Wave energy monitoring data on-demand incremental synchronization method based on power generation working condition judgment

Abstract

The present application relates to a kind of wave energy monitoring data on-demand incremental synchronization method based on power generation condition judgment.First, build and train condition coupling gate cycle unit prediction model, input historical monitoring data output future power generation trend probability, as the configuration basis of synchronization check scheduling and block, retry strategy;Again, based on power generation trend probability dynamic adjustment synchronization check task trigger period, probability is higher than preset threshold, activate synchronization preparation in advance, determine synchronization start and stop time according to incremental promotion rule after task start, generate time window;Subsequently, in window, generator unit power generation power data is preferentially synchronized incrementally, and power time series is obtained;With the sequence as only basis, condition gate judgment is carried out, and the output result controls non-power monitoring parameter synchronization execution;When determining that there is effective power generation condition, the remaining non-power parameters are synchronized in preset order, and for high-frequency high-data-volume parameter, block granularity is dynamically adjusted based on power generation trend probability, and data capture synchronization is completed block by block.

Description

Technical Field

[0001] This invention relates to the field of ocean energy power generation monitoring technology, specifically to a method for incremental synchronization of wave energy monitoring data based on power generation conditions. Background Technology

[0002] The on-site monitoring parameters for the OWC wave energy power generation demonstration project cover five major categories: hydro-meteorological parameters (wave height, period, wave direction, tidal level, temperature, humidity, wind speed, etc.), OWC hydrodynamic response parameters (liquid level in the gas chamber, gas pressure, etc.), turbine and power generation monitoring parameters (gas flow rate, torque, speed, vibration, etc.), electrical parameters (current, voltage, temperature, etc.), and structural safety monitoring parameters (displacement, strain, wave pressure, etc.). According to the classification and statistics, there are 24 high-frequency sampling points (kHz level), 22 mid-frequency sampling points (10 Hz level), and 79 low-frequency sampling points (1 Hz level), totaling 125 parameter measuring points.

[0003] Regarding the number of raw sampling points: Based on the nominal reference sampling rate of each measuring point (kHz level calculated as 1 kHz, 10 Hz level as 10 Hz, and 1 Hz level as 1 Hz), the daily cumulative calculation is: 24 × 1000 × 86400 + 22 × 10 × 86400 + 79 × 1 × 86400 ≈ 2.096 billion records / day, meaning the number of raw sampling points is approximately 2.1 billion records / day. This value is the theoretical benchmark calculated for each measuring point using a unified nominal reference sampling rate, used to characterize the scale of the raw data. In actual operation, the adaptive frequency conversion sampling mechanism may cause the instantaneous sampling rate of some measuring points to be higher than this benchmark value during periods of active power generation (up to 10 kHz). Therefore, the actual number of raw sampling points fluctuates somewhat compared to the theoretical benchmark value. However, this fluctuation does not change the validity of the background technical conclusion that "the amount of raw data far exceeds the effective storage capacity, and a large amount of invalid data exists." After two stages of processing—thinning (lossy downsampling) and lightweight lossless compression—the system can store approximately 100 million records per day, with a cumulative data volume exceeding 800 GB, and this volume continues to grow as the project progresses.

[0004] Due to the intermittent nature of wave energy, a large amount of monitoring data is generated during periods when the generating units are not generating electricity. Performing full synchronization of all parameters would result in the transmission of a large amount of invalid data, wasted storage resources, and synchronization task timeouts. Traditional manual export methods (USB flash drive / email) suffer from serious efficiency drawbacks such as version inconsistencies, difficulty in tracing data, and time consumption of 2-4 hours. Therefore, there is an urgent need for an on-demand incremental data synchronization method that can intelligently determine the synchronization strategy based on the generating conditions.

[0005] Description of the adaptive frequency conversion sampling mechanism: The monitoring system of this project implements an adaptive frequency conversion sampling function at the data acquisition software layer. Its control logic is as follows: The monitoring system has a built-in sampling control module that reads the power generation data of each unit in real time. When the power exceeds 5% of the rated power, it is determined to be "active power generation." The system automatically increases the sampling rate of key dynamic parameters such as vibration, voltage, and current to the kHz level (maximum 10 kHz) to capture transient characteristics during power generation. When the power remains below this threshold, it is determined to be "shutdown or low power generation." The system automatically reduces the sampling rate of the above parameters to 1 Hz or lower to reduce invalid data acquisition and storage energy consumption. This function is implemented through a configured sampling control table, with a sampling rate switching response delay of no more than 1 second. The existence of this mechanism allows the amount of effective data to change dynamically with the power generation intensity, providing a technical prerequisite for subsequent dynamic segmentation strategies for operating condition perception.

[0006] Current national standards have made relevant provisions for the monitoring and testing of wave energy conversion devices. GB / Z 37551.3-2024 "Ocean Energy, Wave Energy, Tidal Energy and Other Water Flow Energy Conversion Devices Part 3: Mechanical Load Measurement" specifies the requirements for data acquisition, calibration and verification of mechanical load measurement; GB / Z 42153-2022 "Test Procedure for Pre-prototype Wave Energy Conversion Devices" defines the processing method for raw test data; GB / T 41340.2-2022 "Technical Specification for Calculation of Power Generation of Marine Power Stations Part 2: Wave Energy" specifies the method for calculating power generation based on field observation data; GB / Z 40295-2021 "Evaluation of Power Generation Performance of Wave Energy Conversion Devices" provides the evaluation method for power generation performance. This invention, while meeting the monitoring requirements of the above standards, further solves the problem of low data synchronization efficiency. Summary of the Invention

[0007] This invention addresses the problems of large monitoring data volume, high proportion of invalid data, and low efficiency of traditional synchronization methods in the OWC wave energy power generation demonstration project. It proposes an on-demand incremental data synchronization method based on power generation condition judgment. This method uses power generation capacity as the business entry point, reduces invalid data synchronization through a condition gating mechanism, and combines incremental advancement, block capture, anomaly self-healing retries, and synchronization log recording techniques to achieve efficient, stable, and traceable synchronization of massive monitoring data.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A method for incremental synchronization of wave energy monitoring data based on power generation condition assessment, comprising the following: Step S1: Construct and train the operating condition coupled gated cyclic unit prediction model, input historical monitoring data into the prediction model, and output the probability of power generation trend within a preset time period in the future, which serves as the scheduling basis for the synchronous inspection task and the configuration parameters for the block and retry strategy of the synchronous execution stage. Step S2: Dynamically adjust the triggering cycle of the synchronization check task based on the probability of power generation trend. When the probability of power generation trend is higher than the preset scheduling threshold, activate the synchronization preparation process before the synchronization check task is started. After the synchronization check task is triggered and started according to the adjusted cycle, determine the start and end times of this synchronization based on the incremental advancement rule and generate a synchronization time window. Step S3: After the synchronization check task is started, within the synchronization time window, the incremental synchronization of the generator set power generation data is performed first to obtain the power time series within the synchronization time window as the input basis for subsequent operating condition gating judgment. Step S4: Using the power time series as the sole criterion, perform operating condition gating judgment, detect whether there is an effective power generation condition within the synchronization time window where the power generation exceeds the preset effective threshold, and output the operating condition judgment result to control the synchronous execution of subsequent non-power monitoring parameters. Step S5: Based on the operating condition determination result, when a valid power generation operating condition is determined, the remaining non-power monitoring parameters are synchronized in a preset category order; among them, for high-data-volume monitoring parameters sampled at high frequencies, the block granularity is dynamically adjusted based on the probability of power generation trend, and data capture and synchronization are performed block by block.

[0009] Preferably, the method further includes the following steps: Step S6: During the synchronous execution of the non-power monitoring parameters, if a data transmission anomaly is detected, the initial backoff duration is dynamically determined based on the power generation trend probability. The actual backoff duration is extended with each retry based on the initial backoff duration. Starting from the second retry, the block granularity is gradually reduced, and self-healing retry is performed to restore data synchronization. Step S7: After the parameter synchronization process of this synchronization check task is completed, record the full process status information of this synchronization, generate a synchronization log, and add the measured monitoring data obtained in this synchronization to the training samples of the working condition coupled gated cyclic unit prediction model, and perform incremental model updates using a sliding window method.

[0010] Preferably, in step S1, the prediction model of the working condition coupled gated loop unit adds a working condition coupling gate to the traditional gated loop unit; the output of the working condition coupling gate is calculated according to the following formula: ; Where t is the index of the current time step; Input the feature vector for the current moment, which includes the power generation of multiple units and wave characteristic parameters; This is the hidden state from the previous moment; The operating condition gating characteristic is obtained by nonlinear transformation of the current power data and wave characteristics; This is the weight matrix of the coupling gate under operating conditions; This is the bias term for the operating condition coupling gate; This is a vector concatenation operation; This is the Sigmoid activation function.

[0011] Preferably, in step S2, determining the synchronization time window based on the incremental advancement rule specifically involves: Maintain a synchronization status record table to store the time when the last successful synchronization was completed. The start time of this synchronization is taken as the time of the last successful synchronization, and the end time of this synchronization is taken as the smaller value between the current system time, the start time, and the maximum time span of a single synchronization. This ensures that each synchronization only processes the data added after the last synchronization, avoiding the repeated retrieval of historical data.

[0012] Preferably, in step S3, the generator set power generation data is the only business entry point for the synchronization check task, and the synchronization process of all non-power monitoring parameters must not be started before the synchronization of power generation data; the synchronization of power generation data follows the incremental advancement rule, and only the power data added after the last successful synchronization is retrieved.

[0013] Preferably, in step S4, the operating condition gating judgment is based solely on the measured power generation data in the power time series: when the power generation of the generator set exceeds a preset effective threshold within the synchronization time window, it is determined that there is an effective power generation operating condition; otherwise, it is determined that there is no effective power generation operating condition. The judgment result of the operating condition gating has a higher priority than the prediction result of the power generation trend probability. The synchronous execution of non-power monitoring parameters is entirely based on the judgment result of the operating condition gating. If it is determined that there is no effective power generation operating condition, the synchronous execution of all non-power monitoring parameters is skipped.

[0014] Preferably, in step S5, when dynamically adjusting the block granularity based on the power generation trend probability, the adjusted block size of the high-frequency sampling high-data-volume monitoring parameters is calculated and determined according to the following formula: ; in, Let m be the adjusted block size for the m-th monitoring parameter; The base block size is calculated based on the current available network bandwidth, database load factor, single synchronization timeout threshold, and average size of a single record. This represents the probability of power generation trends. This is the probability influence coefficient, used to control the degree to which the predicted probability adjusts the block size; This is the maximum number of records that can be synchronized in a single instance. This is the smallest unit size.

[0015] Preferably, in step S6, when performing self-healing retries, starting from the second retries, the block granularity is gradually reduced according to the following rules: ; Where n is the retry number; For the first The block size used during the next retries; `max` represents the current block size, i.e., the block size being executed when the retry process begins; To perform the maximum value operation, ensure that the block size is at least 1.

[0016] Preferably, in step S6, the initial backoff duration is the backoff duration of the first retry, which is dynamically adjusted according to the power generation trend probability using a piecewise linear rule: the higher the power generation trend probability, the shorter the initial backoff duration, so as to ensure the timeliness of data synchronization under high power generation conditions; the lower the power generation trend probability, the longer the initial backoff duration, so as to reduce the resource occupation caused by invalid retries.

[0017] Preferably, in step S7, the incremental model update using a sliding window method specifically involves: After each synchronization is completed, the newly added measured monitoring data is added to the training set of the prediction model of the working condition coupled gated loop unit. When the number of new samples reaches a preset threshold, the model incremental fine-tuning is performed once. During the fine-tuning stage, a sliding window is used to retain the historical data of the most recent preset duration for training. The learning rate during the fine-tuning stage is lower than that during the initial training stage of the model. The initial training and incremental fine-tuning of the model are both completed on the background server. The field monitoring terminal only calls the prediction model to perform forward inference calculation.

[0018] The present invention provides a method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition assessment, which achieves several technical advantages: 1. Reduce invalid data transmission: Under specific field test conditions (30 consecutive days of operation), this invention, through a condition gating mechanism, skips the synchronization of all non-power parameters during periods without power generation, significantly reducing invalid data synchronization. During the test, non-power generation periods accounted for approximately 65%. After skipping these periods through condition gating, the average daily synchronization data volume decreased from approximately 2.8 GB for automatic full synchronization to approximately 0.8 GB, reducing invalid data transmission by approximately 72% (calculated based on: (2.8-0.8) / 2.8≈71.4%, taking approximately 72%). The reduction in data volume (approximately 72%) is higher than the percentage of time periods (approximately 65%). This is because during active power generation periods, key parameters such as vibration, voltage, and current are sampled at kHz levels due to the adaptive frequency conversion sampling mechanism, resulting in significantly higher data volume per unit time compared to shutdown periods (1 Hz or lower). Therefore, although power generation periods only account for about 35% of the test period, their corresponding data weight is higher. Conversely, while non-power generation periods account for 65% of the test period, the sampling rate drops to 1 Hz, resulting in very low data output per unit time. The combined effect of these two factors causes the reduction in data volume (72%) to be higher than the percentage of time periods (65%). The consistent direction and magnitude of these two differences align with the technical logic of the adaptive frequency conversion sampling mechanism. This effect is primarily achieved through power priority synchronization and operating condition gating.

[0019] 2. Improved Synchronization Efficiency: The power-priority strategy reduces the core judgment step of the synchronization process from the traditional manual extraction method (2-4 hours) to automated completion within minutes (12-18 minutes), effectively freeing up maintenance manpower. In a comparison with "automated full synchronization (without using operating condition gating and incremental strategies)," the on-demand incremental synchronization of this invention further reduces network transmission and processing time caused by fetching invalid data in full.

[0020] 3. Enhanced System Stability: The condition-aware dynamic chunking and adaptive retry mechanism effectively handles network fluctuations and database load changes, improving synchronization success rate. Among these, the chunk size... Retry initial backoff time The probabilities are all predicted by the same WCU-GRU. Coupling adjustment (especially in weak network scenarios), Changes can be directly adjusted (Size), so that the dispatch strategy can have a consistent response under different power generation conditions. Attached Figure Description

[0021] Figure 1 The flowchart shows steps S1-S5 of the wave energy monitoring data incremental synchronization method based on power generation condition judgment of the present invention.

[0022] Figure 2This is a schematic diagram of the internal structure of a single time step of the Working Condition Coupled Gated Cyclic Unit (WCU-GRU) of the present invention.

[0023] Figure 3 This is a complete logic flowchart of the working condition gate control judgment of the present invention.

[0024] Figure 4 This is a schematic diagram of the core process of the working condition perception dynamic segmentation and capture strategy of the present invention.

[0025] Figure 5 The flowchart shows steps S1-S7 of the wave energy monitoring data incremental synchronization method based on power generation condition judgment of the present invention.

[0026] Figure 6 This is a schematic diagram illustrating the complete workflow of the method of the present invention.

[0027] Figure 7 This is a schematic diagram of the complete process of adaptive exception self-healing retry of the present invention. Detailed Implementation

[0028] The following detailed implementation of a wave energy monitoring data incremental synchronization method based on power generation condition judgment according to the present invention will be described in detail with reference to specific embodiments. The embodiments of the present invention are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0029] The method of this invention includes the following key designs: ① Power Priority Synchronization: The power generation data of each unit (50 kW / 100 kW) is used as the sole business entry point. Each synchronization task first and only synchronizes the power field to provide a basis for subsequent operating condition judgment.

[0030] ② Operating condition gating: Based on the power value, determine whether there is a valid power generation condition in the current time window. If no power generation is achieved, skip the synchronization of all other parameters to reduce invalid data transfer.

[0031] ③ Incremental progress: Each synchronization continues from the last successfully completed data point, processing only newly added data and avoiding repeated fetching of historical data.

[0032] ④ Chunky fetching: High data volume parameters are split into several small blocks according to time windows and executed in batches, reducing the impact of a single synchronization on the network and database.

[0033] ⑤ Self-healing retry for anomalies: When synchronization anomalies occur, the synchronization block granularity is automatically reduced and retryed to improve the synchronization success rate in complex network scenarios.

[0034] ⑥ Synchronization Log Recording: Synchronization status, results, and exception information are recorded in a unified manner, supporting full-process traceability and problem localization.

[0035] Example 1: Implementation of a method for incremental synchronization of wave energy monitoring data on demand based on power generation condition judgment.

[0036] Combined with appendix Figures 1-4 As shown, the present invention provides a method for incremental synchronization of wave energy monitoring data on demand based on power generation condition judgment.

[0037] like Figure 1 As shown, Figure 1 This is a flowchart of steps S1-S5 of the wave energy monitoring data incremental synchronization method based on power generation condition judgment in Embodiment 1 of the present invention.

[0038] Step S1, Synchronous Window Adaptive Prediction – Power Generation Trend Prediction Based on Operating Condition Coupled Gated Cyclic Unit.

[0039] Step S1 involves constructing and training a prediction model for a condition-coupled gated cyclic unit. Historical monitoring data is input into the prediction model, which outputs the probability of power generation trends within a preset future timeframe. This probability serves as the scheduling basis for the synchronization check task and the configuration parameters for the block-based and retry strategies during the synchronization execution phase. The specific process is as follows: like Figure 2 As shown, Figure 2 This is a schematic diagram of the internal structure of a single time step of a Working Condition Coupled Gated Cyclic Unit (WCU-GRU), including the input layer (current time feature). The hidden state of the previous moment Operating condition gating characteristics The data flow and connection relationships of the working condition coupling gate (WCG), reset gate, candidate hidden state calculation module, hidden state update module, and output layer (fully connected and Sigmoid).

[0040] To overcome the rigidity of traditional fixed time windows (such as synchronizing data from the previous 24 hours at 0:00 every day), this invention proposes a Working-Condition Coupled Gated Recurrent Unit (WCU-GRU) to adaptively predict the power generation conditions of the next synchronization window and realize predictive scheduling of synchronization tasks.

[0041] (1) Unit structure design of WCU-GRU.

[0042] Traditional gated recurrent units (GRUs) consist of two gated structures: a reset gate and a refresh gate. This invention introduces a working-condition coupled gate (WCG) into the GRU, enabling the gated mechanism to sense power generation trend characteristics.

[0043] Traditional solutions typically feed operating condition features (such as power and wave height) directly into the GRU as ordinary input features, and the gating decision relies solely on a linear combination of the hidden state and the input features. This invention introduces a Operating Condition Coupling Gate (WCG), which adds operating condition gating features to the gating weight calculation. The independent nonlinear projection of the model allows for the explicit differentiation between "condition-related features" and "temporal hidden states" in the gated decision path, enabling the model to dynamically adjust the information retention and update ratio according to the intensity of power generation conditions.

[0044] Because wave energy generation is highly intermittent and sea state dependent, traditional GRU models struggle to distinguish between "no-generation conditions" and "normal time-series fluctuations," leading to a high misjudgment rate in synchronization window predictions. WCU-GRU, through a condition coupling gating system, directly embeds generation condition features into the gating decision, causing the model to retain historical trends during periods of no generation and update new information during periods of generation, thereby improving the accuracy of synchronization window predictions. This is of great significance for adaptive prediction of generation conditions and the predictive scheduling of synchronization tasks.

[0045] The WCU-GRU network structure consists of an input layer, a unidirectional WCU-GRU hidden layer, a fully connected layer, and a Sigmoid output layer. Input layer: Receives historical sequence data with a dimension of L×4 (L is the length of the input sequence, in this embodiment L=48 time steps, with each time step spaced 1 hour apart); WCU-GRU hidden layer: contains There are 1 hidden unit that processes the input sequence in ascending time order. Indicates the dimension of the hidden layer; Fully connected layer: maps the output of the hidden layer to 1 dimension; Sigmoid output layer: Compresses the output to the [0,1] interval, representing the probability of power generation trend.

[0046] Output of the operating condition coupling gate The calculation is as follows: ; The symbols are defined as follows: : Current time step index; The current input feature vector has a dimension of 4×1, and is derived from... constitute; The hidden state at the previous moment, with dimensions of ; : Working condition gating features, dimensions are It is obtained by nonlinear transformation of the power data and wave characteristics at the current moment; : Working condition coupling gate weight matrix, dimension is ; : Operating condition coupling gate bias term, dimension 32×1; The vector concatenation operation results in a dimension of 32 + 4 + 32 = 68. Sigmoid activation function This maps the input to the (0,1) interval.

[0047] Operating condition gating characteristics The calculation method is as follows: ; The symbols are defined as follows: The unit is Normalized power generation at any given time (scalar). , =50 kW; 100 kW unit Normalized power generation at any given time (scalar). , 100 kW; : The effective wave height (scalar) at a given time, in meters (m). : The period of the spectral peak at any given time (s). : Working condition gating feature mapping matrix, dimension is The 4-dimensional input features are mapped to a 32-dimensional hidden layer space of working conditions; : Operating condition gated feature bias term, with a dimension of 32×1; Hyperbolic tangent activation function Map the input to Interval.

[0048] Explanation regarding normalization: , That is, normalization is performed based on the rated power of each unit so that the normalized value falls within the range of [0,1] (it may be slightly greater than 1 when the over-generation is caused by abnormal operating conditions).

[0049] about Design motivation: Yes To be independent of The nonlinear projection of the path makes the splicing vector Preserving the original features Simultaneously, an additional representation of the condition, after dimension alignment and load-related nonlinear transformation, is introduced, and both are used in the calculation of the gating weights. This design enables the load features to have an independent expression path in the gating decision, rather than simply serving as ordinary input during hidden state updates.

[0050] The cell state update rules for WCU-GRU are as follows: ; The symbols are defined as follows: : Reset gate output, with a dimension of 32×1, to control the degree of forgetting of historical information; : Candidate hidden states, with a dimension of 32×1; The current hidden state, with a dimension of 32×1; : Reset the gate weight matrix, with dimensions of 32×68; : Candidate hidden state weight matrix, with dimensions of 32×68; : Reset the gate bias term, with a dimension of 32×1; : Candidate hidden state bias term, with a dimension of 32×1; Hadamard product (element-by-element multiplication); The output of the operating condition coupling gate simultaneously controls the retention of historical information and the writing of new information.

[0051] Operating condition coupling gate output value The closer it is to 1, the more the current operating condition is inclined towards the power generation state, and the higher the model's dependence on new information; The closer the value is to 0, the more likely the current operating condition is to be a state of no power generation, and the more historical trend information the model retains.

[0052] The main difference between WCU-GRU and traditional GRU is: Traditional GRU update gates are based only on the current input. and the hidden state of the previous moment calculate; The working condition coupling gate of this invention additionally introduces working condition gating features. This links the gating decision to the operating characteristics of wave energy power generation; This design enables the model to dynamically adjust the information retention and update ratio based on the intensity of power generation conditions.

[0053] (2) Synchronization window prediction based on WCU-GRU.

[0054] Training samples are constructed using historical monitoring data. Let the current time be t, and take historical data from the past L time steps as the input sequence: ; The feature vector at each time step Each symbol is defined in part (1) of step S1.

[0055] Training label construction: Define the true labels for training samples ,in, The prediction step size is K=6 hours in this embodiment. If, within the next k time steps, the actual power generation of either the 50 kW or 100 kW unit exceeds 5% of its rated power (i.e., 50 kW unit > 2.5 kW or 100 kW unit > 5 kW), then the tag is... ,otherwise The mathematical expression is: ; The symbols are defined as follows: : Indicator function, takes the value 1 when the condition in parentheses is true, otherwise takes the value 0; Future time step index, with a value range of: ; Effective power generation threshold (scalar), in this embodiment That is, a 50kW unit is 2.5 kW, and a 100kW unit is 5 kW; 50 kW unit Actual power generation at any given time (raw value, unnormalized); 100 kW unit Actual power generation at any given time (raw value, unnormalized).

[0056] Output layer: Input the trained WCU-GRU model, perform time-step calculations through the hidden layers, and then extract the hidden state at the last time step. (T=t represents the end of the sequence, with a dimension of 32×1), after mapping through a fully connected layer and a Sigmoid output layer, the probability of power generation trends in the next K time steps is output: ; The symbols are defined as follows: : The predicted probability of an effective power generation condition occurring within the next K time steps, with a value range of [0,1]; : The hidden state at the last time step of the input sequence, with a dimension of 32×1, T=t; Output layer weight matrix, with a dimension of 1×32; Output layer bias (scalar); : Sigmoid activation function.

[0057] Output layer parameters , With WCU-GRU main network , , , Each bias term is obtained through end-to-end joint training.

[0058] Loss function: The model training uses the binary cross-entropy loss function. ; The symbols are defined as follows: : Loss function value (scalar); N: Total number of training samples (scalar); i: Training sample index, ; : The true label of the i-th training sample, which takes the value 0 or 1; : The model prediction probability of the i-th training sample, with a value range of [0,1]; : Natural logarithm function.

[0059] The optimizer uses Adam, with an initial learning rate set to ,in, This represents the initial learning rate. The batch size is set to 64, and the training epochs are set to 100. An early stopping strategy is used during training; training is terminated when the validation set loss no longer decreases for 10 consecutive epochs to prevent overfitting.

[0060] The WCU-GRU hidden layer processes the input sequence step by step in ascending time sequence, and continuously fuses historical hidden states with current operating condition characteristics through operating condition coupling gates, thereby improving the accuracy of predicting future power generation trends.

[0061] The power generation trend probability output by the above model will serve as the unified input for synchronous task scheduling, block-level granularity configuration, and retry backoff calculation. The specific scheduling and window determination rules are as follows.

[0062] Step S2: Synchronization scheduling and incremental window determination.

[0063] Step S2 dynamically adjusts the triggering cycle of the synchronization check task based on the probability of power generation trend. When the probability of power generation trend is higher than the preset scheduling threshold, the synchronization preparation process is activated before the synchronization check task is started. After the synchronization check task is triggered and started according to the adjusted cycle, the start and end times of this synchronization are determined based on the incremental advancement rule, and a synchronization time window is generated.

[0064] In step S2, determining the synchronization time window based on the incremental advancement rule specifically involves: maintaining a synchronization status record table to store the time of the last successful synchronization; the start time of this synchronization is taken as the time of the last successful synchronization, and the end time of this synchronization is taken as the smaller value between the current system time, the start time, and the maximum time span of a single synchronization, ensuring that each synchronization only processes data added since the last synchronization, avoiding repeated fetching of historical data. The specific process is as follows: Based on the power generation trend probability obtained in step S1, the system first dynamically adjusts the triggering cycle of the synchronization check task to achieve predictive scheduling; after the task is officially started, the time window range of this synchronization is locked through the incremental advancement rule, and the two work together to complete the preparation for the start of the synchronization task.

[0065] The relationship between prediction results and scheduling decisions: when the prediction probability When the value is higher than the preset threshold θ (θ = 0.65 in this embodiment, where θ is the scheduling trigger threshold and its value ranges from 0.5 to 0.8), the system activates the synchronization preparation process in advance to complete network link detection and resource pre-allocation before the power generation window arrives; when the predicted probability is consistently lower than the threshold, the system automatically extends the synchronization check cycle to reduce the consumption of system resources by invalid polling.

[0066] It should be noted that: 1. The WCU-GRU prediction results are only used as a scheduling reference for starting the synchronization task, and are used to wake up the synchronization preparation process in advance. They do not replace the subsequent condition gating judgment based on actual power data. Whether to perform parameter synchronization in the end is still determined by the gating function based on the measured power value in step S4. As the standard.

[0067] The division of labor between the prediction mechanism and the incremental advancement mechanism: The prediction mechanism determines "when to initiate a synchronous check task" (i.e., controls the length of the check cycle, extending the check interval when the predicted probability is below a threshold); once the task is initiated, the synchronous window is strictly determined according to the incremental advancement rules. The two do not conflict with each other, and extending the check cycle does not affect... The value of .

[0068] To adapt to short-term changes in sea state, the model adopts a sliding window incremental update strategy. For details on the update rules and execution timing, please refer to step S7, synchronization log recording and model incremental update.

[0069] After determining the start time of the synchronization check task according to the above scheduling rules, when the task is officially triggered for execution, the system calculates the specific start and end times of this synchronization through the incremental advancement mechanism and generates a synchronization time window.

[0070] Incremental advancement mechanism: The system maintains a synchronization status record table. The core fields are shown in Table 1: Table 1

[0071] Each time a synchronization task starts, the system reads last_sync_position and uses it as the starting position for this synchronization: ; The symbols are defined as follows: : The start time of this synchronization window; The last time synchronization was successfully completed is read from the status log table; : The end time of this synchronization window; : Current system time; : The maximum time span (scalar) for a single synchronization, which is set to 24 hours in this embodiment.

[0072] Incremental processing ensures that only new data is processed each time, avoiding the repeated fetching of historical data.

[0073] The incremental advancement logic does not conflict with the prediction mechanism in step S1: the prediction mechanism controls the length of the inspection cycle (determining "when to start the inspection"), while the incremental advancement mechanism controls the start and end range of the synchronization window (determining "which data to synchronize after startup"). The two work independently.

[0074] Step S3: Power Priority Synchronization.

[0075] After the synchronization check task is started in step S3, incremental synchronization of generator power data is performed first within the synchronization time window to obtain the power time series within the synchronization time window, which serves as the input basis for subsequent operating condition gating judgment.

[0076] In step S3, the generator set power generation data is the sole business entry point for the synchronization check task. Synchronization processes for all non-power monitoring parameters must not precede the synchronization of the power generation data. The synchronization of the power generation data follows an incremental advancement rule, only retrieving power data added since the last successful synchronization. The specific process is as follows: Once the synchronization time window is determined, the system strictly follows the power-first execution principle to perform data synchronization, as follows: According to GB / T 41340.2-2022, the calculation of wave energy power generation must be based on power data observed on-site. Power generation is the core parameter for evaluating the power generation performance of the power station and is also the basic input for subsequent power generation calculations and performance evaluations. Therefore, this invention uses power generation data as the sole business entry point for the synchronization task, ensuring the complete synchronization of power data first.

[0077] Power-priority synchronization is a rigid prerequisite for the entire on-demand synchronization method, and has the highest execution priority. Synchronization processes for all non-power parameters are not allowed to start before power synchronization, ensuring that input data for operational condition judgment always arrives before business data, and avoiding the premature transmission of invalid data due to out-of-order parameter synchronization.

[0078] Each time a synchronization task starts, the system first performs synchronization of the power field. Let the current synchronization window be... ,in Indicates the start time of the synchronization window. This indicates the end time of the synchronization window. After power synchronization is completed, the power time series of the two units within this window is obtained: ; The symbols are defined as follows: : Power time series (vector) of a 50 kW unit within the synchronization window; : Power time series (vector) of a 100 kW unit within the synchronization window; 50 kW unit The power value at any given time (scalar); 100 kW unit The power value at any given time (scalar); Sampling time interval (scalar), in this embodiment Second.

[0079] The synchronization of the power field also follows an incremental advancement mechanism. The synchronization start time is taken from the last successful synchronization position in the synchronization status record table. Only the power data added after the last synchronization is completed is retrieved, avoiding repeated reading of historical power data and further reducing the time consumption of the synchronization pre-process.

[0080] Step S4: Working condition gate control judgment.

[0081] Step S4 uses the power time series as the sole criterion for judgment to perform operating condition gating judgment, detects whether there is a valid power generation condition within the synchronization time window where the power generation exceeds the preset effective threshold, and outputs the operating condition judgment result to control the synchronous execution of subsequent non-power monitoring parameters.

[0082] In step S4, the operating condition gating judgment uses the measured power generation data in the power time series as the sole criterion: when the power generation of the generator set exceeds a preset effective threshold within the synchronization time window, it is determined that there is an effective power generation operating condition; otherwise, it is determined that there is no effective power generation operating condition. The judgment result of the operating condition gating has higher priority than the prediction result of the power generation trend probability, and the synchronous execution of non-power monitoring parameters is entirely based on the judgment result of the operating condition gating. If it is determined that there is no effective power generation operating condition, the synchronous execution of all non-power monitoring parameters is skipped. The specific process is as follows: like Figure 3 As shown, Figure 3 The complete logic flowchart for the operating condition gating judgment is as follows: First, the power time series of the two units in the synchronization window is obtained. Then, the power value at each time point is scanned one by one to determine whether the power of any unit exceeds the effective power generation threshold. If it does, it is determined to be a power generation condition and enters the subsequent parameter synchronization. If all times are below the threshold, it is determined to be a no-power generation condition and all non-power parameter synchronization is skipped.

[0083] The operating condition gating system uses measured power data as the sole criterion for determining the start and stop of subsequent parameter synchronization processes. It is the core execution switch of the "on-demand" synchronization logic. The judgment result directly determines the data transmission scale of this synchronization task, which is the core feature that distinguishes this solution from traditional full-scale synchronization solutions.

[0084] The operating condition gating judgment function is defined as follows: ; The symbols are defined as follows: : Operating condition gating function, output 1 indicates that there is a valid power generation condition, output 0 indicates that there is no valid power generation condition; Effective power generation value (scalar), in this embodiment .

[0085] The threshold is determined based on the equipment sensor accuracy (±1% of rated power) and empirical estimation of no-load loss (approximately 1.5%~2% of rated power). Under shutdown conditions, the power reading generally does not exceed 3% of the rated power. Statistical analysis of power data from the past 90 days of shutdown periods shows that the power reading during most shutdown periods was approximately 3% lower than the rated power. To retain a safety margin and avoid false triggering of shutdown noise, this embodiment uses 5% of the rated power as the effective power generation threshold. If the gating function G=0, it indicates that there is no effective power generation within this time window. The system skips the synchronization of all other parameters and records this window as "No power generation skipped" in the log.

[0086] like This indicates the existence of a valid power generation condition, and the system enters the subsequent parameter synchronization process.

[0087] Operating condition gating is the final decision-making basis for the synchronization process, and its result has higher priority than the power generation trend prediction result of the WCU-GRU model. The prediction result is only used for adjusting the scheduling timing of the synchronization task and preparing the prerequisite resources, and does not directly skip parameter synchronization; whether to execute the synchronization of other parameters is entirely based on the measured power judgment result of this stage.

[0088] Step S5: Dynamic segmentation and capture strategy based on working condition perception.

[0089] Step S5, based on the operating condition determination result, when a valid power generation condition is determined, synchronizes the remaining non-power monitoring parameters according to a preset category order; among them, for high-data-volume monitoring parameters sampled at high frequencies, the block granularity is dynamically adjusted based on the probability of power generation trends, and data capture and synchronization are performed block by block. The specific process is as follows: like Figure 4 As shown, Figure 4The core process diagram of the dynamic block capture strategy for operating condition perception is as follows: First, the expected data volume is weighted and estimated using the power generation trend probability predicted by WCU-GRU. Then, the baseline block size is calculated based on constraints such as the estimated data volume, network bandwidth, and database load. Next, the predicted probability is used as a reverse adjustment factor to determine the final block size (smaller blocks for high probability and larger blocks for low probability). Finally, the actual number of blocks is recalculated and capture is performed block by block. Figure 4 China follows The order of operations.

[0090] For parameters with high data volume (such as vibration, voltage, current, etc. sampled at the kHz level), this invention proposes a dynamic block capture strategy for working condition perception.

[0091] Technical Prerequisites: As described in the background section, the monitoring system of this project implements an adaptive frequency conversion sampling function at the data acquisition software layer. This function reads real-time power data through a built-in sampling control module. When the power exceeds the rated power by 5%, the sampling rate of key parameters is automatically increased to the kHz level; when the power is below the threshold, it is reduced to 1Hz or lower, with a sampling rate switching response delay of no more than 1 second. The existence of this function allows the amount of effective data to change dynamically with the power generation intensity, providing a direct basis for the dynamic adjustment of the block strategy.

[0092] Prediction of operating condition sensing data volume: Power generation trend probability output by WCU-GRU model A weighted estimate is performed on the expected data volume. Here, m is the parameter index, representing the m-th monitored parameter. M represents the total number of monitored parameters (M=125 in this embodiment). The actual synchronization window duration is taken into account. The time may not be consistent with the predicted step size K, so a time ratio adjustment is introduced: ; The symbols are defined as follows: : Parameter index, ; :parameter Estimated data volume (number of records, scalar) within the current time window; The parameter m represents the baseline data volume (number of records, scalar) for the same historical period, obtained from statistics in the historical database. : Operating condition influence coefficient (dimensionless scalar), in this embodiment, we take Its value ranges from 0.1 to 0.5; : The probability of power generation trend predicted by WCU-GRU, with a value range of [0,1]; : The actual duration of the current synchronization window, in hours; K: Prediction step size, in this embodiment K=6 hours.

[0093] coefficient The selection criteria are as follows: Regression analysis of the historical data volume and power time series over the past 90 days revealed that the data volume increases approximately linearly with power, with a growth slope of approximately 0.3~0.5. Therefore, [the data volume was selected]. This is a value based on engineering experience. This coefficient can be adjusted according to the sea state characteristics of different stations.

[0094] Block Quantity Prediction: Introducing Parameters This represents the average data size (in bytes) per record for parameter m, determined by the data type and precision of this parameter. It is based on the estimated data volume. Average size of a single record Current network available bandwidth Database load factor and single synchronization timeout threshold Predict the number of blocks: ; The symbols are defined as follows: Predict the number of blocks (scalar); Parameter m: Average data size per record, in bytes per record; Current available network bandwidth, in bytes per second; Database load factor (dimensionless scalar), with a value range of (0,1], reflects the current load status of the database. The closer the value is to 1, the lighter the load (the faster the database response). Single synchronization timeout threshold, in seconds. In this embodiment... =120 seconds; : Floor down operator; : Maximum value operation.

[0095] Block size determination: Block size Directly using predicted probability As a reverse adjustment factor: ; in: ; The symbols are defined as follows: : The final block size of parameter m, in units of strips / block; The maximum number of records that can be synchronized in a single instance is limited in this embodiment. =100,000 records, with a value range of 50,000 to 200,000 records; Minimum block size, in this embodiment =1000 items; The baseline block size determined by network and database resource constraints, in units of stripes / block; : Probability influence coefficient (dimensionless scalar), in this embodiment, we take The value ranges from 0.3 to 0.7, controlling the degree to which the prediction probability adjusts the block size; , : Calculate the minimum and maximum values.

[0096] coefficient The selection criteria are: through different Experimental comparison of synchronization success rates under different conditions. The optimal value is achieved by using sufficiently small block granularity during high-probability periods to avoid timeouts, and sufficiently large block granularity during low-probability periods to reduce synchronization overhead.

[0097] about Explanation of values: In this embodiment, the data record format transmitted synchronously is consistent with the format of locally valid storage records (i.e., no additional structural transformation or secondary compression is performed before transmission). Therefore, the average size of a single record in the network transmission scenario can be approximately estimated by using "daily valid storage data volume / daily valid storage record count". Bytes / strip, as an example value for block-based computation.

[0098] Calculate according to the correct order of the formula: Baseline block: strip; This value far exceeds Upper limit. According to the formula. The order of operations should be calculated first. Then proceed and Clamp.

[0099] when hour, Article, still Therefore =100,000 entries; when hour, Article, still Therefore =100,000 entries.

[0100] In other words, under the given network conditions (10 Mbps), It is far beyond Upper limit Equal to =100,000 records The dynamic regulation effect is in this scenario The upper limit is dominant, but not significant. Under this set of parameters, It serves as the main constraint, representing the maximum safe limit on the number of records that can be synchronized in a single instance.

[0101] The value of 0.8 is an exemplary setting used to demonstrate the calculation process. The actual value is determined in real time according to the on-site database load.

[0102] The actual number of blocks is recalculated: Once confirmed, the system recalculates the actual number of blocks using the following formula: ; The symbols are defined as follows: : Actual number of blocks (scalar); : The round-up operator.

[0103] Original The formula is only a rough estimate of resource consumption during the preliminary stage of the block-based strategy and is not used as the final execution basis. In actual execution, [the formula will be used]. To ensure It covers the entire estimated data volume.

[0104] Example 2, as Figure 5-7 As shown, Figure 5 This is a flowchart of steps S1-S7 of the wave energy monitoring data incremental synchronization method based on power generation condition judgment in Embodiment 2 of the present invention. Figure 6 The diagram illustrates the complete workflow of the method of this invention, which includes three stages: the prediction and scheduling stage (the WCU-GRU model predicts the probability of power generation trends in the next time window and decides whether to trigger the synchronization check in advance), the synchronization execution stage (power priority synchronization → operating condition gating judgment → synchronization of other parameters as needed), and the exception handling and log recording stage (adaptive retry → result recording → incremental model update). The diagram also marks the division of labor between the prediction mechanism and the incremental advancement mechanism.

[0105] Based on the method described in Example 1, the method in Example 2 further includes the following steps: Step S6: Adaptive anomaly self-healing retry.

[0106] In step S6, if a data transmission anomaly is detected during the synchronization of the non-power monitoring parameters, the initial backoff duration is dynamically determined based on the power generation trend probability. The actual backoff duration is extended with each retry based on the initial backoff duration, and the block granularity is gradually reduced from the second retry onwards. Self-healing retry is performed to restore data synchronization.

[0107] In step S6, the initial backoff duration is the backoff duration for the first retry, and it is dynamically adjusted according to the power generation trend probability using a piecewise linear rule: the higher the power generation trend probability, the shorter the initial backoff duration, to ensure the timeliness of data synchronization under high power generation conditions; the lower the power generation trend probability, the longer the initial backoff duration, to reduce the resource consumption caused by invalid retries. The specific process is as follows: like Figure 7 As shown, Figure 7 The diagram illustrates the complete process of adaptive anomaly self-healing retry. After an anomaly is detected during synchronization, the initial backoff time is dynamically determined using a piecewise linear function based on the power generation trend probability predicted by WCU-GRU. Then, the process is executed step by step according to the retry sequence number. The first retry maintains the original block size and the backoff time is the base value. The second retry reduces the block size to 1 / 2 and doubles the backoff time. Subsequent retry continue to exponentially reduce the block size and double the backoff time until the maximum number of retry attempts is reached or the retry is successful. If all attempts fail, the process is marked as "pending manual processing".

[0108] If an exception occurs during synchronization (network timeout, database connection interruption, data verification failure, etc.), the system initiates an adaptive retry mechanism. Let the retry sequence number be... ,in Maximum number of retries: (1) First retry (n=1): After waiting for the initial backoff time, re-execute with the original block size. Initial backoff time Based on the power generation trend probability predicted by WCU-GRU Dynamic adjustment: ; That is, high probability periods ( )use Synchronization is restored in seconds; low probability periods ( )use Seconds to reduce the network resource consumption of invalid retries; intermediate probability Linear interpolation is used. This piecewise linear function is based on the two anchor values ​​and the scheduling threshold mentioned above. The surrounding transitional needs are constructed.

[0109] First retry retreat time for: ; The symbols are defined as follows: Initial retreat time, which is about A piecewise linear function.

[0110] (2) Second retry and thereafter (n≥2): The granularity of the blocks is reduced step by step from the second retry according to the following rules: ; The symbols are defined as follows: : No. The block size used during each retry, in units of stripes / block; Retry number ; : Maximum number of retries, in this embodiment ; : Current block size, i.e., the size of the block being executed when entering the retry process, in units of segments / block; : Take the maximum value operation to ensure that the block size is at least 1.

[0111] That is, the second retry reduces the size to 1 / 2 of the original block, the third retry reduces it to 1 / 4, and so on.

[0112] Retreat time increases exponentially: ; The symbols are defined as follows: : No. The backoff time for each retry, in seconds.

[0113] Maximum number of retries In this embodiment, the maximum number of retries is set to 5. If the retries still fail after reaching the maximum number, the segment is marked as "synchronization failure pending manual handling," and the complete error stack is recorded to the log table.

[0114] It should be noted that, and Although they apply to two different aspects, namely the block size and the retry strategy, they are not set independently: This determines the data throughput of a single synchronization task, while the initial backoff time for retrying... This determines the retry schedule after synchronization failure, and both factors jointly affect the total resource consumption of the same synchronization task under the same predicted conditions. When network bandwidth is low, fall into Within the interval, Changes can be directly adjusted Size, at this point, high prediction probability ( Larger blocks → Smaller blocks Reduced) and shorter retreat ( The combined strategy of reducing (or decreasing) resources per unit time controls the peak resource level while ensuring timeliness; when the prediction probability is low, the combination of "larger blocks and longer backoff" reduces scheduling overhead. This method of coordinating the configuration of multiple scheduling parameters based on the same prediction output is different from the traditional approach where the block strategy and retry strategy are set according to independent rules (such as fixed block thresholds and fixed backoff bases) and are not related to each other. This reflects the unified control effect of the prediction results on the overall resource profile of the synchronization task.

[0115] Step S7: Synchronize log recording and model closed-loop update.

[0116] After the parameter synchronization process of this synchronization check task is completed in step S7, record the full process status information of this synchronization, generate a synchronization log, and add the measured monitoring data obtained in this synchronization to the training samples of the working condition coupled gated cyclic unit prediction model, and perform incremental model updates using a sliding window method.

[0117] In step S7, the incremental model update using a sliding window method specifically involves: after each synchronization, the newly added measured monitoring data is added to the training set of the condition-coupled gated recurrent unit prediction model. When the number of new samples reaches a preset threshold, incremental model fine-tuning is performed. During the fine-tuning phase, a sliding window is used to retain historical data for the most recent preset duration for training. The learning rate during the fine-tuning phase is lower than that during the initial model training phase. Both the initial training and incremental fine-tuning of the model are completed on the background server, and the on-site monitoring terminal only calls the prediction model to perform forward inference calculations. The specific process is as follows: The system establishes a unified logging standard, and the following information for each synchronization operation is written to the log table: Task ID (unique identifier); Synchronize start and end times; Synchronization window range ( ); The probability of power generation trends predicted by WCU-GRU ; Operating condition gating judgment result (G=0 or G=1); The number of records and the amount of data for each parameter are synchronized. Number of retries and detailed results of each retrieval; Exception type and error stack; Final synchronization status (success / skipped / failed).

[0118] After completing the synchronization log recording, the system uses the measured data acquired in this synchronization for incremental iteration of the WCU-GRU model to continuously optimize the prediction accuracy. The specific strategy is as follows: Model Incremental Update: The WCU-GRU model employs a sliding window incremental update strategy. After each synchronization, newly added measured data is added to the training set. When the number of new samples reaches a preset threshold (e.g., 1000), incremental fine-tuning is performed; otherwise, accumulation continues. Assuming daily synchronization with 24 new time-step samples added each time, it takes approximately 42 days to reach the 1000-sample fine-tuning threshold, and incremental fine-tuning is performed approximately every 42 days (about six weeks). During the incremental fine-tuning phase, a sliding window is used to retain the most recent data. Training is performed using daily data, among which This represents the number of days in the sliding window, allowing the model to continuously adapt to recent short-term changes in sea conditions and avoiding excessive reliance on historical data. The learning rate during the fine-tuning phase is set to 1 / 10 of the initial training rate (i.e.,...). ).

[0119] The incremental fine-tuning of the model is completed on the backend server / cloud. The on-site monitoring terminal only performs forward inference calculations and does not undertake training tasks, so as to meet the lightweight deployment requirements of the on-site embedded terminal.

[0120] Example 3: Complete application example of OWC wave energy power generation demonstration project.

[0121] Based on the on-demand incremental synchronization main process described in Example 1 and the adaptive abnormal retry and model closed-loop update mechanism described in Example 2, this embodiment provides a detailed explanation of the complete execution process and operational effects of the method in conjunction with the actual deployment scenario of an OWC wave energy power generation demonstration project.

[0122] Taking an OWC wave energy power generation demonstration project as an example, one 50 kW and one 100 kW generator set were configured on-site. Monitoring parameters covered 125 measurement points (24 at the kHz level, 22 at the 10 Hz level, and 79 at the 1 Hz level), with sampling frequencies covering three levels: 1 Hz, 10 Hz, and kHz. The experimental environment was as follows: the on-site monitoring terminal was deployed in the OWC demonstration project control cabinet, connected to a cloud synchronization server via a 4G / 5G network, with an average network bandwidth of approximately 10 Mbps (approximately 1.25 MB / s). (Bytes / second); the database uses MySQL 8.0, and the server is configured with a 16-core CPU and 64 GB of memory. The monitoring system strictly follows the requirements of GB / Z42153-2022 for the processing methods of raw test data, and the selection of all sensors and data acquisition software conforms to the selection recommendations provided in this standard.

[0123] Step 1: Model initialization and training.

[0124] Historical monitoring data (including power, wave height, period, etc.) from the past 90 days was collected to construct the training set for the WCU-GRU model. With an input window of L=48 hours and a prediction step size of K=6 hours, approximately 90×24-48-6≈2100 training samples can be generated. The model input sequence length is L=48 (i.e., data from the past 48 hours, one time step per hour), the prediction step size is K=6 (predicting whether power will be generated within the next 6 hours), and the number of hidden layer neurons is... The Adam optimizer (learning rate) is used. The training phase (batch size 64, epoch 100) continues until the validation set loss converges. This initial training phase is only used to capture the basic patterns of power generation conditions; adaptation to seasonal sea state changes depends on continuous accumulation through subsequent long-term incremental fine-tuning.

[0125] Step 2: Synchronize window adaptive prediction.

[0126] The system's backend server performs synchronization window prediction checks according to a preset fixed time schedule (e.g., every 6 hours at 4 AM, 10 AM, 4 PM, and 10 PM daily). This baseline period is a configurable parameter. The WCU-GRU model is invoked, taking the feature data from the past 48 hours as input, and outputting the probability of power generation trends for the next 6 hours. .like (θ=0.65), the system will start the synchronization preparation process in advance at the next check time; if The system maintains the current inspection cycle and does not trigger synchronization prematurely. After the synchronization task is triggered, the start and end time windows for this synchronization are determined according to the incremental advancement rules.

[0127] Step 3: Power-priority synchronization.

[0128] After the synchronization task is started, the power data of the 50 kW and 100 kW units are synchronized first, and the synchronization range is from the last successful synchronization time to the current time.

[0129] Step 4: Working condition gating judgment.

[0130] A threshold judgment is performed on the power data obtained during synchronization. If any unit has a power value exceeding 5% of the rated power within the synchronization window (i.e., 50 kW unit > 2.5 kW or 100 kW unit > 5 kW), then the gating function G=1, and proceed to step 5; otherwise, G=0, skip all non-power parameter synchronization, record the log, and end the current task.

[0131] Step 5: Parameter sequence synchronization and working condition perception block.

[0132] The system executes synchronously in sequence according to parameters such as liquid level, air pressure, wind speed, torque, rotational speed, vibration, current, voltage, temperature, displacement, strain, and wave pressure, fully covering five major categories of parameters: hydrometeorology, OWC hydrodynamic response, turbine and power generation monitoring, electrical parameters, and structural safety monitoring.

[0133] For high sampling rate parameters in the kHz range (such as vibration and voltage), the power generation trend probability is based on the output of the WCU-GRU. According to the formula After dynamically determining the block size, capture the blocks one by one, and then... Execute after recalculating the actual number of blocks. Under the specific network conditions (10Mbps), Far exceeding Upper limit Equal to =100,000 records The dynamic regulation effect is in this scenario Upper limit dominant, by This serves as a primary constraint, setting a safety limit on the maximum number of records that can be synchronized in a single instance. In weak network scenarios, The dynamic adjustment effect will be manifested.

[0134] Step 6: Exception handling and logging.

[0135] If an exception is encountered during synchronization, it will be handled according to the adaptive retry mechanism in step 6 of Method Implementation Example 2 (see step 6 of Method Implementation Example 2 for the correspondence between the retry number n and the block reduction). All operation results (success / failure / skipped) and detailed status information are written to the log table.

[0136] Step 7: Incremental model update.

[0137] After each synchronization, the newly acquired measured data is added to the training set of the WCU-GRU model. Incremental fine-tuning is performed on the background server when the number of new samples reaches 1000. Based on daily synchronization and adding 24 time-step samples each time, it takes approximately 42 days for the new samples to reach the fine-tuning threshold, and incremental fine-tuning is performed approximately every 42 days (about six weeks). During the fine-tuning phase, a sliding window is used to retain the most recent data. Tian data, learning rate set to This allows the model to continuously adapt to recent short-term changes in sea conditions. The on-site monitoring terminal only performs forward inference and does not undertake training tasks.

[0138] Actual measured results of this embodiment: The following table shows the comparative data between the method of this invention and the traditional full-scale synchronization method during 30 consecutive days of on-site operation testing in the demonstration project: Table 2

[0139] Experimental environment description: The on-site monitoring terminal is deployed in the control cabinet of the OWC demonstration project and is connected to the cloud synchronization server via a 4G / 5G network, with an average network bandwidth of approximately 10 Mbps; the database uses MySQL 8.0, and the server is configured with a 16-core CPU and 64 GB of memory.

[0140] Comparison method description: Traditional full synchronization (manual export): Maintenance personnel export all 125 parameter data from the on-site database daily via USB drive and transmit them to the cloud server via email, which takes an average of 2-4 hours.

[0141] Automated full synchronization (without adopting the strategy of this invention): A scheduled task script is used to automatically pull all parameter data at 0:00 every day, without adopting the working condition gating and incremental strategies.

[0142] Explanation of statistical criteria: "Synchronization timeout" is defined as a single synchronization task taking more than 30 minutes; The “retry success recovery rate” is defined as the proportion of synchronization anomalies that occur within 30 days that are successfully recovered through the adaptive retry mechanism.

[0143] Comparison and explanation: Automated full synchronization (intermediate control) eliminates the time-consuming manual intervention compared to manual export, but still requires the transmission of all data (including invalid data), so it takes 25-35 minutes.

[0144] This invention, based on on-demand incremental synchronization, skips invalid parameter synchronization during periods without power generation by using operating condition gating. This reduces the average daily synchronization data volume by approximately 72% (from 2.8 GB to 0.8 GB) and further shortens the time to 12-18 minutes. This gain is directly attributed to the core technical features of this invention, such as operating condition gating, incremental advancement, and dynamic segmentation. In this process, the WCU-GRU primarily serves the predictive scheduling of the synchronization window (parameter optimization of segmentation and retry strategies) rather than directly participating in invalid data removal.

[0145] The statistical basis for the 92% retry success rate is that 25 synchronization anomalies occurred during the 30-day test period, of which 23 were successfully recovered through the adaptive retry mechanism.

[0146] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0147] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for incremental synchronization of wave energy monitoring data based on power generation condition judgment, characterized in that, Includes the following: Step S1: Construct and train the operating condition coupled gated cyclic unit prediction model, input historical monitoring data into the prediction model, and output the probability of power generation trend within a preset time period in the future, which serves as the scheduling basis for the synchronous inspection task and the configuration parameters for the block and retry strategy of the synchronous execution stage. Step S2: Dynamically adjust the triggering cycle of the synchronization check task based on the probability of power generation trend. When the probability of power generation trend is higher than the preset scheduling threshold, activate the synchronization preparation process before the synchronization check task is started. After the synchronization check task is triggered and started according to the adjusted cycle, determine the start and end times of this synchronization based on the incremental advancement rule and generate a synchronization time window. Step S3: After the synchronization check task is started, within the synchronization time window, the incremental synchronization of the generator set power generation data is performed first to obtain the power time series within the synchronization time window as the input basis for subsequent operating condition gating judgment. Step S4: Using the power time series as the sole criterion, perform operating condition gating judgment, detect whether there is an effective power generation condition within the synchronization time window where the power generation exceeds the preset effective threshold, and output the operating condition judgment result to control the synchronous execution of subsequent non-power monitoring parameters. Step S5: Based on the operating condition determination result, when a valid power generation operating condition is determined, the remaining non-power monitoring parameters are synchronized in a preset category order; among them, for high-data-volume monitoring parameters sampled at high frequencies, the block granularity is dynamically adjusted based on the probability of power generation trend, and data capture and synchronization are performed block by block.

2. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment as described in claim 1, characterized in that, The method further includes the following steps: Step S6: During the synchronous execution of the non-power monitoring parameters, if a data transmission anomaly is detected, the initial backoff duration is dynamically determined based on the power generation trend probability. The actual backoff duration is extended with each retry based on the initial backoff duration. Starting from the second retry, the block granularity is gradually reduced, and self-healing retry is performed to restore data synchronization. Step S7: After the parameter synchronization process of this synchronization check task is completed, record the full process status information of this synchronization, generate a synchronization log, and add the measured monitoring data obtained in this synchronization to the training samples of the working condition coupled gated cyclic unit prediction model, and perform incremental model updates using a sliding window method.

3. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment according to claim 1, characterized in that, In step S1, the prediction model of the working condition coupled gated loop unit adds a working condition coupling gate to the traditional gated loop unit; the output of the working condition coupling gate is calculated according to the following formula: ； Where t is the index of the current time step; Input the feature vector for the current moment, which includes the power generation of multiple units and wave characteristic parameters; This is the hidden state from the previous moment; The operating condition gating characteristic is obtained by nonlinear transformation of the current power data and wave characteristics; This is the weight matrix of the coupling gate under operating conditions; This is the bias term for the operating condition coupling gate; This is a vector concatenation operation; This is the Sigmoid activation function.

4. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment as described in claim 1, characterized in that, In step S2, determining the synchronization time window based on the incremental advancement rule specifically involves: Maintain a synchronization status record table to store the time when the last successful synchronization was completed. The start time of this synchronization is taken as the time of the last successful synchronization, and the end time of this synchronization is taken as the smaller value between the current system time, the start time, and the maximum time span of a single synchronization. This ensures that each synchronization only processes the data added after the last synchronization, avoiding the repeated retrieval of historical data.

5. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment according to claim 1, characterized in that, In step S3, the generator set power generation data is the only business entry point for the synchronization check task. The synchronization process of all non-power monitoring parameters must not be started before the synchronization of power generation data. The synchronization of power generation data follows the incremental advancement rule and only pulls the power data added after the last successful synchronization.

6. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment according to claim 1, characterized in that, In step S4, the operating condition gating judgment is based solely on the measured power generation data in the power time series: when the power generation of the generator set exceeds the preset effective threshold within the synchronization time window, it is determined that there is an effective power generation operating condition; otherwise, it is determined that there is no effective power generation operating condition. The judgment result of the operating condition gating has a higher priority than the prediction result of the power generation trend probability. The synchronous execution of non-power monitoring parameters is entirely based on the judgment result of the operating condition gating. If it is determined that there is no effective power generation operating condition, the synchronous execution of all non-power monitoring parameters is skipped.

7. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment according to claim 1, characterized in that, In step S5, when dynamically adjusting the block granularity based on the power generation trend probability, the adjusted block size of the high-frequency sampling high-data-volume monitoring parameters is calculated and determined according to the following formula: ； in, Let m be the adjusted block size for the m-th monitoring parameter; The base block size is calculated based on the current available network bandwidth, database load factor, single synchronization timeout threshold, and average size of a single record. This represents the probability of power generation trends. This is the probability influence coefficient, used to control the degree to which the prediction probability adjusts the block size; This is the maximum number of records that can be synchronized in a single instance. This is the smallest unit size.

8. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment according to claim 2, characterized in that, In step S6, when performing self-healing retries, starting from the second retry, the block granularity is reduced step by step according to the following rules: ； Where n is the retry number; This refers to the block size used during the nth retry; `max` represents the current block size, i.e., the block size being executed when the retry process begins; To perform the maximum value operation, ensure that the block size is at least 1.

9. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment according to claim 2, characterized in that, In step S6, the initial backoff duration is the backoff duration of the first retry, which is dynamically adjusted according to the power generation trend probability using a piecewise linear rule: the higher the power generation trend probability, the shorter the initial backoff duration, so as to ensure the timeliness of data synchronization under high power generation conditions; the lower the power generation trend probability, the longer the initial backoff duration, so as to reduce the resource occupation of invalid retries.

10. The method for on-demand incremental synchronization of wave energy monitoring data based on power generation condition judgment according to claim 2, characterized in that, In step S7, the incremental model update using the sliding window method specifically involves: After each synchronization is completed, the newly added measured monitoring data is added to the training set of the prediction model of the working condition coupled gated loop unit. When the number of new samples reaches a preset threshold, a model incremental fine-tuning is performed. During the fine-tuning phase, a sliding window is used to retain historical data for the most recent preset duration for training. The learning rate during the fine-tuning phase is lower than that during the initial training phase of the model. Both the initial training and incremental fine-tuning of the model are completed on the background server. The on-site monitoring terminal only calls the prediction model to perform forward inference calculations.

Images