Navigation equipment health management method based on SAITS algorithm and digital twin platform

Abstract

The invention relates to a navigation equipment health management method based on an SAITS algorithm and a digital twin platform. The method comprises the following steps: acquiring navigation equipment operation data; adopting an improved SAITS algorithm to carry out interpolation on missing data, and predicting to obtain future navigation equipment operation data; identifying current and future abnormal states of the navigation equipment by adopting a self-adaptive anomaly detection algorithm; constructing an equipment fault mode knowledge graph according to the identified equipment exception; simulating the operation of the navigation system by using a digital twin platform, verifying the confidence coefficient of equipment abnormity, and generating a health assessment report; if the confidence coefficient is not smaller than a confidence coefficient threshold value, generating an exception coping strategy in combination with an equipment fault mode knowledge graph; incremental learning or retraining is carried out on an adaptive anomaly detection algorithm to improve the detection accuracy. The method has remarkable advantages in the aspects of improving the navigation equipment management efficiency, reducing the fault risk, optimizing the maintenance cost and the like, and is particularly suitable for water transportation scenes with strict requirements on reliability and real-time performance.

Description

Technical Field

[0001] The present invention belongs to the field of data processing and fault detection, and particularly relates to a health management method for navigation equipment based on the SAITS algorithm and a digital twin platform. Background Art

[0002] With the continuous increase of navigation equipment in modern shipping, the abnormal detection and maintenance of navigation equipment become more and more important. The operating environments of navigation equipment such as water level gauges and ship locks are complex and changeable, and the aging of the equipment itself, the interference of the external environment, and the fluctuation of the operating load will all cause the operating state of the equipment to be abnormal. The existing equipment health methods lack comprehensiveness and the prediction of the future state of the equipment, resulting in sudden failures of the equipment.

[0003] Traditional health management methods for navigation equipment usually adopt fixed thresholds or simple rule models, which are simple but have significant defects when facing a large number of different navigation equipment and complex operating environments. Fixed thresholds are difficult to adapt to the dynamic changes of equipment states and may also cause false alarms during normal fluctuations. At the same time, most existing navigation equipment is built with microcomputers to obtain different types of data, and traditional methods cannot make full use of the above heterogeneous data for accurate analysis. In addition, traditional methods only perform static detection on the existing data and cannot reasonably predict the trend of equipment health. Summary of the Invention

[0004] The object of the present invention is to solve the above problems and provide a health management method for navigation equipment based on the SAITS algorithm and a digital twin platform. The method uses edge computing nodes to perform real-time pre-screening on channel topology data, combines an environmental factor correction matrix to dynamically compensate sensor data, improves the real-time performance, integrity, and anti-environmental interference ability of data collection, and ensures the reliability of input data; improves the Self-Attention-based Imputation for Time Series (SAITS) model, performs high-precision imputation on missing data through the improved SAITS algorithm, and predicts the future operating state of the equipment to provide a high-quality data basis for abnormal detection; adopts a dynamically adjusted abnormal detection algorithm, combines a fault mode knowledge graph, and realizes the accurate identification of current and future abnormal states and the rapid positioning of the root cause of the fault; constructs a multi-dimensional digital twin model of "ship-lock-channel", verifies the confidence of abnormal results through simulation, compares the physical entity data with the multi-dimensional digital twin model, reduces the false alarm rate, and improves the credibility of fault diagnosis results.

[0005] To achieve the above object, the technical solution provided by the present invention is as follows: A health management method for navigation equipment based on the SAITS algorithm and a digital twin platform, comprising the following steps: S1: Obtain the operation data of the navigation equipment; use the edge computing node to pre-screen the channel topology perception data, and perform spatial correlation verification on the cluster sensor data; establish an environmental factor correction matrix to dynamically compensate the sensor data affected by environmental parameters; clean, integrate, and store the operation data of the navigation equipment, and synchronize it to the digital twin platform; S2: Adopt an improved SAITS algorithm to interpolate the missing data and predict the future operation data of the navigation equipment; S3: Based on the predicted data obtained in step S2, adopt an adaptive anomaly detection algorithm to identify the current and future abnormal states of the navigation equipment; construct a knowledge graph of equipment failure modes according to the identified equipment anomalies; S4: Construct a digital twin model of the navigation system. For the equipment anomalies identified in step S3, use the digital twin platform to simulate the operation of the navigation system, verify the confidence level of the equipment anomalies identified in step S3, and generate a health assessment report; if the confidence level is not less than the confidence level threshold, combine the equipment failure mode knowledge graph to generate an anomaly response strategy; S5: According to the actual detection feedback of the navigation equipment, perform incremental learning or retraining on the adaptive anomaly detection algorithm to improve the detection accuracy; S6: Generate a maintenance plan according to the anomaly response strategy or fault handling plan, and use the digital twin platform to preview the impact of the maintenance plan on the traffic volume and efficiency.

[0006] Preferably, step S4 further includes calculating the deviation between the predicted data of the navigation equipment and the simulation result. If the deviation exceeds the deviation threshold, the domain expert further diagnoses and confirms the fault, and combines the expert opinions to form a fault handling plan using a fault reasoning model; Further, in step S2, the improvement of the existing SAITS algorithm by the improved SAITS algorithm includes the following steps: 1) Use STL decomposition for seasonal data. The operation of the navigation system is closely related to the water level, and there is a certain relationship between the traffic volume and the flood season and dry season. Therefore, the operation of some equipment shows an obvious seasonal trend; through the STL decomposition technology, the original time series data of the navigation system is decomposed into three parts: trend, seasonality, and residual, which are used as the input of the SAITS model together, and can more effectively perform missing value interpolation on seasonal data and noisy data; 2) Anomaly sample enhancement. The SAITS model will lose accuracy when facing outliers; generate anomaly samples through data generation technology and increase the weight of anomaly samples when training the SAITS model, which can effectively strengthen the learning ability of abnormal data and enhance the accuracy of predicting abnormal data; 3) Introduce a feedback mechanism. If the prediction data error of the SAITS model is large, it will lead to incorrect anomaly detection. Introducing a feedback mechanism can improve the accuracy and generalization ability of the SAITS model; adaptively adjust the strategy and retrain according to the error type such as false detection and missed detection, including adjusting the masking ratio, adding attention layers, and retraining with abnormal samples.

[0007] Furthermore, step S2 specifically includes the following sub-steps: Step S201: Masking process; Assume the input of the SAITS model is , introduce an indicator mask vector for the original missing value ; for each batch of input , randomly select observed values according to a proportion for artificial masking; the time series after artificial masking is the masked feature vector , and the missing mask vector is ; Step S202: Establish a joint optimization imputation task MIT and a reconstruction task ORT. MIT is a prediction task for artificially masked values, and ORT is a reconstruction task for observed values; The concatenation of the masked feature vector and its missing mask vector is linearly transformed and used as the input feature data e of the first diagonal masked self-attention block DMSA1; after position encoding the input feature data e of DMSA1, it passes through stacked attention layers. Each attention layer contains a diagonal masked multi-head attention layer and a feed-forward network, and then a linear layer is used for projection to obtain the first reconstruction feature vector ; Take the concatenation of and as the input feature data of the second diagonal masked self-attention block DMSA2. In the second diagonal masked self-attention block DMSA2, take as the input of the linear layer, then after position encoding, input it into N stacked attention layers, and then sequentially pass through a linear layer, an activation layer, and a linear layer to obtain the second reconstruction feature vector ; In the weighted combination block, dynamically weight and according to time dependence and missing information to obtain the third reconstruction feature vector ; replace the missing values in with to obtain the imputed data ; Step S203: Joint optimization training; The final loss function is the weighted sum of the MIT and ORT losses. By minimizing the loss function, the network hyperparameters are adjusted to improve the accuracy and generalization ability of the SAITS model; Step S204: Data prediction; Use the trained SAITS model for prediction; regard the future time series as missing values in the latest time series data and connect them after the latest time series; input the entire time series into the trained SAITS model, and through interpolation and reconstruction, generate the prediction results for future time series data, which are used for the evaluation of future device operation.

[0008] Preferably, in step S3, the deep Q-network DQN is used as one of the adaptive anomaly detection algorithms.

[0009] Furthermore, in step S3, the construction of the equipment failure mode knowledge graph specifically includes: Data collection and processing: Collect logs, call chain data, papers, network articles, and operation manuals generated by the navigation system, screen the collected data, and after sentence segmentation of the screened data, use it as the input of the large language model; Knowledge extraction: Based on the large language model, perform knowledge extraction, and extract knowledge into entity classes, relationship classes, and attribute classes respectively; Knowledge fusion and ontology construction: Perform knowledge fusion on the extracted knowledge, integrate the information of the same entity in multi-source data, for example, merge the same devices from different data sources into one entity, and integrate the relationships into duplicate or contradictory relationship descriptions to form a consistent relationship expression; Graph storage and application: Use the neo4j database to store entity knowledge and connect entity relationships. When the large model searches for relevant content later, it can directly access the neo4j database.

[0010] Furthermore, step S4 specifically includes the following sub-steps: S401: Build a multi-dimensional digital twin model; Based on the "geometry-physics-behavior-rule" multi-dimensional model framework, establish a multi-dimensional joint model including the geometry model of ships-locks-channels, the hydrodynamic physics model, the lock operation behavior model, and the navigation rules model to depict the geometric features, physical properties, behavior coupling relationships, and behavior guidelines of physical entities; S402: Generate twin data; The twin data is formed by the fusion of the working data generated by the operation of the physical entity and the simulation data of the virtual model. The twin data is used to drive the virtual model to control, optimize, and predict the working state of the physical entity; S403: Fault simulation verification; Inject the abnormal types and parameters identified in step S3 into the digital twin model, and use the digital twin simulation engine for dynamic simulation to verify the confidence of the equipment abnormalities identified by the adaptive anomaly detection algorithm, and generate a health assessment report; Compare the deviation between the simulation results and the predicted data of the navigation equipment; S404: Abnormality response; If the confidence is not less than the confidence threshold, combine the equipment failure mode knowledge graph to generate an abnormality response strategy; Calculate the deviation between the predicted data of the navigation equipment and the simulation results. If the deviation exceeds the deviation threshold, the domain expert further diagnoses and confirms the fault, and combines the expert opinions to adopt a fault reasoning model to form a fault handling plan.

[0011] Furthermore, the digital twin platform includes a data floor that integrates basic data, shared data, business data, 3D data, and monitoring data to provide data support for the multi-dimensional digital twin model.

[0012] As another invention object of the present invention, a navigation equipment health management system is provided, including the following modules: A multi-source data acquisition module that acquires multi-source data from the navigation system, and the multi-source data includes sensor data, navigation monitoring data, network monitoring data, navigation scheduling system log data, and manually entered data; A data processing module that cleans, integrates, and standardizes the acquired multi-source data and stores it in a time series format; the data processed by the data processing module is synchronized to the data floor of the digital twin platform through the digital twin module; An interpolation prediction module that uses an improved SAITS algorithm to interpolate missing data and simultaneously predicts the operation data of the navigation equipment in the future time period; A digital twin platform that integrates basic data, shared data, business data, 3D data, and monitoring data through a data floor; constructs a multi-dimensional digital twin model based on a "geometry-physics-behavior-rule" multi-dimensional model framework; the working data generated by the operation of the physical entity and the simulation data of the multi-dimensional digital twin model are fused to form twin data, and the twin data is used to drive the multi-dimensional digital twin model to control, optimize, and predict the working state of the physical entity; An anomaly detection module that uses an adaptive anomaly detection algorithm to identify anomalies in the current operation data and future operation data of the navigation equipment; A root cause analysis module used to build a knowledge graph model and analyze the root cause of the anomalies identified by the anomaly detection module; A digital twin module, which is used to upload the operation data of the waterway navigation system to the digital twin platform; upload the anomalies identified by the anomaly detection module to the digital twin platform for simulation to verify the confidence level of the predicted anomalies; calculate the deviation between the predicted data of the waterway navigation equipment and the simulation results, and if the deviation exceeds the deviation threshold, combine the opinions of domain experts to judge and confirm the faults. An early warning module, which generates real-time alarm information according to the anomaly detection results and generates a health assessment report for possible future anomalies. The health assessment report includes the current operating status of the equipment, prediction of fault types, remaining life assessment, and potential faults that may be induced. An anomaly handling module, which generates anomaly response strategies for the anomalies identified by the anomaly detection module; for the faults judged by the digital twin module, combines a fault reasoning model to generate fault handling solutions and maintenance plans. A model correction module, which optimizes the detection adaptive anomaly model according to the feedback of the operator on the detection results. The optimization process includes adjusting the detection threshold, updating the anomaly sample library, and retraining the adaptive anomaly detection algorithm.

[0013] Compared with the prior art, the beneficial effects of the present invention include: 1) The present invention deeply combines edge computing, improved SAITS algorithm, digital twin, knowledge graph and incremental learning to form an integrated closed-loop system of "perception - analysis - simulation - optimization"; through environmental compensation, algorithm self-optimization and expert collaboration mechanism, it realizes strong adaptability to complex working conditions of waterway navigation operation; adopts a dynamically adjusted anomaly detection algorithm, combines with a fault mode knowledge graph, realizes accurate identification of current and future abnormal states of the waterway navigation system and rapid positioning of the root cause of faults, and improves detection sensitivity and intelligent level; by constructing a multi-dimensional digital twin model of "ship - lock - waterway", using simulation to verify the confidence level of abnormal results, and combining physical entity data with virtual simulation comparison, it reduces the false alarm rate and improves the credibility of diagnosis results. The present invention has significant advantages in improving the management efficiency of waterway navigation equipment, reducing fault risks, optimizing maintenance costs, etc., and is especially suitable for water transportation scenarios with strict requirements for reliability and real-time performance.

[0014] 2) The present invention uses edge computing nodes to perform real-time pre-screening on the waterway topology data, and combines an environmental factor correction matrix to dynamically compensate sensor data, significantly improving the real-time performance, integrity and anti-environmental interference ability of data collection, and ensuring the reliability of input data.

[0015] 3) The present invention performs high-precision interpolation on missing data through the improved SAITS algorithm and predicts the future operating status of equipment, providing a high-quality data basis for anomaly detection.

[0016] 4) The present invention provides an expert collaborative decision-making mechanism: when the simulation deviation exceeds the threshold, domain experts are introduced for joint diagnosis with the fault reasoning model to form a closed-loop decision-making of "machine intelligence + human experience", ensuring the scientificity and accuracy of complex fault handling.

[0017] 5) Based on the actual detection feedback, the present invention performs incremental learning or retraining on the anomaly detection algorithm to achieve dynamic optimization of the model, adapt to long-term challenges such as equipment aging and environmental changes, and ensure the continuous high-precision operation of the system.

[0018] 6) The present invention previews the impact of the maintenance plan on the navigation efficiency through the digital twin platform, quantifies key indicators such as downtime and loss of navigation volume, assists in formulating the optimal maintenance strategy, and minimizes the risk of operation interruption to the greatest extent.

[0019] 7) The present invention provides a closed-loop link from data to decision-making: covering the entire process of "data collection → prediction → detection → simulation → optimization → execution", realizing real-time monitoring, early warning, accurate diagnosis and proactive maintenance of equipment status, and being able to extend the equipment life and reduce the operation and maintenance costs.

[0020] 8) By constructing a knowledge graph of equipment fault modes, the present invention correlates historical fault data, expert experience and real-time detection results, supports multi-dimensional fault reasoning and generation of disposal plans, and can improve the decision-making efficiency in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The present invention will be further described below in conjunction with the drawings and embodiments.

[0022] Figure 1 It is a schematic flowchart of the navigation equipment health management method according to the embodiment of the present invention.

[0023] Figure 2 It is a schematic diagram of the prediction model of the improved SAITS algorithm according to the embodiment of the present invention.

[0024] Figure 3 It is a schematic diagram of the DQN according to the embodiment of the present invention.

[0025] Figure 4 It is a schematic diagram of the modeling of the knowledge graph according to the embodiment of the present invention.

[0026] Figure 5 It is a schematic diagram of the digital twin platform according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] As Figure 1 shown, the navigation equipment health management method based on the SAITS algorithm and the digital twin platform includes: The navigation equipment health management method based on the SAITS algorithm and the digital twin platform includes the following steps: Step S1: Obtain the operation data of the navigation equipment; use the edge computing node to pre-screen the channel topology perception data, and perform spatial correlation verification on the cluster sensor data; establish an environmental factor correction matrix to dynamically compensate the sensor data affected by environmental parameters; clean, integrate, and store the operation data of the navigation equipment, and synchronize it to the digital twin platform.

[0028] In the embodiment, multi-source data is obtained through the navigation business system, including physical indicators of mechanical equipment (such as stress, deformation, damage), electrical parameters of electronic equipment (voltage, current, temperature, etc.), operation and maintenance data of navigation equipment such as the server log of the navigation equipment, the hydraulic characteristics of the ship lock, and navigation monitoring. Integrate, clean the collected data, record the missing values, and upload them to the digital twin platform.

[0029] Taking the motor parameter signal as an example, its sample format is shown in Table 1.

[0030] Table 1 Motor Parameter Signal

[0031] Where NaN represents the missing measured value.

[0032] Establish an environmental factor correction matrix to dynamically compensate the sensor data under extreme environments.

[0033] An example of the correction matrix is shown in Table 2.

[0034] Table 2 Sensor Data Correction Algorithm

[0035] In Table 2, C_sand represents the measured sediment concentration value in real time, P represents the theoretical hydraulic pressure under the design condition, v represents the real-time water flow velocity, P_adj represents the corrected pressure, ΔT represents the temperature change, α represents the coefficient of thermal expansion of the material, L represents the original length of the structure, ΔL represents the deformation of the metal structure, S represents the original output value of the sensor, k_fouling represents the biofouling coefficient, and S_corrected represents the corrected sensor data.

[0036] In the embodiment, the drain algorithm is used to convert the log data, and the data in the database is converted into a time series using the Pandas library of Python. The sample data is divided into a training set and a sample set according to a ratio, where 70% is the training set and 30% is the test set, and parameters such as the batch size are set.

[0037] The z-score is used to standardize the data.

[0038] Step S2: Adopt an improved SAITS algorithm, such as Figure 2As shown, impute the missing data and predict the future operation data of the navigation equipment; Step S201: Masking process; Pre-define hyperparameters such as the number of attention heads and the number of layers. For each batch of inputs , randomly artificially mask 15% of the observed values. The time series after artificial masking is the masked feature vector , and the missing mask vector is ; Step S202: Establish a joint optimization imputation task MIT and a reconstruction task ORT. MIT is a prediction task for the artificially masked values, and ORT is a reconstruction task for the observed values; The concatenation of the masked feature vector and its missing mask vector , after being transformed by a linear layer, serves as the input feature data e for the first diagonal masked self-attention block DMSA1. After position encoding the input feature data e of DMSA1, it passes through stacked attention layers. Each attention layer contains a diagonal masked multi-head attention layer and a feed-forward network, and then a linear layer is used for projection to obtain the first reconstruction feature vector ; ; In the formula, We is the weight matrix, be and p are the bias terms; Z is the output after N attention layers, is the reconstructed matrix; FFN( ) represents the feed-forward neural network function; DiagMaskedMHA( ) represents the diagonal masked multi-head attention function; are the weights and biases of the linear layer respectively; Take the concatenation of and as the input feature data for the second diagonal masked self-attention block DMSA2, ; In the second diagonal masked self-attention block DMSA2, take as the input of the linear layer, then after position encoding, input it into N stacked attention layers, and then successively pass through a linear layer, an activation layer, and a linear layer to obtain the second reconstruction feature vector , ; is the input feature data for the second diagonal masked self-attention block DMSA2, is the weight matrix of the first linear layer of DMSA2, and p are the bias terms of this linear layer; is the output feature vector after N attention layers of DMSA2, which is the second reconstructed feature vector; In the weighted combination block, according to the time dependence and certain information, dynamically and are weighted to obtain the third reconstructed feature vector ; use to replace the missing values in, and the imputed data is obtained; ; ; The imputation error represents the mean absolute error calculated between the artificial missing values and their respective imputations, ; wherein is the mean absolute error function; The reconstruction loss represents the mean absolute error calculated between the observed values and their respective reconstructions, ; Step S203: Joint optimization training; The final loss function is the weighted sum of the MIT and ORT losses , is the scaling factor, and the network hyperparameters are adjusted by minimizing the loss function to improve the accuracy and generalization ability of the SAITS model; Step S204: Data prediction; Use the trained SAITS model for prediction; take the future time series as missing values in the latest time series data and connect them after the latest time series; input the entire time series into the trained SAITS model, and through imputation and reconstruction, generate the prediction results for the future time series data, which are used to evaluate the future operation of the device.

[0039] Taking the motor parameter signal as an example, the data sample after its imputation and prediction is shown in Table 3.

[0040] Table 3 Data of the motor parameter signal after imputation

[0041] Step S3: Based on the prediction data obtained in Step S2, use the adaptive anomaly detection algorithm to identify the current and future abnormal states of the navigation equipment; construct a knowledge graph of equipment fault modes according to the identified equipment anomalies, as shown in Figure 4 ;

[0042] In the embodiment, taking the motor as an example, the deep Q - network (DQN) is used as the adaptive anomaly detection algorithm for the predicted parameter signal of the motor, which specifically includes: (1) Define the state space S and the action space A: The state space is the multi - dimensional historical sensing data corresponding to the device, and the action space set A is defined as: judge the current state as "normal", judge the current state as "abnormal", and adjust the detection threshold; (2) Set the reward function: The reward function includes the reward for correctly detecting the device anomaly , the penalty for wrongly detecting the device anomaly , the reward for correctly judging the device as normal , and the penalty for missing the device anomaly ; The final reward function is the weighted sum of each reward: ; is the final reward result; are all weight coefficients.

[0043] The action - value function Q - value function is composed of the current benefit and the future benefit. Let represent the impact of the current action on the future, then the target Q - value function is: ; In the formula, represents the discount factor; respectively represent the rewards at time t, t + 1, t + 2... t + n; respectively represent the states at time t and t + 1; respectively represent the actions at time t and t + 1.

[0044] (3) Iterative training; Execute the action at each time step to obtain the reward , and enter the next state , then put the sample tuple back into the experience replay pool; In the iterative training, randomly select a batch of samples from the experience replay pool for training; Calculate the current state value and the target according to the main network. The loss function is the mean square error between the current state value and the target , ; Minimize the loss function and iterate until convergence; (4) Anomaly detection; Use the future time series predicted by SAITS interpolation as the input of DQN to obtain the states of each time node of the time series, that is, the detection and prediction of anomalies.

[0045] Input the time series predicted by the improved SAITS algorithm described above into the trained DQN model to obtain the prediction of anomalies. Taking the motor signal as an example, the anomaly detection results are shown in Table 4.

[0046] Table 4 Motor Anomaly Detection Results

[0047] In the embodiment, for data types with almost no abnormal values such as power supply equipment and lock gates (deformation and damage), the K-means clustering method is used as the anomaly detection algorithm.

[0048] Construct a knowledge graph of equipment failure modes, specifically including: Data collection and processing: Collect logs, call chain data, papers, online articles, and operation manuals generated by the navigation system, screen the collected data, and after sentence segmentation of the screened data, use it as the input of the large language model; Knowledge extraction: Based on the large language model, perform knowledge extraction, and extract knowledge into entity classes, relationship classes, and attribute classes respectively; Knowledge fusion and ontology construction: Perform knowledge fusion on the extracted knowledge, integrate the information of the same entity in multi-source data, for example, merge the same equipment from different data sources into one entity, and integrate the relationships into duplicate or contradictory relationship descriptions to form a consistent relationship expression; Graph storage and application: Use the neo4j database to store entity knowledge and connect entity relationships. When the large model searches for relevant content later, it can directly access the neo4j database.

[0049] Step S4: Construct a digital twin model of the navigation system. For the equipment anomalies identified in step S3, use the digital twin platform to simulate the operation of the navigation system, verify the confidence level of the equipment anomalies identified in step S3, and generate a health assessment report; if the confidence level is not less than the confidence level threshold, combine the equipment failure mode knowledge graph to generate an anomaly response strategy.

[0050] Step S401: Construct a multi-dimensional digital twin model and a data floor; Based on the "geometry-physics-behavior-rule" multi-dimensional model framework, establish a multi-dimensional joint model including the ship-lock-channel geometry model, hydrodynamic physics model, lock operation behavior model, and navigation rule model to comprehensively describe the geometric features, physical properties, behavior coupling relationships, and behavior criteria of physical entities.

[0051] The data backplane integrates basic data, shared data, business data, 3D data, and monitoring data, providing data support for the multi-dimensional digital twin model.

[0052] Step S402: Generate twin data; The working data generated by the operation of the physical entity and the simulation data of the virtual model are fused to form twin data, which is used to drive the virtual model to control, optimize, and predict the working state of the physical entity; Step S403: Fault simulation verification; Inject the abnormal types and parameters identified in Step S3 into the digital twin model, perform dynamic simulation using the digital twin simulation engine, calculate the confidence level of the fault prediction based on the coincidence degree between the predicted data of the navigation equipment and the simulation results, and generate a health assessment report for possible future abnormalities. The health assessment report includes the current operating state of the equipment, predicted fault types, and potential faults that may be induced, etc.

[0053] Step S404: Abnormality response; If the confidence level is not less than the confidence threshold, an abnormality response strategy is generated in combination with the equipment fault mode knowledge graph; calculate the deviation between the predicted data of the navigation equipment and the simulation results. If the deviation exceeds the deviation threshold, domain experts further diagnose and confirm the fault, and in combination with expert opinions, a fault reasoning model is used to form a fault handling plan.

[0054] In the embodiment, the Deepseek-R1 model under DeepSeek Company is used as the fault reasoning model.

[0055] The Deepseek-R1 model has the characteristics of small parameters and convenient local deployment. After being deployed on the local server and combined with the above knowledge graph, it can be used as an auxiliary tool for fault response to generate a fault handling plan.

[0056] A patterned input example of the fault reasoning model is as follows: [Clearly describe the abnormal phenomena occurring in the equipment / system] "For the hydraulic hoist of the gate of Gezhouba No. 2 Shiplock under the condition of high water level difference (ΔH>24m): - The amplitude of pressure pulsation exceeds the threshold (+32%) - The closing time in place is extended (designed 120s → measured 158s) - Accompanied by high-frequency vibration (the 250Hz component increases by 15dB)" Step S5: According to the actual detection feedback of the navigation equipment, perform incremental learning or retraining on the adaptive anomaly detection algorithm to improve the detection accuracy.

[0057] After the operator processes the alarm, the accuracy of the detection result will be fed back. If an anomaly is correctly detected, the sample will be marked as an abnormal sample and saved. If there is a misdetection, the sensitivity of the adaptive anomaly detection algorithm is increased, and normal data is supplemented for incremental learning to prevent the model from overfitting. If there is a missed detection, the threshold of the adaptive anomaly detection algorithm is adjusted, and the weight of the reward function in the DQN model is adjusted. Then, the abnormal event data is appended with an abnormal event label and incremental learning is performed.

[0058] Step S6: Generate a maintenance plan according to the abnormal response strategy or fault handling plan. At the same time, synchronize the maintenance plan to the in-station dispatching system and re-make the dispatching plan; synchronously upload it to the digital twin platform, and based on the virtual commissioning function, preview the impact of the maintenance plan on the traffic volume and efficiency.

[0059] Combined with the implementation results, the present invention can accurately and comprehensively evaluate and predict the operating status of equipment by using equipment sensors, monitoring systems, operation and maintenance logs, etc. as sample data. Combined with the improved SAITS algorithm, it can effectively complete the missing values of multivariate time series and efficiently predict the future operation data of shipping equipment, providing accurate data support for the fault detection and prediction of the Three Gorges shipping equipment. The adaptive anomaly detection algorithm can perform dynamic learning and adaptive adjustment, accurately identify the anomalies in the SAITS prediction data, and effectively reduce the risk of misdetection and missed detection. Combined with the Three Gorges equipment knowledge graph, it can quickly and accurately locate the fault type and potential associated faults. The feedback mechanism can enhance the fitting degree of the data, further improve the accuracy and generalization ability of the model. The access of the digital twin platform provides full-element support for the equipment health management of the Three Gorges navigation system, effectively improving the prediction accuracy and collaborative operation and maintenance within the basin. Taking the motor of the Three Gorges ship lock as an example, the accuracy of motor fault detection has increased by about 23%, the false alarm rate has decreased from 18% to 4.5%, the average warning time has increased from 4 days to 21 days, the single maintenance cost has been greatly reduced, the frequency of unplanned outages has decreased by about 65%, and the navigation suspension loss has been reduced by about 75 million yuan per year. In addition, through the simulation and prediction of the digital twin platform, through spare part sharing and expert linkage, the fault recovery efficiency has been increased by 300%.

[0060] Embodiment 2: Based on the method of Embodiment 1, a navigation equipment health management system is provided, including the following modules: A multi-source data acquisition module that acquires multi-source data from the navigation system, where the multi-source data includes sensor data, navigation monitoring data, network monitoring data, navigation dispatching system log data, and manually entered data; A data processing module that cleans, integrates, and standardizes the acquired multi-source data and stores it in a time series format; Among them, the log data is parsed into event sequence data through the Drain algorithm, and the database data is converted into time series data through the Python library; the data processed by the data processing module is synchronized to the data floor of the digital twin platform through the digital twin module; The interpolation prediction module uses the improved SAITS algorithm to interpolate missing data and predict the operation data of navigation equipment in future time periods; The digital twin platform integrates basic data, shared data, business data, 3D data, and monitoring data through the data floor; based on the "geometry-physics-behavior-rule" multi-dimensional model framework, a multi-dimensional digital twin model is constructed, as Figure 5 shown; the working data generated by the operation of the physical entity is fused with the simulation data of the multi-dimensional digital twin model to form twin data, and the twin data is used to drive the multi-dimensional digital twin model to control, optimize, and predict the working state of the physical entity; The anomaly detection module uses an adaptive anomaly detection algorithm to identify anomalies in the current operation data and future operation data of navigation equipment; The root cause analysis module is used to build a knowledge graph model and analyze the root cause of the anomalies identified by the anomaly detection module; The digital twin module is used to upload the operation data of the navigation system to the digital twin platform; upload the anomalies identified by the anomaly detection module to the digital twin platform for simulation to verify the confidence level of the predicted anomalies; calculate the deviation between the predicted data of the navigation equipment and the simulation results, and if the deviation exceeds the deviation threshold, combine the opinions of domain experts to judge and confirm the fault; The early warning module generates real-time alarm information according to the anomaly detection results and generates a health assessment report for possible future anomalies. The health assessment report includes the current operation state of the equipment, prediction of fault types, remaining life assessment, and potential faults that may be induced; The anomaly handling module generates anomaly response strategies for the anomalies identified by the anomaly detection module; for the faults judged by the digital twin module, combines the fault reasoning model to generate fault handling plans and maintenance schedules; The model correction module optimizes the detection adaptive anomaly model according to the feedback of the operator on the detection results. The optimization process includes adjusting the detection threshold, updating the anomaly sample library, and retraining the adaptive anomaly detection algorithm.

[0061] Combined with the implementation results, the present invention significantly improves the operation efficiency of the Three Gorges navigation equipment through the deep integration of the digital twin platform and intelligent health management technology. At the same time, based on basin-level data sharing and multi-hub collaborative operation and maintenance, it promotes resource complementarity and joint dispatching of hubs such as Gezhouba, improves the power generation efficiency of the entire basin, forms a comprehensive benefit of safety, economy, and ecology, and provides a technical benchmark for the intelligent management of the Yangtze River Golden Waterway.

Claims

1. A health management method for general aviation equipment based on the SAITS algorithm and the digital twin platform, characterized in that It includes the following steps: S1: Obtain the operation data of navigation equipment; Use the edge computing node to pre-screen the channel topology perception data and perform spatial correlation verification on the cluster sensor data; Establish an environmental factor correction matrix to dynamically compensate the sensor data affected by environmental parameters; clean, integrate, and store the operation data of navigation equipment; S2: Adopt an improved SAITS algorithm to interpolate the missing data and predict the future operation data of navigation equipment; S3: Based on the prediction data obtained in step S2, adopt an adaptive anomaly detection algorithm to identify the current and future abnormal states of navigation equipment; construct a knowledge graph of equipment failure modes according to the identified equipment anomalies; S4: Construct a digital twin model of the navigation system. For the equipment anomalies identified in step S3, use the digital twin platform to simulate the operation of the navigation system, verify the confidence level of the equipment anomalies identified in step S3, and generate a health assessment report; if the confidence level is not less than the confidence level threshold, combine the knowledge graph of equipment failure modes to generate an abnormal response strategy; S5: According to the actual detection feedback of navigation equipment, perform incremental learning or retraining on the adaptive anomaly detection algorithm to improve the detection accuracy.

2. The method for health management of navigation equipment according to claim 1, wherein In step S2, the improvements of the improved SAITS algorithm on the existing SAITS algorithm include: 1) Use STL decomposition for seasonal data. The operation of the navigation system is closely related to the water level, and the shipping volume has a certain relationship with the flood season and dry season. Therefore, the operation of some equipment shows an obvious seasonal trend; through the STL decomposition technology, the original time series data of the navigation system is decomposed into trend, seasonality, and residuals, which are jointly used as the input of the SAITS model; 2) Abnormal sample enhancement. The SAITS model will lose accuracy when facing outliers; generate abnormal samples through data generation technology, increase the weight of abnormal samples when training the SAITS model, strengthen the learning ability of abnormal data, and enhance the accuracy of predicting abnormal data; 3) Introduce a feedback mechanism. The SAITS model has a large prediction data error. Introducing a feedback mechanism can improve the accuracy and generalization ability of the SAITS model; according to the misdetection and missed detection situations, adaptively adjust the strategy and retrain, including adjusting the masking ratio, adding attention layers, and retraining with abnormal samples.

3. The method for managing the health of navigation equipment according to claim 2, wherein Step S2 specifically includes the following sub-steps: Step S201: Masking processing; Assume that the input of the SAITS model is , and an indicator mask vector is introduced for the original missing values ; for each batch of inputs , randomly select observations for artificial masking in proportion; the time series after artificial masking is the masked feature vector , and the missing masking vector is ; Step S202: Establish a joint optimization interpolation task MIT and a reconstruction task ORT. MIT is a prediction task for artificially masked values, and ORT is a reconstruction task for observed values; Masked feature vector and its missing mask vector After the concatenation is linearly transformed, it serves as the input feature data e for the first diagonal masked self-attention block DMSA1; after position encoding the input feature data e of DMSA1, it passes through stacked attention layers, each of which contains a diagonal masked multi-head attention layer and a feed-forward network, and then a linear layer is used for projection to obtain the first reconstructed feature vector ; ； Where, We is the weight matrix, be and p are the bias terms; Z is the output after passing through N attention layers, is the reconstructed matrix; FFN( ) represents the feed-forward neural network function; DiagMaskedMHA( ) represents the diagonal masked multi-head attention function; are the weights and biases of the linear layer, respectively; Take the and connection as the input feature data of the second diagonal masked self-attention block DMSA2 , ; In the second diagonal masked self-attention block DMSA2, is used as the input of the linear layer, and after position encoding, it is input into N stacked attention layers, and then successively passes through a linear layer, an activation layer, and a linear layer to obtain the second reconstructed feature vector , ； is the input feature data for the second diagonal masked self-attention block DMSA2, is the weight matrix of the first linear layer of DMSA2, and p is the bias of this linear layer; is the output feature vector after passing through the N attention layers of DMSA2, is the second reconstructed feature vector; In the weighted combination block, and are weighted dynamically according to time dependence and certainty information to obtain a third reconstructed feature vector ; the missing values in are replaced with to obtain the imputed data ; ； ； Interpolation error represents the mean absolute error calculated between the artificial missing values and their respective interpolations, ; where is the mean absolute error function; Reconstruction loss represents the mean absolute error calculated between the observations and their respective reconstructions, ; Step S203: Joint optimization training; The final loss function is the weighted sum of the MIT and ORT losses , is the scaling factor. By minimizing the loss function, the network hyperparameters are adjusted to improve the accuracy and generalization ability of the SAITS model; Step S204: Use the trained SAITS model for prediction; In the latest time series data, regard the future time series as missing values and connect them after the latest time series; Input the entire time series into the trained SAITS model, and through interpolation and reconstruction, generate the prediction results of future time series data for the evaluation of future equipment operation.

4. The method for health management of navigation equipment according to claim 1, wherein, In step S3, a deep Q-network DQN is adopted as one of the adaptive anomaly detection algorithms, specifically including: (1) Define the state space S and the action space A: The state space is the multi-dimensional historical sensing data corresponding to the device, and the action space set A is defined as: judge the current state as "normal", judge the current state as "abnormal", and adjust the detection threshold; (2) Set the reward function: The reward function includes the reward for correctly detecting a device anomaly , the penalty for incorrectly detecting a device anomaly , the reward for correctly determining that the device is normal , and the penalty for missing a device anomaly ; The final reward function is the weighted sum of the rewards: ； is the final reward result; are all weight coefficients; The action value function Q-value function is jointly composed of the current benefit and the future benefit. Using to represent the impact of the current action on the future, the target Q-value function is: ； Wherein, represents the discount factor; respectively represent the rewards at time t, time t+1, time t+2... time t+n; respectively represent the states at time t and time t+1; respectively represent the actions at time t and time t+1; (3) Iterative training; Execute an action at each time step Obtain a reward and enter the next state , then put the sample tuple back into the experience replay pool; in iterative training, randomly select a batch of samples from the experience replay pool for training; calculate the current state value and target , and the loss function is the mean squared error between the current state value and the target . ； Iteratively loop through minimizing the loss function until convergence; (4) Anomaly detection; Use the future time series predicted by SAITS interpolation as the input of DQN to obtain the states of each time node in the time series, that is, detect and predict anomalies.

5. The method for health management of navigation equipment according to claim 1, wherein In step S3, the construction of the device fault mode knowledge graph specifically includes: Data collection and processing: Collect the logs, call chain data, and operation manuals generated by the navigation system, screen the collected data, and after sentence splitting the screened data, use it as the input of the large language model; Knowledge extraction: Based on the large language model, perform knowledge extraction, and extract knowledge into entity classes, relationship classes, and attribute classes respectively; Knowledge fusion and ontology construction: Perform knowledge fusion on the extracted knowledge, integrate the information of the same entity in multi-source data, merge the same devices from different data sources into one entity, and integrate the relationships into repeated or contradictory relationship descriptions to form a consistent relationship expression; Graph storage and application: Use the neo4j database to store entity knowledge and connect entity relationships.

6. The method for health management of navigation equipment according to claim 1, wherein, Step S4 also includes calculating the deviation between the predicted data of the navigation device and the simulation result. If the deviation exceeds the deviation threshold, the domain expert further diagnoses and confirms the fault, and combines the expert opinions to adopt a fault reasoning model to form a fault handling plan.

7. The method for health management of navigation equipment according to claim 6, wherein Step S4 specifically includes the following sub-steps: S401: Construct a multi-dimensional digital twin model; Based on the "geometry-physics-behavior-rule" multi-dimensional model framework, establish a multi-dimensional joint model including the ship-lock-channel geometry model, hydrodynamics physical model, lock operation behavior model, and navigation rule model to depict the geometric characteristics, physical properties, behavior coupling relationships, and behavior guidelines of physical entities; S402: Generate twin data; Twin data is formed by fusing the working data generated by the operation of the physical entity and the simulation data of the virtual model. The twin data is used to drive the virtual model to control, optimize, and predict the working state of the physical entity; S403: Fault simulation verification; Inject the anomaly types and parameters identified in step S3 into the digital twin model, and use the digital twin simulation engine for dynamic simulation to verify the confidence of the device anomalies identified by the adaptive anomaly detection algorithm, and generate a health assessment report; S404: Anomaly response; If the confidence is not less than the confidence threshold, combine the device fault mode knowledge graph to generate an anomaly response strategy; Calculate the deviation between the predicted data of the navigation device and the simulation result. If the deviation exceeds the deviation threshold, the domain expert further diagnoses and confirms the fault, and combines the expert opinions to adopt a fault reasoning model to form a fault handling plan.

8. The method for health management of navigation equipment according to claim 7, wherein The digital twin platform includes a data floor that integrates basic data, shared data, business data, 3D data, and monitoring data to provide data support for the multi-dimensional digital twin model.

9. The method for health management of navigation equipment according to claim 1, wherein The method for managing the health of navigation equipment further includes step S6: generating a maintenance plan according to the abnormal response strategy or the fault handling plan, and using the digital twin platform to preview the impact of the maintenance plan on the navigation volume and efficiency.

10. The system for the method for managing the health of navigation equipment according to any one of claims 1-9, characterized in that it includes the following modules: A multi-source data acquisition module that acquires multi-source data from the navigation system, where the multi-source data includes sensor data, navigation monitoring data, network monitoring data, navigation scheduling system log data, and manually entered data; A data processing module that cleans, integrates, and standardizes the acquired multi-source data and stores it in a time series format; The data processed by the data processing module is synchronized to the data floor of the digital twin platform through the digital twin module; An interpolation prediction module that uses an improved SAITS algorithm to interpolate missing data and predict the operation data of navigation equipment in future time periods; A digital twin platform that integrates basic data, shared data, business data, 3D data, and monitoring data through a data floor; constructs a multi-dimensional digital twin model based on a "geometry-physics-behavior-rule" multi-dimensional model framework; forms twin data by fusing the working data generated by the operation of the physical entity with the simulation data of the multi-dimensional digital twin model, and the twin data is used to drive the multi-dimensional digital twin model to control, optimize, and predict the working state of the physical entity; An anomaly detection module that uses an adaptive anomaly detection algorithm to identify anomalies in the current operation data and future operation data of navigation equipment; A root cause analysis module for building a knowledge graph model and analyzing the root cause of anomalies identified by the anomaly detection module; A digital twin module for uploading the operation data of the navigation system to the digital twin platform; uploading the anomalies identified by the anomaly detection module to the digital twin platform for simulation to verify the confidence level of the predicted anomalies; Calculating the deviation between the predicted data of the navigation equipment and the simulation result, and if the deviation exceeds the deviation threshold, combining the opinions of domain experts to determine and confirm the fault; An early warning module that generates real-time alarm information according to the anomaly detection result and generates a health assessment report for possible future anomalies. The health assessment report includes the current operation state of the equipment, prediction of fault types, remaining life assessment, and potential faults that may be induced; An anomaly handling module that generates an abnormal response strategy for the anomalies identified by the anomaly detection module; For the faults determined by the digital twin module, combining with the fault inference model to generate a fault handling plan and a maintenance plan; A model correction module that optimizes the detection adaptive anomaly model according to the feedback of the operator on the detection result. The optimization process includes adjusting the detection threshold, updating the anomaly sample library, and retraining the adaptive anomaly detection algorithm.

Images