Method for predicting cathode protection potential of FPSO (floating production storage and offloading) riser support structure

Abstract

The invention belongs to the technical field of corrosion protection and prediction of ocean engineering structures, and particularly relates to a method for predicting the cathode protection potential of an FPSO stand pipe supporting structure. According to the prediction method, the advantages of the seasonal autoregression integral moving average model and the long short-term memory neural network are combined, collaborative modeling and high-precision prediction of linear and nonlinear characteristics in the potential data are achieved, the prediction precision is high, and the prediction result robustness is high. The method for predicting the cathode protection potential of the FPSO riser support structure comprises the following steps: collecting historical potential time sequence data of the FPSO riser support structure under the condition of external cathode protection potential, and preprocessing the historical potential time sequence data; constructing and training a seasonal autoregressive integral moving average model; constructing and training a long-short-term memory neural network model; and performing combined prediction on the target time period to obtain a prediction result of the target time period of the cathode protection potential of the FPSO riser support structure.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of corrosion protection and prediction of marine engineering structures, and particularly relates to a cathodic protection potential prediction method for an FPSO riser support structure. BACKGROUND

[0002] FPSO riser support structures are long-term exposed to a marine environment with high salinity, high humidity and strong electrochemical corrosion, and the corrosion protection effect thereof is directly related to the safety and service life of the platform structure. Among them, impressed current cathodic protection (ICCP) is the main anticorrosion means for protecting such structures. However, due to uneven current distribution, shielding effect and other factors, local areas of the cathodic protection may be under-protected or over-protected.

[0003] In view of the above, the prior art mainly uses the following two methods to predict the impressed current cathodic protection potential: one is numerical simulation based on electrochemical theory, and the other is a statistical method relying on field experience. However, the above techniques have certain technical limitations, for example: the numerical simulation process is complex and time-consuming, and it is difficult to meet the needs of real-time prediction; and the accuracy of the experience statistical method is limited, and it is difficult to effectively describe the nonlinear dynamic characteristics of the system.

[0004] In addition, technical personnel have designed and developed various prediction models in the time series prediction field to help predict the impressed current cathodic protection potential. However, the inventors have found in further research that for the specific scene of FPSO riser support structure ICCP potential prediction, the related technology still has obvious shortcomings. For example, although the existing time series model can handle seasonal characteristics, its linear assumption cannot describe complex nonlinear processes; deep learning models such as LSTM models have strong nonlinear fitting capabilities, but they often have difficulty accurately capturing the inherent seasonal periodicity of potential data.

[0005] Therefore, it is urgent for technical personnel in the field to further provide a new and reliable FPSO riser support structure cathodic protection potential prediction method to overcome the complex time series problems existing in the FPSO riser support structure potential prediction process. SUMMARY

[0006] The application provides an FPSO riser support structure cathodic protection potential prediction method. The FPSO riser support structure cathodic protection potential prediction method combines the advantages of the seasonal autoregressive integrated moving average model and the long short-term memory neural network, realizes collaborative modeling and high-precision prediction of linear and nonlinear characteristics in potential data, has high prediction accuracy, and the robustness of the prediction result is strong.

[0007] To solve the above technical problems, the application adopts the following technical solutions: A FPSO riser support structure cathodic protection potential prediction method, comprising the following steps: Step S1: Collecting historical potential time series data of the FPSO riser support structure under the condition of an applied cathodic protection potential, and pre-processing the historical potential time series data; Step S2: Building and training a seasonal autoregressive integrated moving average model; Step S3: Building and training a long short-term memory neural network model; Step S4: Combining the prediction of the FPSO riser support structure cathodic protection potential target period to obtain the prediction result of the FPSO riser support structure cathodic protection potential target period.

[0008] More preferably, the pre-processing process of the historical potential time series data in step S1 specifically comprises the following steps: Step S11: Using ADF unit root test method to test the stationarity of the historical potential time series data; Step S12: First-order difference processing of the original sequence obtained after stationarity test; Step S13: ADF test on the sequence obtained after first-order difference processing to ensure that the first-order difference sequence is converted into a stationary sequence; Step S14: Drawing autocorrelation and partial autocorrelation graphs of the stationary sequence; Step S15: Dividing the stationary sequence into training set and test set according to the ratio of 8:2.

[0009] More preferably, the process of building and training a seasonal autoregressive integrated moving average model in step S2 is specifically described as follows: Using BIC information criterion to optimize the model parameters (p, d, q) x (P, D, Q) s of the seasonal autoregressive integrated moving average model; Based on the optimization result, the training set data is used for fitting to obtain the prediction value and residual sequence.

[0010] More preferably, the process of building and training a long short-term memory neural network model in step S3 is specifically described as follows: Normalizing the residual sequence for pre-processing; Taking the residual sequence as input to build a long short-term memory neural network model; Using Adam optimizer to train the long short-term memory neural network model, taking mean square error as loss function, and iterating and optimizing network parameters through back propagation algorithm to complete the training of the long short-term memory neural network model.

[0011] More preferably, the step S4 combines the prediction of the FPSO riser support structure cathodic protection potential target period to obtain the prediction result of the FPSO riser support structure cathodic protection potential target period, specifically comprising the following steps: Step S41: using a seasonal autoregressive integrated moving average model, a linear prediction component of the FPSO riser support structure cathodic protection potential target period is predicted; Step S42: using a long short-term memory neural network model, a nonlinear residual prediction component of the FPSO riser support structure cathodic protection potential target period is predicted; Step S43: superimposing the linear prediction component of the FPSO riser support structure cathodic protection potential target period and the nonlinear residual prediction component of the FPSO riser support structure cathodic protection potential target period, to obtain the prediction result of the FPSO riser support structure cathodic protection potential target period.

[0012] The present application provides a kind of FPSO riser support structure cathodic protection potential prediction method.Therein, the FPSO riser support structure cathodic protection potential prediction method specifically includes the following steps: step S1: collect the historical potential time series data of FPSO riser support structure under the condition of impressed cathodic protection potential, and the historical potential time series data is preprocessed;Step S2: build and train seasonal autoregressive integrated moving average model;Step S3: build and train long short-term memory neural network model;Step S4: the FPSO riser support structure cathodic protection potential target period is combined and predicted, to obtain the prediction result of the FPSO riser support structure cathodic protection potential target period.

[0013] The FPSO riser support structure cathodic protection potential prediction method with the above step characteristics has at least the following technical advantages compared with the prior art: (1), high prediction accuracy: through the data fusion of seasonal autoregressive integrated moving average model and long short-term memory neural network model, the comprehensive fitting ability of each model is significantly improved. Experimental results show that the average absolute percentage error (MAPE) of the FPSO riser support structure cathodic protection potential prediction method provided by the present application is about 1.08%, which is significantly lower than the error of the existing single model; (2), strong prediction stability: the FPSO riser support structure cathodic protection potential prediction method provided by the present application combines the FPSO riser support structure cathodic protection potential in the whole cycle to improve the fitting degree between the prediction curve and the actual value, the error fluctuation is small, and the prediction result has strong robustness and generalization ability; (3) High practicality: The cathodic protection potential prediction method for FPSO riser support structure provided by this invention provides more reliable data support for corrosion protection and intelligent operation and maintenance of FPSO marine platforms, which helps to achieve preventive maintenance and reduce operational risks and costs. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the following drawings: Figure 1 A flowchart illustrating a method for predicting the cathodic protection potential of an FPSO riser support structure provided by the present invention; Figure 2 A schematic diagram of the original sequence sample of historical potential time series data of FPSO riser support structure under applied cathodic protection potential conditions; Figure 3 To Figure 2 The diagram shows the sequence after first-order difference processing of the historical potential time series data. Figure 4a This is a schematic diagram of the autocorrelation plot (ACF) of a stationary series. Figure 4b A schematic diagram of the partial autocorrelation plot (PACF) of a stationary series; Figure 5 This is a diagram showing the difference between the actual and predicted values ​​when only the seasonal autoregressive integral moving average model is considered. Figure 6 This is a diagram showing the difference between the actual and predicted values ​​when only considering the Long Short-Term Memory neural network model. Figure 7 This is a diagram showing the difference between the actual and predicted values ​​after superimposing the linear prediction component and the nonlinear residual prediction component. Figure 8 For comprehensive evaluation Figure 5 , Figure 6 , Figure 7 The diagram shows a comparison of the differences between the predicted curves and the actual values ​​for the three prediction results. Detailed Implementation

[0015] This invention provides a method for predicting the cathodic protection potential of an FPSO riser support structure. This method combines the advantages of a seasonal autoregressive integral moving average model and a long short-term memory neural network to achieve collaborative modeling and high-precision prediction of linear and nonlinear characteristics in the potential data, resulting in high prediction accuracy and strong robustness.

[0016] like Figure 1As shown, the FPSO riser support structure cathodic protection potential prediction method provided by the present application specifically comprises the following steps: Step S1: Collecting historical potential time series data of the FPSO riser support structure under the condition of an applied cathodic protection potential, and pre-processing the historical potential time series data.

[0017] First, the historical potential time series data of the FPSO riser support structure under the condition of an applied cathodic protection potential is collected. Specifically, as a specific embodiment, a set of historical potential time series data is provided herein, in which 360 sample points are obtained, and the original sequence can be referred to as Figure 2 As shown, the annual periodicity (i.e. the seasonal parameter s=12) is obvious.

[0018] Then, the historical potential time series data is further pre-processed. As a more preferred embodiment of the present application, the process of pre-processing the historical potential time series data specifically comprises the following steps: Step S11: Using the ADF unit root test method to perform stationarity test on the historical potential time series data.

[0019] It is worth noting that, for the original historical potential time series data shown in Figure 2 , it can be found that the test result shows that the p-value is about 0.2963 (greater than the significance level of 0.05), so it is determined that the historical potential time series data is a non-stationary sequence.

[0020] Step S12: First-order difference processing is performed on the original sequence obtained after the stationarity test.

[0021] Step S13: ADF test is performed again on the sequence obtained after the first-order difference processing to ensure that the first-order difference sequence is converted into a stationary sequence.

[0022] Specifically, first-order difference processing is performed on the above historical potential time series data, and the sequence after the first-order difference processing can be referred to as Figure 3 As shown. Then, ADF test is performed again on the sequence obtained after the first-order difference processing, and it can be found that the p-value becomes 0.001 (less than 0.05), indicating that the sequence after the first-order difference processing has been converted into a stationary sequence.

[0023] Step S14: Drawing the autocorrelation graph and partial autocorrelation graph of the stationary sequence.

[0024] On the basis of completing steps S11-S13, the autocorrelation graph (ACF) and the partial autocorrelation graph (PACF) of the stationary sequence are further drawn, as shown in Figure 4a , Figure 4b It can be found that the tailing characteristics meet the modeling conditions of the seasonal autoregressive integrated moving average model.

[0025] Step S15: dividing the stationary sequence into a training set and a test set in a ratio of 8:2.

[0026] Specifically, the stationary sequence is divided into a training set and a test set. The division ratio is implemented in a ratio of 8:2. That is, the first 288 data points are used for the training set for model training, and the last 72 data points are used for the test set for testing and verifying the prediction performance of the model.

[0027] Step S2: constructing and training a seasonal autoregressive integrated moving average model.

[0028] On the basis of completing step S1, step S2 is further implemented. As a more preferred embodiment of the present application, the process of constructing and training a seasonal autoregressive integrated moving average model in step S2 is specifically described as follows: The BIC information criterion is used to optimize the model parameters (p, d, q) x (P, D, Q) s of the seasonal autoregressive integrated moving average model.

[0029] Specifically, the optimization range can be referred to as follows: the maximum values of p and q are set to 4, the maximum values of P and Q are set to 3, d = 1, D = 1, and s = 12. The BIC information criterion is used for determination, and the optimal seasonal autoregressive integrated moving average model structure is SARIMA(1, 1, 2) x (3, 1, 2) 12.

[0030] Based on the optimization result, the training set data is used for fitting to obtain a prediction value and a residual sequence.

[0031] On the basis of completing the above optimization operation, the optimization result is fitted to obtain a corresponding prediction value and calculate a residual sequence, i.e., a difference between the true value and the prediction value. The optimization result is obtained by comparing the prediction result with the true value on the test set, which can be referred to as shown in Figure 5 Specifically, the prediction evaluation index is: RMSE = 0.0125, MAE = 0.0101, and MAPE = 1.33%.

[0032] Step S3: constructing and training a long short-term memory neural network model.

[0033] On the basis of completing step S2, step S3 is further implemented. As a more preferred embodiment of the present application, the process of constructing and training a long short-term memory neural network model in step S3 is specifically described as follows: First, the residual sequence is normalized and preprocessed.

[0034] It is worth noting that the long short-term memory neural network model is constructed to capture the nonlinear fluctuation components that cannot be explained by the seasonal autoregressive integrated moving average model in step S2.

[0035] Then, the residual sequence is used as input to construct a long short-term memory neural network model.

[0036] The long short-term memory neural network model has a three-layer structure, and the key hyperparameters are set as follows: the input sequence length is set to 60, the number of hidden layer units is set to 200, and the Dropout ratio is set to 0.2.

[0037] Finally, the long short-term memory neural network model is trained using the Adam optimizer, with mean square error as the loss function, and the network parameters are iteratively optimized by the backpropagation algorithm to complete the training of the long short-term memory neural network model.

[0038] Specifically, the long short-term memory neural network model is trained using the Adam optimizer, with an initial learning rate of 0.001, mean square error as the loss function, and the network parameters are iteratively optimized by the backpropagation algorithm. After training the long short-term memory neural network model, the residual in the test set period is used for prediction, and the prediction results are de-normalized to obtain the final prediction value of the nonlinear residual component. As a comparison, the difference between the prediction value and the true value of the original potential sequence predicted by the long short-term memory neural network model alone is shown in FIG. Figure 6 The prediction evaluation indicators are RMSE = 0.0288, MAE = 0.0237, and MAPE = 3.12%.

[0039] Step S4: Combining the prediction of the FPSO riser support structure cathodic protection potential target period to obtain the prediction result of the FPSO riser support structure cathodic protection potential target period.

[0040] Based on the completion of steps S1-S3, step S4 is further implemented. As a more preferred embodiment of the present application, step S4 specifically includes the following steps: Step S41: Using the seasonal autoregressive integrated moving average model to predict the linear prediction component of the FPSO riser support structure cathodic protection potential target period.

[0041] Step S42: Using the long short-term memory neural network model to predict the nonlinear residual prediction component of the FPSO riser support structure cathodic protection potential target period.

[0042] Step S43: superimposing the linear prediction component of the FPSO riser support structure cathodic protection potential target period and the nonlinear residual prediction component of the FPSO riser support structure cathodic protection potential target period to obtain the prediction result of the FPSO riser support structure cathodic protection potential target period.

[0043] For the FPSO riser support structure cathodic protection potential target period, the difference between the predicted value and the true value of the final prediction result of the FPSO riser support structure cathodic protection potential target period can be obtained after superimposing the linear prediction component output in step S2 and the nonlinear residual prediction component output in step S3, which can be compared as shown in Figure 7 It is worth noting that compared with the prediction result using a single mode, the prediction method provided by the present application has the highest fitting degree of the prediction curve and the actual value and the smallest error fluctuation range. Further reference can be made to Figure 8 as shown. Figure 8 Figure 5 , Figure 6 , Figure 7 The comparison between the prediction curve and the true value can be found in the schematic diagram, which fully proves the comprehensive advantages of the FPSO riser support structure cathodic protection potential prediction method in capturing the linear trend, seasonal regularity and nonlinear dynamics of the potential sequence, and has the characteristics of high prediction accuracy and strong robustness.

[0044] The present application provides a FPSO riser support structure cathodic protection potential prediction method. Wherein, the FPSO riser support structure cathodic protection potential prediction method specifically comprises the following steps: step S1: collecting historical potential time series data of FPSO riser support structure under the condition of impressed cathodic protection potential, and preprocessing the historical potential time series data; step S2: constructing and training a seasonal autoregressive integrated moving average model; step S3: constructing and training a long short-term memory neural network model; step S4: combining prediction of FPSO riser support structure cathodic protection potential target period to obtain the prediction result of FPSO riser support structure cathodic protection potential target period.

[0045] The FPSO riser support structure cathodic protection potential prediction method with the above step characteristics has at least the following technical advantages compared with the prior art: (1) high prediction accuracy: through data fusion of the seasonal autoregressive integrated moving average model and the long short-term memory neural network model, the comprehensive fitting ability of each model is significantly improved. Experimental results show that the average absolute percentage error (MAPE) of the FPSO riser support structure cathodic protection potential prediction method provided by the present application is about 1.08%, which is significantly lower than the error of the existing single model; ​(2) Strong prediction stability: the FPSO riser support structure cathodic protection potential prediction method provided by the application improves the fitting degree between the prediction curve and the actual value by combining the FPSO riser support structure cathodic protection potential in the whole cycle, the error fluctuation is small, and the prediction result has strong robustness and generalization ability; (3) Strong practicability: the FPSO riser support structure cathodic protection potential prediction method provided by the application provides more reliable data support for the corrosion protection and intelligent operation and maintenance of the FPSO offshore platform, which helps to realize preventive maintenance and reduces the operation risk and cost.

[0046] The above is only a specific embodiment of the application, but the protection scope of the application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the application, which should be covered within the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A method of predicting cathodic protection potential for an FPSO riser support structure, characterized by, The method comprises the following steps: Step S1: collecting historical potential time series data of the FPSO riser support structure under the condition of an applied cathodic protection potential, and pre-processing the historical potential time series data; Step S2: constructing and training a seasonal autoregressive integrated moving average model; Step S3: constructing and training a long short-term memory neural network model; Step S4: combining prediction of the target period of the FPSO riser support structure cathodic protection potential to obtain the prediction result of the target period of the FPSO riser support structure cathodic protection potential.

2. The method of claim 1, wherein the FPSO riser support structure cathodic protection potential prediction method is characterized by, The pre-processing process of the historical potential time series data in step S1 comprises the following steps: Step S11: using ADF unit root test method to perform stationarity test on the historical potential time series data; Step S12: performing first-order difference processing on the original sequence obtained after the stationarity test; Step S13: performing ADF test again on the sequence obtained after the first-order difference processing to ensure that the first-order difference sequence is converted into a stationary sequence; Step S14: drawing autocorrelation and partial autocorrelation graphs of the stationary sequence; Step S15: dividing the stationary sequence into a training set and a test set according to a ratio of 8:

2.

3. The method of claim 1, wherein the FPSO riser support structure cathodic protection potential prediction method is characterized by, The process of constructing and training the seasonal autoregressive integrated moving average model in step S2 is specifically described as follows: Using BIC information criterion to optimize the model parameters (p, d, q) × (P, D, Q) s of the seasonal autoregressive integrated moving average model; Based on the optimization result, the training set data is used for fitting to obtain a prediction value and a residual sequence.

4. The method of claim 3, wherein the method further comprises: The process of constructing and training the long short-term memory neural network model in step S3 is specifically described as follows: Performing normalization preprocessing on the residual sequence; Taking the residual sequence as input, a long short-term memory neural network model is constructed; Using Adam optimizer to train the long short-term memory neural network model, taking mean square error as a loss function, and iteratively optimizing network parameters through a back propagation algorithm to complete the training of the long short-term memory neural network model.

5. The method of claim 4, wherein the method further comprises: The process of combining prediction of the target period of the FPSO riser support structure cathodic protection potential to obtain the prediction result of the target period of the FPSO riser support structure cathodic protection potential in step S4 comprises the following steps: Step S41: using the seasonal autoregressive integrated moving average model to predict a linear prediction component of the target period of the FPSO riser support structure cathodic protection potential; Step S42: using the long short-term memory neural network model to predict a nonlinear residual prediction component of the target period of the FPSO riser support structure cathodic protection potential; Step S43: superimposing the linear prediction component of the target period of the FPSO riser support structure cathodic protection potential and the nonlinear residual prediction component of the target period of the FPSO riser support structure cathodic protection potential to obtain the prediction result of the target period of the FPSO riser support structure cathodic protection potential.

Images