An unmanned underwater vehicle semi-physical simulation and performance verification method and system

Abstract

The application discloses a kind of unmanned underwater vehicle semi-physical simulation and performance verification method and system, it is related to unmanned underwater vehicle test verification technical field;Its technical points are: according to the structure data of underwater vehicle body, the virtual simulation model matched with the whole characteristic of real machine hardware is constructed, the whole link operation logic of each system is restored, and the underwater virtual environment model that can simulate the dynamic change of underwater physical characteristics is constructed;Then the two-way data interaction channel of virtual end and real machine is built by the whole link of software and hardware, the cooperative debugging of the two is started based on the real machine operation data of power control system, the real-time correction of virtual model parameter by real machine operation feedback is realized, and the dynamic adaptation of real machine operation state by virtual environment change;Finally, through stability determination rule continuous monitoring, until parameter correction convergence, stability is up to standard after determining final simulation environment, let real machine hardware and virtual simulation model no longer disjointed, and simulation environment can highly restore underwater dynamic actual operation scene.

Description

Technical Field

[0001] This invention relates to the field of unmanned underwater vehicle testing and verification technology, specifically to a hardware-in-the-loop simulation and performance verification method and system for unmanned underwater vehicles. Background Technology

[0002] As core equipment for underwater operations, the performance and reliability of unmanned underwater vehicles (UUVs) directly determine the effectiveness of applications such as underwater exploration, marine engineering, and underwater security. Hardware-in-the-loop simulation and performance verification are crucial steps in the research and development and mass production of UUVs. These simulations can mimic actual underwater conditions in a laboratory environment, allowing for comprehensive testing of the overall performance and the performance of each system within the UUV. However, their limitations are as follows: Firstly, when building the semi-physical simulation environment, a virtual simulation model and an underwater virtual environment model are statically constructed based on the submersible's design parameters. After the models are built, the parameters are not dynamically adjusted. The actual hardware of the unmanned underwater vehicle is simply connected to the virtual environment through a simple communication interface. After a single run parameter matching and debugging, a two-way data interaction channel between the actual and virtual ends is not established, nor is the collaborative operation status of the two continuously monitored. This makes it impossible to achieve real-time parameter correction and dynamic adaptation of the operation status between the actual hardware and the virtual simulation model. As a result, the actual hardware and the virtual simulation model are severely disconnected, and the underwater virtual environment cannot reproduce the dynamic physical characteristics changes in actual operations. The simulation environment has a very low degree of fit with the actual underwater operation scenario and cannot truly reflect the actual operating performance of the submersible. Secondly, when conducting comparative analysis of simulation operation data and actual calibration standard data, a fixed-weight analysis method was used to uniformly compare various performance dimensions. The weighting of different performance dimensions was not dynamically adjusted according to the type of test conditions and fault characteristics, resulting in a serious mismatch between the quantitative results of the data comparison deviation and the verification requirements of actual operating conditions. Furthermore, after completing the data comparison, only a simple pass / fail judgment was given based on the deviation value, without considering the correlation between the underwater vehicle's design parameters and the operating performance of each system to pinpoint performance shortcomings, generating targeted performance optimization suggestions, or designing a re-simulation verification process after parameter adjustments. This resulted in only simple performance judgments, failing to provide specific optimization directions for the development and improvement of the underwater vehicle, creating a break in the verification process, and preventing performance improvement of the unmanned underwater vehicle through iterative testing. The performance verification lost its guiding value for research and development. Therefore, a semi-physical simulation and performance verification method for unmanned underwater vehicles is urgently needed. Summary of the Invention

[0003] To achieve the above objectives, the present invention provides the following technical solution: A method for hardware-in-the-loop simulation and performance verification of an unmanned underwater vehicle includes the following steps: Acquire multi-source heterogeneous basic data required for hardware-in-the-loop simulation and performance verification of unmanned underwater vehicles, and preprocess the multi-source heterogeneous basic data; Based on preprocessed multi-source heterogeneous basic data, a semi-physical simulation operating environment for unmanned underwater vehicles is built, which dynamically adapts the physical hardware and the virtual simulation model, and enables bidirectional collaborative response between the physical and virtual ends. Based on the design performance indicators and actual application scenario requirements of the unmanned underwater vehicle, a gradient test condition simulation test task containing normal working conditions and fault working conditions is generated, and the full-process simulation test of each working condition is executed in the completed hardware-in-the-loop simulation environment. Full-series runtime data during the simulation test and preset real-machine calibration standard data were collected separately. A dynamic weight allocation and iterative optimization method based on working condition characteristics was used to conduct multi-dimensional deviation quantitative comparative analysis of the two types of data. Based on the quantitative comparative analysis of data deviations, and combined with the correlation between the design parameters of the unmanned underwater vehicle and the operation of each system, performance optimization suggestions are generated, and the performance verification and evaluation results of the unmanned underwater vehicle are calculated. The performance verification evaluation results are compared with the preset performance verification standard threshold and a loop iterative verification process is performed. If the evaluation result meets the preset standard, the performance verification is deemed passed and archived. Otherwise, the parameters are adjusted according to the performance optimization suggestions, and the simulation and performance verification process is restarted until the evaluation result meets the preset standard.

[0004] Furthermore, acquiring and preprocessing multi-source heterogeneous basic data includes: The process involves identifying end-to-end data acquisition nodes covering the actual unmanned underwater vehicle (UUV), environmental simulation, test control, and historical data collection points. Data related to UUV operation, environment, working conditions, and historical data are collected. Multiple sets of raw heterogeneous data are acquired from each end-to-end data acquisition node. Basic processing, including anomaly removal and missing data supplementation, is performed on the raw heterogeneous data. Based on the physical attributes and spatiotemporal dimensions of the data, the processed raw heterogeneous data are categorized into different types of target basic data. The categorized target basic data undergoes spatiotemporal dimension matching and data format standardization conversion to complete the heterogeneous data fusion process. Data validity verification rules are established, and the fused target basic data is verified according to these rules, eliminating invalid data to obtain multi-source heterogeneous basic data.

[0005] Furthermore, building a hardware-in-the-loop simulation environment that dynamically adapts between the physical machine and the virtual model includes: Based on data related to the underwater vehicle's structure, a virtual simulation model matching the full characteristics of the actual hardware is constructed to recreate the full-link operational logic mapping of each system of the underwater vehicle. Based on data related to the underwater environment, a virtual environment model recreating the underwater physical characteristics and changing patterns is constructed. The actual hardware of the unmanned underwater vehicle is connected to the constructed virtual simulation model and underwater virtual environment model through a full-link hardware and software communication connection, establishing a two-way data interaction channel between the actual and virtual ends. Based on the actual operation data of the underwater vehicle's power control system, collaborative debugging of the actual hardware and virtual simulation model is initiated, and real-time parameter correction and dynamic adaptation of the operating status of the actual and virtual ends are performed. Rules for judging the stability of the simulation environment are formulated, and the collaborative operation status of the actual and virtual ends is continuously monitored. When the stability index reaches the preset standard and the parameter correction tends to be stable, it is recorded as the final operating environment of the unmanned underwater vehicle's semi-physical simulation.

[0006] Furthermore, generating gradient-based test condition simulation test tasks and executing the full-process simulation test includes: Based on the performance indicators of unmanned underwater vehicles (UUVs) and the risk characteristics of actual application scenarios, a test indicator system covering all performance dimensions for both routine and fault conditions is established. Based on underwater fault condition characteristic data, a test condition sequence is generated, gradually extending from routine navigation conditions to single-factor fault conditions and multi-factor composite fault conditions. Corresponding test execution parameters are set for each condition in the gradient test condition sequence, along with control requirements for condition execution. Simulation test tasks are initiated in a semi-physical simulation environment according to the set test execution parameters, and the connection between different conditions is performed based on the condition switching trigger conditions. Throughout the entire simulation test process, real-time time-series operational data of each system of the UUV are collected, classified and labeled according to test condition type and fault level, forming a full-volume time-series operational dataset corresponding to the gradient test condition sequence.

[0007] Further, the comparative analysis of deviation quantification includes: Based on the performance dimensions and system classification of the underwater vehicle, features are extracted from the full-scale time-series operational data of the simulation test to construct a simulation feature dataset. Pre-set real-vehicle calibration standard data is retrieved, and features are extracted according to the same rules as the simulation feature dataset to construct a real-vehicle standard feature dataset. Based on the type of test conditions and fault characteristics, a dynamic weight allocation method for condition adaptation is constructed, setting corresponding weight percentages for each performance dimension feature. The simulation feature dataset and the real-vehicle standard feature dataset are compared dimension by dimension to calculate the deviation values ​​and deviation ratios for each performance dimension. Combined with historical verification data from similar underwater vehicles, the dynamic weight allocation method is iteratively optimized, updating the weight percentages for each performance dimension feature. Based on the optimized weight percentages, the deviation values ​​and deviation ratios for each performance dimension are weighted and calculated to obtain the quantitative result of the deviation between the simulation data and the real-vehicle standard data.

[0008] Furthermore, the generated performance verification evaluation results include: Based on the data deviation quantification results, and according to the preset test index system scoring rules, the performance of each system of the unmanned underwater vehicle (UUV) is scored. A correlation analysis is conducted between the UUV's design parameters and the operational performance of each system to confirm the degree of influence of various design parameters on system performance. Based on the performance scores of each system, the data deviation quantification results, and the degree of influence of design parameters, performance optimization suggestions are generated, defining optimization targets, adjustment ranges, and control methods, and quantitatively predicting the optimization effects. The basic weight ratio of each system in the overall operation of the UUV is analyzed, and the basic weight ratio is further corrected based on the actual application scenario weights of the test conditions. The performance scores of each system are weighted and calculated with the corrected weight ratios to obtain the performance verification and evaluation results of the UUV. The evaluation results are then standardized to form quantitative values ​​that can be directly compared with preset standard thresholds.

[0009] A hardware-in-the-loop simulation and performance verification system for an unmanned underwater vehicle includes: The data processing module is used to acquire multi-source heterogeneous basic data required for the hardware-in-the-loop simulation and performance verification of unmanned underwater vehicles, and to preprocess the multi-source heterogeneous basic data. The environment setup module is used to build a semi-physical simulation operating environment for unmanned underwater vehicles that dynamically adapts to the real hardware and the virtual simulation model based on preprocessed multi-source heterogeneous basic data. The test execution module is used to generate gradient test condition simulation test tasks based on the performance indicators of the unmanned underwater vehicle and the requirements of actual application scenarios, and to execute full-process simulation tests and collect relevant operational data in a semi-physical simulation environment. The comparative analysis module is used to retrieve preset actual machine calibration standard data and adopts a dynamic weight allocation and iterative optimization method based on working condition characteristics to perform multi-dimensional deviation quantitative comparative analysis between simulation operation data and actual machine standard data. The optimization and evaluation module is used to generate performance optimization suggestions based on the data comparison and analysis results. At the same time, it combines the design parameters of the underwater vehicle body with the correlation between the operation of each system to calculate the performance verification and evaluation results of the unmanned underwater vehicle. The judgment module is used to compare the performance verification evaluation results with the preset performance verification standard threshold, and perform iterative verification judgment operations such as archiving and storing the results after they meet the standard, and adjusting parameters and re-simulating and verifying the results after they fail to meet the standard.

[0010] Furthermore, the data processing module includes: The node determination unit is used to determine the full-link data acquisition nodes covering the actual submersible end, environmental simulation end, test control end, and historical data end, and to plan the full-dimensional data acquisition range; The preprocessing unit is used to acquire raw heterogeneous data from each acquisition node and perform basic processing operations such as removing abnormal data and supplementing missing data. The classification unit is used to classify and determine the raw heterogeneous data after basic processing into different types of target basic data according to the data's physical attributes and the spatiotemporal dimensions of data collection. The format unification unit is used to perform spatiotemporal dimension matching and data format standardization conversion on the classified target basic data, and to perform heterogeneous data fusion processing. The validity verification unit is used to verify the fused target basic data according to the data validity verification rules, remove invalid data, and output multi-source heterogeneous basic data.

[0011] Furthermore, the environment setup module includes: The virtual model building unit is used to construct a virtual simulation model that matches the full characteristics of the actual hardware based on the relevant data of the submarine's body structure, and to perform full-link operation logic mapping of the submarine; The underwater environment model building unit is used to construct a virtual environment model that recreates the underwater physical characteristics and changing patterns based on relevant underwater environment data. The docking unit is used to establish a full-link communication connection between the actual hardware of the unmanned underwater vehicle and the virtual simulation model and underwater virtual environment model, and to build a two-way data interaction channel between the actual and virtual ends. The collaborative calibration unit is used to initiate collaborative debugging between the actual and virtual ends based on the actual operating data of the submersible's power control system, and to perform real-time parameter correction and dynamic adaptation of the operating status of the two. The stability determination unit is used to continuously monitor the collaborative operation status of the physical machine and the virtual terminal according to the stability determination rules of the simulation environment, perform bidirectional calibration, and determine the semi-physical simulation operation environment.

[0012] Furthermore, the comparative analysis module includes: The simulation feature extraction unit is used to extract features from the full time-series running data of the simulation test according to the performance dimension and system classification of the submarine, and to construct the simulation feature dataset. The actual machine standard feature construction unit is used to retrieve the preset actual machine calibration standard data, extract features according to the same rules, and construct the actual machine standard feature dataset. The working condition adaptation weight allocation unit is used to construct a dynamic weight allocation method for working condition adaptation based on the type of test working condition and fault characteristics, and to set the weight ratio of each performance dimension feature. The multi-dimensional deviation calculation unit is used to compare the simulation feature dataset with the actual standard feature dataset in each dimension, and calculate the deviation value and deviation ratio of each performance dimension. The weight iteration optimization unit is used to iteratively optimize the dynamic weight allocation method by combining historical verification data of similar underwater vehicles, and update the weight ratio of each performance dimension feature. The deviation calculation unit is used to perform weighted calculations on the deviation values ​​and deviation ratios of each performance dimension based on the optimized weight ratios, so as to obtain the quantification results of the deviation between simulation and actual data.

[0013] This invention provides a method and system for hardware-in-the-loop simulation and performance verification of unmanned underwater vehicles, which has the following advantages: 1. Based on the submersible's structural data, a virtual simulation model matching the full characteristics of the actual hardware is constructed to recreate the full-link operation logic of each system. Simultaneously, an underwater virtual environment model simulating dynamic changes in underwater physical properties is built. Then, a two-way data interaction channel between the actual and virtual ends is established through full-link hardware and software integration. Based on the actual operating data of the power control system, collaborative debugging between the two is initiated, enabling real-time correction of virtual model parameters based on actual operating feedback, and dynamic adaptation of the virtual environment to the actual operating state. Finally, stability judgment rules are continuously monitored until parameter correction converges and stability meets the standards, thus determining the final simulation environment. This ensures that the actual hardware and the virtual simulation model are no longer disconnected, and the simulation environment can highly reproduce the actual underwater dynamic operating scenario, effectively guaranteeing the consistency between simulation test results and the actual operation of the submersible.

[0014] 2. Combining the core performance indicators of the unmanned underwater vehicle (UUV) with the risk characteristics of actual application scenarios, a test indicator system covering all performance dimensions for both routine and fault conditions was developed. A test condition sequence was generated, gradually extending from routine navigation conditions to single-factor fault conditions and multi-factor composite fault conditions, comprehensively covering all types of conditions that the UUV may encounter in actual operations. Standardized test execution parameters were set for each condition, and the automatic connection of conditions was achieved based on triggering conditions. The collected time-series operation data was classified and labeled according to condition type and fault level. This solved the problems of single test conditions and non-standard execution in existing solutions. It can not only verify the routine navigation performance of the UUV, but also fully test its operational capabilities and fault robustness in complex composite fault scenarios, achieving comprehensive performance verification.

[0015] 3. Based on the comprehensive deviation quantification results of simulation and actual data, the performance of each system component is scored to pinpoint the performance shortcomings of the submersible. Then, a correlation analysis is conducted between the main body design parameters and the operational performance of each system to identify the key design parameters affecting performance. Subsequently, combining the component scores, deviation results, and the degree of influence of design parameters, targeted performance optimization suggestions are generated, specifying the optimization targets, adjustment ranges, and control methods, with quantitative predictions of the optimization effects. Finally, a comprehensive performance evaluation is completed through a secondary correction using basic weights combined with application scenario weights. A threshold is set for cyclical processing; if the threshold is met, all data is archived to support subsequent R&D; if the threshold is not met, parameters are adjusted according to the optimization suggestions, and the entire verification process is restarted. This approach solves the problems of one-sided performance evaluation, lack of specific optimization suggestions, and broken verification processes in existing solutions. It can accurately identify the root causes of performance problems and provide feasible optimization directions for R&D improvements, significantly improving the practical value and execution efficiency of the verification work. Attached Figure Description

[0016] Figure 1 This is an overall flowchart of an embodiment of the present invention; Figure 2 This is a flowchart of the data acquisition and preprocessing process in this invention; Figure 3 This is a flowchart of the simulation environment setup process in this invention; Figure 4 This is a flowchart of the gradient condition test in this invention; Figure 5 This is a flowchart of the deviation quantification analysis in this invention; Figure 6 This is a flowchart of the performance evaluation and optimization process in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Please refer to Figures 1 to 6 As shown, this embodiment provides a hardware-in-the-loop simulation and performance verification method for unmanned underwater vehicles. The method uses an autonomous unmanned underwater vehicle as the unified verification object. The vehicle is designed with a maximum diving depth of 500m, a rated maximum speed of 4kN, and a full-load endurance of 100km. The method forms an uninterrupted data loop and verification loop according to the principle of data-driven, layer-by-layer progression, covering the entire life cycle of performance verification in the R&D stage of unmanned underwater vehicles.

[0019] I. Hardware-in-the-loop simulation and performance verification of unmanned underwater vehicles: The system is mainly divided into six categories: acquisition and preprocessing of multi-source heterogeneous basic data, construction of a semi-physical simulation operating environment, execution of gradient-type test condition simulation tests, quantitative comparison and analysis of deviations between simulation and actual machine data, generation of performance optimization suggestions, and iterative verification and processing of comprehensive evaluation results. It follows a logical sequence of data foundation building, environment construction, operating condition testing, deviation analysis, performance evaluation, and iterative optimization. The output of each step is standardized and directly used as the input for the next step, forming an end-to-end uninterrupted data loop. Data flow is achieved through a unified data format, synchronized timestamps, and standardized interfaces. Ultimately, it realizes a full-process semi-physical simulation and performance verification encompassing testing, evaluation, optimization, and retesting, enabling precise verification and optimization of the entire system performance, including dynamic control, perception, propulsion, and fault response for unmanned underwater vehicles. The technical system includes multi-source heterogeneous... This system incorporates data processing technology, hardware-in-the-loop (HIL) modeling technology, automated gradient testing technology, dynamic weight deviation analysis technology, targeted performance evaluation technology, and iterative optimization technology. Designed to meet the performance verification needs throughout the entire lifecycle of unmanned underwater vehicles (UUVs), the system utilizes multi-source heterogeneous data processing technology to provide a high-quality data foundation for all subsequent operations. Hardware-in-the-loop (HIL) modeling technology achieves high-fidelity replication of actual underwater operating scenarios. Automated gradient testing technology enables standardized unmanned testing across all operating conditions. Dynamic weight deviation analysis technology accurately quantifies the discrepancies between simulation and actual data. Targeted performance evaluation technology precisely identifies performance weaknesses in UUVs. Iterative optimization technology continuously improves performance. Furthermore, this technology system is versatile and adaptable to the performance verification of small and medium-sized UUVs of different models and application scenarios.

[0020] like Figure 2As shown, the acquisition and preprocessing of multi-source heterogeneous basic data involves multi-dimensional acquisition, preprocessing, fusion, and calibration to output effective, fully standardized multi-source heterogeneous basic data. This data, after being synchronized with the coordinate system via timestamps, serves as the input for building the hardware-in-the-loop (HIL) simulation environment. Based on this standardized data, a 1:1 virtual model of the actual machine is constructed, enabling underwater dynamic environment modeling and bidirectional adaptation between the virtual and actual machines. This creates a dynamically adapted HIL simulation environment. After stability testing, the matching degree of the operating state is effectively improved, serving as the execution environment for gradient-based test condition simulation. This dedicated test environment allows for flexible adjustment of environmental parameters according to test requirements. Gradient-based test condition simulation execution involves completing automated tests across all gradient conditions, from normal to faulty, using standardized parameters within this simulation environment. Simultaneous acquisition of operating data from both the actual machine and the virtual model outputs a full-volume time-series test dataset categorized by fault level. This dataset, after data cleaning and format standardization, serves as the input for comparative analysis of the quantitative comparison of simulation and actual machine data deviations. Deviation quantification and comparative analysis: Based on the classification and labeled dataset, a dual-feature dataset is constructed, and a dynamic weighting method is used to complete the deviation quantification calculation, outputting the comprehensive deviation under all working conditions. This deviation result, after validity verification, serves as the core basis for performance evaluation in generating performance optimization suggestions and calculating comprehensive evaluation results. Based on the deviation quantification results, the generation of performance optimization suggestions and the calculation of comprehensive evaluation results complete the performance scoring of each system component, comprehensive performance evaluation, and design parameter-operational performance correlation analysis, outputting a standardized comprehensive performance evaluation value of 0.9707 and feasible targeted performance optimization suggestions. This evaluation value and optimization suggestions, after rationality verification, serve as the verification judgment and performance optimization basis for iterative verification processing. Iterative verification processing: By comparing the standardized comprehensive performance evaluation value with the qualified threshold of preset industry standards plus scenario requirements, the compliance / non-compliance judgment is completed. If compliance is achieved, the entire process verification data is standardized and archived; if non-compliance is not achieved, the parameters of the actual machine and virtual model are strictly adjusted according to the targeted performance optimization suggestions, and the entire process iterative process is initiated until the standardized comprehensive performance evaluation value reaches the preset threshold requirement.

[0021] II. Acquisition and Preprocessing of Multi-Source Heterogeneous Basic Data: The acquisition and preprocessing of multi-source heterogeneous basic data is the foundation of the entire verification method. Its implementation quality directly determines the subsequent accuracy. It consists of five parts: full-link acquisition node planning, raw heterogeneous data preprocessing, heterogeneous data classification, spatiotemporal fusion and format unification, and data validity verification. Focusing on improving the quality of basic data and solving technical problems such as multi-source heterogeneous data, inconsistent formats, spatiotemporal asynchrony, and noise, it is divided into full-link multi-dimensional acquisition technology, raw data anomaly removal technology, raw data missing supplementation technology, two-dimensional classification technology, spatiotemporal correlation fusion technology, heterogeneous data format unification technology, and data quality triple verification technology. The selection criteria, operation procedures, and quantitative standards are clearly defined to ensure the standardization and accuracy of data processing.

[0022] The entire data acquisition process is planned and implemented, deploying acquisition nodes according to the principles of full coverage, high accuracy, and traceability, covering four dimensions: the actual aircraft end, the environmental simulation end, the test and control end, and the historical data end. Specifically, at the actual aircraft end, 28 acquisition nodes are deployed at monitoring points of key components such as the submersible's power system, control system, sensing system, and propulsion system. These nodes utilize high-precision industrial sensors and employ an RS485 bus for data transmission, with the acquisition frequency set at 10Hz. This frequency is selected and adapted to the dynamic changes during low-speed navigation of the submersible, and the acquired data includes structured operating parameters such as battery voltage, battery current, speed, diving depth, attitude angle, servo motor angle, and thruster speed. At the environmental simulation end, key parameters such as flow field, pressure field, and temperature field are simulated on an underwater simulation test platform. Eight data acquisition nodes are evenly deployed in the key area, with a acquisition frequency set to 5Hz. This frequency is suitable for the slow-changing characteristics of the underwater environment. The acquired data includes simulated environmental data such as water flow velocity (0-2 m / s), water pressure (0-10 MPa), and water temperature (0-30℃). The test control terminal uses industrial control software on the simulation test console to acquire structured control data in real time, including test commands, operating condition switching parameters, test duration, and equipment operating status. The acquisition frequency is synchronized with the actual test terminal. The historical data terminal connects to the historical verification database of similar submersibles, classifies and extracts test conditions according to data type, and collects historical data from 100 simulation tests and 20 actual sea tests, including historical operating data, deviation data, and performance evaluation data, providing data support for subsequent weighted iterative correlation analysis.

[0023] For the preprocessing of raw heterogeneous data, a layered approach is adopted to address the issues of pulse outliers, numerical outliers, and missing data in the collected raw data. First, a moving average filter with a window size of 5 is used to remove pulse outliers. This method is suitable for removing sudden anomalies caused by high-frequency noise. The window size is selected based on the acquisition frequency and data variation characteristics. Specifically, the logic is as follows: using 5 consecutive sampling points as a sliding window, the arithmetic mean of the data within the window is calculated. Sampling points within the window whose deviation from the mean exceeds 20% are identified as pulse outliers, and the window mean is used to replace the value of these outliers. The window slides continuously along the time axis with a step size of 1 sampling point, completing the outlier removal for the entire time series data. Next, a 3σ criterion with a confidence level of 99.73% is used to remove numerical outliers. This method, based on the normal distribution characteristics of the data, removes outliers exceeding the mean ± 3 standard deviations. Specifically, the logic is as follows: [Further details on single-condition, single-acquisition-section processing are needed]. The data is grouped, and the arithmetic mean μ and standard deviation σ of each group are calculated. The effective data interval [μ-3σ, μ+3σ] is defined, and values ​​outside the interval are identified as outliers and removed. After completion, μ and σ of the remaining data are recalculated, and the verification is repeated twice to ensure that there are no residual outliers. For the problem of missing data, a classification-based supplementation method is adopted. For datasets with consecutive missing lengths not exceeding 5, linear interpolation is used to supplement the data. This method is suitable for scenarios with continuous data changes and ensures the smoothness of the data. That is, the two effective sampling points before and after the missing segment are used as the start and end nodes, and the numerical change is linearly distributed according to the sampling time interval to fill in the continuous data of the missing segment. For discrete missing datasets, the nearest neighbor mean method is used to supplement the data. In scenarios with gradual data changes, the authenticity of the data is ensured. That is, the arithmetic mean of the three effective sampling points before and after the missing point is used as the supplementary value for the missing point to supplement the missing data. The preprocessed data has no anomalies and no missing data and can be directly entered into subsequent processing.

[0024] Heterogeneous data classification involves classifying all preprocessed data into five categories based on their physical attributes: body structure design data, underwater environment data, power control operation data, fault condition characteristic data, and historical verification data. Each category of data is labeled with a millisecond-level timestamp and acquisition location coordinates. The timestamps are synchronized across all nodes using the NTP network time synchronization protocol to ensure time consistency of all data. The actual device is labeled with GPS coordinates in the WGS84 coordinate system, while the environmental simulation device is labeled with the location coordinates in the local coordinate system of the experimental platform, achieving full-dimensional traceability of the data.

[0025] Spatiotemporal fusion and format unification are achieved by prioritizing time and using location as a secondary factor to fuse data from four sources. A three-dimensional correlated dataset of time, location, and data values ​​is constructed using the synchronized millisecond-level timestamp as the core index and the collection location coordinates as the secondary index. A data association algorithm is then used to achieve deep spatiotemporal fusion of data from the actual machine, environmental simulation, test control, and historical data sources, resolving the issue of spatiotemporal asynchrony among multi-source data. The data association algorithm employs a two-dimensional spatiotemporal matching logic, specifically implemented as follows: First, coarse matching in the time dimension: using 10ms as the minimum time granularity, sampling points with an absolute difference in timestamps ≤10ms are grouped into the same time slice, completing the time dimension alignment of the data. Second, fine matching in the spatial dimension: for data within the same time slice, a pre-defined transformation matrix between the actual machine's WGS84 GPS coordinates and the environmental simulation's local coordinate system is used to unify the coordinate system, associating sampling points with spatial location deviations ≤0.1m. Third, correlation verification: Pearson correlation is calculated for the bound multi-source data. Data with a correlation coefficient ≥ 0.9 is considered valid and included in the 3D association dataset, while data with a correlation coefficient < 0.9 is considered invalid and removed, ultimately achieving deep spatiotemporal fusion of data from four sources. Simultaneously, heterogeneous data of different types and formats undergoes standardized format conversion. Structured data is directly converted to CSV format, semi-structured log data is converted to JSON format after extracting core fields, and unstructured image and video data are converted to CSV format after extracting core features such as texture, contour, and motion trajectory using feature extraction algorithms. The feature extraction algorithm is implemented using the OpenCV vision processing library, extracting the submersible contour and underwater terrain texture features from image / video frames using the Canny edge detection algorithm, and calculating the submersible's motion trajectory and relative displacement features using optical flow. The extracted feature values ​​are sorted by timestamp and converted to structured CSV format. All data is uniformly stored and encoded in UTF-8, with a unified data naming rule of acquisition end – data type – timestamp, achieving format unification and storage standardization for heterogeneous data.

[0026] Data validity verification employs a clear three-tiered standardized verification standard. A batch verification algorithm is used to automatically verify the fused and unified data. The integrity requirement is that the missing rate of a single data field does not exceed 5%. The accuracy requirement is that the deviation between numerical data and calibration values ​​does not exceed ±2%, with this standard based on sensor accuracy and industry standards. The temporal continuity requirement is that the acquisition interval deviation does not exceed ±100ms, with this standard based on timestamp synchronization accuracy. Data that fails verification is marked and removed, specifically data that fails to meet the integrity, accuracy, and temporal continuity standards. The final output is valid, standardized, multi-source heterogeneous basic data. This data, after verification, is anomaly-free, has no missing data, and is spatiotemporally synchronized with a unified format, meeting the data requirements for subsequent hardware-in-the-loop simulation environment construction and can be directly applied without additional processing.

[0027] III. Setting up a hardware-in-the-loop simulation environment: like Figure 3 The diagram illustrates a key technology carrier for achieving a high degree of realism in underwater operational scenarios, and is also a crucial component of hardware-in-the-loop (HIL) simulation. It comprises five parts: 1:1 physical machine-matched virtual simulation model construction, underwater dynamic environment model construction, end-to-end hardware and software integration, collaborative debugging of physical and virtual ends, and simulation environment stability assessment. Addressing technical challenges such as disconnect between physical and virtual environments, fixed environment simulation, and asynchronous communication, the technology utilizes the following techniques: 1:1 full-feature modeling technology, multi-software joint modeling technology, underwater dynamic environment simulation technology, end-to-end hardware and software communication technology, industrial-grade data interaction technology, two-way dynamic adaptation technology between physical and virtual environments, and triple stability monitoring technology for the simulation environment. It clearly defines the modeling basis, communication protocol, debugging process, and judgment criteria to ensure high fidelity and stability of the simulation environment.

[0028] A virtual simulation model matching the actual aircraft was constructed based on multi-source heterogeneous basic data. It acquired and preprocessed the body structure design data and power control operation data, employing a joint modeling approach and adhering to the 1:1 full-characteristic modeling principle to complete the construction of the underwater vehicle virtual simulation model. The modeling content included geometric modeling, dynamic modeling, and control modeling, ensuring the virtual model matched the characteristics of the actual aircraft. Geometric modeling was completed using Vortex Studio, rigorously reproducing the underwater vehicle's external dimensions, titanium alloy structural materials, and the geometry and connection relationships of internal components. The model resolution was set to 0.01m based on simulation accuracy requirements, while also reproducing the underwater vehicle's mass, inertia, center of gravity, center of buoyancy, and other physical parameters. Based on the actual aircraft's hydrodynamic test data, Vortex Studio was used to further refine the model. Dynamics modeling was completed in Studio, calibrating key parameters such as hydrodynamic coefficients, drag coefficients, and propulsion efficiency to recreate the underwater dynamics characteristics of the submersible, including navigation, turning, and descent. Control modeling was completed using MATLAB Simulink, replicating the PID trajectory control, fault emergency control, and power distribution control algorithms of the actual aircraft, with control parameters completely consistent with the main control board of the actual aircraft. The geometric model, dynamics model, and control model were jointly debugged and integrated to construct a five-dimensional virtual simulation model including a power sub-model, a control sub-model, a perception sub-model, a propulsion sub-model, and a body dynamics sub-model. The model embedded actual aircraft operating data, and the model parameter update frequency was set to 10Hz to match the acquisition frequency of the actual aircraft, fully recreating the full-link operating logic of each subsystem of the submersible and ensuring that the behavior characteristics of the virtual model are highly consistent with the actual aircraft.

[0029] The underwater dynamic environment model, also known as the underwater virtual environment model, is constructed using Fluent fluid simulation software based on underwater environmental data acquired and preprocessed from multi-source heterogeneous basic data. This model achieves high-fidelity dynamic simulation of the underwater environment. The model includes flow field, pressure field, temperature field, and topographic models, representing the underwater dynamic physical characteristics. Specifically, the flow field model dynamically adjusts the water velocity from 0 to 2 m / s, supporting various flow field types such as uniform and turbulent flow, with a flow field accuracy set to 0.01 m / s. The pressure field model, based on the hydrostatic pressure formula P=ρgh, realizes real-time changes in water pressure with depth from 0 to 500 m. At a depth of 500 m, ρ=1000 kg / m³. 3 g = 9.8 m / s 2 Given h = 500m, we calculate P = 1000 × 9.8 × 500 = 4.9 × 10⁻⁶. 6 The pressure accuracy is set to 0.01 MPa (Pa=4.9 MPa). The temperature field model allows for free adjustment of water temperature from 0 to 30℃, with a temperature accuracy of 0.1℃. The terrain model, based on underwater terrain data, constructs three typical terrains: flat, rugged, and reef areas, with a terrain resolution of 0.1 m, supporting free switching and combination of terrains. The flow field, pressure field, temperature field, and terrain model are coupled and integrated to complete the construction of an underwater dynamic environment model. This model achieves real-time data exchange with a 1:1 real-machine matched virtual simulation model through a data interface, and environmental parameters can be adjusted with one click according to test requirements, accurately reproducing the complex dynamic underwater operating environment.

[0030] The entire hardware and software chain is integrated, adhering to the principles of bidirectional communication and real-time interoperability. This completes the integration between the physical hardware, simulation software, and test console, resolving technical issues such as communication asynchrony and protocol incompatibility. The submersible hardware is connected to the simulation server via industrial Ethernet TCPIP protocol and serial RS232 protocol. The simulation server is equipped with an Intel Xeon Gold 6330 processor, with the serial communication baud rate set to 115200bps and the industrial Ethernet bandwidth set to 1000Mbps to ensure high-speed and stable data transmission. Simultaneously, an OPC UA industrial-grade data interaction middleware is deployed on the simulation server to complete OPC UA node configuration data mapping and security settings. The first step involves node classification configuration, configuring three types of OPC based on data type. The UA nodes are configured as follows: data acquisition nodes (corresponding to sensor operation data on the actual machine), command control nodes (corresponding to control commands issued by the virtual terminal), and status monitoring nodes (corresponding to system operation status and fault information). Each node is configured with a unique node ID, data type, read / write permissions, and update frequency. The update frequency of both data acquisition and command control nodes is set to 10Hz, consistent with the actual machine acquisition frequency. The second step involves bidirectional data mapping, establishing a one-to-one mapping relationship between the actual machine hardware register addresses and the OPC UA data acquisition nodes, and establishing a one-to-one mapping relationship between the virtual simulation model control variables and the OPC UA command control nodes. The mapping deviation is controlled within one sampling period. The third step involves security settings, employing a username + password authentication mechanism to hierarchically control node read / write permissions. The actual machine hardware control nodes are only accessible to authorized users of the simulation test console. Simultaneously, AES-256 encryption algorithm is used for end-to-end encryption of transmitted data to prevent data tampering and leakage. This is implemented using MATLAB Simulink and Vortex. The software interoperability between simulation software such as Studio and Fluent and the actual hardware test console establishes a data feedback channel from the actual machine to the virtual end and an instruction control channel from the virtual end to the actual machine. The data feedback channel is used for the real-time transmission of actual machine running data to the virtual model, and the instruction control channel is used for the real-time transmission of control instructions from the virtual model to the actual machine main control board. The transmission delay of the two-way channel is controlled at the millisecond level.

[0031] The real-machine and virtual-machine collaborative debugging process begins with the real-machine operating data of the power control system output from the multi-source heterogeneous basic data acquisition and preprocessing steps as the initial value. A 30-minute real-machine and virtual-machine collaborative debugging session is initiated, employing an iterative fine-tuning method to achieve bidirectional dynamic adaptation between the two, ensuring a high degree of synchronization in operating states. During the debugging process, real-machine operating parameters are collected in real time through the data feedback channel and compared with the operating parameters of the virtual model. When the deviation between the real-machine speed and the virtual model speed is not less than 0.1 knots, the propulsion power parameters are automatically corrected through the power sub-model of the virtual model until the deviation decreases. The speed is as low as 0.1 kN. When the water flow velocity in the underwater dynamic environment model is adjusted from 0.5 m / s to 1.5 m / s, the virtual model generates a servo adjustment command through the control sub-model. The command is then sent to the main control board of the actual aircraft via the command control channel to adjust the servo angle from 5° to 15°. The actual aircraft automatically adjusts its operating state to adapt to the environmental change. For key parameters such as attitude, diving depth, and servo angle, single-parameter debugging and multi-parameter joint debugging are completed in sequence. After multiple iterations of fine-tuning, the operating states of the actual aircraft and the virtual model are highly synchronized, and the adaptation accuracy meets the requirements of subsequent testing.

[0032] The simulation environment stability was determined by establishing a clear three-fold quantitative stability standard, namely the triple rule. A high-frequency monitoring method was used, sampling once every 10ms, to continuously monitor the stability of the collaboratively debugged simulation environment for 30 minutes. Monitoring indicators included data transmission latency, parameter correction error, and operational state matching degree. The data transmission latency was required to be no more than 50ms, the parameter correction error no more than ±1%, and the operational state matching degree no less than 98%. Real-time analysis and statistics of the monitoring data were performed. Actual monitoring data showed an average data transmission latency of 28ms, an average parameter correction error of 0.6%, and an average operational state matching degree of 99.1%, all meeting the preset standards. The simulation environment underwent overall integration testing, completing trials under normal operating conditions and single-fault operating conditions. After confirming that the environment had no lag, no disconnections, and no data loss, this environment was determined as the final semi-physical simulation operating environment, meeting all environmental requirements for subsequent gradient-type test simulation execution steps and directly supporting automated testing of various gradient conditions.

[0033] IV. Execution of Gradient-Type Test Condition Simulation Test: like Figure 4As shown, the core components of achieving standardized and automated testing across all operating conditions are comprised of five parts: the formulation of a test indicator system, the generation of a gradient test condition sequence, the standardized setting of test execution parameters, the automated connection of operating conditions, and the classification and labeling of test data according to the fault level of the operating condition. This design addresses the problems of single operating conditions, non-standard testing, messy data, and excessive manual intervention in existing technologies. It sequentially employs performance indicator quantification technology, gradient operating condition design technology, test parameter standardization technology, trigger-condition automated control technology, and dual-classification and labeling technology for data. It clearly defines the trigger logic and labeling specifications based on design principles to ensure the standardization and automation of the testing process and the orderliness of the data.

[0034] The testing indicator system is formulated based on three main criteria: industry standards, underwater vehicle design specifications, and application scenario requirements. It establishes quantitative testing standards covering the entire underwater vehicle system, providing clear judgment criteria for operational condition testing. Based on industry standards for performance testing and evaluation of underwater unmanned vehicles, and considering the five core performance design indicators of underwater vehicles—power output, navigation accuracy, control response, endurance, and fault robustness—and taking into account the risk characteristics of sudden changes in water flow, sensor failures, and equipment failures in marine resource exploration applications, quantitative testing standards are formulated for different operational conditions. Under normal operating conditions, the requirements are: thruster power deviation not exceeding ±2%, track deviation not exceeding ±0.5m, servo response delay not exceeding 100ms, actual endurance to theoretical endurance ratio not less than 95%, and no abnormalities in the operation of each system. Under fault conditions, the requirements are: power retention rate not less than 90% under single-factor faults, track deviation not exceeding ±1.0m, response delay not exceeding 200ms, timely fault alarms, no secondary faults, and no system collapse under multi-factor composite faults, with emergency return operation completed and fault recovery time not exceeding 5 seconds. All indicators have clearly defined quantitative thresholds and judgment methods.

[0035] A gradient-based test condition sequence is generated, following the design principles of gradual progression from simple to complex and ensuring equipment safety. This results in a test sequence of twelve sub-conditions across three categories, ensuring the rationality and safety of the tests and preventing damage to the actual equipment due to sudden changes in conditions. The first gradient represents conventional navigation conditions, including four sub-conditions: still water 2kN, still water 4kN, slow current 0.5m / s, and rapid current 1.5m / s, covering typical routine scenarios in the daily operation of the submersible. The second gradient represents single-factor failure conditions, including five sub-conditions: depth gauge no data, power battery voltage drop of 10%, servo motor jamming of 5°, single thruster power drop of 20%, and short-term communication system disconnection of 10s, covering the submersible's sensing, power, control, propulsion, and communication systems. The system presents typical single-fault scenarios for core systems such as communication, with fault parameters set based on the probability and severity of faults in actual applications. The third gradient represents multi-factor composite fault conditions, including three sub-conditions: depth gauge failure plus sudden water flow change, servo motor jamming plus single thruster power reduction, and power battery voltage drop plus communication disconnection. These cover common composite fault scenarios in practical applications, with fault combinations adhering to the principles of relevance and typicality. Furthermore, fault levels are categorized into mild, moderate, and severe based on performance deviation: mild (performance deviation not exceeding 10%), moderate (performance deviation 10% to 20%), and severe (performance deviation exceeding 20%). Single-factor faults are classified as mild to moderate, while multi-factor composite faults are classified as moderate to severe, achieving a balance between operating conditions and... Precise matching of fault levels and execution of test sequences in ascending order of gradient ensure the stability of the testing process. All fault conditions are implemented using a step-by-step soft injection method. The first step is fault pre-verification, in which the actual hardware operating status is continuously monitored for 30 seconds before fault injection to confirm that the operating parameters of each system are within the safe threshold range and there are no native faults before fault injection can be performed. The second step is step-by-step parameter loading. For numerical faults (such as power battery voltage drop, thruster power drop, servo motor jamming), the fault parameters are loaded in 10 step cycles, with each step cycle lasting 1 second and loading 10% of the fault target value in each cycle to avoid parameter abrupt changes from impacting the actual hardware. For switch-type faults (such as no data from the depth gauge or communication disconnection), the system first blocks the data channel and then triggers the fault logic. After the fault duration ends, the data channel and normal operating logic are automatically restored. The third step is fault state maintenance, which maintains the consistency of fault parameters without parameter drift within the single-condition test duration. The fourth step is fault safety exit. After the single-condition test ends, the normal parameters are gradually restored by reversing the step loading process. Only after confirming that all systems of the actual machine have returned to normal operating status can the next condition be entered. An emergency stop button is set throughout the process. When the actual machine encounters situations such as parameters exceeding the threshold or hardware abnormalities, the fault injection can be terminated with one click, and the actual machine can be restored to a safe operating state.

[0036] Standardized test execution parameters are set for all operating conditions, ensuring consistent and reproducible testing. Tests are executed in ascending order of gradient, with each operating condition lasting 10 minutes, based on the submarine's stabilization time. The total test duration is 120 minutes. Each sub-operating condition is tested twice to eliminate the influence of random factors. The data acquisition frequency is set to 10Hz, consistent with the acquisition frequency of the physical machine and virtual model, and frequency synchronization of all acquisition nodes is achieved through the NTP time synchronization protocol. Test data is stored in real-time in a dedicated database on the simulation server using distributed storage to ensure no data loss or delay. A test anomaly handling mechanism is also implemented. When abnormal situations such as parameter exceeding thresholds or equipment failure occur in the physical machine or virtual model, the test is automatically stopped and an alarm is triggered to ensure the safety of the testing process.

[0037] The system features automated operation mode transitions, achieved through a real-time polling monitoring mechanism based on an OPC UA server. This ensures seamless transitions without human intervention, guaranteeing timeliness and accuracy. The simulation test control software is configured with operation mode switching trigger conditions: a single operation mode test duration of 10 minutes with complete and anomaly-free data acquisition. The control software polls these trigger conditions every 10ms. When the trigger conditions are met, it automatically sends an operation mode switching command to both the physical machine main control board and the virtual model. The physical hardware and virtual model synchronously adjust to the operating parameters of the next operation mode, including environmental, fault, and control parameters. During operation mode switching, data acquisition and transmission remain uninterrupted, ensuring the continuity of test data and guaranteeing the efficiency and continuity of the testing process.

[0038] Test data is categorized and labeled according to operating condition and fault level. Full-series operational data of all systems of the submersible are simultaneously collected through acquisition nodes on both the physical and virtual ends, including physical operating parameters, virtual model operating parameters, environmental parameters, fault parameters, and control parameters. An automated yet manually reviewed approach is used for dual classification and labeling based on operating condition type and fault level. Operating condition types are categorized as routine, single-factor fault, and multi-factor composite fault; fault levels are categorized as mild, moderate, and severe. Labeling information includes operating condition type, fault level, sub-operating condition name, number of tests, acquisition time, and acquisition node, directly embedded in the data file naming and attributes. The data file naming rule is: operating condition type – fault level – sub-operating condition name – number of tests, and the data format is CSV. A hierarchical folder structure is established according to operating condition and fault level to classify and store the labeled data, and data metadata is added to achieve orderly data management. Finally, the labeling results are manually reviewed to ensure the accuracy of classification and labeling, forming an ordered dataset corresponding one-to-one with the twelve sub-operating conditions. This provides high-quality data input for subsequent quantitative comparison and analysis of the deviation between simulation and physical data.

[0039] V. Quantitative Comparative Analysis of Deviations between Simulation and Actual Data, Generation of Performance Optimization Suggestions, and Comprehensive Evaluation Results: like Figure 5 As shown, this method achieves accurate deviation quantification and targeted performance evaluation, comprising seven parts: dual-feature dataset construction, dynamic weight allocation adapted to operating conditions, multi-dimensional deviation calculation, weight iterative optimization, correlation analysis of design parameters and operational performance, secondary weight correction, and generation of targeted optimization suggestions. It addresses issues in existing technologies such as fixed weights, singular analysis, one-sided evaluation, and unclear optimization direction. The method sequentially employs unified feature extraction technology, dynamic weight allocation adapted to operating conditions, multi-dimensional deviation quantification technology, iterative weight optimization technology, correlation analysis technology, scenario-based secondary weight correction technology, and targeted optimization suggestion generation technology. Each technique clearly defines extraction rules, setting basis, calculation methods, analysis processes, and optimization principles to ensure the accuracy of deviation analysis, the scientific nature of performance evaluation, and the feasibility of optimization suggestions.

[0040] A dual-feature dataset was constructed based on the classified and labeled ordered dataset output from the gradient-based test condition simulation test execution steps. A unified extraction rule was used to construct both the simulation feature dataset and the actual aircraft standard feature dataset, ensuring the comparability of the two datasets. Twelve sub-operating conditions and five performance dimensions were used as core extraction dimensions: power output, navigation accuracy, control response, endurance, and fault robustness. For each dimension, five core feature values—mean, maximum, minimum, standard deviation, and deviation rate—were extracted. During the extraction process, the data underwent secondary cleaning to remove transitional data generated by operating condition switching. A structured dataset was constructed using Python Pandas. Both datasets contain five fields: operating condition type, fault level, sub-operating condition, performance dimension, and feature value, totaling 60 feature dimensions. The row and column indices of the datasets were kept consistent. The actual aircraft standard data used calibration data from multiple sea tests of the submersible. This data was verified by industry standards to ensure the authority and accuracy of the comparison benchmark. After the two datasets were constructed, data format and consistency verification were performed to ensure no missing data and no field errors.

[0041] The system employs dynamic weight allocation tailored to different operating conditions, abandoning the traditional fixed weight model. Based on expert experience, submersible design requirements, and operating condition characteristics, it assigns differentiated dynamic weights to five performance dimensions, with a total weight of 1, achieving precise weight matching with operating conditions. Under normal operating conditions, the submersible's navigation accuracy and power output are core performance parameters, thus receiving higher weights. Under single-factor failure conditions, the performance dimensions of the faulty system are the core evaluation objects, therefore, higher weights are assigned to the corresponding performance dimensions of the faulty system. Under multi-factor composite failure conditions, fault robustness... As core performance parameters, higher weights are allocated to these parameters. Specifically, the weights are set as follows: for the 2kN calm water normal operating condition, the weights for power output, navigation accuracy, control response, endurance, and fault robustness are 0.25, 0.3, 0.2, 0.25, and 0, respectively; for the combined fault condition of depth gauge failure and sudden change in water flow, the weights for power output, navigation accuracy, control response, endurance, and fault robustness are 0.15, 0.15, 0.15, 0.2, and 0.35, respectively. The weights for all operating conditions have been reviewed and verified by experts to ensure the scientific nature of the weight allocation.

[0042] Multi-dimensional deviation calculation involves comparing and analyzing the same feature value of the simulation feature dataset and the actual machine standard feature dataset dimensionally and under each working condition. It simultaneously calculates both the deviation value and the deviation ratio, achieving precise quantification of the deviation. These two indicators complement each other, comprehensively reflecting the degree of deviation. The deviation value calculation formula is ΔX = X 仿真 -X 标准 This is used to reflect the absolute magnitude of the deviation; the formula for calculating the deviation ratio is ΔP=|X 仿真 -X 标准 | / X 标准 ×100% is used to reflect the relative magnitude of the deviation; a batch calculation script written in Python is used to complete the deviation calculation of all feature values, and the calculation results are classified and stored according to the performance dimension of the operating condition; taking the navigation accuracy index of still water 2kn operating condition as an example, the simulated track deviation is 0.4m, the standard track deviation is 0.38m, and the calculated deviation value is ΔX=X 仿真 -X 标准 =0.4-0.38=0.02m, the deviation ratio is ΔP=|X 仿真 -X 标准 | / X 标准 ×100%=|0.4-0.38| / 0.38×100%=5.26%, which intuitively reflects the degree of deviation in navigation accuracy under this working condition. The simulation feature dataset and the actual standard feature dataset are dual-source data.

[0043] Weight iterative optimization, combining historical verification data of similar underwater vehicles from multi-source heterogeneous basic data acquisition and preprocessing steps, employs the Analytic Hierarchy Process (AHP) to iteratively optimize the initial dynamic weights, improving the accuracy of deviation analysis. The first step involves constructing a hierarchical model, with overall underwater vehicle performance verification as the target layer, five performance dimensions (power output, navigation accuracy, control response, endurance, and fault robustness) as the criterion layer, and the deviation quantification results under various operating conditions as the solution layer. The second step involves constructing a judgment matrix, using a 1-9 scale to compare the relative importance of each performance dimension in the criterion layer pairwise, where scale 1 indicates... Both dimensions are equally important, with a scale of 9 indicating that one dimension is extremely important relative to the other. Based on historical data from 100 simulation tests and 20 real-sea tests, the impact of deviations in each performance dimension on the overall operational effectiveness of the submersible is statistically analyzed, and a 5th-order judgment matrix A = (a_ij)5×5 is constructed, where a_ij is the importance scale of the i-th performance dimension relative to the j-th performance dimension. The third step involves calculating the weight coefficients. The judgment matrix A is normalized column-wise, and the arithmetic mean of the elements in each row is calculated to obtain the initial weight vector ω = (ω1, ω2, ω3, ω4, ω5) for each performance dimension. T The fourth step is consistency testing. Calculate the maximum eigenvalue λmax of the judgment matrix. This is the most commonly used manual calculation method in the Analytic Hierarchy Process (AHP) that meets engineering requirements in terms of accuracy. Then calculate the consistency index CI = (λmax - n) / (n - 1), where n = 5 is the matrix order. Combining this with the average random consistency index RI of matrices of the same order (RI = 1.12 when n = 5), calculate the consistency ratio CR = CI / RI. Next, perform a consistency test on the judgment matrix. If the test coefficient CR < 0.1, the judgment matrix is ​​considered valid; otherwise, the matrix is ​​readjusted. Finally, calculate the weight coefficients of each performance dimension based on the tested judgment matrix, and adjust the initial dynamic weights. Adjustments and optimizations were made, increasing the fault robustness weight of the composite fault condition from 0 to 0.35 and decreasing the navigation accuracy weight from 0.2 to 0.15. After three historical data backtests, the optimized weights improved the matching degree between the deviation quantification results and the actual sea test, ensuring that the deviation analysis results were more realistic. The historical data backtest verification involved substituting the optimized dynamic weights into existing historical simulation and actual sea test data of similar underwater vehicles to verify the rationality and accuracy of the weight allocation. This ensured that the optimized weights could truly reflect the impact of different performance dimensions on the overall operational performance of the underwater vehicle, ultimately improving the matching degree between the deviation quantification results and the actual sea test.

[0044] The comprehensive deviation quantification calculation for all operating conditions is based on optimized weights, and a weighted summation method is used to calculate the comprehensive deviation ratio for all operating conditions, thus achieving comprehensive quantification of the deviation for all operating conditions. First, the single-dimensional weighted deviation W is calculated. i The calculation formula is W i =ω i×ΔP i , where ω i For the optimized weights, ΔP i The deviation ratio is calculated; then, the weighted deviations of all single dimensions under each working condition are summed to obtain the comprehensive deviation of a single working condition; finally, the arithmetic mean of the comprehensive deviations of the single working conditions of the twelve sub-working conditions is taken to obtain the comprehensive deviation of the entire working condition; this result intuitively reflects the overall degree of deviation between the simulation model and the actual machine, providing a core basis for performance evaluation.

[0045] A correlation analysis of design parameters and operational performance was conducted using Pearson correlation coefficient analysis. Combined with significance testing, the root cause design parameters of performance bottlenecks in each system were precisely identified, providing direction for performance optimization. First, the core design parameters of the submersible were extracted, including the aspect ratio of 6.4:1, depth gauge accuracy ±0.1m, PID trajectory control algorithm parameter P=0.8, fault emergency response algorithm response threshold, and thruster speed coefficient. Then, the Pearson correlation coefficients between each design parameter and the deviations in operational performance of each system were calculated; the closer the absolute value of the correlation coefficient is to 1, the stronger the correlation. Simultaneously, a significance test was performed, with a significance level of P<0.05 indicating a significant correlation. The analysis results showed that performance bottlenecks in the navigation system were significantly correlated with depth gauge accuracy and PID trajectory control algorithm parameter P=0.8, with correlation coefficients of 0.92 and 0.88, respectively. Performance bottlenecks in the fault response system were significantly correlated with a high fault emergency response threshold, with a correlation coefficient of 0.95. The root cause design parameters of performance bottlenecks in each system were identified, providing a precise basis for subsequent targeted optimization.

[0046] like Figure 6As shown, the performance scoring and weighting are adjusted twice. Performance scores for each system component are based on a 100-point scale. The weights are then adjusted based on application scenario requirements to achieve a scientific assessment of the submersible's overall performance. For example, the scoring rules are based on deviation percentages: a deviation of no more than 5% earns a full score of 100 points; 5% to 10% deducts 10 points; 10% to 20% deducts 20 points; and exceeding 20% ​​deducts 50 points. The scoring results are categorized by system and operating condition. The scores are as follows: power system 98.2 points, navigation system 96.5 points, control system 97.8 points, endurance system 98.5 points, and fault response system 95.3 points. Simultaneously, considering the core requirements of navigation accuracy and fault robustness in marine resource exploration applications, the basic weights are adjusted twice. The basic weights for the submersible are: power 0.2, navigation 0. 25. The original weights were 0.2 for handling, 0.2 for range, and 0.15 for fault handling. The revised weights are: Power 0.19, Navigation 0.3, Handling 0.19, Range 0.12, and Fault Handling 0.2, with the total weight remaining at 1. These revised weights better reflect real-world application scenarios. A weighted comprehensive performance score is calculated based on these revised weights using the formula S = 0.19 × 98.2 + 0.3 × 96.5 + 0.19 × 97.8 + 0.12 × 98.5 + 0.2 × 95.3, resulting in a weighted comprehensive performance score of 97.07. To facilitate standardized evaluation, the weighted comprehensive performance score is converted to a standardized comprehensive performance evaluation value ranging from 0 to 1 using the formula V = S / 100. The standardized comprehensive performance evaluation value is 0.9707, achieving a precise comprehensive performance assessment.

[0047] Targeted optimization recommendations are generated by combining sub-performance scores, deviation quantification results, and operational performance correlation analysis results of design parameters. Based on the principles of feasibility, economy, and effectiveness, targeted performance optimization recommendations are generated, clearly defining the optimization targets, adjustment ranges, and control methods, and quantifying the predicted optimization effects to ensure the feasibility of the recommendations. For example, targeting the performance shortcomings of the navigation system, the optimization targets are depth gauge accuracy and PID trajectory control algorithm parameter P. The adjustment range is to improve depth gauge accuracy from ±0.1m to ±0.05m and adjust PID trajectory control algorithm parameter P from 0.8 to 0.9. The control method is to further improve the accuracy of the depth gauge accuracy from ±0.1m to ±0.05m and adjust PID trajectory control algorithm parameter P from ±0.8 to ±0.9. The process involved replacing the depth gauge with a high-precision industrial-grade one, debugging the PID algorithm parameters via MATLAB Simulink, and re-flashing the data to the actual main control board. The first step was virtual simulation pre-verification, where the depth gauge accuracy parameters and the PID proportional coefficient P were synchronously updated in the virtual simulation model. Three repeated simulation tests were performed under full gradient conditions to verify the improvement in track deviation after parameter adjustment and to confirm the absence of secondary control oscillations or overshoot. The second step was offline debugging on the actual machine, where the optimized PID algorithm parameters were compiled into a hex format file and connected to the simulation test bench via the JTAG interface of the actual main control board. First, the original parameters of the main control board are read and backed up. Then, the optimized parameter file is burned to the main control board's Flash storage area. Third, online verification is performed on the actual aircraft. After burning, the main control board is restarted, and a 10-minute flight test is completed under normal operating conditions (2km in still water) to verify the parameter effectiveness and control effect. After confirming there are no abnormalities, parameter adjustments are completed. The estimated optimization effect is a reduction in the track deviation ratio to within 3% and an improvement in the navigation system score to above 98.5 points. Addressing the performance shortcomings of the fault response system, the optimization target is the fault emergency response threshold, and the adjustment range includes the emergency response thresholds for various faults. The adjustment was reduced by 20%. The adjustment method involved debugging the fault emergency handling algorithm and optimizing the fault judgment logic on a simulation test bench and re-burning it to the main control board. This process was consistent with the PID parameter adjustment process. First, the fault response speed and false alarm rate after threshold adjustment were verified in a virtual environment. After confirming that the fault response time was ≤500ms and the false alarm rate was 0, the algorithm was burned and verified on the actual machine. All optimization suggestions underwent feasibility analysis, taking into account the cost of hardware replacement and the operability of software parameter adjustment, to ensure that they could be implemented in actual R&D and to provide a clear performance optimization direction for subsequent iterative verification of processing steps.

[0048] VI. Iterative Verification Process: To achieve continuous optimization of underwater vehicle performance and form a complete verification cycle, the system incorporates four key technical features: preset performance verification standard thresholds, comparison and judgment of comprehensive evaluation results, archiving and storing compliant data, adjustment of non-compliant parameters, and full-process iterative re-verification. All of these features are designed to address issues such as fragmented verification processes, lack of optimization cycles, and non-standard data archiving in existing technologies. The technical means employed are, in order: industry standard-based threshold setting technology combined with scenario requirements; automatic comparison and judgment technology of evaluation values; distributed data encryption archiving technology; parameter adjustment technology based on targeted optimization suggestions; and full-process iterative execution technology. Each feature clearly defines the threshold basis, judgment logic, archiving specifications, adjustment process, and iteration termination conditions to ensure the integrity and effectiveness of the verification cycle.

[0049] The performance verification standard threshold is preset based on industry standards for performance testing and evaluation of underwater unmanned vehicles (UUVs) and the practical application needs of underwater vehicles in marine resource exploration. It also considers the error range of real-sea testing and presets a standardized comprehensive performance evaluation value threshold to ensure the scientific validity and practicality of the threshold. The industry standard specifies a passing threshold of 0.8 for UUV performance verification. Considering the higher requirements for navigation accuracy and fault robustness in marine resource exploration scenarios, the passing threshold is raised to 0.85. That is, a standardized comprehensive performance evaluation value of not less than 0.85 is considered passing performance verification, and a value below 0.85 is considered failing. This threshold has been verified through expert review and real-sea testing, balancing industry standards with practical application needs, and providing a clear judgment standard for performance verification.

[0050] The comprehensive evaluation results are compared and judged using an automated comparison method. The standardized comprehensive performance evaluation value of the submersible is compared and analyzed with a preset pass threshold to achieve accurate and rapid judgment of the verification results. A judgment script is written in the simulation test control software. The standardized comprehensive performance evaluation value of 0.9707, generated from the performance optimization suggestions and the comprehensive evaluation result calculation steps, is imported into the script and automatically compared with the preset threshold of 0.85. If the evaluation value is greater than or equal to the threshold, it is judged as compliant; if the evaluation value is less than the threshold, it is judged as non-compliant. In this verification, 0.9707 is higher than 0.85, and the submersible is judged to have passed the performance verification. The judgment results are displayed in real time on the test console, and a judgment report is automatically generated, including the evaluation value, threshold, judgment result, and performance scores of each system, ensuring the traceability of the judgment results.

[0051] The compliance data archiving and storage system standardizes and encrypts the entire process of verification data, from the acquisition and preprocessing of multi-source heterogeneous basic data to the generation of performance optimization suggestions and the calculation of comprehensive evaluation results, ensuring data security, integrity, and traceability. Archived data includes multi-source heterogeneous basic data, test datasets, deviation quantification data, performance scoring data, gradient test condition sequences, targeted performance optimization suggestions, comprehensive performance evaluation results, and judgment reports. A Hadoop distributed database is used as the data storage medium, deployed on a dedicated server. Data encryption is employed to process the archived data and prevent data leakage. Simultaneously, data metadata specifications are established, adding metadata information such as data source, collection time, test condition, processing personnel, and judgment results to each data set. A hierarchical archiving structure is established according to the test process and data type to achieve orderly data management. The database has tiered access permissions, open only to R&D personnel, with different R&D personnel assigned different operating permissions, including viewing, downloading, and editing, ensuring data security. The archived data serves as core data support for the subsequent mass production and modification of the submersible, and also provides a reference for simulation verification of similar submersibles.

[0052] For scenarios where the standardized comprehensive performance evaluation value fails to meet the standard (e.g., 0.82), a cyclical iterative re-verification process is implemented, following the logic of parameter adjustment – ​​simulation verification – evaluation judgment, to ensure that the submersible's performance ultimately reaches the qualified threshold. Simultaneously, an iteration termination condition is set to avoid invalid iterations. First, based strictly on the performance optimization suggestion generation and the targeted optimization suggestions generated from the comprehensive evaluation result calculation steps, synchronous parameter adjustments are made to both the submersible's physical hardware and virtual model. During the adjustment process, the principle of "simulation first, then physical" is followed. Parameter adjustments and simulation verification are first completed in the virtual model. After confirming the effectiveness of the parameter adjustments, adjustments are made to the physical hardware, such as replacing the high-precision depth gauge, modifying PID algorithm parameters, and lowering the fault emergency response threshold. The virtual model is synchronously updated with the corresponding parameters to ensure consistency with the physical hardware. After parameter adjustment, a full-process simulation verification is initiated, from multi-source heterogeneous basic data acquisition and preprocessing to performance optimization suggestion generation and comprehensive evaluation result calculation. During the verification process, all test parameters and judgment criteria remain unchanged to ensure the comparability of iterations. After the full-process verification is completed, the standardized comprehensive performance evaluation value is recalculated and compared with the qualified threshold. If the threshold is met, the iteration is completed. If the threshold is not met, the parameters are adjusted again based on the new performance evaluation results and optimization suggestions until the standardized comprehensive performance evaluation value reaches the qualified threshold of not less than 0.85. The iteration termination condition is that the standardized comprehensive performance evaluation value reaches the qualified threshold for two consecutive iterations. If the threshold is not met after five iterations, the iteration is paused and the root cause design parameters of the performance bottleneck are re-analyzed. If the performance improvement rate is less than 1%, that is, the improvement rate of the evaluation value between two adjacent iterations is less than 1%, it is determined that the performance has reached the bottleneck.

[0053] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0054] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for hardware-in-the-loop simulation and performance verification of an unmanned underwater vehicle, characterized in that, Includes the following steps: Acquire multi-source heterogeneous basic data required for hardware-in-the-loop simulation and performance verification of unmanned underwater vehicles, and preprocess the multi-source heterogeneous basic data; Based on preprocessed multi-source heterogeneous basic data, a semi-physical simulation operating environment for unmanned underwater vehicles is built, which dynamically adapts the physical hardware and the virtual simulation model, and enables bidirectional collaborative response between the physical and virtual ends. Based on the design performance indicators and actual application scenario requirements of the unmanned underwater vehicle, a gradient test condition simulation test task containing normal working conditions and fault working conditions is generated, and the full-process simulation test of each working condition is executed in the completed hardware-in-the-loop simulation environment. Full-series runtime data during the simulation test and preset real-machine calibration standard data were collected separately. A dynamic weight allocation and iterative optimization method based on working condition characteristics was used to conduct multi-dimensional deviation quantitative comparative analysis of the two types of data. Based on the quantitative comparative analysis of data deviations, and combined with the correlation between the design parameters of the unmanned underwater vehicle and the operation of each system, performance optimization suggestions are generated, and the performance verification and evaluation results of the unmanned underwater vehicle are calculated. The performance verification evaluation results are compared with the preset performance verification standard threshold and a loop iterative verification process is performed. If the evaluation results meet the preset standard, the performance verification is deemed to have passed and the data is archived and stored. Conversely, if the parameters are not optimized, the simulation and performance verification process will be restarted until the evaluation results meet the preset standards.

2. The method for hardware-in-the-loop simulation and performance verification of an unmanned underwater vehicle according to claim 1, characterized in that, Acquiring and preprocessing multi-source heterogeneous basic data includes: The process involves identifying end-to-end data acquisition nodes covering the actual unmanned underwater vehicle (UUV), environmental simulation, test control, and historical data collection. Data related to UUV design, operation, environment, operating conditions, and historical data are collected. Multiple sets of raw heterogeneous data are obtained from each node, and basic processing is performed, including anomaly removal and missing data supplementation. Based on the data's physical attributes and spatiotemporal dimensions, the processed raw heterogeneous data are categorized into different types of target basic data. The categorized target basic data undergoes spatiotemporal dimension matching and data format standardization to complete the heterogeneous data fusion process. Data validity verification rules are established, and the fused target basic data is verified according to these rules, eliminating invalid data to obtain multi-source heterogeneous basic data.

3. The method for hardware-in-the-loop simulation and performance verification of an unmanned underwater vehicle according to claim 1, characterized in that, Building a hardware-in-the-loop simulation environment that dynamically adapts between the physical machine and the virtual model includes: Based on data related to the underwater vehicle's structure, a virtual simulation model matching the full characteristics of the actual hardware is constructed to recreate the full-link operational logic mapping of each system of the underwater vehicle. Based on data related to the underwater environment, a virtual environment model recreating the underwater physical characteristics and changing patterns is constructed. The actual hardware of the unmanned underwater vehicle is connected to the constructed virtual simulation model and underwater virtual environment model through a full-link hardware and software communication connection, establishing a two-way data interaction channel between the actual and virtual ends. Based on the actual operation data of the underwater vehicle's power control system, collaborative debugging of the actual hardware and virtual simulation model is initiated, and real-time parameter correction and dynamic adaptation of the operating status of the actual and virtual ends are performed. Rules for judging the stability of the simulation environment are formulated, and the collaborative operation status of the actual and virtual ends is continuously monitored. When the stability index reaches the preset standard and the parameter correction tends to be stable, it is recorded as the final operating environment of the unmanned underwater vehicle's semi-physical simulation.

4. The method for hardware-in-the-loop simulation and performance verification of an unmanned underwater vehicle according to claim 1, characterized in that, Generating gradient-based test condition simulation tasks and executing full-process simulation tests includes: Based on the performance indicators of unmanned underwater vehicles (UUVs) and the risk characteristics of actual application scenarios, a test indicator system covering all performance dimensions for both routine and fault conditions is established. Based on underwater fault condition characteristic data, a test condition sequence is generated, gradually extending from routine navigation conditions to single-factor fault conditions and multi-factor composite fault conditions. Corresponding test execution parameters are set for each condition in the gradient test condition sequence, along with control requirements for condition execution. Simulation test tasks are initiated in a semi-physical simulation environment according to the set test execution parameters, and the connection between different conditions is performed based on the condition switching trigger conditions. Throughout the entire simulation test process, real-time time-series operational data of each system of the UUV are collected, classified and labeled according to test condition type and fault level, forming a full-volume time-series operational dataset corresponding to the gradient test condition sequence.

5. The method for hardware-in-the-loop simulation and performance verification of an unmanned underwater vehicle according to claim 1, characterized in that, Deviation quantification and comparative analysis includes: Based on the performance dimensions and system classification of the underwater vehicle, features are extracted from the full-scale time-series operational data of the simulation test to construct a simulation feature dataset. Pre-set real-vehicle calibration standard data is retrieved, and features are extracted according to the same rules as the simulation feature dataset to construct a real-vehicle standard feature dataset. Based on the type of test conditions and fault characteristics, a dynamic weight allocation method for condition adaptation is constructed, setting corresponding weight percentages for each performance dimension feature. The simulation feature dataset and the real-vehicle standard feature dataset are compared dimension by dimension to calculate the deviation values ​​and deviation ratios for each performance dimension. Combined with historical verification data from similar underwater vehicles, the dynamic weight allocation method is iteratively optimized, updating the weight percentages for each performance dimension feature. Based on the optimized weight percentages, the deviation values ​​and deviation ratios for each performance dimension are weighted and calculated to obtain the quantitative result of the deviation between the simulation data and the real-vehicle standard data.

6. The method for hardware-in-the-loop simulation and performance verification of an unmanned underwater vehicle according to claim 1, characterized in that, The generated performance verification evaluation results include: Based on the data deviation quantification results, and according to the preset test index system scoring rules, the performance of each system of the unmanned underwater vehicle (UUV) is scored. A correlation analysis is conducted between the UUV's design parameters and the operational performance of each system to confirm the degree of influence of various design parameters on system performance. Based on the performance scores of each system, the data deviation quantification results, and the degree of influence of design parameters, performance optimization suggestions are generated, defining optimization targets, adjustment ranges, and control methods, and quantitatively predicting the optimization effects. The basic weight ratio of each system in the overall operation of the UUV is analyzed, and the basic weight ratio is further corrected based on the actual application scenario weights of the test conditions. The performance scores of each system are weighted and calculated with the corrected weight ratios to obtain the performance verification and evaluation results of the UUV. The evaluation results are then standardized to form quantitative values ​​that can be directly compared with preset standard thresholds.

7. A hardware-in-the-loop simulation and performance verification system for an unmanned underwater vehicle, characterized in that, include: The data processing module is used to acquire multi-source heterogeneous basic data required for the hardware-in-the-loop simulation and performance verification of unmanned underwater vehicles, and to preprocess the multi-source heterogeneous basic data. The environment setup module is used to build a semi-physical simulation operating environment for unmanned underwater vehicles that dynamically adapts to the real hardware and the virtual simulation model based on preprocessed multi-source heterogeneous basic data. The test execution module is used to generate gradient test condition simulation test tasks based on the performance indicators of the unmanned underwater vehicle and the requirements of actual application scenarios, and to execute full-process simulation tests and collect relevant operational data in a semi-physical simulation environment. The comparative analysis module is used to retrieve preset actual machine calibration standard data and adopts a dynamic weight allocation and iterative optimization method based on working condition characteristics to perform multi-dimensional deviation quantitative comparative analysis between simulation operation data and actual machine standard data. The optimization and evaluation module is used to generate performance optimization suggestions based on the data comparison and analysis results. At the same time, it combines the design parameters of the unmanned underwater vehicle with the correlation between the operation of each system to calculate the performance verification and evaluation results of the unmanned underwater vehicle. The judgment module is used to compare the performance verification evaluation results with the preset performance verification standard threshold, and perform iterative verification judgment operations such as archiving and storing the results after they meet the standard, and adjusting parameters and re-simulating and verifying the results after they fail to meet the standard.

8. The hardware-in-the-loop simulation and performance verification system for an unmanned underwater vehicle according to claim 7, characterized in that, The data processing module includes: The node determination unit is used to determine the full-link data acquisition nodes covering the actual submersible end, environmental simulation end, test control end, and historical data end, and to plan the full-dimensional data acquisition range; The preprocessing unit is used to acquire raw heterogeneous data from each acquisition node and perform basic processing operations such as removing abnormal data and supplementing missing data. The classification unit is used to classify and determine the raw heterogeneous data after basic processing into different types of target basic data according to the data's physical attributes and the spatiotemporal dimensions of data collection. The format unification unit is used to perform spatiotemporal dimension matching and data format standardization conversion on the classified target basic data, and to perform heterogeneous data fusion processing. The validity verification unit is used to verify the fused target basic data according to the data validity verification rules, remove invalid data, and output multi-source heterogeneous basic data.

9. The hardware-in-the-loop simulation and performance verification system for an unmanned underwater vehicle according to claim 7, characterized in that, The environment setup module includes: The virtual model building unit is used to construct a virtual simulation model that matches the full characteristics of the actual hardware based on the relevant data of the submarine's body structure, and to perform full-link operation logic mapping of the submarine; The underwater environment model building unit is used to construct a virtual environment model that recreates the underwater physical characteristics and changing patterns based on relevant underwater environment data. The docking unit is used to establish a full-link communication connection between the actual hardware of the unmanned underwater vehicle and the virtual simulation model and underwater virtual environment model, and to build a two-way data interaction channel between the actual and virtual ends. The collaborative calibration unit is used to initiate collaborative debugging between the actual and virtual ends based on the actual operating data of the submersible's power control system, and to perform real-time parameter correction and dynamic adaptation of the operating status of the two. The stability determination unit is used to continuously monitor the collaborative operation status of the physical machine and the virtual terminal according to the stability determination rules of the simulation environment, perform bidirectional calibration, and determine the semi-physical simulation operation environment.

10. The hardware-in-the-loop simulation and performance verification system for an unmanned underwater vehicle according to claim 7, characterized in that, The comparative analysis module includes: The simulation feature extraction unit is used to extract features from the full time-series running data of the simulation test according to the performance dimension and system classification of the submarine, and to construct the simulation feature dataset. The actual machine standard feature construction unit is used to retrieve the preset actual machine calibration standard data, extract features according to the same rules, and construct the actual machine standard feature dataset. The working condition adaptation weight allocation unit is used to construct a dynamic weight allocation method for working condition adaptation based on the type of test working condition and fault characteristics, and to set the weight ratio of each performance dimension feature. The multi-dimensional deviation calculation unit is used to compare the simulation feature dataset with the actual standard feature dataset in each dimension, and calculate the deviation value and deviation ratio of each performance dimension. The weight iteration optimization unit is used to iteratively optimize the dynamic weight allocation method by combining historical verification data of similar underwater vehicles, and update the weight ratio of each performance dimension feature. The deviation calculation unit is used to perform weighted calculations on the deviation values ​​and deviation ratios of each performance dimension based on the optimized weight ratios, so as to obtain the quantification results of the deviation between simulation and actual data.

Images