Marine flexible riser motion response prediction method based on EMD-LSTM
By combining EMD decomposition and LSTM model, the modal characteristics of the flexible riser are extracted, which solves the problems of low prediction accuracy and frequency domain sensitivity in the existing technology and realizes high-precision and stable prediction of the flexible riser motion response.
Patent Information
- Application Number
- CN202510697886.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-28
- Publication Date
- 2025-09-12
AI Technical Summary
The existing technology for predicting the motion response of flexible risers has the following problems: low prediction accuracy, prediction results are sensitive to frequency domain changes, and the model is prone to falling into local optimality.
The flexible riser strain data is decomposed based on empirical mode decomposition (EMD) to extract the natural modal frequency, damping ratio and modal shape information. The EMD-LSTM model is constructed in combination with the long short-term memory neural network (LSTM). Each modal component is modeled and predicted, and the final response prediction value is obtained through linear superposition and weighted fusion.
The prediction accuracy is improved, the sensitivity to frequency domain changes is reduced, the model is prevented from falling into local optimality, and a more stable prediction of the flexible riser motion response is achieved.
Smart Images
Figure CN120632342A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of flexible riser motion response prediction, and in particular to an EMD-LSTM-based marine flexible riser motion response prediction method. Background Art
[0002] Flexible risers are key components in offshore oil and gas production systems, connecting subsea production units to floating platforms. They are primarily used to transport media such as oil, gas, water, or gas. Due to their extremely complex service environments, they are constantly subjected to the coupling effects of multiple environmental loads, including waves, wind, and currents. Their structural responses exhibit significant nonlinear, non-stationary, and multi-scale dynamic characteristics, often leading to vortex-induced vibrations, alternating bending, and fatigue failure, posing a serious threat to the safety and economic viability of deep-sea oil and gas field development.
[0003] In order to achieve real-time monitoring and health assessment of the structural status of flexible risers, high-precision motion response prediction technology has become a research hotspot. Traditional methods are mainly based on numerical simulation methods of physical modeling and prediction methods based on empirical formulas or statistical methods. Using mechanical modeling methods such as finite element analysis (FEA), CFD coupling simulation, and vibration analysis, the response behavior of flexible risers under various working conditions can be simulated. However, these methods have shortcomings such as complex modeling process, strong parameter dependence, long calculation time, unsuitability for real-time application, and lack of adaptive ability when actual working conditions change. Methods based on empirical formulas, such as autoregressive (AR), moving average (MA), Kalman filtering, etc., have certain prediction capabilities when data features are stable, but when faced with the problems of non-stationary marine environment and diverse structural responses, their prediction accuracy drops significantly, making it difficult to meet the accuracy and timeliness requirements of online monitoring of marine engineering.
[0004] In recent years, with the development of artificial intelligence, the application of deep learning methods in structural dynamics prediction has gradually increased. Long short-term memory (LSTM) neural networks, with their superior time series modeling capabilities, have been widely used in fields such as structural response prediction and ocean state assessment. However, most existing LSTM applications directly use raw signals as input, ignoring the multi-scale dynamic characteristics of the signals. This results in prediction results that are sensitive to frequency domain changes, the model is prone to falling into local optimality, and it is difficult to explain the model's learning mechanism in different response frequency bands. Summary of the Invention
[0005] The purpose of the present invention is to provide a method for predicting the motion response of marine flexible risers based on EMD-LSTM, so as to solve the problems of low prediction accuracy, sensitivity of prediction results to frequency domain changes, and easy fall into local optimum of the existing technology.
[0006] To achieve the above objectives, the present invention provides a method for predicting the motion response of a marine flexible riser based on EMD-LSTM, comprising the following steps:
[0007] S1. Obtain historical flexible riser strain data from the offshore platform structural health monitoring system, and perform outlier cleaning preprocessing on the historical flexible riser discrete point strain data to obtain a flexible riser strain data group;
[0008] S2. Decomposing the flexible riser strain data set dynamically based on the empirical mode decomposition (EMD) algorithm to obtain an independent modal data set, and extracting features from the independent modal data set to obtain the natural modal frequency, damping ratio, and modal shape information of the flexible riser;
[0009] S3, constructing an EMD-LSTM model based on the natural modal frequency, damping ratio and modal shape information of the flexible riser and the independent modal data set;
[0010] S4. Input the current flexible riser strain data obtained from the offshore platform structural health monitoring system into the EMD-LSTM model to obtain the IMF prediction value output by each sub-model of the EMD-LSTM model, and linearly superimpose the IMF prediction value output by each sub-model to obtain the flexible riser motion response prediction value.
[0011] In some embodiments of the present application, in S1, obtaining historical flexible riser strain data in an offshore platform structural health monitoring system and performing outlier cleaning preprocessing on the historical flexible riser discrete point strain data includes:
[0012] S11. Obtaining discrete point strain data of a flexible riser in an offshore platform structural health monitoring system measured by an optical fiber strain sensor or a strain gauge;
[0013] S12. Use the Z-score algorithm or the IQR algorithm to detect and eliminate abnormal data points in the flexible riser discrete point strain data, and use the interpolation method to fill in the missing data points;
[0014] S13. Convert the strain data of the discrete points of the flexible riser into the displacement data of the discrete points of the flexible riser. The conversion expression is:
[0015] u(x)=∫ε(x)dx+C;
[0016] Where u(x) is the displacement data of the discrete points of the flexible riser, ε(x) is the strain data of the discrete points of the flexible riser, and C is the integral constant obtained based on the boundary conditions;
[0017] S14. Using wavelet transform to perform denoising on the displacement data of discrete points of the flexible riser to obtain a flexible riser strain data group.
[0018] In some embodiments of the present application, in S14, wavelet transform is used to perform denoising on the flexible riser discrete point displacement data, and the obtained flexible riser strain data group includes:
[0019] S141, performing multi-layer wavelet decomposition on the original response sequence using wavelet basis functions to obtain displacement components in different frequency bands;
[0020] S142, using a threshold denoising algorithm to remove noise from high-frequency detail components in displacement components of different frequency bands to obtain high-frequency denoised components;
[0021] S143. Use wavelet reconstruction method to process the high-frequency denoising component to obtain a flexible riser strain data group.
[0022] In some embodiments of the present application, in S2, the flexible riser strain data set is subjected to dynamic response decomposition based on an empirical mode decomposition (EMD) algorithm to obtain an independent modal data set including:
[0023] S21. Analyze and obtain the local maximum and minimum values of the extreme points of the flexible riser strain data set. Use interpolation to connect the local maximum and minimum values to obtain the upper envelope U(t) and lower envelope L(t), and calculate the local average value. The calculation formula is:
[0024]
[0025] Where U(t) is the upper envelope, L(t) is the lower envelope, and m1(t) is the local average;
[0026] S22. Extract the candidate value of the first IMF component to be decomposed. The expression is:
[0027] h1(t)=x(t)-m1(t);
[0028] Where x(t) is the displacement component and h1(t) is the candidate value of the first IMF component;
[0029] S23, based on the relationship between h1(t) and the IMF preset condition, calculate the displacement component c1(t) of the first IMF;
[0030] S24. Remove the first IMF component from the original displacement data to obtain the residual component r1(t), which is expressed as:
[0031] r1(t)=x(t)-c1(t);
[0032] Where c1(t) is the displacement component of the first IMF, and r1(t) is the residual component;
[0033] S25. Use the residual component r1(t) as the new displacement data and repeat steps S21-S24 until the residual component becomes a monotonic function. The decomposition result is:
[0034]
[0035] Among them, c i (t) is the displacement component of the i-th IMF, r n (t) is the remainder of the residual component, and x(t) is the displacement component;
[0036] An independent modal data set is obtained based on x(t).
[0037] In some embodiments of the present application, in S23, based on the relationship between h1(t) and the IMF preset condition, calculating the displacement component c1(t) of the first IMF includes:
[0038] Determine whether h1(t) meets the IMF preset conditions. If h1(t) meets the IMF preset conditions, obtain the first IMF displacement component c1(t). If h1(t) does not meet the IMF preset conditions, repeat steps S21-S22 to continue decomposing h1(t) until it meets the IMF preset conditions, and obtain the first IMF displacement component c1(t).
[0039] IMF preset conditions include: the number of zero crossings is the same as the number of extreme values and the local mean is zero.
[0040] In some embodiments of the present application, in S3, constructing an EMD-LSTM model based on the natural modal frequency, damping ratio, and modal shape information of the flexible riser and the independent modal data set includes:
[0041] The independent modal data group is split into multiple IMF components, and an LSTM model is constructed for each IMF component;
[0042] The LSTM model of each IMF component was trained using supervised learning. The root mean square error (RMS) loss function and Adam optimizer were used. The ratio of training set, validation set, and test set was 6:2:2. The number of training rounds was adjusted based on the model convergence.
[0043] The EMD-LSTM model is obtained based on the trained LSTM model of each IMF component.
[0044] In some embodiments of the present application, in S4, the current flexible riser strain data obtained in the offshore platform structural health monitoring system is input into the EMD-LSTM model, and the IMF prediction value output by each sub-model of the EMD-LSTM model is obtained, including:
[0045] S41. Obtain current flexible riser strain data in the offshore platform structural health monitoring system;
[0046] S42. The EMD-LSTM model includes multiple LSTM sub-models. Each LSTM sub-model corresponds to an IMF component. Assuming there are k LSTM sub-models and k corresponding IMF components, the calculation formula for the IMF prediction value is:
[0047]
[0048] in, represents the predicted value of the kth IMF component at time t+1.
[0049] In some embodiments of the present application, in S4, linearly superimposing the IMF prediction values output by each sub-model to obtain the flexible riser motion response prediction value includes:
[0050] S43. Based on the EMD reconstruction principle, the IMF prediction values output by each sub-model are linearly superimposed to obtain the flexible riser motion response prediction value. The calculation formula is:
[0051]
[0052] in, is the predicted value of the flexible riser motion response at time t+1, is the estimated value of the EMD residual term.
[0053] In some embodiments of the present application, S4 further includes:
[0054] S44, using weighted fusion strategy to optimize S43, the expression is:
[0055]
[0056] Among them, w i is the fusion weight of the i-th IMF component, is the predicted value of the flexible riser motion response obtained by weighted fusion;
[0057] right Perform time series restoration to obtain the predicted value of the flexible riser motion response at the target time.
[0058] The advantages and beneficial effects of the present invention over the prior art are:
[0059] 1. The flexible riser strain data is decomposed using the EMD algorithm to obtain independent modal data sets, and key features such as natural modal frequency, damping ratio, and modal morphology are extracted. The powerful time series modeling capabilities of the LSTM network are also applied to accurately model and predict the decomposed modal components. This more comprehensively and accurately reflects the dynamic characteristics of the flexible riser, provides rich information for the model, and significantly improves prediction accuracy.
[0060] 2. The EMD algorithm decomposes complex signals into multiple relatively simple modal components, reducing signal complexity. When building the EMD-LSTM model, each sub-model only needs to model and predict a single modal component, significantly reducing model complexity and the number of parameters. This makes it easier to find the global optimal solution during training and avoid being trapped in a local optimum.
[0061] 3. The EMD algorithm can decompose the flexible riser strain data into multiple intrinsic mode functions (IMFs). Each IMF corresponds to a component in a different frequency range in the signal, enabling the model to model and predict different frequency components separately, thereby reducing the sensitivity of the prediction results to frequency domain changes and improving the stability of the prediction.
[0062] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] Figure 1 Schematic diagram of the steps of a method for predicting motion response of a marine flexible riser based on EMD-LSTM in an embodiment of the present invention;
[0064] Figure 2 This is a flow chart of a method for predicting motion response of a marine flexible riser based on EMD-LSTM in an embodiment of the present invention;
[0065] Figure 3 is a schematic diagram of the change of each IMF component over time in an embodiment of the present invention;
[0066] Figure 4 Schematic diagram comparing the predicted results and actual results of the flexible riser motion response according to an embodiment of the present invention. DETAILED DESCRIPTION
[0067] In the description of the present invention, it should be noted that the terms "upper", "lower", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, or the orientations or positional relationships in which the inventive product is usually placed when in use. These are only for the convenience of describing the present invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on the present invention. In the description of the present invention, it should also be noted that, unless otherwise expressly specified and limited, the terms "setting", "installation" and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the internal parts of two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0068] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0069] like Figure 1-Figure 2 As shown, the present invention provides a method for predicting the motion response of a marine flexible riser based on EMD-LSTM, comprising the following steps:
[0070] S1. Obtain historical flexible riser strain data from the offshore platform structural health monitoring system, and perform outlier cleaning preprocessing on the historical flexible riser discrete point strain data to obtain a flexible riser strain data group.
[0071] S2. Based on the empirical mode decomposition (EMD) algorithm, the dynamic response of the flexible riser strain data group is decomposed to obtain an independent modal data group. Feature extraction is then performed on the independent modal data group to obtain the natural modal frequency, damping ratio, and modal shape information of the flexible riser.
[0072] S3. Construct an EMD-LSTM model based on the natural modal frequency, damping ratio and modal shape information of the flexible riser and the independent modal data group.
[0073] S4. Input the current flexible riser strain data obtained from the offshore platform structural health monitoring system into the EMD-LSTM model to obtain the IMF prediction value output by each sub-model of the EMD-LSTM model, and linearly superimpose the IMF prediction value output by each sub-model to obtain the flexible riser motion response prediction value.
[0074] The advantages and beneficial effects of the present invention over the prior art are:
[0075] 1. The flexible riser strain data is decomposed using the EMD algorithm to obtain independent modal data sets, and key features such as natural modal frequency, damping ratio, and modal morphology are extracted. The powerful time series modeling capabilities of the LSTM network are also applied to accurately model and predict the decomposed modal components. This more comprehensively and accurately reflects the dynamic characteristics of the flexible riser, provides rich information for the model, and significantly improves prediction accuracy.
[0076] 2. The EMD algorithm decomposes complex signals into multiple relatively simple modal components, reducing signal complexity. When building the EMD-LSTM model, each sub-model only needs to model and predict a single modal component, significantly reducing model complexity and the number of parameters. This makes it easier to find the global optimal solution during training and avoid being trapped in a local optimum.
[0077] 3. The EMD algorithm can decompose the flexible riser strain data into multiple intrinsic mode functions (IMFs). Each IMF corresponds to a component in a different frequency range in the signal, enabling the model to model and predict different frequency components separately, thereby reducing the sensitivity of the prediction results to frequency domain changes and improving the stability of the prediction.
[0078] In some embodiments of the present application, in S1, obtaining historical flexible riser strain data in an offshore platform structural health monitoring system and performing outlier cleaning preprocessing on the historical flexible riser discrete point strain data includes:
[0079] S11. Obtaining discrete point strain data of a flexible riser in an offshore platform structural health monitoring system measured by an optical fiber strain sensor or a strain gauge;
[0080] S12. Use the Z-score algorithm or IQR algorithm to detect and eliminate abnormal data points in the flexible riser discrete point strain data, and use interpolation to fill in missing data points to ensure the continuity and integrity of the input data and prevent abnormal points from affecting subsequent analysis;
[0081] S13. Convert the strain data of the discrete points of the flexible riser into the displacement data of the discrete points of the flexible riser. The conversion expression is:
[0082] u(x)=∫ε(x)dx+C;
[0083] Where u(x) is the displacement data of the discrete points of the flexible riser, ε(x) is the strain data of the discrete points of the flexible riser, and C is the integral constant obtained based on the boundary conditions;
[0084] S14. In order to effectively eliminate the interference of high-frequency noise on model training while retaining the mutation and non-stationary characteristics of the flexible riser motion, the present invention adopts wavelet transform for denoising, including selecting appropriate wavelet basis functions, performing multi-layer wavelet decomposition on the original response sequence, and separating displacement components of different frequency bands; applying a threshold denoising algorithm to the high-frequency detail components to remove high-frequency noise; and performing wavelet reconstruction on the processed components to obtain a smooth, noise-free motion response.
[0085] In some embodiments of the present application, in S14, wavelet transform is used to perform denoising on the flexible riser discrete point displacement data, and the obtained flexible riser strain data group includes:
[0086] S141, performing multi-layer wavelet decomposition on the original response sequence using wavelet basis functions to obtain displacement components in different frequency bands;
[0087] S142, using a threshold denoising algorithm to remove noise from high-frequency detail components in displacement components of different frequency bands to obtain high-frequency denoised components;
[0088] S143. Use wavelet reconstruction method to process the high-frequency denoising component to obtain a flexible riser strain data group.
[0089] In some embodiments of the present application, in S2, the flexible riser strain data set is subjected to dynamic response decomposition based on an empirical mode decomposition (EMD) algorithm to obtain an independent modal data set including:
[0090] Decomposing the dynamic response of a flexible riser into multiple independent modes facilitates analysis of its dynamic characteristics under wave, current, and internal fluid loads. Extracting the natural modal frequencies, damping ratios, and modal shapes of the riser allows for assessment of fatigue life, vibration resistance, and fault location. When determining the IMF components, the following two conditions must be met:
[0091] (1) Within the entire range of the flexible riser motion response, the number of zero-crossing points and extreme points is equal or differs by at most one.
[0092] (2) At any point, the average value of the envelope of the local maximum and the envelope of the local minimum is zero.
[0093] S21. Analyze and obtain the local maximum and minimum values of the extreme points of the flexible riser strain data set. Use interpolation to connect the local maximum and minimum values to obtain the upper envelope U(t) and lower envelope L(t), and calculate the local average value. The calculation formula is:
[0094]
[0095] Where U(t) is the upper envelope, L(t) is the lower envelope, and m1(t) is the local average;
[0096] S22. Extract the candidate value of the first IMF component to be decomposed. The expression is:
[0097] h1(t)=x(t)-m1(t);
[0098] Where x(t) is the displacement component and h1(t) is the candidate value of the first IMF component;
[0099] S23, based on the relationship between h1(t) and the IMF preset condition, calculate the displacement component c1(t) of the first IMF;
[0100] S24. Remove the first IMF component from the original displacement data to obtain the residual component r1(t), which is expressed as:
[0101] r1(t)=x(t)-c1(t);
[0102] Where c1(t) is the displacement component of the first IMF, and r1(t) is the residual component;
[0103] S25. Use the residual component r1(t) as the new displacement data and repeat steps S21-S24 until the residual component becomes a monotonic function. The decomposition result is:
[0104]
[0105] Among them, c i (t) is the displacement component of the i-th IMF, r n (t) is the remainder of the residual component, and x(t) is the displacement component;
[0106] An independent modal data set is obtained based on x(t).
[0107] In some embodiments of the present application, in S23, based on the relationship between h1(t) and the IMF preset condition, calculating the displacement component c1(t) of the first IMF includes:
[0108] Determine whether h1(t) meets the IMF preset conditions. If h1(t) meets the IMF preset conditions, obtain the first IMF displacement component c1(t). If h1(t) does not meet the IMF preset conditions, repeat steps S21-S22 to continue decomposing h1(t) until it meets the IMF preset conditions, and obtain the first IMF displacement component c1(t).
[0109] IMF preset conditions include: the number of zero crossings is the same as the number of extreme values and the local mean is zero.
[0110] In some embodiments of the present application, in S3, constructing an EMD-LSTM model based on the natural modal frequency, damping ratio, and modal shape information of the flexible riser and the independent modal data set includes:
[0111] In order to accurately predict the motion response of flexible risers in complex marine environments, the present invention constructs independent LSTM models for each IMF component. The input of each LSTM model is the corresponding IMF component. In each IMF component, the first 60% of the values are selected as the training set, the middle 20% as the validation set, and the last 20% as the test set. Each LSTM model uses several memory units and fully connected layers, and uses the historical information of the time series for state updates and future predictions. Each model uses the training set for supervised learning, and the loss function uses the root mean square error (RMSE). The accuracy of the prediction is verified by RMSE, mean absolute error (MAE), symmetric mean absolute percentage error (SMAPE), and variance (R2).
[0112] (1) Model structure
[0113] Each LSTM model uses a three-layer LSTM unit with a memory and forgetting mechanism, which can effectively extract the long-term and short-term dependencies of the time series; each LSTM layer consists of 64 hidden units; a Dropout layer is set with a parameter of 0.2; the last layer is a fully connected output layer, which outputs the predicted value of the current IMF component.
[0114] (2) Training methods
[0115] All models were trained using supervised learning; the root mean square error (RMSE) was used as the loss function. Adam was used as the optimizer; the training set: validation set: test set ratio was 6:2:2, and the number of training rounds was adjusted dynamically based on model convergence.
[0116] The independent modal data group is split into multiple IMF components, and an LSTM model is constructed for each IMF component;
[0117] The LSTM model of each IMF component was trained using supervised learning. The root mean square error (RMS) loss function and Adam optimizer were used. The ratio of training set, validation set, and test set was 6:2:2. The number of training rounds was adjusted based on the model convergence.
[0118] The EMD-LSTM model is obtained based on the trained LSTM model of each IMF component.
[0119] In some embodiments of the present application, in S4, the current flexible riser strain data obtained in the offshore platform structural health monitoring system is input into the EMD-LSTM model, and the IMF prediction value output by each sub-model of the EMD-LSTM model is obtained, including:
[0120] S41. Obtain current flexible riser strain data in the offshore platform structural health monitoring system;
[0121] S42. The EMD-LSTM model includes multiple LSTM sub-models. Each LSTM sub-model corresponds to an IMF component. Assuming there are k LSTM sub-models and k corresponding IMF components, the calculation formula for the IMF prediction value is:
[0122]
[0123] in, represents the predicted value of the kth IMF component at time t+1.
[0124] In some embodiments of the present application, in S4, linearly superimposing the IMF prediction values output by each sub-model to obtain the flexible riser motion response prediction value includes:
[0125] S43. Based on the EMD reconstruction principle, the IMF prediction values output by each sub-model are linearly superimposed to obtain the flexible riser motion response prediction value. The calculation formula is:
[0126]
[0127] in, is the predicted value of the flexible riser motion response at time t+1, is the estimated value of the EMD residual term.
[0128] In some embodiments of the present application, S4 further includes:
[0129] S44. In order to improve the reconstruction accuracy and the robustness of the model, the weighted fusion strategy is used to optimize the modal fusion process. The weighted fusion strategy is used to optimize S43. The expression is:
[0130]
[0131] Among them, w i is the fusion weight of the i-th IMF component, is the predicted value of the flexible riser motion response obtained by weighted fusion;
[0132] right Perform time series restoration to obtain the predicted value of the flexible riser motion response at the target time.
[0133] The beneficial effects of the present invention are as follows: In summary, the present invention proposes a method for predicting the motion response of marine flexible risers based on long short-term memory neural networks (LSTM). Aiming at the nonlinear and non-stationary characteristics of the flexible riser response in the marine environment, the empirical mode decomposition (EMD) and multiple LSTM models are combined to finely model the modal characteristics of each order. By introducing data preprocessing strategies such as wavelet transform denoising, outlier processing and standardization, the quality of model input and prediction accuracy are effectively improved; by adopting the method of independent prediction and fusion reconstruction of multiple sub-models, a high-precision prediction of the future response trend of the flexible riser is achieved. Experimental verification shows that this method has good generalization ability and engineering adaptability, can provide reliable technical support for marine structure safety assessment and health monitoring, and has broad practical application prospects and promotion value.
[0134] The following describes the implementation of the present invention in detail with reference to specific examples.
[0135] Step 1: Data acquisition and preprocessing
[0136] The strain monitoring data of discrete points on the dynamic flexible riser are obtained by sensors arranged on the riser. The Z-score method is used to process the abnormal values of the monitoring data. The results are shown in Table 1.
[0137] Table 1 Strain data of discrete points of flexible riser
[0138]
[0139] Through the relationship between strain and displacement, the strain of discrete points of the flexible riser is converted into its displacement. The results are shown in Table 2.
[0140] Table 2 Calculation results of the displacement of discrete points of flexible risers
[0141]
[0142]
[0143] Step 2: Empirical mode decomposition, taking the data collected by sensor 57 as an example, the displacement data of the discrete point is decomposed into 8 intrinsic mode function (IMF) components by the empirical mode decomposition method, as shown in Table 3. The specific IMF components change with time as shown in Figure 3 shown.
[0144] Table 3 IMF components
[0145]
[0146] Step 3: LSTM-based prediction of the flexible riser motion response. Using the data collected by sensor 57 as an example, a separate LSTM model was constructed for each IMF component (as shown in Table 3). For each IMF component, the first 60% of the values were selected as the training set, the middle 20% as the validation set, and the last 20% as the test set for analysis.
[0147] Step 4: Use the trained LSTM model to perform time series prediction on each IMF component and linearly superimpose the prediction results to obtain the predicted value of the overall motion response of the flexible riser. The actual value and predicted value of the motion response of the flexible riser are shown in Table 4. The prediction results of the overall motion response of the flexible riser are shown in Table 4. Figure 4 shown.
[0148] Table 4 Actual and predicted values of flexible riser motion response
[0149] Time(s) Actual displacement (m) Predicted displacement (m) 143.110 -0.3888 -0.3789 143.130 -0.3894 -0.3788 143.149 -0.3910 -0.3801 143.169 -0.3933 -0.3822 143.189 -0.3950 -0.3840 143.209 -0.3960 -0.3846 143.229 -0.3966 -0.3847 143.248 -0.3961 -0.3850 143.268 -0.3942 -0.3837 143.288 -0.3919 -0.3811 …… …… ……
[0150] The prediction accuracy of the model is evaluated by the root mean square error (RMSE), mean absolute error (MAE), symmetric mean absolute percentage error (SMAPE), and variance (R2), and the results are shown in Table 5. As shown in Table 5, the MAE is 0.0062, indicating that the model has high accuracy in predicting the motion response of the flexible riser.
[0151] Table 5 Error analysis of prediction results
[0152]
[0153]
[0154] In this application, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. In the event of any inconsistency, the meaning described in this specification or the meaning derived from the contents recorded in this specification shall prevail. In addition, the terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.
[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for predicting the motion response of marine flexible riser based on EMD-LSTM, characterized in that: The following steps are involved: S1. Obtain historical flexible riser strain data from the offshore platform structural health monitoring system, and perform outlier cleaning preprocessing on the historical flexible riser discrete point strain data to obtain a flexible riser strain data group; S2. Decomposing the flexible riser strain data set dynamically based on the empirical mode decomposition (EMD) algorithm to obtain an independent modal data set, and extracting features from the independent modal data set to obtain the natural modal frequency, damping ratio, and modal shape information of the flexible riser; S3, constructing an EMD-LSTM model based on the natural modal frequency, damping ratio and modal shape information of the flexible riser and the independent modal data set; S4. Input the current flexible riser strain data obtained from the offshore platform structural health monitoring system into the EMD-LSTM model to obtain the IMF prediction value output by each sub-model of the EMD-LSTM model, and linearly superimpose the IMF prediction value output by each sub-model to obtain the flexible riser motion response prediction value.
2. The method for predicting motion response of a marine flexible riser based on EMD-LSTM according to claim 1, characterized in that: In S1, obtaining historical flexible riser strain data in the offshore platform structural health monitoring system and performing outlier cleaning preprocessing on the historical flexible riser discrete point strain data include: S11. Obtaining discrete point strain data of a flexible riser in an offshore platform structural health monitoring system measured by an optical fiber strain sensor or a strain gauge; S12. Use the Z-score algorithm or the IQR algorithm to detect and eliminate abnormal data points in the flexible riser discrete point strain data, and use the interpolation method to fill in the missing data points; S13. Convert the strain data of the discrete points of the flexible riser into the displacement data of the discrete points of the flexible riser. The conversion expression is: u(x)=∫ε(x)dx+C; Where u(x) is the displacement data of the discrete points of the flexible riser, ε(x) is the strain data of the discrete points of the flexible riser, and C is the integral constant obtained based on the boundary conditions; S14. Using wavelet transform to perform denoising on the displacement data of discrete points of the flexible riser to obtain a flexible riser strain data group.
3. The method for predicting motion response of a marine flexible riser based on EMD-LSTM according to claim 2, characterized in that: In S14, wavelet transform is used to perform denoising on the flexible riser discrete point displacement data, and the obtained flexible riser strain data group includes: S141, performing multi-layer wavelet decomposition on the original response sequence using wavelet basis functions to obtain displacement components in different frequency bands; S142, using a threshold denoising algorithm to remove noise from high-frequency detail components in displacement components of different frequency bands to obtain high-frequency denoised components; S143. Use wavelet reconstruction method to process the high-frequency denoising component to obtain a flexible riser strain data group.
4. The method for predicting motion response of a marine flexible riser based on EMD-LSTM according to claim 3, characterized in that: In S2, the flexible riser strain data set is dynamically decomposed based on the empirical mode decomposition (EMD) algorithm to obtain an independent modal data set including: S21. Analyze and obtain the local maximum and minimum values of the extreme points of the flexible riser strain data set. Use interpolation to connect the local maximum and minimum values to obtain the upper envelope U(t) and lower envelope L(t), and calculate the local average value. The calculation formula is: Where U(t) is the upper envelope, L(t) is the lower envelope, and m1(t) is the local average; S22. Extract the candidate value of the first IMF component to be decomposed. The expression is: h1(t)=x(t)-m1(t); Where x(t) is the displacement component and h1(t) is the candidate value of the first IMF component; S23, based on the relationship between h1(t) and the IMF preset condition, calculate the displacement component c1(t) of the first IMF; S24. Remove the first IMF component from the original displacement data to obtain the residual component r1(t), which is expressed as: r1(t)=x(t)-c1(t); Where c1(t) is the displacement component of the first IMF, and r1(t) is the residual component; S25. Use the residual component r1(t) as the new displacement data and repeat steps S21-S24 until the residual component becomes a monotonic function. The decomposition result is: Among them, c i (t) is the displacement component of the i-th IMF, r n (t) is the remainder of the residual component, and x(t) is the displacement component; Based on x(t), independent modal data sets are obtained.
5. The method for predicting motion response of a marine flexible riser based on EMD-LSTM according to claim 4, characterized in that: In S23, based on the relationship between h1(t) and the IMF preset condition, the displacement component c1(t) of the first IMF is calculated, which includes: Determine whether h1(t) meets the IMF preset conditions. If h1(t) meets the IMF preset conditions, obtain the first IMF displacement component c1(t). If h1(t) does not meet the IMF preset conditions, repeat steps S21-S22 to continue decomposing h1(t) until it meets the IMF preset conditions, and obtain the first IMF displacement component c1(t). IMF preset conditions include: the number of zero crossings is the same as the number of extreme values and the local mean is zero.
6. The method for predicting motion response of a marine flexible riser based on EMD-LSTM according to claim 5, characterized in that: In S3, constructing an EMD-LSTM model based on the natural modal frequency, damping ratio, and modal shape information of the flexible riser and the independent modal data set includes: The independent modal data group is split into multiple IMF components, and an LSTM model is constructed for each IMF component; The LSTM model of each IMF component was trained using supervised learning. The root mean square error (RMS) loss function and Adam optimizer were used. The ratio of training set, validation set, and test set was 6:2:
2. The number of training rounds was adjusted based on the model convergence. The EMD-LSTM model is obtained based on the trained LSTM model of each IMF component.
7. The method for predicting motion response of a marine flexible riser based on EMD-LSTM according to claim 6, characterized in that: In S4, the current flexible riser strain data obtained in the offshore platform structural health monitoring system is input into the EMD-LSTM model to obtain the IMF prediction value output by each sub-model of the EMD-LSTM model, including: S41. Obtain current flexible riser strain data in the offshore platform structural health monitoring system; S42. The EMD-LSTM model includes multiple LSTM sub-models. Each LSTM sub-model corresponds to an IMF component. Assuming there are k LSTM sub-models and k corresponding IMF components, the calculation formula for the IMF prediction value is: in, represents the predicted value of the kth IMF component at time t+1.
8. The method for predicting motion response of a marine flexible riser based on EMD-LSTM according to claim 7, characterized in that: In S4, the IMF prediction values output by each sub-model are linearly superimposed to obtain the flexible riser motion response prediction value including: S43. Based on the EMD reconstruction principle, the IMF prediction values output by each sub-model are linearly superimposed to obtain the flexible riser motion response prediction value. The calculation formula is: in, is the predicted value of the flexible riser motion response at time t+1, is the estimated value of the EMD residual term.
9. The method for predicting motion response of a marine flexible riser based on EMD-LSTM according to claim 8, characterized in that: Said S4 further includes: S44, using weighted fusion strategy to optimize S43, the expression is: Among them, w i is the fusion weight of the i-th IMF component, is the predicted value of the flexible riser motion response obtained by weighted fusion; right Perform time series restoration to obtain the predicted value of the flexible riser motion response at the target time.