously monitors the vibration characteristics, temperature change trends, and power generation efficiency decay rate of key components to calculate the equipment health index, and collects data pairs of control commands and actual power generation effects for online calibration of the proxy model.

[0017] In a third aspect, a control device is provided, including a processor and a storage device, the storage device being adapted to store a plurality of program codes, the program codes being adapted to be loaded and run by the processor to perform the method described in the second aspect.

[0018] In a fourth aspect, a computer-readable storage medium is provided having a plurality of program codes stored thereon, the program codes being adapted to be loaded and run by a processor to perform the method described in the second aspect.

[0019] The above-described technical solutions of the present invention have at least one or more of the following beneficial effects: 1. Unification of Real-Time Performance and Global Optimization Capability: Through a decoupled architecture with two time scales, population evolution is placed in a slow layer at the minute level (cycle 2-5 minutes), while the fast layer directly outputs control commands within each control cycle (e.g., 2 seconds) through a radial basis function interpolation mapping model, eliminating the need for online iteration. Experimental results show that the slow layer's average optimization time is 10.2 seconds, and the fast layer's interpolation time is an average of 0.6 milliseconds, with uninterrupted control while retaining the global optimization capability of the metaheuristic algorithm.

[0020] 2. Significantly improved power generation efficiency: Under typical irregular wave sea conditions (significant wave height 1.2~2.1m, spectral peak period 6.5~9.8s), compared with conventional fixed speed control, the average power generation during the control period of this invention is increased by about 7.5%.

[0021] 3. Effective reduction of equipment wear: Through adaptive weighted multi-objective decision-making and dynamic adjustment of health index, the root mean square value of spindle vibration during the control period of this invention is reduced by about 12%, the number of torque impacts is reduced by about 15%, and the service life of the device is extended.

[0022] 4. Strong environmental adaptability: The fuzzy adaptive mechanism based on wave change trend automatically balances global exploration and local development; the rotational speed restriction constraint with robust boundary effectively avoids the resonance risk caused by measurement error; the fitness evaluation assisted by the surrogate model combined with recursive least squares online calibration adapts to the long-term characteristic drift of the equipment.

[0023] 5. High engineering reliability: The cold start mechanism ensures no control vacuum period; the extrapolation protection and output smoothing mechanism ensure safe operation under extreme fluctuations; the 95th percentile of the composite control deviation is about 13%, which meets the engineering control accuracy requirements. Attached Figure Description

[0024] The disclosure of this invention will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Furthermore, similar numbers in the drawings are used to denote similar components, wherein: Figure 1This is a schematic diagram of the structure of an intelligent monitoring and optimization system for the operating parameters of a marine energy device according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the RBF mapping model and node distribution in an embodiment of the present invention; Figure 3 This is the ocean energy simulation test site in this embodiment of the invention; Figure 4 This is a schematic diagram comparing the control effects of embodiments of the present invention; Figure 5 This is a flowchart illustrating a method for intelligent monitoring and optimization of operating parameters of a marine energy device according to an embodiment of the present invention. Detailed Implementation

[0025] Some embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0026] In the description of this invention, "module" and "processor" can include hardware, software, or a combination of both. A module can include hardware circuitry, various suitable sensors, communication ports, memory, and may also include software components, such as program code, or a combination of software and hardware. A processor can be a central processing unit, microprocessor, image processor, digital signal processor, or any other suitable processor. The processor has data and / or signal processing capabilities. The processor can be implemented in software, in hardware, or a combination of both. Non-transitory computer-readable storage media includes any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The terms "at least one A or B" or "at least one of A and B" have a similar meaning to "A and / or B" and can include only A, only B, or A and B. The singular terms "a" or "this" can also include plural forms.

[0027] The purpose of this invention is to provide a system and method for real-time optimization of the operating parameters of marine energy devices in highly stochastic marine environments. This invention draws upon the architectural concept of hybrid optimization and fuzzy adaptation in metaheuristic algorithms, and makes the following substantial improvements to address the multi-timescale characteristics, complex physical constraints, and high environmental stochasticity of marine energy devices: (1) A two-layer time-scale decoupled optimization architecture is proposed—the population evolution is placed in a slow layer at the minute level, and the fast layer outputs control commands in real time through a mapping model, thereby decoupling global optimization from real-time control. Simulation results show that under typical irregular wave conditions, the average deviation between the slow layer optimization results with statistical sea state characteristics as input and the instantaneous optimal control parameters is less than 8%, which meets the engineering control accuracy requirements; (2) Design a fuzzy adaptive mechanism based on wave change trend - its logical chain is: the change trend of wave statistical characteristics reflects the degree of environmental uncertainty; the greater the environmental uncertainty, the more necessary it is to enhance the global exploration ability of the population (increase the inertia weight); when the environment tends to be stable, the focus is on local fine development (decrease the inertia weight). (3) A method for handling rotational speed restriction with robust boundaries is proposed: the restriction boundary is extended based on the sensor measurement accuracy, which effectively avoids the risk of mechanical resonance under the condition of engineering measurement error; (4) Establish a rapid fitness evaluation and online calibration mechanism assisted by a surrogate model—verified by an irregular wave high-fidelity simulation model, the root mean square error of the surrogate model in the vicinity of the rated operating condition is within 5%. The recursive least squares method is used for online calibration of the output layer, and global retraining is triggered when the cumulative error continuously exceeds the threshold; (5) Construct an adaptive weighted multi-objective decision-making mechanism for dynamic trade-off between efficiency and loss—the equipment health index affects the weight allocation in the weighted objective function in real time, so as to achieve automatic balance between power generation efficiency and equipment protection.

[0028] like Figure 1 As shown, the present invention provides an intelligent monitoring and optimization system for the operating parameters of marine energy devices, comprising: The data sensing and edge preprocessing layer 110 is used to collect real-time operational data of marine energy devices and environmental wave data, and to preprocess the data to extract wave statistical features and real-time wave features. Specifically, the data sensing and edge preprocessing layer 110 includes: a sensor array (barometric pressure sensor, wave height / tide gauge, speed / torque sensor, vibration accelerometer, temperature sensor, voltage / current transformer) deployed on the marine energy device and an edge computing unit. Specifically: the barometric pressure sensor is used to collect chamber pressure; the wave height / tide gauge is used to collect real-time wave height and tide level; the speed / torque sensor is used to collect turbine speed and rotor torque; the vibration accelerometer is used to collect spindle vibration acceleration; the temperature sensor is used to collect bearing temperature; and the voltage / current transformer is used to collect generator output voltage and current. The edge computing unit performs data cleaning, wavelet denoising, and wave feature extraction, calculating characteristic quantities such as spectral peak period and wave energy flux from the wave height time series. In optional embodiments, the calculated characteristic quantities include: significant wave height, spectral peak period, wave energy flux, and the mean and standard deviation of each statistical quantity within a sliding time window.

[0029] In one specific embodiment, the operating data includes: spindle vibration acceleration, turbine speed, rotor torque, bearing temperature, generator voltage and current; environmental data includes: wave height, tide level and current velocity; wave statistical characteristics include the following quantities calculated within a sliding time window: mean effective wave height. Spectral peak period mean mean wave energy flux and the standard deviation of significant wave height Real-time wave characteristics include: real-time effective wave height. Real-time spectral peak period and real-time wave energy flux .

[0030] The dual-layer timescale optimization engine 120 includes: a slow-layer optimization module 121, configured to be triggered at a first preset period, performing population evolution optimization based on the wave statistical characteristics and the current state of the marine energy device to generate an optimal control parameter mapping model that reflects the mapping relationship between wave characteristics and control parameters; and a fast-layer mapping module 122, configured to directly calculate and output control commands based on the real-time wave characteristics according to the optimal control parameter mapping model in each control period. The dual-layer time-scale optimization engine 120 is deployed on a local industrial control computer or a shore-based server and includes the following two collaborative layers: Slow Layer Optimization Module 121: Every preset period (2-5 minutes, 180 seconds in this example) Perform a population evolution optimization calculation based on the wave statistical feature vector within a past time window (e.g., 15 minutes). Based on the current state of the device, the optimal control parameter mapping relationship is generated. During the slow-layer calculation, the fast-layer mapping module continues to use the mapping model generated in the previous cycle to ensure seamless and uninterrupted control logic.

[0031] Fast layer mapping module 122: In each control cycle (In this example, the timeframe is 2 seconds) Within this timeframe, based on the control parameter mapping relationship generated in the slow layer, and according to the real-time wave measurement values... Direct calculation of control commands No online iterative optimization is required.

[0032] The fast layer uses instantaneous features to reduce computational complexity, while the slow layer uses statistical features to improve robustness. The two layers have different input dimensions but achieve consistent association through a mapping model.

[0033] Specifically, the core technical features of the dual-layer time-scale optimization engine 120 will be explained in detail below.

[0034] (I) Two-layer time-scale decoupling architecture Slow layer optimization module 121 in each cycle Within, based on the wave statistical feature vector within a past time window (15 minutes in this example), Perform population evolution and output the optimal combination of control parameters under this sea state condition. Control parameter vector Includes turbine speed setpoint Air chamber valve opening Controllable variables, etc.

[0035] Fast layer mapping module 122 in each control cycle Inside, based on real-time measured wave characteristics Control commands are obtained from the mapping relationship established in the slow layer through radial basis function interpolation. The fast layer mapping module 122 constructs the mapping relationship using the radial basis function interpolation method and includes a sliding window node management mechanism. The sliding window node management mechanism is used to limit the upper limit of the number of nodes used in the radial basis function interpolation, and when the addition of a new node causes the number of nodes to exceed the upper limit, nodes with similar sea conditions are clustered and merged.

[0036] 1. Verification of the relationship between statistical optimality and instantaneous optimality To verify the engineering rationality of the two-layer architecture, simulation verification was conducted on the OWC device at a certain marine energy simulation test site. Fifty sets of irregular wave sea states (each set containing 15 minutes of continuous wave data) were selected, and the control parameters optimized based on the 15-minute statistical characteristics were calculated. And the control parameter sequence obtained based on the instantaneous wave height and period optimization at each time step. (Theoretical benchmark). [The following is a list of parameters:] ... It operates with fixed parameters under actual wave conditions, recording the power generation at each moment.

[0037] Statistical results: Using the statistically optimal parameters, the average power generation at each time point reached 92.3% to 96.8% of the theoretical instantaneous optimal average power, with an average deviation of 7.2%. Considering that real-time point-by-point optimization calculation is not feasible, this deviation level is within the acceptable range for engineering control accuracy.

[0038] 2. Error superposition analysis In actual control, three types of errors may coexist: the average deviation between statistical optimum and instantaneous optimum (approximately 7.2%), RBF interpolation error (<4.2% in the normal node region, reaching 8%~12% in the sparse region), and surrogate model prediction error (RMSE approximately 4.3%). Since these three types of errors originate independently and have different distribution characteristics, the probability of them simultaneously reaching their maximum values ​​is extremely low. Simulation statistics show that under the most unfavorable operating condition combination, the 95th percentile of the composite control deviation is approximately 13%, and the 90th percentile is approximately 10%, mainly occurring during periods of abrupt sea state changes (approximately 5% of the total operating time). During periods of stable sea state (approximately 75% of the total operating time), the composite control deviation is typically within the range of 5%~8%. The system's protection mechanisms further suppress the impact of extreme deviations.

[0039] 3. RBF Interpolation Error Analysis and Boundary Treatment The fast layer mapping module 122 also includes an extrapolation protection mechanism and an output smoothing mechanism. The extrapolation protection mechanism is used to backtrack to the control parameter value of the nearest neighbor node when the real-time wave characteristics fall outside the effective range of the mapping model. The output limiting and smoothing mechanism is used to impose a rate-of-change limit on the control commands output by the fast layer mapping module 122. Specifically: Radial basis function interpolation has an interpolation error of no more than 4.2% in regions with normal node density (node ​​spacing ≤ normalized distance 0.15 in the input space); however, in sparsely populated regions (node ​​spacing > 0.25), the interpolation error may rise to 8%~12%. The fast layer mapping module is equipped with the following protection mechanisms: Extrapolation protection mechanism: When real-time wave characteristics When the event falls outside the convex hull of an RBF node, the control parameter value is retraced to the nearest neighbor node, and the event is marked.

[0040] Node density monitoring: Record the normalized distance between the input feature and the nearest RBF node during each fast layer query. When the distance exceeds the threshold for multiple consecutive queries, new exploration individuals are preferentially generated in this sparse region during the next slow layer cycle.

[0041] Output limiting and smoothing mechanism: The rate of change of the control command output by RBF is limited to avoid impact on the actuator caused by interpolation jumps.

[0042] To prevent the mapping model nodes from accumulating indefinitely, a sliding window node management mechanism is adopted: only non-dominated solution nodes generated within the last 24 hours are retained, with a maximum number of nodes. When the addition of a new node causes the number of nodes to exceed the upper limit, the K-means clustering method is used to merge nodes with similar sea conditions. The two nodes with the closest Euclidean distance are merged into one, and the position of the merged node is taken as the average of the two points to ensure that the computational complexity of the fast layer remains constant. Figure 2 This is a schematic diagram of the RBF mapping model and node distribution. Figure 2 The solid blue circle in the middle represents the nodes of the optimal control parameters accumulated after each population evolution in the slow layer, which are concentrated near the rated operating conditions to form dense clusters; the dashed circle on the right represents the nodes in the sparse region, which are sparsely distributed.

[0043] 4. System cold start mechanism The dual-layer timescale optimization engine 120 also includes a cold start mechanism, which is configured such that: when the system starts for the first time or the optimal control parameter mapping model is invalid, the fast layer mapping module 122 outputs control commands using preset conservative control parameters; after the slow layer optimization module 121 completes the first population evolution and generates an effective mapping model, the source of control commands for the fast layer mapping module 122 is smoothly switched to the calculation results based on the mapping model. Specifically: When the system is first powered on or restarted after a long period of inactivity, the slow-layer optimization module 121 has not yet completed its first population evolution, and the RBF mapping model is empty. During this period (approximately the first 3 minutes), control commands originate from the following sources: ① Preset conservative parameter mode: During the system configuration phase, a set of conservative control parameters is preset based on the device's historical operating data and engineering experience. This parameter, provided all safety constraints (speed limits, actuator amplitude limits, etc.) are met, allows for stable operation under common fluctuation conditions. After system startup, the fast-layer controller directly adopts... Output control commands.

[0044] ② Transition to Optimization Mode: After the slow-layer optimization module completes its first evolution, it generates an initial RBF mapping model and notifies the fast-layer controller via a flag. The fast-layer controller smoothly switches to the RBF mapping output mode in the next control cycle. The switching process uses a ramp function transition (linear interpolation within 5 seconds in this example) to avoid step changes in control commands.

[0045] ③ Historical mapping reuse (optional): If the system restarts briefly due to communication interruption or other reasons, and the most recent valid RBF mapping model is retained in local storage, the model can be directly loaded as the initial mapping to shorten the transition time.

[0046] This cold start mechanism ensures that the device remains under effective control throughout its entire lifespan, without any control vacuum period.

[0047] (II) Fuzzy Adaptive Mechanism Based on Wave Change Trend The slow-layer optimization module 121 employs a hybrid optimization algorithm combining particle swarm optimization and differential evolution, and includes a fuzzy adaptive mechanism based on wave change trends. The fuzzy adaptive mechanism uses wave height deviation, wave energy flux change rate, and efficiency change as input variables, and dynamically adjusts the inertia weights of the particle swarm optimization through fuzzy logic rules. Slow-level optimization employs a hybrid structure combining particle swarm optimization and differential evolution. In particle swarm optimization, inertia weights... The value of determines the search behavior: larger Enhanced global exploration, smaller Enhance local development.

[0048] The function of the fuzzy logic controller is to automatically determine the required search strategy based on the changing trend of the current sea state, and then adjust the change in inertia weight. The input variables and membership function parameters are shown in Table 1. The input variables are normalized to the interval [0,1], and the membership functions use a combination of triangular, Z-shaped, and S-shaped functions.

[0049] Table 1. Parameters of Membership Functions for Fuzzy Input Variables

[0050] Note: The parameters for the Z-shaped function are left boundary a and right boundary b; the parameters for the S-shaped function are left boundary a and right boundary b; the parameters for the triangular function are left foot a, peak b, and right foot c.

[0051] Table 2. Membership Function Parameters for Fuzzy Output Variables

[0052] Table 3. Fuzzy Rule Base (Complete Set of 27 Rules)

[0053] The centroid method is used to resolve fuzziness. The inertia weight update formula is: ; in For the updated inertia weights, This is the current inertia weight. The inertia weight is constrained to... Within the range.

[0054] (III) Fitness assessment and online calibration assisted by surrogate models The slow layer optimization module 121 also includes a fitness evaluation mechanism assisted by a surrogate model, that is, to quickly evaluate the fitness of individuals in the population through a pre-trained neural network surrogate model.

[0055] The surrogate model employs a three-layer backpropagation (BP) neural network: the number of nodes in the input layer corresponds to the dimension of the input features (4-dimensional wave statistical features + 2-3-dimensional control parameters, totaling 6-7 dimensions), the hidden layer has 12 nodes (activation function tansig), and the output layer has 2 nodes (activation function purelin), which output the predicted power generation value. and loss prediction value .

[0056] (1) Sources of offline training data The training data comes from two parts: (1) long-term operating data (including irregular wave conditions) collected at the OWC device at a certain ocean energy simulation test site; (2) high-fidelity CFD model (irregular wave boundary conditions are generated using JONSWAP spectrum, and the turbulence model is SST) for the typical operating range of the device. Simulation samples were generated (with approximately 1.2 million grid cells). The two sets of data total... =800 sets, covering 50%~120% of the rated operating conditions.

[0057] (2) Accuracy verification of the surrogate model The accuracy of the surrogate model was evaluated using an independent validation set (200 sets of irregular wave operating condition data): the root mean square error (RMSE) for power generation prediction was 4.3% (relative to rated power), and the maximum absolute percentage error (MAPE) was 7.8%; the RMSE for equipment loss prediction was 5.1%, and the maximum APE was 9.5%. This accuracy meets the fitness assessment requirements for slow-layer population evolution.

[0058] Regarding the adaptability of irregular wave extrapolation: The training data already includes irregular wave conditions with different spectral peak periods and effective wave heights, and the generalization error of the surrogate model within the training range has been confirmed on the validation set. For extreme wave conditions beyond the training range, the system sets up an extrapolation protection mechanism (see the aforementioned "Extrapolation Protection Mechanism" section) to fall back to the nearest neighbor control parameters to ensure safety.

[0059] (3) RLS online calibration After each slow layer cycle, actual running data is collected, and the weights of the neural network output layer are updated online using the recursive least squares method. The forgetting factor λ = 0.98, and the covariance matrix is ​​initialized to... ( (This is the identity matrix). The update formula is: ; in: This is the hidden layer output vector at the current time. This is an estimate of the output layer weight vector; The actual measured output (power generation or loss index); It is the gain vector; Let be the covariance matrix.

[0060] (4) Global retraining trigger mechanism: When the prediction error exceeds the preset threshold for 5 consecutive slow-layer cycles (each cycle is not a sudden change in sea state). At this time, a global retraining process is triggered in the background. This global retraining uses the 800 most recently accumulated valid datasets and is executed in a separate thread, without blocking the current slow-layer optimization. Real-world testing on a shore-based industrial control computer (Intel i7-10700, 32 GB RAM) showed an average retraining time of 42 seconds and a maximum of 68 seconds, both within the slow-layer cycle. Completed within the specified time.

[0061] (5) Explanation of the application boundaries of RLS RLS's online calibration of output layer weights is suitable for scenarios where equipment characteristics undergo localized, slow changes (such as sensor drift or minor scaling on blade surfaces). When equipment characteristics undergo global, structural changes (such as changes in chamber geometry or severe blade damage), it is necessary to rely on global retraining triggered by cumulative error exceeding a threshold or continuous monitoring of equipment health indices to issue maintenance warnings.

[0062] (iv) Quantitative evaluation function for equipment loss Equipment loss assessment value It is composed of the following weighted average of measurable physical quantities: ; in: Effective value of spindle vibration acceleration (unit: ); Vibration limits are set with reference to the vibration limits for horizontal axis rotating machinery below 2 MW in ISO 10816-3 standard; Rotor torque change rate (unit: Nm / s). Take the maximum absolute value within the sliding window; The maximum allowable torque variation rate is determined by taking the manufacturer's design value from the generator design manual. As a conservative limit for long-term operation; Temperature rise of bearing during the evaluation period (unit: °C); The maximum allowable temperature rise is set according to the parameters on the equipment's nameplate. , , The corresponding weight coefficients, and satisfying The values ​​are all in the range of (0,1), and optionally in one embodiment The value is determined by the analytic hierarchy process combined with expert experience.

[0063] The value range of this function is usually [0,1] or slightly higher than 1 (when a certain indicator exceeds the limit). The higher the value, the more serious the equipment damage.

[0064] (v) Handling of rotational speed-limited areas with robust boundaries The slow-layer optimization module 121 also includes a robust boundary-based speed restriction constraint processing mechanism. Based on the measurement accuracy of the speed sensor, the boundary of the original speed restriction zone is extended outward by a safety margin to form an effective restriction zone. Individuals whose control parameters fall into the effective restriction zone are subject to a fitness penalty proportional to the intrusion depth. Specifically: Turbines have one or more speed limits, the first Each restricted area is designated as an interval. Considering that speed measurement errors may cause the actual operating point to unexpectedly enter the restricted area and trigger resonance, the effective restricted area boundary is extended based on sensor accuracy.

[0065] Let the estimated standard deviation of the speed measurement sensor corresponding to its accuracy class be _____. (Originally obtained from the sensor datasheet; typical value is 0.5~1% of rated speed), robustness coefficient The effective restricted area boundary is: ; in: : Index of the restricted speed range; : The lower boundary value of the z-th original rotational speed restriction zone; : The upper boundary value of the z-th original rotational speed restriction zone; The standard deviation estimate of the rotational speed sensor reflects the measurement accuracy; : Robustness coefficient, a factor used to expand the boundaries of restricted areas; : The lower boundary value of the z-th effective speed limit after considering measurement error; : The upper boundary value of the z-th effective speed limit after considering measurement error.

[0066] For control parameter combinations falling into the effective restricted area (i.e.) ) individuals, calculate their invasion depth : ; in: The turbine rotational speed corresponding to the current individual; The depth of an individual's rotational speed within the effective restricted zone, i.e., the distance from the nearest boundary.

[0067] Intrusion depth represents the distance of the individual from the nearest boundary of the effective restricted area. The applied fitness penalty is: ; in: Penalty coefficient, ranging from 0.5 to 1.0; Speed ​​adjustment range .

[0068] The penalty term increases with the intrusion depth, encouraging the optimization algorithm to push the rotation speed outside the restricted area. When multiple restricted areas overlap after expansion, the overlapping areas are merged into a single continuous restricted area, and the intrusion depth is calculated based on the merged boundary.

[0069] (vi) Health index-driven adaptive weighted multi-objective decision making Slow-level optimization uses a dynamic weighting method to transform the dual objective (maximizing power generation efficiency and minimizing equipment losses) into a single objective function. : ; in: Control parameter combination Output power (unit: kW); Rated power of the device (unit: kW); Equipment loss assessment value; Efficiency preference weighting coefficient, derived from equipment health index Determined dynamically.

[0070] (1) Equipment health index Calculation Equipment Health Index Calculated using the following formula: ; in: : 7-day moving average trend value of the effective value of vibration (unit: ); The percentage decrease in power generation efficiency relative to the initial baseline; : The maximum allowable percentage of efficiency degradation, set to 20%; : Fault warning signal, 0 indicates no warning, 1 indicates a warning; weighting coefficient , , The determination was made based on historical fault data statistics and expert experience.

[0071] The value range is [0,1], with values ​​closer to 1 indicating better device health. All components have been normalized to the [0,1] interval, and the linearly weighted result is obtained through... The form is converted into a health index.

[0072] (2) Weighting coefficients Dynamic determination Efficiency preference weighting coefficient The calculation formula is: ; Among them, the benchmark value Adjustment range Substituting the values, we get: ; when exist When changing within a range, The theoretical range is In practical applications, a hard limiting constraint consistent with the theoretical range is applied. .

[0073] when When it is higher (e.g., >0.7), The bias is towards 0.6~0.75, with decisions focusing on power generation efficiency; when When the value is low (e.g., <0.5), The bias is towards 0.45~0.60, and the decision-making focuses on reducing equipment wear and tear.

[0074] The intelligent control execution layer 130 is used to receive control commands output by the fast layer mapping module 122 and adjust the controllable operating parameters of the marine energy device; the controllable operating parameters include one or more of the following: turbine speed setpoint, air chamber valve opening, generator torque, propeller pitch angle and yaw angle; Specifically: The intelligent control execution layer 130 receives control commands and adjusts the controllable operating parameters of the marine energy device through a PLC or real-time controller, including but not limited to: turbine speed setpoint. Air chamber valve opening Pitch angle Yaw angle Generator torque The control parameter vector is uniformly denoted as... .

[0075] The Equipment Health Assessment and Online Learning Module 140 is used to continuously monitor the vibration characteristics, temperature change trends, and power generation efficiency decay rate of key components, calculate the equipment health index, and collect data pairs between control commands and actual power generation effects for online calibration of the proxy model.

[0076] Specifically, the equipment health assessment and online learning module 140 is used to continuously monitor the vibration characteristics, temperature change trends, and power generation efficiency degradation rate of key components, and calculate the equipment health index. Simultaneously, data pairs of control commands and actual power generation effects are collected for online calibration of the slow-layer proxy model. The key components include at least: a spindle, bearings, and a generator; the vibration characteristic is the spindle vibration acceleration; the temperature change trend is the bearing temperature rise; and the power generation efficiency decay rate is calculated based on the generator output power.

[0077] Furthermore, the device health assessment and online learning module 140 is also used to perform online calibration of the output layer of the proxy model using the recursive least squares method, and trigger global retraining when the cumulative prediction error exceeds a preset threshold.

[0078] Furthermore, the equipment health assessment and online learning module 140 is also used to dynamically adjust the efficiency preference weights based on the calculated equipment health index; the slow-layer optimization module 121 dynamically weights the two objectives of maximizing power generation efficiency and minimizing equipment losses according to the dynamically adjusted efficiency preference weights, so as to construct a single-objective optimization function. Specific implementation examples: 1. Implementation environment: The 100kW OWC wave energy generation device at a certain marine energy simulation test site was used as the implementation object. The device is equipped with a Wells turbine (1.2m in diameter) and a doubly-fed induction generator. The control system hardware configuration includes: ARM Cortex-A72 edge computing unit, Beckhoff CX5130 fast layer controller, and shore-based industrial control computer (Intel i7-10700, 32GB RAM) for slow layer optimization engine. Figure 3 This is the marine energy simulation test site in this embodiment of the invention.

[0080] 2. Algorithm parameter settings: (1) Slow-level optimization cycle Seconds, fast layer control cycle Second; (2) Population size Maximum number of iterations ; (3) Particle swarm learning factor Initial value of inertial weight ; (4) Differential evolution scaling factor Crossover probability ; (5) The fuzzy rule base is shown in Table 3, and the membership function parameters are shown in Tables 1 and 2; (6) Neural network surrogate model: 6-dimensional input layer, 12-node hidden layer (tansig), 2-node output layer (purelin); RLS forgetting factor Initial covariance matrix ; (7) RBF node sliding window: 24 hours, maximum number of nodes ; (8) Equipment loss weight ; (9) Health Index Weighting , , benchmark weight Adjustment range ; (10) Penalty coefficient Robustness coefficient .

[0081] 3. Execution process example: (1) System cold start phase: This is the first time the device has been powered on; slow-layer optimization is not yet complete. The fast-layer controller uses preset conservative parameters. As the initial control command, the device enters a stable operating state. At the same time, the slow-layer optimization module begins to collect wave characteristic data, and after accumulating enough data for 15 minutes, it triggers the first population evolution.

[0082] The first slow-layer optimization is complete, and the optimal solution is output. Based on the corresponding wave characteristics, an initial RBF mapping model is established. The fast layer controller then transfers control commands from... The transition to RBF output is linear and smooth, with no shock during the switching process. Afterwards, it enters a normal double-layer optimization loop.

[0083] (2) Data preparation: Collect wave characteristic sequences from the past 15 minutes and calculate the mean. =1.8m, =8.2s, =18.5kW / m, standard deviation =0.22 m.

[0084] (3) Population evolution: Forty individuals are randomly generated, and the process iterates for 80 generations. Each evaluation calls a neural network proxy model (average time 0.4ms / evaluation), totaling approximately 1.3 seconds for 3200 evaluations. Additional computations such as particle updates, differential evolution operators, and sorting take approximately 8.9 seconds. The total time for a single slow layer optimization is approximately 10.2 seconds. During this time, the fast layers continue to use the RBF model from the previous cycle.

[0085] (4) Decision output: Current equipment health index =0.88. Calculate the efficiency preference weight: ; Choose the objective function that maximizes Solution The solution predicts a power generation of 90.8 kW and a predicted loss value. L =0.18.

[0086] (5) Mapping update: Will The RBF mapping model is updated using a sliding window mechanism, along with its corresponding wave feature vectors.

[0087] 4. Calculation time verification: Statistics on the time taken for 50 consecutive slow-level optimizations: average 10.4 seconds, standard deviation 1.1 seconds, maximum 13.2 seconds, all less than [previous value]. =180 seconds. The average time for fast layer RBF interpolation is 0.6 ms, and the maximum time is 1.1 ms, which satisfies... =2-second cycle requirement.

[0088] To verify the control effect of the method of this invention, a comparative experiment was conducted with a constant speed control method under the same sea state conditions. A typical irregular wave condition with a significant wave height of 1.2–2.1 m and a spectral peak period of 6.5–9.8 s was selected, and 6 hours of operating data were recorded. Figure 4 The diagram shown is a comparison of the control effects of an embodiment of the present invention. Figure 4 The diagram includes two sub-graphs, A and B, where A represents the comparison results of average power generation, and B represents the comparison results of dynamic control performance. The output power of the method described in this invention is generally higher than that of constant-speed control, and the fluctuations are more stable. Statistical results show that the average power generation of this invention is 87.6 kW, while that of constant-speed control is 81.5 kW, representing an improvement of approximately 7.5%.

[0089] Figure 5 This is a flowchart illustrating an intelligent monitoring and optimization method for the operating parameters of a marine energy device according to an embodiment of the present invention. The method, applied to the system described in the foregoing embodiment, includes the following steps: S1: Real-time acquisition of operational data and environmental wave data of marine energy devices through the data sensing and edge preprocessing layer, and extraction of wave statistical features and real-time wave features; S2: Triggered by the slow layer optimization module at a first preset period, population evolution optimization is performed based on the wave statistical characteristics and the current state of the ocean energy device to generate an optimal control parameter mapping model that reflects the mapping relationship between wave characteristics and control parameters; S3: In each control cycle, the fast layer mapping module directly calculates and outputs control commands based on the optimal control parameter mapping model and the real-time wave characteristics, without the need for online iterative optimization; S4: The intelligent control execution layer adjusts the controllable operating parameters of the marine energy device according to the control commands; S5: The equipment health assessment and online learning module continuously monitors the vibration characteristics, temperature change trends, and power generation efficiency decay rate of key components to calculate the equipment health index, and collects data pairs of control commands and actual power generation effects for online calibration of the proxy model.

[0090] Furthermore, the intelligent monitoring and optimization system for marine energy device operating parameters according to an embodiment of the present invention also includes a control device. In one embodiment of the control device according to the present invention, the control device includes a processor and a storage device. The storage device can be configured to store a program for executing the intelligent monitoring and optimization method for marine energy device operating parameters according to the above-described method embodiments. The processor can be configured to execute the program in the storage device, which includes, but is not limited to, a program for executing the intelligent monitoring and optimization method for marine energy device operating parameters according to the above-described method embodiments. For ease of explanation, only the parts related to the embodiments of the present invention are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of the present invention. The control device can be a control device device device comprising various electronic devices.

[0091] Furthermore, the intelligent monitoring and optimization system for marine energy device operating parameters of the present invention also includes a computer-readable storage medium. In one embodiment of the present invention, the computer-readable storage medium can be configured to store a program that executes the intelligent monitoring and optimization method for marine energy device operating parameters described in the above-described method embodiments. This program can be loaded and run by a processor to implement the aforementioned intelligent monitoring and optimization method for marine energy device operating parameters. For ease of explanation, only the parts related to the embodiments of the present invention are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of the present invention. The computer-readable storage medium can be a storage device comprising various electronic devices. Optionally, in the embodiments of the present invention, the computer-readable storage medium is a non-transitory computer-readable storage medium.

[0092] Furthermore, it should be understood that since the various modules are only provided to illustrate the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the processor's software, hardware, or a combination of software and hardware. Therefore, the number of modules shown in the figures is merely illustrative.

[0093] Those skilled in the art will understand that the various modules in the device can be adaptively split or combined. Such splitting or combining of specific modules will not cause the technical solution to deviate from the principles of the present invention; therefore, the technical solutions after splitting or combining will fall within the protection scope of the present invention.

[0094] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A smart monitoring and optimization system for the operating parameters of a marine energy device, characterized in that, include: The data sensing and edge preprocessing layer is used to collect real-time operational data and environmental wave data from marine energy devices, and to preprocess the data to extract wave statistical features and real-time wave features. The dual-layer timescale optimization engine includes: a slow-layer optimization module, configured to be triggered at a first preset period, performing population evolution optimization based on the wave statistical characteristics and the current state of the ocean energy device to generate an optimal control parameter mapping model that reflects the mapping relationship between wave characteristics and control parameters; and a fast-layer mapping module, configured to directly calculate and output control commands based on the real-time wave characteristics within each control period, according to the optimal control parameter mapping model. The intelligent control execution layer is used to receive control commands output by the fast layer mapping module and adjust the controllable operating parameters of the marine energy device; the controllable operating parameters include one or more of the following: turbine speed setpoint, air chamber valve opening, generator torque, propeller pitch angle and yaw angle; The equipment health assessment and online learning module is used to continuously monitor the vibration characteristics, temperature change trends and power generation efficiency degradation rate of key components, calculate the equipment health index, and collect data pairs between control commands and actual power generation effects for online calibration of the proxy model. The slow-layer optimization module employs a hybrid optimization algorithm combining particle swarm optimization and differential evolution, and includes a fuzzy adaptive mechanism based on wave change trends. The fuzzy adaptive mechanism uses wave height deviation, wave energy flux change rate, and efficiency change as input variables, and dynamically adjusts the inertia weight of particle swarm optimization through fuzzy logic rules. The slow layer optimization module also includes: a robust boundary speed restriction constraint processing mechanism. The robust boundary speed restriction constraint processing mechanism is based on the measurement accuracy of the speed sensor, expands the boundary of the original speed restriction area outward by a safety margin to form an effective restriction area, and applies a fitness penalty proportional to the invasion depth to individuals whose control parameters fall into the effective restriction area. The effective restricted area boundary is: ; in: : Index of the restricted speed range; : The lower boundary value of the z-th original rotational speed restriction zone; : The upper boundary value of the z-th original rotational speed restriction zone; The standard deviation estimate of the rotational speed sensor reflects the measurement accuracy; : Robustness coefficient, a factor used to expand the boundaries of restricted areas; : The lower boundary value of the z-th effective speed limit after considering measurement error; : The upper boundary value of the z-th effective speed limit after considering measurement error; For individuals in the population whose rotational speed falls within the expanded effective restricted zone, calculate the invasion depth: ; in: The turbine rotational speed corresponding to the current individual; The depth of an individual's rotational speed within the effective restricted zone, i.e., the distance from the nearest boundary; Intrusion depth represents the distance of the individual from the nearest boundary of the effective restricted area, and the fitness penalty applied is: ; in: Penalty coefficient, ranging from 0.5 to 1.0; Speed ​​adjustment range ; The equipment health assessment and online learning module is also used to dynamically adjust the efficiency preference weights based on the calculated equipment health index; the slow-layer optimization module dynamically weights the two objectives of maximizing power generation efficiency and minimizing equipment losses according to the dynamically adjusted efficiency preference weights, so as to construct a single-objective optimization function; The equipment loss is quantified by the equipment loss assessment value L, and its calculation formula is as follows: ; in, The effective value of the spindle vibration acceleration. The rotor torque variation rate, The temperature rise of the bearing during the evaluation period. For vibration limits, To allow the maximum rate of torque change, To allow the maximum temperature rise, , , These are the weighting coefficients, and ; The fast layer mapping module uses radial basis function interpolation to construct the mapping relationship and includes a sliding window node management mechanism. The sliding window node management mechanism is used to limit the upper limit of the number of nodes used in radial basis function interpolation, and when the addition of a new node causes the number of nodes to exceed the upper limit, nodes with similar sea conditions are clustered and merged. The slow layer optimization module also includes a fitness evaluation mechanism assisted by a surrogate model, that is, to quickly evaluate the fitness of individuals in the population through a pre-trained neural network surrogate model; the device health assessment and online learning module is also used to perform online calibration of the output layer of the surrogate model using the recursive least squares method, and to trigger global retraining when the cumulative prediction error exceeds a preset threshold. The online calibration process involves collecting actual operating data after each slow layer cycle, updating the weights of the neural network output layer online using the recursive least squares method, setting the forgetting factor λ = 0.98, and initializing the covariance matrix as follows: , For the identity matrix; the update formula is: ; in: This is the hidden layer output vector at the current time. This is an estimate of the output layer weight vector; The actual measured output, i.e., power generation or loss indicators; It is the gain vector; It is the covariance matrix; Slow-level optimization uses a dynamic weighting method to transform maximizing power generation efficiency and minimizing equipment losses into a single objective function. : ; in: Control parameter combination Output power at the following levels; Rated power of the device; Equipment loss assessment value; Efficiency preference weighting coefficient, derived from equipment health index Dynamically determined; Equipment Health Index Calculated using the following formula: ; in: : The 7-day moving average trend value of the effective value of vibration; The percentage decrease in power generation efficiency relative to the initial baseline; : The maximum permissible percentage of efficiency degradation; : Fault warning signal, 0 indicates no warning, 1 indicates a warning; weighting coefficient , , The determination was made based on historical fault data statistics and expert experience. Efficiency preference weighting coefficient The calculation formula is: ; Among them, the benchmark value Adjustment range .

2. The intelligent monitoring and optimization system for the operating parameters of marine energy devices according to claim 1, characterized in that, The operational data includes: spindle vibration acceleration, turbine speed, rotor torque, bearing temperature, generator voltage and current; the environmental wave data includes: wave height, tide level and current velocity; the wave statistical features include the following quantities calculated within a sliding time window: mean significant wave height, mean spectral peak period, mean wave energy flux and standard deviation of significant wave height; the real-time wave features include: real-time significant wave height, real-time spectral peak period and real-time wave energy flux.

3. The intelligent monitoring and optimization system for the operating parameters of marine energy devices according to claim 1, characterized in that, The dual-layer timescale optimization engine also includes a cold start mechanism, which is configured to: when the system starts for the first time or the optimal control parameter mapping model is invalid, control the fast layer mapping module to output control commands using preset conservative control parameters; after the slow layer optimization module completes the first population evolution and generates an effective mapping model, smoothly switch the source of control commands for the fast layer mapping module to the calculation results based on the mapping model.

4. The intelligent monitoring and optimization system for the operating parameters of marine energy devices according to claim 1, characterized in that, The fast layer mapping module also includes an extrapolation protection mechanism and an output limiting and smoothing mechanism; the extrapolation protection mechanism is used to backtrack to the control parameter value of the nearest neighbor node when the real-time wave feature falls outside the effective range of the mapping model; the output limiting and smoothing mechanism is used to impose a rate of change limit on the control commands output by the fast layer mapping module.

5. A method for intelligent monitoring and optimization of operating parameters of a marine energy device, applied to the system as described in any one of claims 1 to 4, characterized in that, The method includes the following steps: S1: Real-time acquisition of operational data and environmental wave data of marine energy devices through the data sensing and edge preprocessing layer, and extraction of wave statistical features and real-time wave features; S2: Triggered by the slow layer optimization module at a first preset period, population evolution optimization is performed based on the wave statistical characteristics and the current state of the ocean energy device to generate an optimal control parameter mapping model that reflects the mapping relationship between wave characteristics and control parameters; S3: In each control cycle, the fast layer mapping module directly calculates and outputs control commands based on the optimal control parameter mapping model and the real-time wave characteristics, without the need for online iterative optimization; S4: The intelligent control execution layer adjusts the controllable operating parameters of the marine energy device according to the control commands; S5: The equipment health assessment and online learning module continuously monitors the vibration characteristics, temperature change trends, and power generation efficiency decay rate of key components to calculate the equipment health index, and collects data pairs of control commands and actual power generation effects for online calibration of the proxy model.

Citation Information

Patent Citations

  • Wave energy power generation efficiency optimization control method and device based on artificial intelligence

    CN120630739A

  • Ocean extreme wave measurement method based on multi-source data driving

    CN121829466A