Underwater choke residual life prediction method based on adaptive near neighbor enhancement

Abstract

This application belongs to the field of marine oil and gas production equipment condition monitoring technology, specifically involving a method for predicting the remaining life of underwater throttle valves based on adaptive nearest neighbor enhancement. The method includes: S1 constructing an underwater throttle valve erosion and degradation database; S2 obtaining erosion and degradation status information of the current underwater throttle valve under test; S3 constructing erosion and degradation enhancement features; S4 establishing an erosion and degradation similarity evaluation model; S5 constructing a nearest neighbor sample enhancement training set; S6 establishing a two-layer random forest remaining life prediction model; and S7 constructing the remaining life prediction results. This invention can fully utilize the observed erosion and degradation status of the current throttle valve under test, improve the model's adaptability to the current valve's individual degradation characteristics, and alleviate prediction bias caused by inconsistencies between historical and current sample distributions.

Description

Technical Field

[0001] This application belongs to the field of marine oil and gas production equipment condition monitoring technology, specifically involving a method for predicting the remaining life of underwater throttle valves based on adaptive nearest neighbor enhancement. Background Technology

[0002] Submersible throttle valves are critical flow control components in offshore oil and gas production systems, typically operating in complex environments with high pressure, high flow velocity, and media containing sand and particles. During actual service, particles in the solid-liquid two-phase flow pass at high speed through the internal flow channels, valve core, valve seat, cage, or throttling orifice of the throttle valve, continuously impacting, cutting, and abrading the valve wall, causing localized erosion damage. As the erosion process progresses, the maximum erosion depth in critical internal components gradually increases, altering the flow channel structure and throttling characteristics. This can lead to decreased valve control accuracy, deteriorated sealing performance, and even the inability to withstand normal operating pressure after reaching the failure threshold. Therefore, identifying the erosion degradation state of submersible throttle valves and predicting their remaining service life is crucial for the safe operation, maintenance decisions, and optimized opening control of submersible production equipment.

[0003] In existing technologies, research on erosion problems in throttle valves mainly focuses on erosion rate calculation, erosion location identification, and short-term erosion characteristic analysis. For example, erosion rates or maximum erosion depths under different operating conditions such as valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, and mass flow rate can be obtained through numerical simulation, experimental testing, or historical operating data, and the current degree of erosion of the throttle valve can be determined accordingly. While these methods can reflect the instantaneous erosion state or short-term degradation trend of the throttle valve to some extent, throttle valve erosion is not a static wear problem at a single moment, but a long-term, cumulative, dynamically evolving, and significantly nonlinear degradation process. If the remaining life is calculated solely based on the current erosion rate or the erosion results under a fixed operating condition, the nonlinear characteristics of the erosion degradation process over time are easily overlooked, making it difficult to accurately reflect the true life evolution process of the valve from its current state to failure. Summary of the Invention

[0004] The technical problem this invention aims to solve is to address the shortcomings of existing methods for predicting the remaining service life of underwater throttle valves, such as insufficient consideration of dynamic nonlinearity in erosion degradation, difficulty in accurately characterizing the coupling relationship of multiple operating conditions, inadequate utilization of the real-time degradation state of the throttle valve under test, and insufficient adaptability of the prediction model to sample distribution shifts and local operating condition changes. This invention provides a method for predicting the remaining service life of underwater throttle valves based on adaptive nearest neighbor enhancement. The technical solution is as follows: A method for predicting the remaining life of an underwater throttle valve based on adaptive nearest neighbor enhancement includes the following steps: S1. Construct a database of underwater throttle valve erosion degradation; S2. Obtain information on the erosion and degradation status of the underwater throttle valve under test; S3. Construct erosion degradation enhancement features; Based on the operating parameters of the current underwater throttle valve under test and historical throttle valve samples, an enhanced feature is constructed to characterize the erosion degradation state. The enhancement features include one or more of the following: erosion energy density, aperture-velocity interaction term, and particle concentration-pressure drop ratio; The enhanced features, along with valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, mass flow rate, erosion time, erosion rate, and maximum erosion depth, are used as input features for the similarity evaluation model and the remaining life prediction model. S4. Establish a similarity evaluation model for erosion degradation; Erosion degradation sequences and corresponding operating parameters were extracted from historical life samples. A similarity evaluation model is established based on the difference between the current erosion degradation state sequence of the underwater throttle valve under test and historical life samples. The similarity evaluation model considers both the erosion degradation evolution law and the matching relationship of operating parameters. The overall similarity between the current underwater throttle valve under test and each historical lifespan sample is calculated using a similarity evaluation model. S5. Construct a training set to enhance nearest neighbor samples; S6. Establish a two-layer random forest remaining lifetime prediction model; Train the first-layer random forest model using an expanded training set; An initial mapping relationship between erosion degradation characteristics and remaining useful life is established using a first-layer random forest model. The prediction residuals are calculated based on the difference between the prediction results of the first-layer random forest model and the actual lifetime values. The predicted residuals are used to construct a second-layer random forest model; A mapping relationship between input features and prediction residuals is established using a second-layer random forest model; S7. Construct the remaining useful life prediction results.

[0005] Preferably, erosion degradation data of multiple historical underwater throttle valves are collected, including simulated erosion data and actual operational erosion data; For each historical underwater throttle valve, record its erosion degradation information throughout the entire process from initial operation to failure, and form a corresponding lifetime sample; The lifespan sample includes at least valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, mass flow rate, erosion time, erosion rate, maximum erosion depth, and corresponding remaining lifespan information. Construct a historical erosion degradation database; the historical erosion degradation database includes simulated erosion data and historical operational erosion data, and each historical throttle valve sample can be represented as: ; in, Indicates the first A sequence of erosion degradation of a historical throttling valve. Indicates the first The throttle valve in the first Erosion performance indicators for each monitoring cycle Indicates the first The lifespan of a historical throttle valve; the erosion performance indicators include erosion rate, maximum erosion depth, wear amount, or other parameters that can characterize the erosion degradation state of the valve.

[0006] Preferably, real-time operating parameters and erosion degradation parameters of the underwater throttle valve under test are collected during its operation. The real-time operating parameters include one or more of the following: valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, and mass flow rate. The erosion degradation parameters include erosion rate, erosion depth, wear amount, or other parameters that can characterize the erosion degradation state. Construct a sequence of erosion and degradation states of the underwater throttle valve under test based on time sequence; The erosion degradation sequence of the underwater throttle valve under test can be represented as follows: ; in, This represents the sequence of erosion degradation states that have been observed in the underwater throttle valve currently under test. This represents the erosion degradation state under the current monitoring period; the remaining service life prediction problem is expressed as: ; in, This indicates the remaining service life of the underwater throttle valve currently being tested. This represents a set of historical erosion and degradation samples. This represents the lifetime prediction mapping relationship determined jointly by historical samples and the current degradation state.

[0007] Preferably, the erosion energy density is used to characterize the impact energy of solid particles on the throttle valve wall, and its expression is as follows: ; In the formula, Indicates erosion energy density. Indicates fluid density, Indicates fluid velocity. Indicates particle volume fraction. Indicates particle size; During data preprocessing, particle volume fraction, particle size, fluid velocity, and mass flow rate are standardized.

[0008] Preferably, the temporal similarity of erosion degradation is calculated using dynamic time warping distance, and its expression is: ; In the formula, This indicates the current erosion degradation sequence of the throttle valve under test. Compared with historical sample erosion degradation sequence The dynamic time-normalized distance between them; Indicates the time series alignment path; This indicates that the throttle valve under test is in the [number]th ...]. Erosion state values ​​at each time point; Indicates the historical throttle valve in the first Erosion state values ​​at each time point; The similarity of operating parameters is calculated using cosine similarity, and its expression is: ; In the formula, This represents the vector of operating parameters for the underwater throttle valve currently under test. This represents a vector of operating parameters for historical throttle valve samples; the vector includes one or more of the following: valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, mass flow rate, and erosion time. The larger the value, the closer the operating conditions are to the current throttle valve under test and historical samples; based on this, an erosion-operating condition joint similarity is constructed: ; In the formula, Indicates the current throttle valve under test and the first The overall similarity between historical samples This represents the current erosion degradation state sequence of the throttle valve under test. Indicates the first A sequence of historical throttle valve erosion and degradation states. This represents the vector of operating parameters of the throttle valve under test. Indicates the first A vector of operating condition parameters for a historical sample. Indicates the weighting coefficient of the operating condition; Constructing a hybrid distance function that integrates multiple operating conditions: ; In the formula, Indicates the mixed distance. Indicates the similarity distance of operating parameters. Indicates the time distance of erosion degradation. Indicates the weighting coefficient; Mahalanobis distance can be used for calculation.

[0009] Preferably, step S5 adaptively expands the nearest neighbor samples. The number of augmented nearest neighbor samples is expressed as: ; In the formula, Indicates the first The number of nearest neighbor samples added; Indicates the first The overall similarity between the nearest neighbor samples and the current throttle valve under test; and These represent the maximum and minimum similarity in the nearest neighbor sample set, respectively; Indicates the expansion scaling factor; Indicates the original sample size; This represents the floor function; The expression for non-uniform weight allocation to nearest neighbor samples is as follows: ; In the formula, Indicates the first The weight coefficients of the nearest neighbor samples, It represents the average similarity of the nearest neighbor sample set.

[0010] Preferably, after completing the nearest neighbor sample retrieval, the remaining lifetime decay trend of the nearest neighbor samples is further extracted as an input feature for residual correction in the subsequent two-layer random forest model; the nearest neighbor remaining lifetime decay trend is used to describe the degradation changes in the lifetime dimension of historical samples similar to the current underwater throttle valve under test, and its expression is: ; In the formula, TRUL represents the decay trend of the remaining lifetime of the nearest neighbor. Indicates the number of nearest neighbor samples. Indicates the first The remaining lifetime of the nearest neighbor samples. Indicates the initial remaining useful life. Indicates the erosion time.

[0011] Preferably, the two-layer random forest model includes a first-layer base random forest model (Base-RF) and a second-layer residual random forest model (Residual-RF). During the training phase, the enhanced training sample set is input into the first layer of the base random forest model: each input sample is represented as... ,in Indicates length is The erosion degradation time series, This represents the parameter vector corresponding to the operating condition. The first layer of the basic random forest model establishes the initial mapping relationship between the input features and the remaining service life. ; In the formula, This represents the output of the first layer of the basic random forest model. Initial lifetime prediction for each sample This represents the mapping function of the first-layer basic random forest model; Based on the actual remaining service life Compared with the initial lifetime prediction The difference between them is used to calculate the predicted residual. : ; Predicting residuals and input features Input the second-layer residual random forest model and establish the mapping relationship between the input features and the predicted residuals: ; In the formula, This represents the residual prediction value output by the second-layer residual random forest model. This represents the residual correction mapping function of the second-layer residual random forest model.

[0012] Preferably, in step S7, during the prediction phase, the erosion degradation state sequence and operating parameters of the current underwater throttle valve under test are input into the first-layer basic random forest model to obtain the initial remaining lifetime prediction value. Simultaneously, input the second-layer residual random forest model to obtain residual correction values. The final predicted remaining useful life is expressed as follows: ; In the formula, This represents the predicted final remaining service life of the underwater throttle valve currently under test.

[0013] Preferred features include online updates of S8 models and case storage; Once the underwater throttle valve under test reaches the preset failure condition, its actual remaining service life data is obtained. The model parameters are updated based on the deviation between the predicted lifespan and the actual lifespan. The operating parameters, erosion degradation sequence, and actual lifespan information of the underwater throttle valve under test are stored in the historical erosion degradation database. Used for continuous updates and optimization of the remaining useful life prediction model.

[0014] Compared with the prior art, the beneficial effects of this application are as follows: First, this invention fuses historical erosion data, simulated erosion data, and the real-time erosion degradation sequence of the throttle valve currently under test. Instead of relying solely on a single moment's erosion rate or a single erosion index for remaining service life estimation, it incorporates the long-term erosion degradation evolution process of the throttle valve from its current state to failure into the prediction process. Therefore, this invention can more fully characterize the cumulative, dynamic, and nonlinear characteristics of erosion damage in underwater throttle valves, which helps improve the rationality of the remaining service life prediction results.

[0015] Second, this invention constructs a similarity evaluation method that combines erosion degradation characteristics with operating parameters. When calculating the similarity between the current underwater throttle valve under test and historical life samples, it simultaneously considers the evolution law of the erosion degradation sequence and the matching relationship of operating parameters such as valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, and mass flow rate. Compared with methods that match samples based solely on a single operating parameter or a single degradation index, this invention can more comprehensively reflect the similarity of degradation patterns between different throttle valve samples, thereby improving the accuracy of historical case retrieval.

[0016] Third, this invention integrates a hybrid distance function across multiple operating conditions to comprehensively measure the similarity of operating parameters and the temporal similarity of erosion rates, thereby reducing the impact of different dimensions, scales, and time series lengths on similarity judgment. This approach considers both the influence of operating parameters on flow field distribution, particle transport, and erosion intensity, as well as the dynamic characteristics of erosion degradation sequences over time, thus enabling more accurate identification of historical samples that closely resemble the current erosion degradation state of the throttle valve under test.

[0017] Fourth, this invention employs a nearest-neighbor sample adaptive enhancement strategy. Based on the comprehensive similarity between the current underwater throttle valve under test and historical lifespan samples, highly similar samples are adaptively expanded and weighted, causing the training sample distribution to shift towards the actual degradation mode of the current throttle valve under test. Compared with traditional prediction methods using fixed training samples, this invention can fully utilize the observed erosion degradation state of the current throttle valve under test, improve the model's adaptability to the individual degradation characteristics of the current valve, and mitigate prediction bias caused by inconsistencies between historical and current sample distributions.

[0018] Fifth, this invention establishes a two-layer random forest remaining life prediction model. The first-layer random forest model is used to obtain the initial prediction results of the remaining life, and the second-layer random forest model is used to model and correct the prediction residuals generated by the first-layer model. Compared with single-layer random forest models or other single regression prediction models, this invention can further compensate for systematic errors caused by operating condition fluctuations, feature coupling, and unmodeled degradation factors based on the initial life estimate, thereby improving the stability and reliability of the remaining life prediction results of underwater throttle valves.

[0019] Sixth, this invention can adapt to practical engineering problems such as the limited number of underwater throttle valve erosion data samples, sparse high-risk operating condition samples, differences between simulation data and actual operating data, and noise in monitoring data. Through steps such as case retrieval, nearest neighbor enhancement, residual correction, and online updates, this invention can continuously correct the prediction model under multi-source heterogeneous data conditions, enhancing the model's robustness to complex operating conditions and local degradation differences.

[0020] Seventh, this invention can output the predicted remaining service life of the underwater throttle valve currently under test, providing a life constraint basis for underwater throttle valve operation status assessment, maintenance and repair decisions, failure early warning, and dynamic optimization of valve opening. By predicting the life of the throttle valve through erosion and degradation, the problems of maintenance delays or premature maintenance caused by relying solely on experience judgment or periodic maintenance can be reduced, which helps to improve the safety of underwater oil and gas production equipment operation and the pertinence of maintenance decisions.

[0021] Eighth, after the underwater throttle valve under test reaches the preset failure condition, this invention can re-store its actual lifespan information, operating parameters, and erosion degradation sequence into the historical erosion degradation database and use it for subsequent model updates. This forms a closed-loop prediction mechanism of "data acquisition - similarity retrieval - lifespan prediction - residual correction - case storage," enabling the remaining lifespan prediction model to be continuously improved with new operating data, thereby enhancing the applicability of lifespan prediction for similar underwater throttle valves. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the CBR closed-loop process of the underwater throttle valve remaining service life prediction method of the present invention. Figure 2 This is a schematic diagram of a two-layer random forest prediction structure based on adaptive nearest neighbor enhancement. Figure 3 This is a flowchart illustrating the adaptive nearest neighbor enhancement method. Figure 4 This is a schematic diagram comparing the traditional training method with the nearest neighbor reinforcement training method of the present invention; Figure 5 A schematic diagram of the structure of a two-layer random forest remaining lifetime prediction model; Figure 6 A comparison chart showing the prediction results of different prediction models for the erosion rate of underwater throttle valves; Figure 7 This is a comparison chart showing the prediction results of different prediction models for the maximum erosion depth of underwater throttle valves. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Specific implementation details are set forth in the following description to provide a full understanding of the invention. However, the invention can be implemented in many ways other than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] This embodiment provides a method for predicting the remaining service life of an underwater throttle valve based on adaptive nearest neighbor enhancement. This method uses erosion degradation data of underwater throttle valves under solid-liquid two-phase flow conditions as a basis, integrating historical erosion samples, simulated erosion samples, the current erosion state sequence of the throttle valve under test, and multi-condition parameters. Through a case-based reasoning framework, adaptive nearest neighbor enhancement, and two-layer random forest residual correction, the predicted remaining service life value of the current underwater throttle valve under test is obtained.

[0026] like Figure 1 As shown, the present invention generally includes the following steps: constructing an underwater throttle valve erosion degradation database, acquiring the current erosion degradation state of the underwater throttle valve under test, constructing erosion degradation enhancement features, evaluating erosion degradation similarity, constructing a nearest neighbor sample enhanced training set, establishing a two-layer random forest remaining lifetime prediction model, constructing remaining lifetime prediction results, and online model updating and case storage. The following describes each step in detail with reference to the accompanying drawings.

[0027] S1 constructs a database of underwater throttle valve erosion degradation: In one specific embodiment, the historical erosion degradation instance library includes simulated erosion data and historical operational erosion data. Each historical throttle valve sample can be represented as: ; in, Indicates the first A sequence of erosion degradation of a historical throttling valve. Indicates the first The throttle valve in the first Erosion performance indicators for each monitoring cycle Indicates the first The lifespan of a historical throttle valve. The erosion performance indicators include erosion rate, maximum erosion depth, wear amount, or other parameters that can characterize the erosion degradation state of the valve.

[0028] S2 acquires information on the erosion and degradation status of the underwater throttle valve currently under test.

[0029] Real-time operating parameters and erosion degradation parameters of the underwater throttle valve under test are collected during operation. The real-time operating parameters include one or more of the following: valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, and mass flow rate; the erosion degradation parameters include erosion rate, erosion depth, wear amount, or other parameters that can characterize the erosion degradation state.

[0030] The erosion degradation sequence of the underwater throttle valve under test can be represented as follows: ; in, This represents the sequence of erosion degradation states that have been observed in the underwater throttle valve currently under test. This represents the erosion degradation state during the current monitoring period. The remaining service life prediction problem can be expressed as: ; in, This indicates the remaining service life of the underwater throttle valve currently under test. This represents a set of historical erosion and degradation samples. This represents the lifetime prediction mapping relationship determined jointly by historical samples and the current degradation state.

[0031] S3 constructs erosion degradation enhancement features In this embodiment, sample features include, but are not limited to, valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, mass flow rate, erosion time, erosion rate, and maximum erosion depth. To improve the model's ability to represent complex operating conditions, erosion degradation enhancement features can be further constructed. These enhancement features include one or more of erosion energy density, opening-velocity interaction term, and particle concentration-pressure drop ratio. These enhancement features can be directly constructed from operating condition parameters such as valve opening, fluid velocity, particle size, particle volume fraction, and pressure drop, and used as input features for subsequent similarity evaluation models and remaining life prediction models.

[0032] The erosion energy density is used to characterize the impact energy of solid particles on the throttle valve wall, and its expression can be given as: ; In the formula, Indicates erosion energy density. Indicates fluid density, Indicates fluid velocity. Indicates particle volume fraction. This indicates the particle size. This feature allows us to incorporate the combined effects of particle kinetic energy, particle concentration, and particle size on erosion damage into lifetime prediction models.

[0033] During data preprocessing, parameters with different dimensions, such as particle volume fraction, particle size, fluid velocity, and mass flow rate, are standardized. Preferably, the Z-score standardization method is used, and its expression is: ; in, Represents the original feature values. This represents the mean of the feature. The standard deviation of this feature is represented by... This represents the standardized feature values. For parameters with large dimensional spans, such as pressure drop, logarithmic transformation can be used to reduce the impact of extreme values ​​on the model training process. For features with strong correlations, such as valve opening and flow coefficient, principal component analysis can be used for dimensionality reduction, reducing input dimensionality and model redundancy while retaining key feature information.

[0034] S4 establishes a similarity evaluation model for erosion degradation: like Figure 2 and Figure 3 As shown, in this step, the present invention performs erosion-operating condition joint similarity calculation on the current underwater throttle valve to be tested and the historical throttle valve samples. Figure 2 This is used to illustrate the data transfer relationship between similarity evaluation results and subsequent nearest neighbor sample enhancement and two-layer random forest prediction models; Figure 3 This step illustrates the connection between the erosion-operating condition joint similarity calculation and the adaptive expansion of nearest neighbor samples. The main focus of this step is to evaluate the similarity between the current underwater throttle valve under test and historical samples; nearest neighbor sample expansion and model prediction will be performed subsequently.

[0035] In the erosion-operating condition joint similarity calculation step, the erosion degradation time series similarity and operating condition parameter similarity between the current underwater throttle valve under test and historical samples are calculated separately. The erosion degradation time series similarity can be calculated using the dynamic time warping distance, and its expression is: ; In the formula, This indicates the current erosion degradation sequence of the throttle valve under test. Compared with historical sample erosion degradation sequence The dynamic time-normalized distance between them; Indicates the time series alignment path; This indicates that the throttle valve under test is in the [number]th ...]. Erosion state values ​​at each time point; Indicates the historical throttle valve in the first The erosion state values ​​at each time point. By using dynamic time warping distance, the impact caused by differences in time length, phase shift, and degradation rate among different erosion sequences can be reduced.

[0036] The similarity of operating parameters can be calculated using cosine similarity, and its expression is: ; In the formula, This represents the vector of operating parameters for the underwater throttle valve currently under test. This represents a vector of operating parameters for a historical throttle valve sample. The operating parameter vector includes one or more of the following: valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, mass flow rate, and erosion time. The larger the value, the closer the operating conditions of the current throttle valve under test are to those of historical samples.

[0037] Based on this, a joint similarity between erosion and working conditions is constructed: ; In the formula, Indicates the current throttle valve under test and the first The overall similarity between historical samples This represents the current erosion degradation state sequence of the throttle valve under test. Indicates the first A sequence of historical throttle valve erosion and degradation states. This represents the vector of operating parameters of the throttle valve under test. Indicates the first A vector of operating condition parameters for a historical sample. This represents the operating condition weighting coefficient. The formula considers both the temporal similarity of the erosion degradation sequence and the spatial similarity of the operating condition parameters, enabling a more comprehensive evaluation of the similarity in degradation patterns between the current throttle valve under test and historical samples.

[0038] In this embodiment, the overall similarity This is used to evaluate the overall similarity between the current throttle valve under test and historical samples; during the nearest neighbor sample retrieval process, a hybrid distance function can be further employed. Supplementary constraints or corrections are applied to the ranking results of similar samples.

[0039] Furthermore, to improve the reliability of similar sample retrieval under multiple operating conditions, a hybrid distance function that integrates multiple operating conditions is constructed: ; In the formula, Indicates the mixed distance. Indicates the similarity distance of operating parameters. Indicates the time distance of erosion degradation. This represents the weighting coefficient. Preferably, It can be determined through grid search, cross-validation, or Bayesian optimization.

[0040] in, The Mahalanobis distance can be used for calculation: ; In the formula, and These represent the operating condition parameter vectors for the two samples, respectively. This represents the covariance matrix of the operating parameters. Compared to Euclidean distance, Mahalanobis distance can take into account the correlation between different operating parameters, reducing the impact of different dimensions and feature correlations on the distance metric. The dynamic time warping distance can be used to measure the temporal similarity between two erosion rate sequences or erosion depth sequences.

[0041] S5 constructs a nearest neighbor sample augmentation training set: After completing the similarity calculation, based on the comprehensive similarity... The historical samples are sorted, and the K most similar historical samples are selected as nearest neighbors. To make the training sample distribution closer to the actual degradation mode of the throttle valve under test, the nearest neighbors are adaptively expanded. The number of expanded j-th nearest neighbors can be expressed as: ; In the formula, Indicates the first The number of nearest neighbor samples added; Indicates the first The overall similarity between the nearest neighbor samples and the current throttle valve under test; and These represent the maximum and minimum similarity in the nearest neighbor sample set, respectively; Indicates the expansion ratio factor; N represents the original sample size. This represents the floor function. As can be seen from this formula, historical samples with higher similarity receive more augmentation, causing the enhanced training set to shift towards the degradation mode of the current throttle valve under test.

[0042] To further enhance the impact of highly similar samples on the model training process, non-uniform weight allocation can be applied to nearest neighbor samples, as expressed in the following expression: ; In the formula, Indicates the first The weight coefficients of the nearest neighbor samples, This represents the average similarity of the nearest neighbor sample set. This weighting function gives higher training weights to highly similar samples, thereby amplifying the contribution of samples that are closer to the degradation pattern of the current throttle valve under test.

[0043] After completing the nearest neighbor sample retrieval, the remaining lifetime decay trend of the nearest neighbor samples is further extracted as an input feature for residual correction in the subsequent two-layer random forest model. The nearest neighbor remaining lifetime decay trend describes the degradation changes in the lifetime dimension of historical samples similar to the current underwater throttle valve under test, and its expression is: ; In the formula, TRUL represents the decay trend of the remaining lifetime of the nearest neighbor, and K represents the number of nearest neighbor samples. This represents the remaining lifetime of the k-th nearest neighbor sample. Indicates the initial remaining useful life. This indicates the erosion time. This feature can provide information on the lifetime decay trend for subsequent residual correction.

[0044] like Figure 4 As shown, traditional training methods typically use only historical erosion samples of throttle valves to train the prediction model, with the current erosion degradation sequence of the underwater throttle valve under test only used as input during the prediction phase. This invention introduces the current erosion degradation state sequence of the throttle valve under test into the training phase and reconstructs the training sample distribution through similar sample retrieval and nearest-neighbor sample enhancement, enabling the training samples to dynamically adapt to the actual degradation state of the current throttle valve under test. This approach can mitigate prediction bias caused by the inconsistency between the historical sample distribution and the current sample distribution.

[0045] S6 establishes a two-layer random forest remaining lifetime prediction model: like Figure 5 As shown, this invention employs a two-layer random forest remaining lifetime prediction model. The two-layer random forest model includes a first-layer base random forest model (Base-RF) and a second-layer residual random forest model (Residual-RF).

[0046] During the training phase, the augmented training sample set is input into the first layer of the base random forest model. Each input sample can be represented as... ,in Indicates length is The erosion degradation time series, This represents the parameter vector corresponding to the operating condition. The first layer of the basic random forest model establishes the initial mapping relationship between the input features and the remaining useful life: ; In the formula, This represents the output of the first layer of the basic random forest model. Initial lifetime prediction for each sample This represents the mapping function of the first-layer basic random forest model.

[0047] Based on the actual remaining service life Compared with the initial lifetime prediction The difference between them is used to calculate the predicted residuals: ; In the formula, Indicates the first The prediction residuals for each sample. Then, the prediction residuals... and input features Input the second-layer residual random forest model and establish the mapping relationship between the input features and the predicted residuals: ; In the formula, This represents the residual prediction value output by the second-layer residual random forest model. This represents the residual correction mapping function of the second-layer residual random forest model.

[0048] S7 build remaining useful life prediction results: In the prediction phase, the erosion degradation state sequence and operating parameters of the underwater throttle valve under test are input into the first-layer basic random forest model to obtain the initial remaining lifetime prediction value. Simultaneously, input the second-layer residual random forest model to obtain residual correction values. The final predicted remaining useful life is expressed as follows: ; In the formula, This represents the predicted final remaining service life of the underwater throttle valve currently under test. This represents the initial lifetime prediction result output by the first-layer basic random forest model. This represents the residual correction result output by the second-layer residual random forest model. Through this two-layer structure, the first-layer model is responsible for capturing the main mapping relationship between erosion degradation characteristics and lifetime, while the second-layer model is responsible for compensating for local operating condition fluctuations, feature coupling, and model bias that the first-layer model fails to fully represent.

[0049] During model training, hyperparameters such as the number of trees, maximum depth, minimum number of leaf node samples, and number of feature subsets in the random forest model can be determined using grid search, cross-validation, or Bayesian optimization methods. To avoid model evaluation bias caused by a single training set partition, historical data can be divided into a basic training set, an augmented training set, a validation set, and a test set. The basic training set is used to build the initial prediction model, the augmented training set is used for training the nearest neighbor augmentation and residual correction model, the validation set is used for model parameter tuning, and the test set is used to evaluate the model's predictive ability.

[0050] In one specific embodiment, the mean absolute percentage error (MAPE), root mean square error (RMSE), and score can be used to evaluate the prediction results. The MAPE expression is: ; In the formula, Indicates the first The true value of each sample Indicates the first The predicted value for each sample, This represents the sample size. The RMSE expression is: .

[0051] The score can be represented using an asymmetric penalty function: ; In the formula, Indicates the first Predicted remaining useful life for each sample Indicates the first The predicted lifespan is the actual remaining lifespan of each sample. When the predicted lifespan is greater than the actual lifespan, it may lead to delayed warnings; therefore, the Score metric imposes a higher penalty on this type of error. When the predicted lifespan is less than the actual lifespan, it constitutes an early warning, and the penalty is relatively lower. This metric is suitable for risk assessment in underwater throttle valve remaining lifespan prediction scenarios.

[0052] like Figure 6 As shown, in the erosion rate prediction embodiment, the ANETRF model proposed in this invention is compared with the ordinary random forest model RF, the linear regression model LR, the multilayer perceptron MLP, the support vector machine SVM, and the long short-term memory network LSTM. Figure 6 The ANETRF curve in the figure represents the prediction result of the method of the present invention, and the erosion rate historical data curve represents the real or historical control data in the test sample. Figure 6 It can be seen that the ANETRF model can track the changing trend of historical erosion rate data well, and the deviation between the predicted curve and the historical data curve is small, indicating that the present invention can effectively characterize the changing law of erosion rate of throttle valve under different operating conditions.

[0053] like Figure 7 As shown, in the maximum erosion depth prediction embodiment, the ANETRF model is compared with RF, LR, MLP, SVM and LSTM. Figure 7 The ANETRF curve in the figure represents the prediction result of the maximum erosion depth by the method of the present invention, and the erosion depth historical data curve represents the real or historical control data in the test sample. Figure 7 It can be seen that the ANETRF model can track the historical data change trend of the maximum erosion depth well, indicating that the present invention can not only predict the erosion rate, but also predict the maximum erosion depth which is closely related to the valve failure risk.

[0054] In one embodiment, 20 sets of historical erosion data for throttle valves were used as a test set to test the ANETRF, RF, LR, MLP, SVM, and LSTM models. For the erosion rate prediction task, the ANETRF model achieved a score of 1.123, a mean MAPE of 4.56%, and a mean RMSE of 1.15 × 10⁻⁶. -7 For the maximum erosion depth prediction task, the ANETRF model achieved a score of 1.234, a mean MAPE of 5.67%, and a mean RMSE of 0.045. These results demonstrate that the adaptive nearest neighbor enhancement and two-layer random forest residual correction mechanism employed in this invention can improve the model's predictive stability and accuracy for the underwater throttle valve erosion degradation process.

[0055] After obtaining the predicted erosion rate and maximum erosion depth, the remaining service life can be further extrapolated. If mass loss is used as the failure criterion, the remaining service life can be calculated based on the initial mass of the key components of the throttle valve, the allowable mass loss ratio, and the predicted erosion rate. If the maximum erosion depth is used as the failure criterion, the remaining service life can be estimated based on the difference between the maximum allowable erosion depth threshold of the key component and the current erosion depth, combined with the predicted erosion depth growth trend. Therefore, this invention can provide a life constraint basis for the operational status assessment, maintenance and repair decision-making, failure early warning, and dynamic optimization of valve opening of underwater throttle valves.

[0056] S8 Model Online Updates and Case Storage Once the underwater throttle valve under test reaches a preset failure state or subsequent actual lifespan data is obtained, the valve's operating parameters, erosion degradation state sequence, predicted lifespan, and actual lifespan are rewritten into the historical erosion degradation instance database. Furthermore, the similarity evaluation model, nearest neighbor enhanced sample set, and two-layer random forest model parameters are updated based on newly added instances. Through this method, the invention can continuously optimize the prediction model as underwater throttle valve operating data accumulates, improving the applicability and robustness of predicting the remaining lifespan of similar throttle valves.

[0057] Figure 1 This invention illustrates the overall process of constructing an instance library based on historical erosion state sequences of throttle valves, and sequentially performing similarity calculations, similar historical sample retrieval, two-layer random forest prediction, remaining lifetime estimation, and historical record saving and updating for the current state sequence of the throttle valve under test. The arrows in the figure indicate the data transfer direction between erosion degradation data, similar historical samples, and remaining lifetime prediction results.

[0058] Figure 2 To illustrate this invention, the erosion degradation time series of the current underwater throttle valve under test and the historical erosion degradation time series of throttle valves are input into an adaptive nearest neighbor enhancement module. After sample enhancement, training samples are formed and further input into a two-layer random forest prediction model. Finally, the remaining service life prediction result of the current underwater throttle valve under test is output.

[0059] Figure 3 The adaptive nearest neighbor enhancement method includes two processes: erosion-operating condition joint similarity calculation and adaptive expansion of nearest neighbor samples. Specifically, after calculating the similarity between the current underwater throttle valve erosion state sequence and the historical throttle valve erosion state sequence, the nearest historical samples are selected based on the similarity results, and the high similarity samples are expanded to form an enhanced training sample set.

[0060] Figure 4 This invention illustrates that traditional training methods only utilize historical erosion samples of throttle valves to train random forest models, while the present invention introduces the erosion degradation state sequence of the current throttle valve under test during the training phase and reconstructs the training sample distribution through nearest neighbor enhancement, making the training samples closer to the actual degradation state of the current throttle valve under test.

[0061] Figure 5 The two-layer random forest model is described as consisting of a first-layer basic random forest model and a second-layer residual random forest model. The first-layer basic random forest model is used to obtain the initial prediction results of the remaining useful life, and the second-layer residual random forest model is used to model and correct the prediction residuals of the first layer. The initial prediction results and the residual correction results are then fused to obtain the final prediction value of the remaining useful life.

[0062] Figure 6 This invention is used to illustrate the comparison between the prediction method used in this invention and models such as random forest, linear regression model, multilayer perceptron, support vector machine and long short-term memory network in terms of erosion rate prediction results under test sample conditions, and to verify the predictive ability of this invention for the changing trend of erosion rate of throttle valve.

[0063] Figure 7This invention is used to illustrate the comparison between the prediction method used in this invention and other models such as Random Forest, Linear Regression Model, Multilayer Perceptron, Support Vector Machine, and Long Short-Term Memory Network in predicting the maximum erosion depth under test sample conditions. It is also used to verify the predictive ability of this invention for key indicators related to the erosion degradation degree and lifespan prediction of throttle valves. Wherein, RF stands for Ordinary Random Forest; MLP for Multilayer Perceptron; SVM for Support Vector Machine; LSTM for Long Short-Term Memory Network; the linear regression model is a traditional regression comparison; and ANETRF is the method proposed in this application.

Claims

1. A method for predicting the remaining life of an underwater throttle valve based on adaptive nearest neighbor enhancement, characterized in that, Includes the following steps: S1. Construct a database of underwater throttle valve erosion degradation; S2. Obtain information on the erosion and degradation status of the underwater throttle valve under test; S3. Construct erosion degradation enhancement features; Based on the operating parameters of the current underwater throttle valve under test and historical throttle valve samples, an enhanced feature is constructed to characterize the erosion degradation state. The enhancement features include one or more of the following: erosion energy density, aperture-velocity interaction term, and particle concentration-pressure drop ratio; The enhanced features, along with valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, mass flow rate, erosion time, erosion rate, and maximum erosion depth, are used as input features for the similarity evaluation model and the remaining life prediction model. S4. Establish a similarity evaluation model for erosion degradation; Erosion degradation sequences and corresponding operating parameters were extracted from historical life samples. A similarity evaluation model is established based on the difference between the current erosion degradation state sequence of the underwater throttle valve under test and historical life samples. The similarity evaluation model considers both the erosion degradation evolution law and the matching relationship of operating parameters. The overall similarity between the current underwater throttle valve under test and each historical lifespan sample is calculated using a similarity evaluation model. S5. Construct a training set to enhance nearest neighbor samples; S6. Establish a two-layer random forest remaining lifetime prediction model; Train the first-layer random forest model using an expanded training set; An initial mapping relationship between erosion degradation characteristics and remaining useful life is established using a first-layer random forest model. The prediction residuals are calculated based on the difference between the prediction results of the first-layer random forest model and the actual lifetime values. The predicted residuals are used to construct a second-layer random forest model; A mapping relationship between input features and prediction residuals is established using a second-layer random forest model; S7. Construct the remaining useful life prediction results.

2. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 1, characterized in that, Erosion degradation data of multiple historical underwater throttle valves were collected, including simulated erosion data and actual operational erosion data. For each historical underwater throttle valve, record its erosion degradation information throughout the entire process from initial operation to failure, and form a corresponding lifetime sample; The lifespan sample includes at least valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, mass flow rate, erosion time, erosion rate, maximum erosion depth, and corresponding remaining lifespan information. Construct a historical erosion degradation database; the historical erosion degradation database includes simulated erosion data and historical operational erosion data, and each historical throttle valve sample can be represented as: ； in, Indicates the first A sequence of erosion degradation of a historical throttling valve. Indicates the first The throttle valve in the first Erosion performance indicators for each monitoring cycle Indicates the first The lifespan of a historical throttle valve; the erosion performance indicators include erosion rate, maximum erosion depth, wear amount, or other parameters that can characterize the erosion degradation state of the valve.

3. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 1, characterized in that, Collect real-time operating parameters and erosion degradation parameters of the underwater throttle valve under test during its operation; The real-time operating parameters include one or more of the following: valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, and mass flow rate. The erosion degradation parameters include erosion rate, erosion depth, wear amount, or other parameters that can characterize the erosion degradation state. Construct a sequence of erosion and degradation states of the underwater throttle valve under test based on time sequence; The erosion degradation sequence of the underwater throttle valve under test can be represented as follows: ； in, This represents the sequence of erosion degradation states that have been observed in the underwater throttle valve currently under test. This represents the erosion degradation state under the current monitoring period; the remaining service life prediction problem is expressed as: ； in, This indicates the remaining service life of the underwater throttle valve currently being tested. This represents a set of historical erosion and degradation samples. This represents the lifetime prediction mapping relationship determined jointly by historical samples and the current degradation state.

4. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 1, characterized in that, Erosion energy density is used to characterize the impact energy of solid particles on the wall of a throttle valve, and its expression is as follows: ； In the formula, Indicates erosion energy density, Indicates fluid density, Indicates fluid velocity. Indicates particle volume fraction. Indicates particle size; During data preprocessing, particle volume fraction, particle size, fluid velocity, and mass flow rate are standardized.

5. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 1, characterized in that, The temporal similarity of erosion degradation is calculated using dynamic time warping distance, and its expression is as follows: ； In the formula, This indicates the current erosion degradation sequence of the throttle valve under test. Compared with historical sample erosion degradation sequence The dynamic time-normalized distance between them; Indicates the time series alignment path; This indicates that the throttle valve under test is in the [number]th ...]. Erosion state values ​​at each time point; Indicates the historical throttle valve in the first Erosion state values ​​at each time point; The similarity of operating parameters is calculated using cosine similarity, and its expression is: ； In the formula, This represents the vector of operating parameters of the underwater throttle valve currently under test. This represents a vector of operating parameters for historical throttle valve samples; the vector includes one or more of the following: valve opening, fluid velocity, particle size, particle volume fraction, pressure drop, mass flow rate, and erosion time. The larger the value, the closer the operating conditions are to the current throttle valve under test and historical samples; based on this, an erosion-operating condition joint similarity is constructed: ； In the formula, Indicates the current throttle valve under test and the first The overall similarity between historical samples This represents the current erosion degradation state sequence of the throttle valve under test. Indicates the first A sequence of historical throttle valve erosion and degradation states. This represents the vector of operating parameters of the throttle valve under test. Indicates the first A vector of operating condition parameters for a historical sample. Indicates the weighting coefficient of the operating condition; Constructing a hybrid distance function that integrates multiple operating conditions: ； In the formula, Indicates the mixing distance. Indicates the similarity distance of operating parameters. Indicates the time distance of erosion degradation. Indicates the weighting coefficient; Mahalanobis distance can be used for calculation.

6. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 1, characterized in that, Step S5 performs adaptive augmentation on nearest neighbor samples. The number of augmented nearest neighbor samples is expressed as: ； In the formula, Indicates the first The number of nearest neighbor samples added; Indicates the first The overall similarity between the nearest neighbor samples and the current throttle valve under test; and These represent the maximum and minimum similarity in the nearest neighbor sample set, respectively; Indicates the expansion scaling factor; Indicates the original sample size; This represents the floor function; The expression for non-uniform weight allocation to nearest neighbor samples is as follows: ； In the formula, Indicates the first The weight coefficients of the nearest neighbor samples, It represents the average similarity of the nearest neighbor sample set.

7. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 6, characterized in that, After completing the nearest neighbor sample retrieval, the remaining lifetime decay trend of the nearest neighbor samples is further extracted as the input feature for residual correction in the subsequent two-layer random forest model. The nearest neighbor remaining lifetime decay trend is used to describe the degradation changes in the lifetime dimension of historical samples similar to the current underwater throttle valve under test, and its expression is: ； In the formula, TRUL represents the decay trend of the remaining lifetime of the nearest neighbor. Indicates the number of nearest neighbor samples. Indicates the first The remaining lifetime of the nearest neighbor samples. Indicates the initial remaining useful life. Indicates the erosion time.

8. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 1, characterized in that, The two-layer random forest model consists of a first-layer base random forest model (Base-RF) and a second-layer residual random forest model (Residual-RF). During the training phase, the enhanced training sample set is input into the first layer of the base random forest model: each input sample is represented as... ,in Indicates length is The erosion degradation time series, This represents the parameter vector corresponding to the operating condition. The first layer of the basic random forest model establishes the initial mapping relationship between the input features and the remaining service life. ; In the formula, This represents the output of the first layer of the basic random forest model. Initial lifetime prediction for each sample This represents the mapping function of the first-layer basic random forest model; Based on the actual remaining service life Compared with the initial lifetime prediction The difference between them is used to calculate the predicted residual. : ; Predicting residuals and input features Input the second-layer residual random forest model and establish the mapping relationship between the input features and the predicted residuals: ； In the formula, This represents the residual prediction value output by the second-layer residual random forest model. This represents the residual correction mapping function of the second-layer residual random forest model.

9. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 1, characterized in that, In step S7, during the prediction phase, the erosion degradation state sequence and operating parameters of the current underwater throttle valve under test are input into the first-layer basic random forest model to obtain the initial remaining lifetime prediction value. Simultaneously, input the second-layer residual random forest model to obtain residual correction values. The final predicted remaining useful life is expressed as follows: ； In the formula, This represents the predicted final remaining service life of the underwater throttle valve currently under test.

10. The underwater throttle valve remaining life prediction method based on adaptive nearest neighbor enhancement according to claim 1, characterized in that, Includes online updates of S8 models and case storage; Once the underwater throttle valve under test reaches the preset failure condition, its actual remaining service life data is obtained. The model parameters are updated based on the deviation between the predicted lifespan and the actual lifespan. The operating parameters, erosion degradation sequence, and actual lifespan information of the underwater throttle valve under test are stored in the historical erosion degradation database. Used for continuous updates and optimization of the remaining useful life prediction model.

Images