Method, device and equipment for predicting life of key components of offshore wind power equipment and medium

Abstract

This invention relates to the field of wind power equipment life prediction technology, and discloses a method, device, equipment, and medium for predicting the life of key components of offshore wind power equipment. The invention determines the degree of environmental impact based on weather parameter data and divides the expected lifespan impact scenarios into scenarios with and without expected lifespan impact scenarios. It also expands the data for the expected lifespan impact scenarios, effectively solving the problems of difficult data acquisition and scarce fault samples in offshore wind power equipment monitoring. By constructing an adaptive combined prediction model composed of a Long Short-Term Memory (LSTM) network and a Prophet prediction model, and introducing an adaptive weighting mechanism to dynamically adjust the contributions of the two models, the invention fully leverages the advantages of LSTM in nonlinear time series modeling and Prophet in trend capture, achieving complementary model advantages and significantly improving the prediction accuracy of the remaining lifespan of key components.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment life prediction technology, specifically to methods, devices, equipment, and media for predicting the life of key components of offshore wind power equipment. Background Technology

[0002] Offshore wind turbines, especially critical components such as blades and blade bolts, are subjected to harsh marine environments including high salt spray, high humidity, strong wind loads, and wave impact. The combined effects of these environmental factors can easily lead to cumulative damage such as corrosion and fatigue in critical components, thereby affecting the safe operation and economic benefits of the entire wind farm. Therefore, accurately predicting the remaining lifespan of critical components of offshore wind turbines is of paramount importance for implementing preventative maintenance and reducing operation and maintenance costs.

[0003] Currently, data-driven methods for equipment life prediction mostly employ single statistical or machine learning models. These methods typically rely on single types of monitoring data (such as vibration data or temperature data only), making it difficult to comprehensively integrate multi-source environmental information such as salt spray concentration and humidity with mechanical damage data. This results in prediction models failing to accurately reflect the combined impact of complex marine environments on component degradation. Furthermore, offshore wind power equipment has a relatively short operational lifespan and a sparse number of failure samples. Existing models often exhibit overfitting and insufficient generalization ability when processing such small-sample time-series data. In the face of sudden operating conditions such as extreme weather, single models have poor adaptability, and their prediction accuracy is insufficient to meet the needs of engineering applications. Summary of the Invention

[0004] This invention provides a method, device, equipment, and medium for predicting the lifespan of key components of offshore wind power equipment, in order to solve the problems of low accuracy and poor model adaptability in predicting the lifespan of key components of offshore wind power in complex marine environments due to the difficulty in effectively integrating multi-source environmental data and mechanical damage data and the lack of processing mechanisms for small sample time series.

[0005] In a first aspect, the present invention provides a method for predicting the lifespan of key components of offshore wind power equipment, comprising: acquiring historical monitoring data of key components of offshore wind power equipment, the historical monitoring data including weather parameter data at each monitoring time, damage parameter data of key components, and corresponding remaining lifespan labels; determining the degree of environmental impact at each monitoring time based on the weather parameter data, and dividing the historical monitoring data into data under expected lifespan impact scenarios and data under no expected lifespan impact scenarios based on the degree of environmental impact; expanding the data under expected lifespan impact scenarios, and combining the expanded data under expected lifespan impact scenarios with the data under no expected lifespan impact scenarios to form a training dataset; training an adaptive combined prediction model using the training dataset to obtain a trained adaptive combined prediction model, wherein the adaptive combined prediction model is composed of a long short-term memory network and a Prophet prediction model, and dynamically adjusting the weights of the prediction results of the long short-term memory network and the prediction results of the Prophet prediction model in the final prediction result through an adaptive weight mechanism; inputting the monitoring data of the key component to be predicted into the trained adaptive combined prediction model, and outputting the remaining lifespan prediction result of the key component to be predicted.

[0006] This invention determines the degree of environmental impact based on weather parameter data and divides the expected lifespan impact scenario into scenarios with and without expected lifespan impact. It also expands the data for the expected lifespan impact scenario, effectively solving the problems of difficult data acquisition and scarce fault samples in offshore wind power equipment monitoring. By constructing an adaptive combined prediction model composed of a Long Short-Term Memory (LSTM) network and a Prophet prediction model, and introducing an adaptive weighting mechanism to dynamically adjust the contribution of the two models, it fully leverages the advantages of LSTM in nonlinear time series modeling and Prophet in trend capture, achieving complementary model advantages and significantly improving the prediction accuracy of the remaining lifespan of key components. This invention effectively solves the problems of low prediction accuracy and poor model adaptability of the lifespan of key offshore wind power components in complex marine environments due to the difficulty in effectively integrating multi-source environmental data and mechanical damage data, and the lack of processing mechanisms for small-sample time series. It significantly improves the accuracy of lifespan prediction and model adaptability.

[0007] In one optional implementation, weather parameter data includes at least one of salt spray concentration, humidity, temperature, wind speed, and wave height; damage parameter data includes at least one of stress level, vibration amplitude, corrosion rate, and fatigue damage accumulation. This implementation comprehensively collects multiple marine environmental parameters such as salt spray concentration, humidity, temperature, wind speed, and wave height, as well as multi-dimensional mechanical damage parameters such as stress level, vibration amplitude, corrosion rate, and fatigue damage accumulation, achieving comprehensive perception and monitoring of the operating status of key components of offshore wind power equipment. The fusion of multi-source environmental parameters can accurately characterize the comprehensive impact of the marine environment on components, while the collection of multi-dimensional damage parameters fully reflects the degradation process of components under complex operating conditions. This multi-level, multi-dimensional data collection strategy provides a rich data foundation for subsequent environmental impact assessment, scenario segmentation, and life prediction, enabling the prediction model to more comprehensively capture the coupling relationship between environmental factors and mechanical damage, thereby effectively improving the accuracy and reliability of remaining life prediction.

[0008] In one optional implementation, the historical monitoring data also includes historical statistical feature data corresponding to each monitoring moment. The historical statistical feature data includes at least one of historical average stress, historical vibration extreme values, and damage trend slope. This implementation introduces historical statistical feature data, fully mining the deep information in the historical operating data of key components, effectively supplementing the time dimension information missing from single-moment monitoring data. This enables the prediction model to simultaneously grasp the current state and historical evolution patterns, more accurately identifying the degradation patterns and trends of components, thereby significantly improving the accuracy and foresight of remaining service life prediction.

[0009] In one optional implementation, the step of determining the degree of environmental impact at each monitoring time based on weather parameter data, and dividing historical monitoring data into data under the expected lifespan impact scenario and data under the no-expected lifespan impact scenario based on the degree of environmental impact, includes: determining the degree of environmental impact at each monitoring time based on weather parameter data at each monitoring time, as well as the weather parameter threshold and weight coefficient corresponding to each weather parameter; classifying the data corresponding to monitoring times where the degree of environmental impact is greater than the impact threshold as data under the expected lifespan impact scenario; and classifying the data corresponding to monitoring periods where the degree of environmental impact is not greater than the impact threshold and consistently meets the requirement of the degree of environmental impact not greater than the impact threshold for a preset duration as data under the no-expected lifespan impact scenario. This implementation introduces weather parameter thresholds, weight coefficients, and impact thresholds to construct an environmental impact degree assessment mechanism, which can accurately identify the comprehensive impact of different weather parameters on component lifespan. By classifying monitoring periods where the environmental impact exceeds the threshold as expected lifespan impact scenarios, key data under harsh environmental conditions can be prioritized and processed accordingly. Conversely, by classifying monitoring periods where the environmental impact is below the threshold and lasts for a preset duration as scenarios without expected lifespan impact, transient fluctuations in environmental parameters are effectively filtered out, ensuring the stability and reliability of scenario classification. This refined scenario classification strategy allows subsequent data expansion to be processed differently for data under various environmental conditions. This ensures the full utilization of scarce data in expected lifespan impact scenarios while avoiding the over-processing of redundant data in scenarios without impact, significantly improving data utilization efficiency and model training effectiveness.

[0010] In one optional implementation, the step of augmenting data for scenarios affecting expected lifespan includes: using a time-series generative adversarial network (GAN) to augment the data for these scenarios. This implementation effectively addresses the challenges of acquiring monitoring data for key components of offshore wind power equipment and the scarcity of fault samples. The GAN can learn the time-series distribution characteristics of the original small sample data to generate synthetic time-series data that closely approximates the real data distribution, significantly expanding the number of training samples while maintaining the temporal correlation and dynamic characteristics of the data. This targeted data augmentation strategy particularly focuses on scenarios affecting expected lifespan, where the environment is harsh, data is scarcer but more valuable, enabling the model to fully learn key degradation patterns, avoiding overfitting due to insufficient samples, and significantly improving the model's generalization ability and prediction accuracy.

[0011] In one optional implementation, the step of training the adaptive combined prediction model using a training dataset to obtain a trained adaptive combined prediction model includes: pre-training the Long Short-Term Memory (LSTM) network and the Prophet prediction model separately using the training dataset to obtain pre-trained LSM and Prophet prediction models; fixing the parameters of the pre-trained LSM and Prophet prediction models, and training the adaptive weight mechanism using the training dataset to obtain a trained weight network; and performing end-to-end joint fine-tuning of the pre-trained LSM, Prophet prediction model, and trained weight network using the training dataset to obtain the trained adaptive combined prediction model. This implementation first pre-trains the LSM and Prophet prediction models separately, enabling the two sub-models to fully learn their respective strengths in time-series features; secondly, fixing the parameters of the pre-trained sub-models, and training the adaptive weight mechanism separately, enabling the weight network to accurately learn the optimal combination of the two sub-models under different conditions; and finally, performing end-to-end joint fine-tuning to achieve global optimum for the entire combined model. This phased training strategy avoids the convergence difficulties or local optima problems that may result from direct joint training, ensuring that each component can give full play to its advantages, thereby significantly improving the overall performance and prediction accuracy of the adaptive combined prediction model.

[0012] In one optional implementation, the step of inputting monitoring data of the key component to be predicted into a trained adaptive combined prediction model and outputting the remaining lifetime prediction result of the key component includes: inputting the monitoring data of the key component to be predicted into the trained adaptive combined prediction model, and having the Long Short-Term Memory Network (LSTM) and the Prophet prediction model in the adaptive combined prediction model output a first prediction result and a second prediction result, respectively; determining the weights of the first prediction result and the second prediction result through an adaptive weighting mechanism, and weighting and combining the first prediction result and the second prediction result based on the weights to obtain the remaining lifetime prediction result of the key component to be predicted. This implementation dynamically weights and combines the prediction results of the LSM and the Prophet prediction model through an adaptive weighting mechanism, achieving complementary advantages and fusion of the two models. The LSM network is good at capturing nonlinear dynamic characteristics and complex time series patterns, while the Prophet prediction model is strong at identifying trend changes and seasonal patterns. The two models characterize the component degradation process from different dimensions. The adaptive weighting mechanism dynamically adjusts the contributions of the two sub-models to the final prediction result based on the specific characteristics of the input data. This ensures that the prediction results retain the sensitivity of the Long Short-Term Memory network to complex operating conditions while incorporating the Prophet model's accurate grasp of the overall trend, avoiding the biases that may exist with a single model. This dynamic weighted combination strategy significantly improves the accuracy and robustness of remaining lifetime prediction, enabling the model to maintain excellent predictive performance under different environmental conditions and degradation stages.

[0013] In an optional implementation, the method further includes: acquiring the remaining lifetime prediction results of multiple key components; and fusing the remaining lifetime prediction results of multiple key components to obtain the remaining lifetime prediction result of the key component cluster. This implementation fuses the remaining lifetime prediction results of multiple key components, achieving a leap from individual component prediction to cluster-wide prediction, providing more macroscopic and comprehensive reference information for offshore wind farm operation and maintenance decisions. Individual component lifetime prediction can only reflect the health status of a single component, while wind farm operation and maintenance management needs to grasp the overall status of the entire wind farm's key component cluster in order to formulate scientific spare parts procurement plans, rationally arrange maintenance personnel, and schedule maintenance windows. By fusing the prediction results of multiple key components, cluster-level remaining lifetime prediction results can be output, thereby quickly grasping the overall health trend of the entire wind farm's key components, identifying wind turbines or areas requiring key attention, thereby optimizing resource allocation, reducing operation and maintenance costs, and improving the overall operational reliability and economic benefits of the wind farm.

[0014] Secondly, this invention provides a life prediction device for key components of offshore wind power equipment, comprising: a data acquisition module for acquiring historical monitoring data of key components of offshore wind power equipment, the historical monitoring data including weather parameter data at each monitoring time, damage parameter data of key components, and corresponding remaining life labels; a scenario segmentation module for determining the degree of environmental impact at each monitoring time based on weather parameter data, and dividing the historical monitoring data into data under expected life impact scenarios and data under no expected life impact scenarios based on the degree of environmental impact; a data expansion module for expanding the data under expected life impact scenarios, and combining the expanded data under expected life impact scenarios with the data under no expected life impact scenarios to form a training dataset; a model training module for training an adaptive combined prediction model using the training dataset to obtain a trained adaptive combined prediction model, wherein the adaptive combined prediction model is composed of a long short-term memory network and a Prophet prediction model, and the weights of the prediction results of the long short-term memory network and the prediction results of the Prophet prediction model in the final prediction result are dynamically adjusted through an adaptive weight mechanism; and a life prediction module for inputting the monitoring data of the key component to be predicted into the trained adaptive combined prediction model and outputting the remaining life prediction result of the key component to be predicted.

[0015] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method for predicting the lifespan of key components of offshore wind power equipment as described in the first aspect or any corresponding embodiment.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the method for predicting the lifespan of key components of offshore wind power equipment as described in the first aspect or any corresponding embodiment. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the first process of the method for predicting the lifespan of key components of offshore wind power equipment according to an embodiment of the present invention; Figure 2This is a schematic diagram of the second process of the method for predicting the lifespan of key components of offshore wind power equipment according to an embodiment of the present invention; Figure 3 This is a structural block diagram of a life prediction device for key components of offshore wind power equipment according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0021] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0022] Existing data-driven methods for equipment life prediction often employ single statistical or machine learning models. These methods typically rely on single types of monitoring data (such as vibration data or temperature data only), making it difficult to comprehensively integrate multi-source environmental information such as salt spray concentration and humidity with mechanical damage data. This results in prediction models failing to accurately reflect the combined impact of complex marine environments on component degradation. Furthermore, offshore wind power equipment has a short operational lifespan and few failure samples. Existing models often suffer from overfitting and insufficient generalization ability when processing such small-sample time-series data. In the face of sudden operating conditions such as extreme weather, single models exhibit poor adaptability, and their prediction accuracy is insufficient to meet engineering application requirements. Therefore, this invention provides a method, device, equipment, and medium for predicting the lifespan of key components of offshore wind power equipment. This addresses the problems of low prediction accuracy and poor model adaptability in complex marine environments caused by the inability to effectively integrate multi-source environmental data and mechanical damage data, and the lack of processing mechanisms for small-sample time-series data.

[0023] According to an embodiment of the present invention, a method for predicting the lifespan of key components of offshore wind power equipment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0024] This embodiment provides a method for predicting the lifespan of key components of offshore wind power equipment. Figure 1 This is a flowchart of a method for predicting the lifespan of key components of offshore wind power equipment according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain historical monitoring data of key components of offshore wind power equipment. The historical monitoring data includes weather parameter data at each monitoring time, damage parameter data of key components, and corresponding remaining life tags.

[0025] Offshore wind power equipment refers to a complete system of facilities installed in ocean or coastal waters to generate electricity using offshore wind energy resources. It mainly includes wind turbines, towers, electrical systems, and control systems. Compared to onshore wind power, offshore wind power equipment must withstand the harsh marine environment, including salt spray corrosion, high humidity, strong wind loads, and wave impacts, over long periods. Key components are those that play a decisive role in the safe operation of wind power equipment; failure of these components can lead to shutdowns or even major safety accidents. In this invention, this specifically refers to blades and blade bolts. Blades are the core aerodynamic components for capturing wind energy, directly exposed to the marine atmosphere, and susceptible to damage such as salt spray corrosion, leading-edge erosion, and crack propagation. Blade bolts are high-strength fasteners connecting the blades to the hub, bearing the dual effects of alternating loads and a corrosive environment; loosening or breakage of these bolts directly threatens the safety of the unit.

[0026] The purpose of step S101 is to collect basic data for model training. By acquiring weather parameters such as salt spray concentration, humidity, temperature, and wind speed, as well as damage parameters such as stress level, vibration amplitude, and corrosion rate of key components, a multi-dimensional perception of the operating environment and health status of key components of offshore wind power equipment is achieved. Simultaneously, by constructing remaining lifetime labels corresponding to each monitoring moment, a standard output reference is provided for subsequent supervised learning. This step lays the data foundation for the entire lifetime prediction method. Weather parameter data will be used for subsequent environmental impact assessment and scenario segmentation, damage parameter data serves as the core feature of the model input, and remaining lifetime labels provide learning objectives for model training. These three elements together constitute the starting point of the complete chain from data collection to model training, ensuring that subsequent steps have sufficient and accurate data support.

[0027] Step S102: Determine the degree of environmental impact at each monitoring time based on weather parameter data, and divide the historical monitoring data into data under the expected lifespan impact scenario and data under the no expected lifespan impact scenario based on the degree of environmental impact.

[0028] Step S102 is used to quantitatively assess the environmental impact at each monitoring time and classify historical monitoring data into scenarios accordingly. Specifically, this step calculates the comprehensive environmental impact based on weather parameter data (such as salt spray concentration, humidity, temperature, wind speed, etc.) at each monitoring time, as well as the corresponding thresholds and weighting coefficients for each parameter. Then, the data corresponding to monitoring times with environmental impact exceeding a preset threshold are classified as data under the "expected lifespan impact scenario" (i.e., severe environmental periods), while the data corresponding to monitoring periods with environmental impact not exceeding the preset threshold and remaining stable for a certain duration are classified as data under the "no expected lifespan impact scenario". Through this step, a quantitative assessment of the impact of complex marine environments and the classified management of historical data are achieved, laying the foundation for subsequent targeted expansion of scarce severe environmental data.

[0029] In this step, the degree of environmental impact refers to the potential impact of the current environment on the lifespan of critical components, which is quantitatively assessed by integrating multiple weather parameters (such as salt spray concentration, humidity, temperature, wind speed, etc.). It is a numerical indicator; the higher the value, the more severe the environment and the more significant the negative impact on component lifespan.

[0030] In scenarios affecting expected lifespan, harsh environmental conditions (such as typhoons, persistent high salt spray, and extreme temperatures) may accelerate the accumulation of damage to critical components, leading to a shortened lifespan. This type of data is typically scarce but highly valuable, thus requiring further data augmentation.

[0031] In scenarios without anticipated lifespan impacts, environmental conditions are normal, and critical components degrade at a conventional rate, without accelerated damage due to environmental factors. This type of data is relatively abundant and can be directly used for model training without the need for data augmentation.

[0032] Step S103: Expand the data under the expected lifespan impact scenario, and combine the expanded expected lifespan impact scenario data with the data under the no expected lifespan impact scenario to form a training dataset.

[0033] This step augments the data from scenarios impacting expected lifespan. Specifically, it uses algorithms to generate new, synthetic data with a similar distribution to the original small sample data, increasing the number of training samples. In this step, data augmentation is only performed on data from scenarios impacting expected lifespan, as this data is more valuable but scarce. The augmented data from scenarios impacting expected lifespan and data from scenarios without expected lifespan impact are then combined to form the training dataset. This approach addresses the "small sample problem" in offshore wind power equipment monitoring data. Data from scenarios impacting expected lifespan (during severe environmental periods) is typically scarce, making it difficult for models to fully learn the degradation patterns of components under harsh environmental conditions. By augmenting this scarce data, the training dataset achieves sufficient scale while maintaining data diversity, preventing overfitting during model training.

[0034] Step S104: Using the training dataset, the adaptive combined prediction model is trained to obtain the trained adaptive combined prediction model. The adaptive combined prediction model is composed of a long short-term memory network and a Prophet prediction model, and the weights of the prediction results of the long short-term memory network and the prediction results of the Prophet prediction model in the final prediction result are dynamically adjusted through an adaptive weight mechanism.

[0035] This step utilizes the expanded training dataset to train the adaptive combined prediction model, which combines a Long Short-Term Memory (LSTM) network, adept at nonlinear time series modeling, and a Prophet prediction model, skilled at trend capture. An adaptive weighting mechanism dynamically adjusts the contributions of each model to the final prediction result. The training process employs a phased strategy: first, LSTM and Prophet are pre-trained separately to allow them to initially grasp their respective degradation patterns; then, the sub-model parameters are fixed, and the adaptive weighting network is trained individually to learn how to evaluate the reliability of the two sub-models based on input features; finally, the entire combined model undergoes end-to-end joint fine-tuning to achieve global optimum through synergistic optimization of all components. Through this training process, the model not only fully leverages the sensitivity of LSTM to complex operating conditions and the accurate grasp of overall trends by Prophet, but also adaptively adjusts the sub-model weights according to different environmental conditions and degradation stages, thereby significantly improving the accuracy and adaptability of remaining lifetime prediction.

[0036] Step S105: Input the monitoring data of the key component to be predicted into the trained adaptive combined prediction model, and output the remaining life prediction result of the key component to be predicted.

[0037] This step involves monitoring data for the critical component to be predicted, using the same data categories as those used during model training. This data is input into the trained model. First, the Long Short-Term Memory (LSTM) network and the Prophet prediction model output preliminary prediction results. Then, an adaptive weighting mechanism dynamically calculates the weights of both networks based on the input data characteristics and performs a weighted combination, ultimately outputting the remaining life prediction result for the critical component. This step fully leverages the advantages of the trained model: under complex operating conditions such as extreme weather, the weighting mechanism automatically increases the reliance on the LTM network, which is sensitive to sudden changes; in stable environments, it maintains a balance between the two, utilizing Prophet's ability to grasp trends. The output remaining life prediction result provides maintenance personnel with direct quantitative decision-making basis for developing preventative maintenance plans, spare parts procurement schemes, and maintenance window arrangements.

[0038] The method for predicting the remaining lifespan of key components of offshore wind power equipment provided in this embodiment determines the degree of environmental impact based on weather parameter data and divides the scenarios into those with expected lifespan impact and those without. It also expands the data for the scenarios with expected lifespan impact, effectively solving the problems of difficult acquisition of monitoring data and scarce fault samples in offshore wind power equipment. By constructing an adaptive combined prediction model composed of a Long Short-Term Memory (LSTM) network and a Prophet prediction model, and introducing an adaptive weighting mechanism to dynamically adjust the contribution of the two models, it fully leverages the advantages of LSTM in nonlinear time series modeling and Prophet in trend capture, achieving complementary model advantages and significantly improving the prediction accuracy of the remaining lifespan of key components. This invention effectively solves the problems of low prediction accuracy and poor model adaptability of the remaining lifespan of key components of offshore wind power in complex marine environments due to the difficulty in effectively integrating multi-source environmental data and mechanical damage data, and the lack of processing mechanisms for small-sample time series. It significantly improves the accuracy of lifespan prediction and the adaptability of the model.

[0039] This embodiment provides another method for predicting the lifespan of key components of offshore wind power equipment. Figure 2 This is a flowchart of a method for predicting the lifespan of key components of offshore wind power equipment according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain historical monitoring data of key components of offshore wind power equipment. The historical monitoring data includes weather parameter data at each monitoring time, damage parameter data of key components, and corresponding remaining life tags.

[0040] Key components in this embodiment include blade bolts and the blade itself; weather parameter data includes weather forecast data and marine environmental data, such as salt spray concentration, humidity, temperature, wind speed, and wave height. Therefore, in an optional embodiment, the weather parameter data includes at least one of salt spray concentration, humidity, temperature, wind speed, and wave height; damage parameter data includes at least one of stress level, vibration amplitude, corrosion rate, and fatigue damage accumulation.

[0041] The remaining life tag can be obtained through historical failure data backtracking. That is, for components that have failed and been replaced, their remaining life at each historical monitoring point is calculated back based on their commissioning time and actual failure time. It can also be obtained through empirical annotation methods, which are not limited in this invention.

[0042] In another optional implementation, the aforementioned historical monitoring data includes not only weather parameter data, damage parameter data of key components, and remaining life tags at each monitoring time, but also historical statistical characteristic data corresponding to each monitoring time. The historical statistical characteristic data includes at least one of historical average stress, historical vibration extreme values, and damage trend slope.

[0043] Step S202: Determine the degree of environmental impact at each monitoring time based on weather parameter data, and divide the historical monitoring data into data under the expected lifespan impact scenario and data under the no expected lifespan impact scenario based on the degree of environmental impact.

[0044] The purpose of this step S202 is to divide the historical monitoring data into scenarios, that is, to divide it into data collected from key equipment under different scenarios such as the initial state of key wind power components and the state after a period of operation.

[0045] Specifically, step S202 above includes: Step S2021: Determine the degree of environmental impact at each monitoring time based on the weather parameter data at each monitoring time, as well as the weather parameter thresholds and weighting coefficients corresponding to each weather parameter.

[0046] Step S2022: The data corresponding to the monitoring time when the degree of environmental impact is greater than the impact threshold is classified as data under the expected lifespan impact scenario.

[0047] Step S2023: Data corresponding to monitoring periods in which the degree of environmental impact is no greater than the impact threshold and the degree of environmental impact is no greater than the impact threshold for a continuous preset duration are classified as data under the scenario of no expected lifespan impact.

[0048] For example, let the weather parameter vector be W=[w1,w2,...,w n], where w1 is the salt spray concentration, w2 is the humidity, w3 is the temperature, w4 is the wind speed, etc., and n is the number of weather parameter categories; let the threshold vector T = [t1, t2, ..., t n [ ], corresponding to the weather parameter thresholds for each weather parameter. The environmental impact degree function is: I(W) = ∑ i α i ×max(0,w i -t i ) Where, α i The weighting coefficients for each weather parameter can be determined through expert experience or historical data. When I(W) > δ (δ is the impact threshold), it is determined to be a scenario with expected lifespan impact; when I(W) ≤ δ and the situation continues for a preset duration Δt without rebounding, it is determined to be a scenario without expected lifespan impact. The preset duration Δt can be determined through expert experience, statistical analysis of historical data, or actual operating experience of wind farms.

[0049] Step S203: Expand the data under the expected lifespan impact scenario, and combine the expanded expected lifespan impact scenario data with the data under the no expected lifespan impact scenario to form a training dataset.

[0050] In one alternative implementation, a Time Series Generative Adversarial Network (TimeGAN) can be used to augment data in scenarios affecting expected lifespan.

[0051] For example, the TimeGAN augmentation process is as follows: Let the original small sample data be X={x1,x2,...,x...} m}, where x i It is time series data; By training the generator G and the discriminator D, the generated data distribution is made as close as possible to the real data distribution.

[0052] Step S204: Using the training dataset, the adaptive combined prediction model is trained to obtain the trained adaptive combined prediction model. The adaptive combined prediction model is composed of a long short-term memory network and a Prophet prediction model, and the weights of the prediction results of the long short-term memory network and the prediction results of the Prophet prediction model in the final prediction result are dynamically adjusted through an adaptive weight mechanism.

[0053] The adaptive combined prediction model is composed of an LSTM network and a Prophet model, and dynamically adjusts the contribution of each model through an adaptive weighting mechanism.

[0054] On the one hand, data preparation is required before training the relevant model. Based on the training dataset obtained in step S203, an input feature vector is constructed. Taking historical monitoring data, including weather parameter data at each monitoring time, damage parameter data of key components, and historical statistical feature data corresponding to each monitoring time, as an example, the input feature at time t can be expressed as: X(t)=[W(t),D(t),H(t)] Where: W(t) = [salt spray concentration (t), humidity (t), temperature (t), wind speed (t), wave height (t)], i.e., weather data; D(t) = [stress level (t), vibration amplitude (t), corrosion rate (t), cumulative fatigue damage (t)], i.e., damage data; H(t) = [historical mean stress (t-τ:t), historical extreme vibration (t-τ:t), damage trend slope (t-τ:t)], i.e., historical statistical characteristics. The time window τ is determined based on the data frequency and physical characteristics, and is usually taken as τ = 24 hours (for hourly data).

[0055] On the other hand, before training the relevant models, it is necessary to build the basic framework of the adaptive combined prediction model, including the Long Short-Term Memory Network (LSTM network) and the Prophet prediction model, and the adaptive weight mechanism.

[0056] Among them, the LSTM branch design: A three-layer LSTM network structure is adopted: LSTM layer 1: 64 units, return_sequences=True; Dropout layer 1: dropout_rate=0.2; LSTM layer 2: 32 units, return_sequences=False; Dropout layer 2: dropout_rate=0.2; Dense layer: 16 units, activation='relu'; Output layer: 1 unit, activation='linear' (remaining lifetime).

[0057] LSTM layer state update formula: Forget gate f_t=sig(W_f·[h_{t-1},x_t]+b_f); Input gate i_t = sig(W_i·[h_{t-1},x_t]+b_i); Candidate state _t=tanh(W_C·[h_{t-1},x_t]+b_C); Cell state C_t = f_t⊙C_{t-1} + i_t⊙ _t; Output gate o_t = sig(W_o·[h_{t-1},x_t]+b_o); Hidden state h_t = o_t ⊙ tanh(C_t); Where W_f,W_i,W_C,W_o are the weight matrices of each gate structure in the LSTM network, which need to be learned through subsequent training; b_f,b_i,b_C,b_o are the bias terms of each gate structure in the LSTM network, which are learned through subsequent training; h_{t-1} is the hidden state of the previous time step; x_t is the input of the current time step, that is, X(t)=[W(t),D(t),H(t)]; ⊙ represents element-wise multiplication.

[0058] Prophet branch design: The additive form of the Prophet model: y(t) = g(t) + s(t) + h(t) + ε_t Where g(t) = kt + m is the trend term, k is the growth rate, and m is the offset, obtained by fitting historical data; s(t) = ∑[a n cos(2πnt / T)+b n sin(2πnt / T)] is the seasonal term, and a_n and b_n are the Fourier series coefficients of the seasonal term, which are obtained through training; T The period is denoted as h(t) (e.g., 365.25 days is an annual period); h(t) represents the holiday effect (for special weather events at sea); ε_t is the error term that follows a normal distribution.

[0059] Adaptive weighting mechanism: Design an adaptive weight network based on an attention mechanism: Attention score calculation: e(t)=V_a^T·tanh(W_a·h_{lstm}(t)+U_a·h_{prophet}(t)+b_a) Where e(t) represents the attention score (scalar) at time step t, which measures how much information the model should focus on at time step t; V_a represents a column vector, and its transpose V_a^T is a row vector used to map the result of tanh activation to a scalar score; tanh represents the hyperbolic tangent activation function, used to introduce nonlinearity; W_a represents the weight matrix, used to perform a linear transformation on the hidden state h_{lstm}(t) of the LSTM; h_{lstm}(t) represents the hidden state vector generated by the LSTM model at time step t; U_a represents the weight matrix, used to perform a linear transformation on the output h_{prophet}(t) of the Prophet model; h_{prophet}(t) represents the feature vector generated by the Prophet model at time step t; b_a represents the bias term, which is a vector added to the result of the linear transformation.

[0060] Weight normalization: α(t)=exp(e(t)) / [exp(e(t))+exp(-e(t))] Final weights: ω(t)=sig(W_ω·[α(t),Δy(t),var(t)]+b_ω) The sig function compresses the result of the linear transformation to between 0 and 1, serving as the final weights. α(t) represents the attention weight, typically derived from the attention mechanism, indicating the degree of attention paid to the input information at time step t. It can be a scalar or a vector; if a vector, it needs to be converted to a scalar using methods such as summation or averaging. Δy(t) represents the error between the predicted and actual values ​​at time step t, or the change between adjacent time steps. In degradation prediction, it represents the rate of change of degradation or the prediction residual. var(t) represents the variance or uncertainty at time step t, specifically the recent variance of the target value. W_ω represents the weight vector (or matrix) with dimensions 1x3, mapping the three-dimensional input to a scalar. b_ω represents the bias term, a scalar.

[0061] Loss function design: The combined loss function consists of three parts: L_total=L_pred+λ_smoothL_smooth+λ_uncertL_uncert Predicted loss: L_pred = ∑|y_true - |+0.5∑(y_true- )² Weighted smoothing loss: L_smooth=∑|ω(t)-ω(t-1)|² Uncertainty regularization term: L_uncert=-∑[ω(t)log(p_{lstm})+(1-ω(t))log(p_{prophet})] Where ω(t) and ω(t-1) represent the current monitoring time t and the previous monitoring time t-1, respectively, λ_smooth and λ_uncert are regularization coefficients, and y_true represents the true value (target value). Represents the predicted value, |y_true- | represents the absolute error, (y_true- )² represents the squared error; p_{lstm} and p_{prophet} are the confidence scores of each model's prediction.

[0062] Specifically, the training process in step S204 above includes: Step S2041: Using the training dataset, pre-train the Long Short-Term Memory Network and the Prophet prediction model respectively to obtain the pre-trained Long Short-Term Memory Network and the Prophet prediction model.

[0063] When the training dataset includes weather parameter data, damage parameter data of key components, and corresponding remaining life labels for each monitoring time, the weather parameter data and damage parameter data are used as input features, and the corresponding remaining life labels are used as output labels to pre-train the Long Short-Term Memory Network and the Prophet prediction model, respectively.

[0064] When the training dataset includes weather parameter data for each monitoring time, damage parameter data for key components, historical statistical feature data corresponding to each monitoring time, and corresponding remaining life label, the weather parameter data, damage parameter data, and historical statistical feature data are used as input features, and the corresponding remaining life label is used as output label to pre-train the Long Short-Term Memory Network and the Prophet prediction model respectively.

[0065] Step S2042: Fix the parameters of the pre-trained Long Short-Term Memory network and the Prophet prediction model, and use the training dataset to train the adaptive weight mechanism to obtain the trained weight network.

[0066] In the same step S2041, when the training dataset includes weather parameter data, damage parameter data of key components and corresponding remaining life labels at each monitoring time, the adaptive weight mechanism is trained using the weather parameter data and damage parameter data as input features and the corresponding remaining life labels as output labels.

[0067] When the training dataset includes weather parameter data for each monitoring time, damage parameter data for key components, historical statistical feature data corresponding to each monitoring time, and corresponding remaining life label, the adaptive weight mechanism is trained using weather parameter data, damage parameter data, and historical statistical feature data as input features and the corresponding remaining life label as output label.

[0068] Step S2043: Using the training dataset, perform end-to-end joint fine-tuning of the pre-trained long short-term memory network, the Prophet prediction model, and the trained weight network to obtain the trained adaptive combined prediction model.

[0069] In the same step S2041, when the training dataset includes weather parameter data, damage parameter data of key components and corresponding remaining life labels at each monitoring time, the weather parameter data and damage parameter data are used as input features, and the corresponding remaining life labels are used as output labels to perform end-to-end joint fine-tuning of the pre-trained long short-term memory network, the Prophet prediction model and the trained weight network.

[0070] When the training dataset includes weather parameter data for each monitoring time, damage parameter data for key components, historical statistical feature data corresponding to each monitoring time, and corresponding remaining life label, the weather parameter data, damage parameter data, and historical statistical feature data are used as input features, and the corresponding remaining life label is used as output label. The pre-trained long short-term memory network, the Prophet prediction model, and the trained weight network are then jointly fine-tuned end-to-end.

[0071] Step S205: Input the monitoring data of the key component to be predicted into the trained adaptive combined prediction model, and output the remaining life prediction result of the key component to be predicted.

[0072] Specifically, step S205 includes: Step S2051: Input the monitoring data of the key component to be predicted into the trained adaptive combined prediction model, and the long short-term memory network and the Prophet prediction model in the adaptive combined prediction model will output the first prediction result and the second prediction result respectively.

[0073] Step S2052: The weights of the first prediction result and the second prediction result are determined by an adaptive weighting mechanism, and the first prediction result and the second prediction result are weighted and combined based on the weights to obtain the remaining life prediction result of the key component to be predicted.

[0074] The following is an example of the reasoning process, taking prediction at time t as an example: enter: X(t) = [Salt spray concentration: 0.8 mg / m³, humidity: 85%, temperature: 28°C, wind speed: 12 m / s, stress: 125 MPa, vibration: 0.8 mm / s, corrosion: 0.15 mm / year, damage accumulation: 0.42] Predictions from each model: LSTM prediction: y_lstm(t) = 3652 hours remaining (approximately 152 days) Prophet prediction: y_prophet(t) = 4128 hours remaining (approximately 172 days). Weight calculation: Calculated using the weighted network: Attention score: e(t) = 0.68 Recent forecast discrepancy: Δy(t-1) = |3600-4000| = 400 hours Objective variance: var(t) = 225 (hours²) Final weight: ω(t) = σ(0.8 × 0.68 + 0.1 × 400 / 1000 - 0.2 × 225 / 1000) = 0.62 Combined prediction: (t) = 0.62 × 3652 + 0.38 × 4128 = 3821 hours (approximately 159 days) Finally, the model performance was evaluated, and the performance metrics on the test set are shown in Table 1.

[0075] Table 1

[0076] Special scene handling: Adaptive adjustment under extreme weather conditions: When extreme weather conditions are detected (such as I(W)>δ), the weighted network will automatically adjust: ω_extreme(t)=sig(ω(t)+Δω) Where Δω=tanh(β·I(W)), and β is the adjustment coefficient.

[0077] This makes the model more reliant on the dynamic modeling capabilities of LSTM under harsh conditions.

[0078] Quantification of uncertainty: Output prediction interval: CI( )=[ -z·σ, +z·σ] Where σ²=ω²·σ_lstm²+(1-ω)²·σ_prophet²+2ω(1-ω)ρ·σ_lstm·σ_prophet, z is the quantile corresponding to the confidence level, σ_lstm² is the LSTM prediction variance, σ_prophet² is the Prophet prediction variance, ρ is the correlation coefficient between the LSTM and Prophet prediction results, and σ_lstm and σ_prophet represent the standard deviation of the Long Short-Term Memory Network prediction result and the standard deviation of the Prophet prediction model prediction result, respectively.

[0079] This adaptive combined prediction model intelligently integrates the advantages of deep learning and traditional time series methods, significantly improving prediction accuracy while maintaining interpretability. It is particularly suitable for complex and data-scarce scenarios such as offshore wind power equipment life prediction.

[0080] In an optional implementation, in addition to predicting the remaining lifetime of a single key component according to step S205 above, cluster lifetime prediction can also be performed. Therefore, this embodiment may also include step S206, obtaining the remaining lifetime prediction results of multiple key components; and fusing the remaining lifetime prediction results of multiple key components to obtain the remaining lifetime prediction results of the key component cluster.

[0081] For example, the critical equipment cluster area is divided into different sub-regions using the OPTICS algorithm and the K-Medoids algorithm. Specifically, the OPTICS algorithm is used to identify the data density structure and determine the cluster center, while the K-Medoids algorithm is used to select a benchmark point for each sub-region.

[0082] Calculate the normalized mean absolute error (MAE) and correlation coefficient (R) of the expected lifespan of benchmark points and key equipment cluster areas: MAE no m =MAE / max(MAE); R no m =(R+1) / 2, which normalizes the correlation coefficient to the interval [0,1], where max(MAE) is the maximum value of the mean absolute error among all benchmark points, and (R+1) / 2 is the correlation coefficient normalization formula.

[0083] Calculate the weighting coefficient of the benchmark point: w i =(1-MAE no m , i )×R no m , i / ∑ j(1-MAE no m , j )×R no m , j .

[0084] Predicting the short-term lifetime of the cluster: =∑ i w i ×X i X i Predicted lifetime for the i-th benchmark point 。

[0085] For example, a joint distribution of individual lifetimes and cluster lifetimes can also be established based on a Copula function. Let the individual lifetime be X and the cluster lifetime be Y, and the edge distributions F be connected through a Copula function C. x (x) and F y (y): F(x,y)=C(F x (x),F y (y);θ), where θ is the parameter of the Copula function. Predicting the short-term lifetime of the cluster: =F(X1,X2,...,X n ).

[0086] The method for predicting the lifespan of key components of offshore wind power equipment provided in this embodiment improves data utilization efficiency and prediction accuracy through multi-source data collaborative processing and scenario segmentation; it uses the TimeGAN algorithm to solve the small sample problem and combines it with the isolated forest algorithm to ensure data quality; the adaptive combined prediction model can dynamically adjust the weights of each sub-model to adapt to different environmental conditions; and it uses the Copula statistical algorithm to extend lifespan prediction from single units to clusters, improving the practicality of the prediction; it is particularly suitable for the lifespan prediction needs of offshore wind power equipment in harsh marine environments.

[0087] Taking the prediction of blade bolt life in an offshore wind farm as an example, weather data including salt spray concentration, humidity, temperature, and wind speed, as well as damage data such as stress, vibration, and corrosion of the blade bolts, are collected. After preprocessing and scene segmentation using the method of this invention, the original 200 small sample data points are expanded to 2000 data points using the TimeGAN algorithm. After removing outliers using the Isolation Forest algorithm, 1850 valid data points are obtained.

[0088] An adaptive combined prediction model was constructed and trained, achieving a prediction accuracy of 92.3% on the test set, which is 15.7% higher than the single LSTM model and 21.2% higher than the single Prophet model.

[0089] Finally, using the Copula statistical algorithm, the lifespan of the entire wind farm bolt cluster / wind turbine blades was predicted based on the predicted lifespan of 20 individual bolts / wind turbine blades, with an error of less than 8% compared to the actual observed value.

[0090] It is evident that this invention is not only applicable to life prediction of key components such as blades and bolt fractures, but also has wide applicability and good predictive effect.

[0091] This embodiment also provides a lifespan prediction device for key components of offshore wind power equipment. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0092] This embodiment provides a lifespan prediction device for key components of offshore wind power equipment, such as... Figure 3 As shown, it includes: The data acquisition module 301 is used to acquire historical monitoring data of key components of offshore wind power equipment. The historical monitoring data includes weather parameter data at each monitoring time, damage parameter data of key components, and corresponding remaining life tags. The scenario segmentation module 302 is used to determine the degree of environmental impact at each monitoring time based on weather parameter data, and to divide the historical monitoring data into data under the expected lifespan impact scenario and data under the no expected lifespan impact scenario based on the degree of environmental impact. The data augmentation module 303 is used to augment the data under the expected lifespan impact scenario, and to combine the augmented data under the expected lifespan impact scenario with the data under the no expected lifespan impact scenario to form a training dataset. The model training module 304 is used to train the adaptive combined prediction model using the training dataset to obtain the trained adaptive combined prediction model. The adaptive combined prediction model is composed of a long short-term memory network and a Prophet prediction model, and the weights of the prediction results of the long short-term memory network and the prediction results of the Prophet prediction model in the final prediction result are dynamically adjusted through an adaptive weight mechanism. The life prediction module 305 is used to input the monitoring data of the key component to be predicted into the trained adaptive combined prediction model and output the remaining life prediction result of the key component to be predicted.

[0093] The offshore wind power equipment key component life prediction device provided in this embodiment of the invention can execute the offshore wind power equipment key component life prediction method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0094] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0095] The following is a detailed reference. Figure 4 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0096] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0097] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a memory 408, or installed from a ROM 402. When the computer program is executed by the processor 401, it performs the functions defined in the offshore wind power equipment critical component life prediction method of the embodiments of the present invention.

[0098] Figure 4The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0099] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the method for predicting the lifespan of key components of offshore wind power equipment shown in the above embodiments is implemented.

[0100] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0101] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A method for predicting the lifespan of key components of offshore wind power equipment, characterized in that, The method includes: Acquire historical monitoring data of key components of offshore wind power equipment. The historical monitoring data includes weather parameter data at each monitoring time, damage parameter data of key components, and corresponding remaining life tags. The degree of environmental impact at each monitoring time is determined based on the weather parameter data, and the historical monitoring data is divided into data under the expected lifespan impact scenario and data under the no expected lifespan impact scenario based on the degree of environmental impact. The data under the expected lifespan impact scenario is augmented, and the augmented data under the expected lifespan impact scenario and the data under the expected lifespan without impact scenario are combined to form a training dataset; Using the training dataset, the adaptive combined prediction model is trained to obtain the trained adaptive combined prediction model. The adaptive combined prediction model is composed of a long short-term memory network and a Prophet prediction model, and the weights of the prediction results of the long short-term memory network and the prediction results of the Prophet prediction model in the final prediction result are dynamically adjusted through an adaptive weight mechanism. The monitoring data of the key components to be predicted are input into the trained adaptive combined prediction model, and the remaining life prediction results of the key components to be predicted are output.

2. The method for predicting the lifespan of key components of offshore wind power equipment according to claim 1, characterized in that, The weather parameter data includes at least one of salt spray concentration, humidity, temperature, wind speed, and wave height; the damage parameter data includes at least one of stress level, vibration amplitude, corrosion rate, and fatigue damage accumulation.

3. The method for predicting the lifespan of key components of offshore wind power equipment according to claim 1, characterized in that, The step of determining the degree of environmental impact at each monitoring time based on the weather parameter data, and dividing the historical monitoring data into data under the expected lifespan impact scenario and data under the no expected lifespan impact scenario based on the degree of environmental impact, includes: Based on the weather parameter data at each monitoring time, as well as the weather parameter thresholds and weighting coefficients corresponding to each weather parameter, the degree of environmental impact at each monitoring time is determined. Data corresponding to monitoring times when the degree of environmental impact exceeds the impact threshold are classified as data under the expected lifespan impact scenario; Data corresponding to monitoring periods in which the degree of environmental impact is no greater than the impact threshold and remains at the threshold for a preset duration are classified as data under the scenario of no expected lifespan impact.

4. The method for predicting the lifespan of key components of offshore wind power equipment according to claim 1, characterized in that, The step of augmenting the data under the expected lifespan impact scenario includes: using a time-series generative adversarial network to augment the data under the expected lifespan impact scenario.

5. The method for predicting the lifespan of key components of offshore wind power equipment according to claim 1, characterized in that, The step of training the adaptive combined prediction model using the training dataset to obtain the trained adaptive combined prediction model includes: Using the training dataset, the Long Short-Term Memory Network and the Prophet prediction model were pre-trained respectively to obtain the pre-trained Long Short-Term Memory Network and the Prophet prediction model. With the parameters of the pre-trained Long Short-Term Memory network and the Prophet prediction model fixed, the adaptive weight mechanism is trained using the training dataset to obtain a trained weight network. Using the training dataset, the pre-trained long short-term memory network, the Prophet prediction model, and the trained weight network are jointly fine-tuned end-to-end to obtain a trained adaptive combined prediction model.

6. The method for predicting the lifespan of key components of offshore wind power equipment according to claim 1, characterized in that, The step of inputting monitoring data of the key component to be predicted into the trained adaptive combined prediction model and outputting the remaining life prediction result of the key component to be predicted includes: The monitoring data of the key components to be predicted are input into the trained adaptive combined prediction model, and the long short-term memory network and the Prophet prediction model in the adaptive combined prediction model output the first prediction result and the second prediction result, respectively. The weights of the first prediction result and the second prediction result are determined by the adaptive weighting mechanism, and the first prediction result and the second prediction result are weighted and combined based on the weights to obtain the remaining life prediction result of the key component to be predicted.

7. The method for predicting the lifespan of key components of offshore wind power equipment according to claim 1, characterized in that, The method further includes: Obtain the remaining life prediction results for multiple key components; The remaining lifetime prediction results of the multiple key components are fused to obtain the remaining lifetime prediction results of the key component cluster.

8. A life prediction device for key components of offshore wind power equipment, characterized in that, The device includes: The data acquisition module is used to acquire historical monitoring data of key components of offshore wind power equipment. The historical monitoring data includes weather parameter data at each monitoring time, damage parameter data of key components, and corresponding remaining life tags. The scenario segmentation module is used to determine the degree of environmental impact at each monitoring time based on the weather parameter data, and to divide the historical monitoring data into data under the expected lifespan impact scenario and data under the no expected lifespan impact scenario based on the degree of environmental impact. The data augmentation module is used to augment the data under the expected lifespan impact scenario, and to combine the augmented data under the expected lifespan impact scenario with the data under the no expected lifespan impact scenario to form a training dataset. The model training module is used to train the adaptive combined prediction model using the training dataset to obtain the trained adaptive combined prediction model. The adaptive combined prediction model is composed of a long short-term memory network and a Prophet prediction model, and the weights of the prediction results of the long short-term memory network and the prediction results of the Prophet prediction model in the final prediction result are dynamically adjusted through an adaptive weight mechanism. The life prediction module is used to input the monitoring data of the key components to be predicted into the trained adaptive combined prediction model and output the remaining life prediction results of the key components to be predicted.

9. An electronic device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the computer instructions to perform the life prediction method for key components of offshore wind power equipment as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the life prediction method for key components of offshore wind power equipment as described in any one of claims 1 to 7.

Images