Acoustic array adaptive calibration correction system applied to underwater moving target

Abstract

The application provides an acoustic array adaptive calibration correction system applied to an underwater moving target, relates to the technical field of marine equipment, and comprises a multi-source excitation and environment perception module, an intelligent array perception and diagnosis module, an adaptive position inversion and uncertainty quantification module and a closed-loop calibration execution and self-learning optimization module; the multi-source excitation and environment perception module is responsible for providing a reference signal required for calibration and establishing an association model of environment and array distortion, the intelligent array perception and diagnosis module realizes multi-modal data acquisition, array element health state monitoring and array geometry self-perception, the adaptive position inversion and uncertainty quantification module completes signal processing, position estimation and error quantification propagation, and the closed-loop calibration execution and self-learning optimization module executes a compensation strategy, verifies calibration effect and continuously optimizes system performance through online learning; the system solves the problems that cannot be handled by traditional methods, such as environment time variation, multi-sensor conflict and uncertainty quantification.

Description

Technical Field

[0001] This invention relates to the field of marine equipment technology, and more specifically to an acoustic array adaptive calibration and correction system for underwater moving targets. Background Technology

[0002] When underwater vehicles perform missions such as ocean exploration, resource surveying, and military reconnaissance, their acoustic arrays serve as core sensor systems, undertaking crucial functions such as target detection, communication, and navigation. The acoustic array receives underwater acoustic signals through multiple elements and performs beamforming to achieve target localization and tracking. However, the complex and ever-changing underwater environment presents recent challenges to maintaining the geometry. Current impacts, pressure changes, temperature gradients, and the dynamic antenna generated by the underwater vehicle itself can cause array element positions to deviate from the design baseline, leading to manifold mismatch. This severely reduces beamforming accuracy, target localization accuracy, and signal gain, directly affecting system performance. Traditional offline deployment methods require specialized equipment in laboratory or dock environments for setup, which cannot adapt to the dynamic changes of the array during missions. Therefore, improving online deployment technology has become a requirement for the robustness and practicality of underwater acoustic systems.

[0003] Existing underwater acoustic array placement techniques mainly include the following three types of algorithms. The first type is the large-scale placement algorithm using the least squares method. This method involves deploying multiple reference sound sources at known locations, measuring the arrival time of the signals received by each array element, establishing an overdetermined system of equations, and applying the least squares algorithm. The advantages of this algorithm are its simplicity, high computational efficiency, and good accuracy in static or quasi-static environments. However, its disadvantages include sensitivity to measurement and multipath interference. When the sound velocity profile is non-uniform or there is strong multipath influence, the displacement of the arrival time measurement will be significantly amplified in the position estimation, leading to a decrease in decision accuracy. This method is a batch processing algorithm and cannot track the dynamic changes in array position in real time, independent of underwater vehicle maneuvers or changes in sea state. The second type is the transfer-based algorithm based on Kalman auger. This method models the array element position as a dynamic state variable, uses the arrival time measurement of the reference sound sources as an observation, and updates the position estimate through extended Kalman auger or unscented Kalman auger. This algorithm can track the array element position slowly in real time, and compared to batch methods, it has better dynamic prediction. However, this method's bias relies on an accurate system model and noise characteristics. When the model is mismatched or the noise characteristics are unknown, the suppression performance deteriorates. More importantly, this method assumes the reliability of single-sensor measurements and cannot handle multi-sensor measurement conflicts. When inertial navigation, acoustic ranging, and angle-of-arrival measurements give inconsistent results, the filter may diverge or converge to an incorrect solution. The third type is a blind algorithm based on subspace methods. This method utilizes the hierarchical characteristics of the coherence matrix of the direct signal to estimate the element position deviation through the orthogonality of the signal subspace and noise subspace. The advantage of this algorithm is that it does not require known signal source positions and can complete inference based solely on received data, making it suitable for scenarios lacking reference sources. However, this method suffers from high computational complexity, high signal-to-noise ratio requirements, and significant performance limitations in low signal-to-noise ratio or strong interference environments. Furthermore, it cannot distinguish between element positions and signal source orientation, resulting in solution ambiguity.

[0004] US Patent 8009516B2 discloses an underwater acoustic positioning system. By deploying multiple reference stations on the water surface, the underwater device receives acoustic pulses from each reference station using a synchronized time reference and calculates its own position based on the time difference of arrival. However, this patent has shortcomings. It assumes the positioning system only estimates the absolute position of a single underwater target, neglecting the relative position adjustment between multiple internal array elements. This makes it unsuitable for handling array distortion caused by rebar reassembly or towed arrays. Furthermore, the system assumes accurate time synchronization for all devices. In real underwater environments, insufficient temperature crystal oscillators and uncertainties in acoustic propagation delays can accelerate clock synchronization, affecting positioning accuracy. The patent also fails to consider the contamination of time of arrival measurements by multipath propagation and lacks robust design for complex acoustic channel environments.

[0005] Chinese patent CN112346102A discloses an underwater acoustic positioning, navigation, and timing system, including a surface segment, an underwater segment, and a user segment. It establishes an underwater spatiotemporal reference through a seabed base station. This patent emphasizes rapid multi-source information fusion, combining inertial navigation, acoustic navigation, and field-matching navigation information to improve the continuity and accuracy of the positioning system. However, this patent has the following shortcomings: First, the deployment method relies on the pre-architecture of its seabed base station, requiring measurements of the seabed physical array, resulting in high deployment costs and complex operation, making it unsuitable for mobile underwater vehicle platforms. Second, while the system mentions multi-source information fusion, it does not provide a mechanism for handling conflicts in sensor measurement results, lacking a robust fusion algorithm for sensor malfunctions or abnormal data. Third, the patent does not address the modeling of the impact of environmental parameter changes; when environmental factors such as ocean currents and thermoclines cause deformation of the support structure, the system cannot predict and compensate for such physical distortions. Fourth, the system lacks a closed-loop verification mechanism; after compensation, the effect cannot be immediately verified and iterative task optimization cannot be performed, making it difficult to guarantee the correct achievement of task requirements. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings by proposing an adaptive calibration and correction system for acoustic arrays applied to underwater moving targets.

[0007] The present invention adopts the following technical solution:

[0008] An acoustic array adaptive calibration and correction system for underwater moving targets includes a multi-source excitation and environmental perception module, an intelligent array perception and diagnosis module, an adaptive position inversion and uncertainty quantification module, and a closed-loop calibration execution and self-learning optimization module.

[0009] The multi-source excitation and environmental perception module is responsible for providing the reference signals required for calibration and establishing a correlation model between the environment and array distortion. The intelligent array perception and diagnosis module realizes multi-modal data acquisition, array element health status monitoring and array geometry self-sensing. The adaptive position inversion and uncertainty quantization module completes signal processing, position estimation and error quantization propagation. The closed-loop calibration execution and self-learning optimization module executes compensation strategies, verifies calibration effects and continuously optimizes system performance through online learning.

[0010] The multi-source excitation and environmental perception module includes a distributed reference sound source network unit, a multi-physics environment modeling unit, and an environmental array coupling analysis unit. The distributed reference sound source network unit is responsible for managing multiple mobile or fixed reference sound sources. The multi-physics environment modeling unit measures and models the spatiotemporal variation of sound velocity profiles, ocean current fields, thermocline distribution, and multipath channel characteristics in real time. The environmental array coupling analysis unit is used to establish a causal relationship model between environmental parameters and array changes.

[0011] The intelligent array sensing and diagnosis module includes a multimodal array signal acquisition unit, an array element state intelligent diagnosis unit, and an array geometry self-sensing unit. The multimodal array signal acquisition unit simultaneously acquires multi-sensor fusion data such as acoustic signals, vibration signals, strain data, and temperature distribution. The array element state intelligent diagnosis unit is used to identify gradual and sudden faults in the array elements. The array geometry self-sensing unit integrates multi-source information such as inertial measurement unit data, acoustic self-cross-correlation measurement, and reference source reverse positioning to estimate the relative positions of the array elements and the array shape change trend.

[0012] The adaptive position inversion and uncertainty quantization module includes a multi-path robust signal processing unit, a hierarchical position estimation unit, and an uncertainty quantization and confidence propagation unit. The multi-path robust signal processing unit performs filtering, matched filtering, and pulse compression processing on the original waveform. The hierarchical position estimation unit is used to provide initial solutions for rapid batch processing and accurately estimate the three-dimensional coordinates of the array elements. The uncertainty quantization and confidence propagation unit quantifies measurement uncertainty, environmental model uncertainty, and algorithm estimation uncertainty.

[0013] The closed-loop calibration execution and self-learning optimization module includes a multi-mode compensation strategy decision-making unit, a closed-loop execution and effect verification unit, and an online learning and knowledge base update unit. The multi-mode compensation strategy decision-making unit is used to intelligently select compensation schemes and realize task-aware adaptive decision-making. The closed-loop execution and effect verification unit issues specific execution instructions and monitors the execution process in real time. The online learning and knowledge base update unit learns the mapping relationship between environmental patterns, array distortion characteristics, and optimal compensation strategies in historical calibration cases, and establishes and continuously updates the knowledge graph.

[0014] Furthermore, the environmental array coupling analysis unit includes a fluid-structure interaction (FSI) calculation processor, a structural response modeling processor, and a distortion source identification processor. The FSI calculation processor combines array material properties and geometric parameters to predict the bending, torsional, and tensile deformations of the array. The structural response modeling processor integrates measurement data from strain sensors, fiber Bragg grating sensors, and accelerometers to establish a dynamic response model of the array support structure under environmental loads, estimating the offset and vibration characteristics of each array element relative to the design reference. The distortion source identification processor identifies the main environmental factors causing array distortion by comparing the correlation between changes in environmental parameters and array deviations.

[0015] The fluid-structure interaction computation processor calculates the deviation vector of all array element positions according to the following formula. :

[0016] ;

[0017] ;

[0018] Where M represents the number of dominant modes. For the m-th order distortion mode, Let be the excitation coefficient for the m-th mode. The residual term introduced for structural damping and nonlinear effects. This is the environment weight vector corresponding to the m-th mode. The magnitude of the ocean current velocity. The direction angle of the ocean current. For pressure gradient, This represents the amplitude of the temperature gradient. Let t be the Reynolds number of the underwater vehicle, and t be the current time.

[0019] Furthermore, the array geometry self-sensing unit includes an inertial navigation fusion processor, an acoustic self-localization processor, and a geometric constraint optimization processor. The inertial navigation fusion processor is used to map the designed position of each array element in the body coordinate system to the global coordinate system to obtain a priori position estimates. The acoustic self-localization processor is used to calculate the relative geometric relationships between array elements. The geometric constraint optimization processor combines the physical constraints of the array, fuses inertial data and acoustic measurement data, and outputs the optimal estimate of the position of each array element and its covariance matrix.

[0020] For the position p of the j-th array element j The geometric constraint optimization processor constructs the optimization objective according to the following formula:

[0021] ;

[0022] ;

[0023] ;

[0024] ;

[0025] ;

[0026] in, For the optimized estimated position of the j-th array element, p j The actual position of the j-th element. For the global estimation of the bias vector, L TOA For the arrival time measurement fitting term, The measured arrival time from reference sound source i to array element j is s i Let i be the position of the reference sound source, and c be the average speed of sound. For adaptive weights of arrival time measurements, L IMU For inertial navigation prior fitting terms, The estimated element positions for the inertial measurement unit. L is the weight of inertial navigation measurements. geom For rigid / semi-rigid connection constraint terms, For a set of rigidly connected array element pairs, To design the reference spacing vector, L is the regularization parameter for geometric constraints. smooth For the smoothing constraint term of the flexible segment, The regularization parameter for smoothing constraints.

[0027] Furthermore, the uncertainty quantification and confidence propagation unit includes an error source decomposition processor, a sensitivity analysis processor, and a confidence synthesis processor. The error source decomposition processor decomposes the total positioning error into independent or related error components such as measurement noise, clock drift, sound velocity profile uncertainty, multipath residuals, array structure deformation, and algorithm approximation errors, and quantifies the contribution ratio of each error source to the final position estimation. The sensitivity analysis processor is used to propagate the input uncertainty to the covariance matrix of the output position and identify the key parameters that have the greatest impact on positioning accuracy. The confidence synthesis processor integrates the statistical test results of the measurement residuals, model matching evaluation, and historical performance statistics to generate a comprehensive confidence score for the position estimation of each array element.

[0028] The confidence synthesis processor calculates the estimated position of the j-th element after fusion according to the following formula. Covariance Matrix :

[0029] ;

[0030] ;

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] in, The fusion weight for the k-th measurement method, Let tr() be the conflict confidence factor for the k-th measurement method, and tr() be the trace function. Let be the covariance matrix of the k-th measurement method. Let k be the average degree of conflict between the k-th measurement and other measurements. For conflict sensitivity parameters, I is the covariance inflation factor, and I3 is the identity matrix. The average degree of conflict across all measurements;

[0036] The confidence synthesis processor calculates the confidence level (Conf) of the estimated position of the j-th array element according to the following formula. j :

[0037] ;

[0038] in, This serves as a reference uncertainty threshold.

[0039] Furthermore, the multi-mode compensation strategy decision unit includes a task awareness processor, a cost-benefit evaluation processor, and a strategy optimization processor. The task awareness processor receives the task type and task priority parameters currently being performed by the underwater vehicle. The cost-benefit evaluation processor quantitatively evaluates the expected calibration effect, execution cost, and failure risk of each candidate scheme, such as digital domain compensation, mechanical geometry adjustment, and redundant array element switching. The strategy optimization processor selects the optimal compensation scheme or a multi-scheme hybrid strategy based on task constraints, confidence thresholds, and cost-benefit evaluation results.

[0040] The preferred processor selects a compensation strategy according to the following formula:

[0041] ;

[0042] ;

[0043] in, For the selected optimal compensation strategy, S1 is digital domain compensation, S2 is mechanical geometric adjustment, and S3 is redundant array element switching. Let S be the expected improvement in positional accuracy of strategy S, R(S) be the inherent risk coefficient of strategy S, C(S) be the execution cost of strategy S, and x be the value of x. t This is the current environment state vector. , and Weighting coefficients are assigned to the three tasks.

[0044] The beneficial effects achieved by this invention are:

[0045] This system constructs a complete adaptive calibration system from environmental perception, multi-source fusion, uncertainty quantification to closed-loop optimization. It solves the technical defects of traditional methods in environmental time-varying adaptability, multi-sensor conflict handling, uncertainty quantization propagation and intelligent decision optimization. It enables underwater acoustic arrays to maintain high-precision geometric state during long-term autonomous operation in complex marine environments, and provides reliable technical support for underwater vehicles to perform high-quality detection, communication and navigation tasks. It has important engineering application value and broad prospects for promotion.

[0046] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the overall structural framework of the present invention;

[0048] Figure 2 This is a schematic diagram of the multi-source excitation and environmental perception module of the present invention;

[0049] Figure 3 This is a schematic diagram of the intelligent array sensing and diagnosis module of the present invention;

[0050] Figure 4 This is a schematic diagram of the adaptive position inversion and uncertainty quantification module of the present invention;

[0051] Figure 5 This is a schematic diagram of the closed-loop calibration execution and self-learning optimization module of the present invention;

[0052] Figure 6 This is a schematic diagram comparing the positioning accuracy of the present invention with other methods under different signal-to-noise ratios;

[0053] Figure 7 This diagram illustrates a comparison of the tracking performance of the present invention with other methods under dynamic conditions. Detailed Implementation

[0054] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated beforehand. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0055] Example 1.

[0056] An acoustic array adaptive calibration and correction system applied to underwater moving targets, combined with Figure 1 It includes a multi-source excitation and environmental perception module, an intelligent array perception and diagnosis module, an adaptive position inversion and uncertainty quantification module, and a closed-loop calibration execution and self-learning optimization module.

[0057] The multi-source excitation and environmental perception module is responsible for providing the reference signals required for calibration and establishing a correlation model between the environment and array distortion. The intelligent array perception and diagnosis module realizes multi-modal data acquisition, array element health status monitoring and array geometry self-sensing. The adaptive position inversion and uncertainty quantization module completes signal processing, position estimation and error quantization propagation. The closed-loop calibration execution and self-learning optimization module executes compensation strategies, verifies calibration effects and continuously optimizes system performance through online learning.

[0058] Combination Figure 2 The multi-source excitation and environmental perception module includes a distributed reference sound source network unit, a multiphysics environment modeling unit, and an environmental array coupling analysis unit. The distributed reference sound source network unit is responsible for managing multiple mobile or fixed reference sound sources, supporting time-division, frequency-division, or code-division multiplexing collaborative working modes, and dynamically selecting the optimal excitation source combination according to the underwater vehicle's maneuvering state to provide high-quality reference signals. The multiphysics environment modeling unit measures and models the spatiotemporal variation of sound velocity profiles, ocean current fields, thermocline distribution, and multipath channel characteristics in real time, using predictive modeling methods rather than simple environmental parameter measurements. The environmental array coupling analysis unit analyzes the distortion effects of environmental factors such as ocean currents, pressure, and temperature on flexible or scalable arrays, establishes a causal relationship model between environmental parameters and array changes, and provides a basis for subsequent physical-driven compensation strategies.

[0059] Combination Figure 3 The intelligent array sensing and diagnosis module includes a multimodal array signal acquisition unit, an array element state intelligent diagnosis unit, and an array geometry self-sensing unit. The multimodal array signal acquisition unit simultaneously acquires multi-sensor fusion data such as acoustic signals, vibration signals, strain data, and temperature distribution, providing cross-validation information sources for subsequent fault diagnosis and position estimation. The array element state intelligent diagnosis unit uses machine learning anomaly detection algorithms to identify gradual and sudden faults in array elements, achieving an upgrade from traditional statistical detection to predictive maintenance. The array geometry self-sensing unit integrates multi-source information such as inertial measurement unit data, acoustic self-cross-correlation measurement, and reference source reverse positioning, and preliminarily estimates the relative positions of array elements and the trend of array shape changes without the need for external precise positioning.

[0060] Combination Figure 4The adaptive position inversion and uncertainty quantification module includes a multi-path robust signal processing unit, a hierarchical position estimation unit, and an uncertainty quantification and confidence propagation unit. The multi-path robust signal processing unit performs filtering, matched filtering, and pulse compression on the original waveform, and uses an adaptive algorithm switching mechanism to separate the direct path and multipath components, extracting high-precision arrival time, time difference of arrival, and angle of arrival features. The hierarchical position estimation unit adopts a two-layer architecture combining coarse estimation and fine estimation. The coarse estimation layer provides an initial solution through fast batch processing, while the fine estimation layer uses iterative optimization algorithms such as extended Kalman filtering, unscented Kalman filtering, or particle filtering to accurately estimate the three-dimensional coordinates of the array elements, balancing real-time performance and accuracy requirements. The uncertainty quantification and confidence propagation unit quantifies measurement uncertainty, environmental model uncertainty, and algorithm estimation uncertainty, and propagates various uncertainties to the final position estimation result through sensitivity analysis and error propagation theory, generating confidence intervals and risk assessment indicators for calibration decision-making.

[0061] Combination Figure 5 The closed-loop calibration execution and self-learning optimization module includes a multi-mode compensation strategy decision-making unit, a closed-loop execution and effect verification unit, and an online learning and knowledge base update unit. The multi-mode compensation strategy decision-making unit intelligently selects compensation schemes such as digital domain compensation, mechanical geometric adjustment, or redundant array element switching based on the confidence level of position estimation, current task priority, energy consumption constraints, and execution time limits, to achieve task-aware adaptive decision-making. The closed-loop execution and effect verification unit issues specific execution instructions and monitors the execution process in real time. After compensation is completed, the calibration effect is immediately remeasured and verified. If the effect does not reach the preset threshold, it automatically triggers iterative calibration or switches alternative strategies to form a complete closed-loop control loop. The online learning and knowledge base update unit learns the mapping relationship between environmental patterns, array distortion characteristics, and optimal compensation strategies in historical calibration cases, establishes and continuously updates the knowledge graph, enabling the system to quickly transfer historical experience to achieve zero-sample adaptation when encountering new environments. At the same time, it detects long-term drift trends, generates maintenance warnings, and supports remote firmware updates.

[0062] The distributed reference sound source network unit includes a sound source scheduling processor, a coded waveform generation processor, and a spatiotemporal synchronization processor. The sound source scheduling processor dynamically selects the active reference sound source combination based on the current position, motion state, and mission mode of the underwater vehicle. It adopts time-division multiplexing, frequency-division multiplexing, or code-division multiplexing strategies to avoid mutual interference between multiple source signals and optimizes the signal coverage and geometric accuracy factor. The coded waveform generation processor generates a broadband linear frequency modulated signal or orthogonal coded sequence carrying a unique identifier and a precise timestamp for each reference sound source. It controls the transmission power, pulse width, and repetition period to adapt to the propagation requirements under different distances and noise environments. The spatiotemporal synchronization processor maintains high-precision time synchronization between all reference sound sources and the underwater vehicle system. It uses two-way acoustic ranging or acoustic-electromagnetic hybrid synchronization protocols to correct the time difference introduced by propagation delay and updates the spatial coordinate information of each reference sound source in real time.

[0063] The multiphysics environment modeling unit includes a sound velocity profile reconstruction processor, an ocean dynamics parameter processor, and a multipath channel prediction processor. The sound velocity profile reconstruction processor integrates real-time measurement data from temperature, salinity, and depth sensors to quickly reconstruct the local sound velocity distribution function and calculate the refraction effect of the sound wave propagation path. The ocean dynamics parameter processor monitors the ocean current velocity field, thermocline depth changes, and water mass movement characteristics, establishes a spatiotemporal evolution model of environmental parameters, and predicts short-term trends, providing physical driving force data for array distortion prediction. The multipath channel prediction processor simulates the multipath propagation structure based on seabed topography, sea surface conditions, and sound velocity profile information, predicts the arrival time, amplitude attenuation, and phase changes of the main reflection paths, and generates a channel impulse response model for use by the signal processing unit.

[0064] The environmental array coupling analysis unit includes a fluid-structure interaction (FSI) computation processor, a structural response modeling processor, and a distortion source identification processor. The FSI computation processor uses a simplified fluid dynamics model to calculate the force distribution of ocean currents, eddies, and pressure fluctuations on the flexible array or scalable support structure. It combines array material properties and geometric parameters to predict the bending, torsional, and tensile deformation of the array. The structural response modeling processor integrates measurement data from strain sensors, fiber Bragg grating sensors, and accelerometers to establish a dynamic response model of the array support structure under environmental loads. It estimates the offset and vibration characteristics of each array element relative to the design reference. The distortion source identification processor identifies the main environmental factors causing array distortion by comparing the correlation between changes in environmental parameters and array deviations, providing a physical explanation and priority ranking for compensation strategies.

[0065] The fluid-structure interaction computation processor calculates the deviation vector of all array element positions according to the following formula. :

[0066] ;

[0067] ;

[0068] Where M represents the number of dominant modes. For the m-th order distortion mode, Let be the excitation coefficient for the m-th mode. The residual term introduced for structural damping and nonlinear effects. This is the environment weight vector corresponding to the m-th mode. The magnitude of the ocean current velocity. The direction angle of the ocean current. For pressure gradient, This represents the amplitude of the temperature gradient. t represents the Reynolds number of the underwater vehicle, and t represents the current time.

[0069] The multimodal array signal acquisition unit includes a synchronous sampling processor, a multi-channel front-end processor, and a data quality assessment processor. The synchronous sampling processor provides a unified high-precision clock reference for all array elements, including hydrophones, vibration sensors, strain gauges, and temperature sensors, enabling multi-channel parallel sampling with microsecond-level time synchronization and adding a global timestamp to each data frame. The multi-channel front-end processor performs low-noise amplification, anti-aliasing filtering, and high-resolution analog-to-digital conversion on the raw analog signals from each sensor. It supports configurable sampling rates and bandwidths to adapt to the characteristics of acoustic and vibration signals in different frequency bands and outputs a time-aligned set of digital waveforms. The data quality assessment processor monitors the signal-to-noise ratio, dynamic range utilization, and sampling integrity of each channel in real time, detects abnormal events such as saturation distortion, impulse interference, and data packet loss, provides data reliability indicators for subsequent diagnostic units, and triggers adaptive gain adjustment or sampling parameter optimization.

[0070] The array element status intelligent diagnostic unit includes an anomaly detection processor, a fault classification processor, and a health measurement processor. The anomaly detection processor is used to analyze the multimodal sensor data stream of each array element, identify abnormal modes that deviate from the normal operating mode, including early fault symptoms such as decreased signal-to-noise ratio, frequency response drift, abnormal vibration, and abnormal temperature. The fault classification processor classifies the detected anomalies into gradual faults, sudden faults, or environmental interference based on the time-domain evolution law, frequency-domain energy distribution, and cross-sensor correlation of the abnormal features, and locates the specific array element or subsystem where the fault source is located. The health measurement processor calculates a comprehensive health index for each array element, integrates fault type, severity, development trend, and remaining service life prediction, generates a color-coded health status report, and triggers corresponding redundancy switching or maintenance alarm strategies.

[0071] The array geometry self-sensing unit includes an inertial navigation fusion processor, an acoustic self-localization processor, and a geometric constraint optimization processor. The inertial navigation fusion processor integrates six-degree-of-freedom attitude data measured by a high-precision IMU, vertical position information from a depth sensor, and cumulative displacement calculated from dead reckoning to estimate the pose of the underwater vehicle's body coordinate system. It also maps the designed positions of each array element in the body coordinate system to the global coordinate system to obtain prior position estimates. The acoustic self-localization processor utilizes cross-correlation measurements between array elements or reverse positioning signals from a reference sound source to calculate the relative geometric relationships between array elements through polygon ranging or angle of arrival measurements. It can sense the local deformation and overall rotation of the array without relying on an external positioning system. The geometric constraint optimization processor combines the physical constraints of the array, fuses inertial data and acoustic measurement data, and outputs the optimal estimate of the position of each array element and its covariance matrix, providing high-quality prior information for the position inversion module.

[0072] For the position p of the j-th array element j The geometric constraint optimization processor constructs the optimization objective according to the following formula:

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] in, For the optimized estimated position of the j-th array element, p j The actual position of the j-th element. For the global estimation of the bias vector, L TOA For the arrival time measurement fitting term, The measured arrival time from reference sound source i to array element j is s i Let i be the position of the reference sound source, and c be the average speed of sound. For adaptive weights of arrival time measurements, L IMU For inertial navigation prior fitting terms, The estimated element positions for the inertial measurement unit. L is the weight of inertial navigation measurements. geom For rigid / semi-rigid connection constraint terms, For a set of rigidly connected array element pairs, To design the reference spacing vector, L is the regularization parameter for geometric constraints. smoothFor the smoothing constraint term of the flexible segment, Regularization parameters for smoothing constraints;

[0079] The multipath robust signal processing unit includes an adaptive filtering processor, a multipath separation processor, and a feature extraction processor. The adaptive filtering processor dynamically selects the signal processing algorithm based on the current signal-to-noise ratio, multipath density, and computational resource constraints. In low signal-to-noise ratio environments, it uses matched filtering and pulse compression to improve the detection probability. In strong multipath environments, it switches to sparse reconstruction to enhance the direct wave separation capability. The multipath separation processor is used to separate overlapping multipath components in the time-delay-Doppler domain or the angle-delay domain, identify direct waves, and suppress contamination of arrival time measurements by non-direct paths such as sea surface reflection and seabed reflection. The feature extraction processor performs accurate peak detection and phase calculation on the separated direct wave signal, extracts the arrival time, arrival time difference, and arrival angle features of each array element, calculates the measurement uncertainty of each feature, and outputs a time-stamped feature vector set for use by the position estimation unit.

[0080] The hierarchical position estimation unit includes a coarse estimation processor, a fine estimation processor, and a multi-hypothesis tracking processor. The coarse estimation processor is used to batch process the time difference of arrival measurements, quickly calculate the initial position estimates of all array elements under relaxed convergence conditions, provide good initial values ​​for fine estimation, and eliminate obvious gross errors. The fine estimation processor establishes a state space model of the array element position, integrates the arrival time, angle of arrival, inertial navigation priors, and environmental model corrections from multiple measurements, and outputs the optimal estimate of the three-dimensional coordinates of the array elements and their covariance matrix through a recursive update mechanism to achieve smooth tracking of slow drift of the array elements. The multi-hypothesis tracking processor handles multimodal position solutions caused by multipath interference or occlusion, maintains multiple candidate position hypotheses, and dynamically updates the probability weights of each hypothesis based on the likelihood of subsequent measurements. After the hypothesis converges or times out, it selects the most likely position solution or triggers additional disambiguation measurements to improve the positioning robustness in complex propagation environments.

[0081] The uncertainty quantification and confidence propagation unit includes an error source decomposition processor, a sensitivity analysis processor, and a confidence synthesis processor. The error source decomposition processor decomposes the total positioning error into independent or related error components such as measurement noise, clock drift, sound velocity profile uncertainty, multipath residuals, array structure deformation, and algorithm approximation errors, and quantifies the contribution ratio of each error source to the final position estimation. The sensitivity analysis processor calculates the Jacobian matrix or gradient vector of the position estimation relative to each input parameter, propagates the input uncertainty to the covariance matrix of the output position, identifies the key parameters that have the greatest impact on positioning accuracy, and provides guidance for sensor optimization. The confidence synthesis processor integrates the statistical test results of measurement residuals, model matching evaluation, and historical performance statistics to generate a comprehensive confidence score and confidence interval for the position estimation of each array element, and outputs a risk assessment report indicating which array element estimation results are reliable and which require additional verification or conservative processing, for the compensation decision unit to refer to.

[0082] The confidence synthesis processor calculates the estimated position of the j-th element after fusion according to the following formula. Covariance Matrix :

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] in, The fusion weight for the k-th measurement method, Let tr() be the conflict confidence factor for the k-th measurement method, and tr() be the trace function. Let be the covariance matrix of the k-th measurement method. Let k be the average degree of conflict between the k-th measurement and other measurements. For conflict sensitivity parameters, I is the covariance inflation factor, and I3 is the identity matrix. The average degree of conflict across all measurements;

[0090] The confidence synthesis processor calculates the confidence level (Conf) of the estimated position of the j-th array element according to the following formula. j :

[0091] ;

[0092] in, For reference uncertainty threshold;

[0093] The multi-mode compensation strategy decision unit includes a task awareness processor, a cost-benefit evaluation processor, and a strategy optimization processor. The task awareness processor receives the task type and priority parameters currently being performed by the underwater vehicle, and determines whether it is permissible to transmit calibration signals, whether mechanical actions are permissible, and the acceptable calibration delay and energy consumption budget. The cost-benefit evaluation processor quantitatively evaluates the expected calibration effect, execution cost, and failure risk of each candidate scheme, such as digital domain compensation, mechanical geometry adjustment, and redundant array element switching, and ranks the overall cost-effectiveness of each scheme. The strategy optimization processor selects the optimal compensation scheme or a multi-scheme hybrid strategy based on task constraints, confidence thresholds, and cost-benefit evaluation results, and generates detailed execution parameters and execution timing plans to ensure that the compensation actions are consistent with the task requirements.

[0094] The preferred processor selects a compensation strategy according to the following formula:

[0095] ;

[0096] ;

[0097] in, For the selected optimal compensation strategy, S1 is digital domain compensation, S2 is mechanical geometric adjustment, and S3 is redundant array element switching. Let S be the expected improvement in positional accuracy of strategy S, R(S) be the inherent risk coefficient of strategy S, C(S) be the execution cost of strategy S, and x be the value of x. t This is the current environment state vector. , and Assess the weighting coefficients for the three tasks;

[0098] The closed-loop execution and effect verification unit includes an execution instruction issuing processor, a real-time monitoring processor, and an effect evaluation processor. The execution instruction issuing processor converts the compensation strategy into specific hardware control instructions, adjusts the delay register, phase compensation coefficient, and channel gain parameters in the FPGA or DSP for digital domain compensation schemes, sends servo position commands or piezoelectric driver voltage control signals for mechanical adjustment schemes, and sets execution timeout and safety limit protection mechanisms. The real-time monitoring processor continuously collects feedback data from position sensors, torque sensors, and current sensors during the compensation execution process, corrects execution deviations, detects abnormal conditions such as mechanical jamming, overload, or digital compensation saturation, and triggers safety stop or backoff strategies. The effect evaluation processor immediately starts the verification measurement of the reference sound source after compensation, re-extracts the arrival time characteristics and calculates the array element positions, compares the compensated position error, beam pointing accuracy, or array manifold matching degree with the pre-compensation and target thresholds. If the effect is not up to standard, the failure reason is marked and iterative calibration or strategy switching is triggered. If the effect is up to standard, a success case is recorded for use by the learning unit.

[0099] The online learning and knowledge base update unit includes a case learning processor, a knowledge graph maintenance processor, and a predictive maintenance processor. The case learning processor collects complete data from each calibration process, extracts the mapping rules between environmental patterns and optimal compensation strategies, and updates the parameters of the decision model, enabling the system to select effective compensation schemes more quickly and accurately in similar environments. The knowledge graph maintenance processor establishes and dynamically updates a knowledge graph containing environmental type nodes, distortion mode nodes, and compensation strategy nodes, realizing zero-shot learning and transfer learning. When encountering a new environment, it quickly retrieves similar historical cases and generates preliminary compensation suggestions. The predictive maintenance processor analyzes the long-term trends of array element health status and calibration frequency, uses time series prediction or survival analysis models to predict the remaining service life and next failure time of each component, generates preventive maintenance plans and spare parts replacement suggestions, and supports sending health reports and receiving firmware update packages to the shore-based command center via low-bandwidth underwater acoustic communication or satellite links.

[0100] The i, j, and k mentioned above are ordinal numbers used to represent sequence numbers and have no actual meaning.

[0101] To verify the beneficial effects of this invention, the applicant conducted a systematic comparative experiment in a real underwater environment. The experiment used a 32-element linear hydrophone array with an element spacing of 0.5 meters and an operating frequency range of 1-10 kHz. The experimental area was located in a sea area of ​​the South China Sea, at a depth of approximately 120 meters, with a muddy or sandy seabed and sea state 2-3. The experiment compared the performance of the method of this invention with traditional least squares, extended Kalman filtering, and subspace blind calibration methods under different signal-to-noise ratios, different multipath environments, and dynamic time-varying scenarios.

[0102] The first step involved system initialization and environmental parameter measurement. After the underwater vehicle descended to its operating depth, seawater temperature, salinity, and pressure distribution were measured in real time using a temperature, salinity, and depth sensor to reconstruct the sound velocity profile. In this experiment, the surface sound velocity was measured at 1523 m / s, the thermocline depth was approximately 45 meters, and the sound velocity below the thermocline dropped to 1485 m / s. Simultaneously, an acoustic Doppler current profiler was activated to measure the ocean current field, measuring a current velocity of 0.3-0.5 knots, primarily flowing southeast. Based on the environmental parameter data, a multiphysics environmental modeling unit established a sound velocity profile model and a spatiotemporal evolution model of the ocean current field, providing physical driving force input for subsequent calibration. The second step involved the deployment of distributed reference sound sources and signal acquisition. Five movable reference sound sources were deployed in front of, to the sides of, and below the underwater vehicle. These reference sound sources emitted linear frequency modulated signals with a center frequency of 5 kHz, a bandwidth of 2 kHz, and a pulse width of 20 ms. Each reference sound source operates in a time-division multiplexing mode with a transmission interval of 100ms to avoid signal interference. The three-dimensional coordinates of each reference sound source are accurately determined using acoustic self-localization and inertial navigation systems, with a position measurement accuracy better than 0.1 meters. 32 array elements synchronously acquire reference sound source signals at a sampling rate of 48kHz and a sampling duration of 60 seconds, obtaining complete multi-channel time-series data. The third step is signal processing and feature extraction. A multipath robust signal processing unit performs bandpass filtering on the original waveform to remove out-of-band noise and interference, followed by matched filtering and pulse compression, significantly improving the signal-to-noise ratio and time resolution. In strong multipath scenarios, the system automatically switches to a sparse reconstruction algorithm, successfully separating the direct wave from multipath components such as sea surface reflection and seabed reflection. The feature extraction processor accurately extracts the arrival time of the direct wave received by each array element, achieving a measurement accuracy at the microsecond level, while simultaneously estimating the angle of arrival information, providing high-quality observational data for position inversion. The fourth step is position inversion and uncertainty quantification. The hierarchical position estimation unit first performs rapid batch processing through a coarse estimation processor, using weighted least squares to solve the overdetermined equations and obtain initial solutions for the positions of each array element. Subsequently, the fine estimation processor uses extended Kalman filtering for iterative optimization, fusing inertial navigation prior information, acoustic ranging data, and geometric constraints to output the three-dimensional coordinates and covariance matrix of each array element. The uncertainty quantification unit systematically decomposes measurement noise, sound velocity profile uncertainty, multipath residuals, and algorithm approximation errors. Through sensitivity analysis, it calculates the contribution ratio of each error source to the final position estimation, generating confidence scores and confidence intervals to provide a quantitative basis for compensation decisions. The fifth step is closed-loop calibration execution and effect verification. Based on the position estimation results and confidence scores, the multi-mode compensation strategy decision unit intelligently selects a digital domain compensation scheme, adjusts the time delay and phase compensation parameters of each array element, and corrects array manifold mismatch. After compensation, the effect evaluation processor immediately initiates verification measurements to recalculate the array element position errors and beamforming performance indicators.If the calibration effect does not reach the preset threshold, the system automatically triggers a second round of iterative calibration until the accuracy requirements are met or the maximum number of iterations is reached. The data is then processed. Figure 6 and Figure 7 .

[0103] Example 2.

[0104] An adaptive calibration and correction system for acoustic arrays applied to underwater moving targets includes a multi-source excitation and environmental perception module, an intelligent array perception and diagnosis module, an adaptive position inversion and uncertainty quantization module, and a closed-loop calibration execution and self-learning optimization module. The multi-source excitation and environmental perception module is responsible for providing the reference signals required for calibration and establishing a correlation model between the environment and array distortion. The intelligent array perception and diagnosis module realizes multi-modal data acquisition, array element health status monitoring, and array geometry self-sensing. The adaptive position inversion and uncertainty quantization module completes signal processing, position estimation, and error quantization propagation. The closed-loop calibration execution and self-learning optimization module executes compensation strategies, verifies calibration effects, and continuously optimizes system performance through online learning.

[0105] The multi-source excitation and environmental perception module includes a distributed reference sound source network unit, a multi-physics environment modeling unit, and an environment array coupling analysis unit. The distributed reference sound source network unit uses the ST1800 acoustic beacon from ELAC Nautik (Germany) as the reference sound source, and each beacon is equipped with a power amplifier to achieve a sound source level of 195 dB re The system operates at 1 μPa@1m, with a frequency band of 2-12kHz, supporting three waveform types: linear frequency modulation (LFM), M-sequence coded, and Costas frequency hopping. The transmission cycle is dynamically adjusted based on the underwater vehicle's speed and maneuverability. In stationary hovering mode, the transmission interval is set to 200ms to reduce energy consumption, while in high-speed maneuvering mode, it is shortened to 50ms to improve the update rate. The reference sound sources are managed using the TDMA (Time Division Multiple Access) protocol, with each source occupying a fixed time slot of 10ms and a guard interval of 2ms. This ensures that the five reference sound sources complete a full signal transmission cycle within 60ms without mutual interference. Compared to the traditional single-source polling method, this implementation can shorten the calibration cycle by 60%. Actual measurement data shows that in a dynamic environment with a current speed of 0.8 knots, the position tracking delay using TDMA is reduced from 350ms to 140ms. The step processor integrates Seascan's PrecisionTime-3000 underwater acoustic timing module. This module uses bidirectional time-of-flight measurement technology to achieve microsecond-level time synchronization. In practice, the underwater vehicle first sends interrogation signals to each reference sound source. After receiving the signals, the reference sound sources immediately return response signals, carrying their local timestamps. The underwater vehicle calculates the propagation delay based on the round-trip time and the known sound speed profile, and corrects the clock deviation between the reference sound sources. Experiments show that the time synchronization accuracy of this bidirectional ranging method is better than 5 microseconds within a 100-meter range, which is 40% higher than the accuracy of the traditional one-way timing method. As an alternative, when precise timing equipment is unavailable, a self-synchronization method based on signal cross-correlation can be used. This method analyzes the relative time delay of different reference sound source signals arriving at each array element to invert the clock deviation. Although the absolute accuracy of this method is slightly lower, the relative accuracy still meets the calibration requirements.

[0106] The multiphysics environment modeling unit integrates the Sea-Bird Electronics SBE 49 FastCATCTD sensor, which has a sampling rate of 16Hz, a temperature resolution of 0.0001℃, a salinity accuracy of 0.003 PSU, and a depth accuracy of 0.1% of full scale. Using real-time acquired temperature, salinity, and depth data, the sound velocity profile reconstruction processor calculates the sound velocity distribution of each water layer using the Mackenzie empirical formula or the UNESCO standard formula. After establishing discrete depth-sound velocity data points, a continuous sound velocity function is generated through cubic spline interpolation. Field measurements show that in the typical thermocline environment of the South China Sea, the root mean square deviation between the sound velocity profile reconstructed using 16Hz high-speed CTD data and the measurement results from the shipborne CTD profiler is less than 0.3 m / s, significantly better than the 8.5 m / s deviation of the sound velocity profile using the historical database. The ocean dynamics parameter processor is equipped with Teledyne RDI's Workhorse Navigator. The DVL Doppler current meter employs a four-beam Janus configuration, operates at a frequency of 600kHz, has a velocity measurement range of ±10m / s, and a resolution of 0.1cm / s. It can simultaneously measure three-dimensional current vectors and output ocean current profile data. In practical applications, the DVL is mounted on the bottom of an underwater vehicle and emits downwards. The vehicle velocity is obtained by measuring the Doppler frequency shift of the backscattered signal from the seabed. Simultaneously, the relative current velocity is measured using a water layer scattering mode. The absolute ocean current field distribution is obtained by vectorially synthesizing the vehicle velocity and the current velocity. As an economical alternative, a single-point current sensor, such as Nortek's Aquadopp acoustic Doppler current meter, can be used. Although it cannot directly measure the current field profile, layered current velocity data can be obtained through fixed-point hovering measurements at different depths by the underwater vehicle. The multipath channel prediction processor is based on the Bellhop ray tracing model or KR... The AKEN normal mode model was used to simulate underwater acoustic propagation. The input parameters included sound velocity profile, seabed topography, seabed sediment type, and sea surface state. The seabed topography data was obtained from real-time side-scan sonar mapping or pre-installed electronic nautical charts. The seabed sediment parameters were set as follows: for muddy and sandy seabeds, the sound velocity was 1550 m / s, the density was 1.5 g / cm³, and the attenuation coefficient was 0.5 dB / λ; for rocky seabeds, the sound velocity was 1800 m / s, the density was 2.0 g / cm³, and the attenuation coefficient was 0.8 dB / λ. The sea surface roughness was estimated from the real-time wind speed based on the Pierson-Moskowitz wave spectrum model. The simulation results output the arrival time, arrival angle, and relative amplitude of the main propagation paths. Experimental verification showed that in a moderate multipath environment, the correlation coefficient between the multipath structure predicted by the Bellhop model and the measured channel impulse response reached 0.78, providing reliable prior information for multipath separation algorithms.

[0107] The environmental array coupling analysis unit employs a fluid-structure interaction (FSI) modeling method based on finite element analysis. A mechanical response model is established for the flexible towed array mounted on the underwater vehicle. The array support structure uses polyurethane-coated multi-strand steel wire ropes with an outer diameter of 25mm, a tensile strength of 50kN, and a unit length mass of 1.2kg / m. Thirty-two hydrophones are installed at equal intervals of 0.5m on the flexible substrate, each encapsulated within a 60mm diameter oil-filled spherical chamber. The FSI calculation processor uses a hybrid strategy of offline simulation and online rapid estimation, employing either the ANSYS Fluent software package or the OpenFOAM open-source tool. In the offline phase, a parameterized deformation database is established for typical ocean current velocity ranges, dividing the current velocity from 0.1 knots to 2.0 knots into 20 operating conditions, with ocean current injection angles ranging from 0 degrees to 180 degrees. The simulation dataset was divided into 20×13=260 sets of typical working conditions by 15-degree division. For each set, the three-dimensional deformation distribution of the array under steady-state ocean current was calculated. In the online phase, the corresponding deformation prediction value was quickly retrieved based on real-time measured ocean current parameters using table lookup and multivariate interpolation methods. The time for a single lookup and interpolation calculation was less than 5ms, which is more than 1000 times faster than real-time finite element solution. Experimental data showed that under a 0.6 knot transverse ocean current, the maximum lateral offset of the 32-element linear array was 15.3cm, and the spacing between the first and last elements was stretched from the design baseline of 15.5m to 15.72m. The deviation from the fluid-structure interaction simulation prediction value was within 8%. The structural response modeling processor deployed eight fiber Bragg grating (FBG) strain sensors at key locations in the array support structure, using the ODiSI from Luna Systems, Inc. The 6000 fiber optic sensor demodulator has a spatial resolution of 2.6 mm, a strain measurement range of ±15000 με, and a sampling rate of 250 Hz. By measuring the axial and bending strain distribution of the supporting structure and combining beam theory from mechanics of materials, the offset of each array element is calculated. In practice, eight FBG sensors are deployed on the upper and lower surfaces of the front, middle, and tail sections of the array. Two sensors in the front section monitor the axial tension caused by the head drag force, four sensors in the middle section monitor the bending deformation caused by the lateral ocean current, and two sensors in the tail section monitor the end vibration. The actual measurements show that the variation trend of FBG strain data under dynamic sea conditions is highly correlated with the vibration signal measured by the accelerometer, with a correlation coefficient of 0.85. By fusing FBG strain and accelerometer data using Kalman filtering, the accuracy of array element position estimation can be improved from 18cm to 9cm using accelerometers alone. As an alternative, when fiber optic sensing systems are unavailable, piezoelectric thin-film sensors or MEMS accelerometer arrays can be used to achieve similar functionality. Although the single-point measurement accuracy is slightly lower, acceptable results can be achieved by increasing sensor density and using data fusion techniques. The distortion source identification processor uses principal component analysis (PCA) to extract the dominant distortion mode from multi-source observation data. The environmental parameter time series, including ocean current velocity, ocean current direction, pressure change, and temperature gradient, constitute the input matrix X. The array element position deviation... The time series constitutes the output matrix Y. The cross-covariance matrix of the input and output is calculated, and singular value decomposition (SVD) is performed to extract the first five principal modes. Experimental data analysis shows that the first principal mode corresponds to a 67% contribution from the direct drag effect of ocean current velocity; the second principal mode corresponds to a 21% contribution from the yaw effect caused by changes in ocean current direction; the third principal mode corresponds to an 8% contribution from the inertial force generated by the underwater vehicle's maneuvering; and the contribution rates of the remaining higher-order modes are all less than 2% and can be ignored. Based on the principal mode decomposition results, targeted compensation strategies can be designed. For example, global coordinate system scaling compensation can be used for the axial tensile deformation corresponding to the first principal mode, and piecewise polynomial fitting compensation can be used for the bending deformation corresponding to the second principal mode.

[0108] The intelligent array sensing and diagnostic module includes a multi-modal array signal acquisition unit, an array element status intelligent diagnostic unit, and an array geometry self-sensing unit. The multi-modal array signal acquisition unit uses Texas Instruments' (TI) ADS1299 eight-channel 24-bit high-precision ADC chip to construct a 32-channel synchronous sampling system. Each set of four channels shares one ADS1299 chip, requiring a total of eight chips. All chips achieve sampling synchronization through a shared master clock signal, provided by a high-stability crystal oscillator (OCXO) with frequency stability better than ±0.1ppm and temperature drift less than ±1ppb / ℃. The synchronous sampling processor reads the conversion results from the eight ADC chips in parallel via an SPI bus. The sampling rate is configured at 48kHz to meet the sampling theorem requirements for 10kHz acoustic signals. Each ADC channel is front-end configured with a programmable gain amplifier (PGA) and an anti-aliasing low-pass filter. The PGA gain can be adjusted in 6dB steps from 0dB to 60dB to adapt to different distances. The reference sound source signal strength was measured, and the anti-aliasing filter adopted an 8th-order Butterworth low-pass filter with a cutoff frequency of 20kHz and a roll-off rate of -48dB / octave. The actual measurement showed that the signal-to-noise ratio of the sampling system reached 108dB and the total harmonic distortion (THD) was less than -100dB, which is significantly better than the 90dB signal-to-noise ratio of commercial acoustic recorders. The data quality assessment processor monitors the statistical characteristics of each channel in real time, calculates the root mean square value, peak factor, spectral energy distribution and inter-channel correlation coefficient of each frame of data. When the RMS value of a certain channel suddenly drops by more than 50% or the spectral energy is concentrated at the power frequency of 50Hz, it is judged as a sensor failure. When the correlation coefficient between a certain channel and the adjacent channel drops sharply from the normal 0.6 or above to below 0.2, it is judged as a cable detachment or open circuit. In the actual sea trial, this intelligent detection mechanism successfully identified the sensitivity reduction fault of the No. 18 hydrophone caused by water ingress due to seal failure, and avoided erroneous calibration results by timely switching to redundant array elements.

[0109] The intelligent diagnostic unit for array element status implements fault early warning based on an anomaly detection algorithm using machine learning. The anomaly detection processor uses the Isolation Forest algorithm to train a statistical distribution model of multimodal sensor data under normal operating conditions. The training dataset contains 200 hours of marine test data covering different sea states and mission scenarios. The feature vectors include time-domain statistics such as mean, variance, skewness, and kurtosis of each channel; frequency-domain characteristics such as dominant frequency, bandwidth, and spectral centroid; and cross-channel cross-correlation coefficients and coherence functions. The Isolation Forest model parameters are set to 100 decision trees, a maximum depth of 8 layers, and a contamination rate of 0.05. After training, the model can score newly acquired data frames for anomalies in real time, with a score range of 0 to 1. Normal data scores are close to 0, and abnormal data scores are close to 1. A threshold of 0.6 is set as the alarm trigger condition. Experimental verification shows that the model has a detection accuracy of 92% for gradual faults such as hydrophone sensitivity aging and a detection accuracy of 98% for sudden faults such as cable breakage. The fault classification processor is based on a support vector machine (SVM) multi-classification model. The fault types were categorized into five types: decreased hydrophone sensitivity, poor cable contact, preamplifier saturation distortion, strong environmental interference, and multipath reverberation anomalies. The SVM model used a radial basis function (RBF) kernel with a regularization parameter C=100 and a kernel parameter γ=0.01. The parameters were optimized through 10-fold cross-validation, achieving a classification accuracy of 89% on the test set. The health metric processor calculated the health index HI for each element based on the anomaly score, fault type, fault duration, and development trend. The HI value ranged from 0 to 100, with 90 points or above indicating an excellent state where normal use was possible, 70-90 points indicating a good state requiring close monitoring, 50-70 points indicating an available state requiring maintenance, and below 50 points indicating a fault state requiring immediate replacement of redundant units. In practical applications, the health metric mechanism provided an early warning of aging failure for the 9th hydrophone 3 days in advance, giving the maintenance team sufficient time to plan the replacement and avoid task interruption.

[0110] The array geometry self-sensing unit fuses multi-source positioning information to achieve high-precision position estimation. The inertial navigation fusion processor integrates a Northrop Grumman LN-200 fiber optic gyroscope inertial measurement unit (IMU), which includes a three-axis fiber optic gyroscope and a three-axis accelerometer. The gyroscope drift is 0.35° / hr, the accelerometer bias is 30μg, and the output frequency is 200Hz. The position, velocity, and attitude of the underwater vehicle are calculated in real time using the strapdown inertial navigation algorithm (SINS). Depth information comes from a Paroscientific 8000 series quartz crystal pressure sensor with an accuracy of 0.01%FS and a resolution of 0.001%FS. GPS position is obtained from a Trimble BD992 dual-frequency GNSS receiver when the underwater vehicle surfaces. With an accuracy better than 1 meter, the dead reckoning (DR) algorithm is used during underwater navigation to accumulate velocity and attitude data from the IMU output to calculate the position. Depth constraints and seabed topography matching are introduced to correct the accumulated error. Experimental results show that after one hour of underwater navigation, the pure inertial navigation accumulated error is approximately 500 meters. Adding depth constraints reduces this to 200 meters, and introducing topography matching further reduces it to 50 meters. The acoustic self-localization processor estimates the relative geometric relationship using cross-correlation measurements between 32 array elements. Specifically, elements 1 and 32 at the beginning and end of the array are selected as reference nodes, and the other 30 elements are used as nodes to be located. The position of the reference nodes is determined by inertial navigation priors and the inference from the reference sound source. To obtain high-precision initial values ​​for positioning, an overdetermined system of equations is established based on TOA (Time of Arrival) or TDOA (Time Difference of Arrival) measurements to solve for the relative coordinates of 30 nodes to be positioned. To improve robustness, an iterative weighted least squares (IRLS) algorithm is used to reduce the impact of gross errors. The initial weights are set to the reciprocal of the ranging accuracy. After each iteration, the weights are adjusted according to the residual size. For measurement points with residuals greater than 3 times the median, the weights are reduced to 0.1 times their original value. Convergence is usually achieved after 3-5 iterations. A geometric constraint optimization processor introduces physical constraints of the array in the position estimation to improve the rationality of the solution. For rigid sections of the towed array, such as the first 10 array elements, rigid supports are used. For rod connections, a rigid connection constraint term is added to the optimization objective function to penalize the deviation of any adjacent array element spacing from the design value of 0.5m. For flexible sections, such as the 12 array elements in the middle connected by flexible cables, a certain degree of bending deformation is allowed, but curvature continuity is required. A smoothness constraint term is added to the objective function to penalize the abrupt change in second-order difference, i.e., curvature. By adjusting the regularization parameter of the constraint term, the weights of data fitting and prior constraints are balanced. Experiments show that the stability of position estimation is significantly improved after adding geometric constraints, especially when the quality of observation data deteriorates. Under a signal-to-noise ratio of 10dB, the position error of unconstrained estimation is 12.7cm, which is reduced to 7.8cm after adding constraints, an improvement rate of 38%.

[0111] The adaptive position inversion and uncertainty quantization module includes a multipath robust signal processing unit, a hierarchical position estimation unit, and an uncertainty quantization and confidence propagation unit. The multipath robust signal processing unit designs an adaptive processing strategy for complex multipath environments. The adaptive filter processor dynamically selects the signal processing algorithm based on the real-time evaluated signal-to-noise ratio (SNR) and multipath density. In high SNR, low multipath scenarios, a computationally less computationally intensive matched filter (MF) is used. The matched filter convolves the time-reversed conjugates of the received and transmitted signals, outputting an SNR gain equal to the time-bandwidth product (TB). For an LFM signal with a pulse width of 20ms and a bandwidth of 2kHz, TB = 40, or 16dB, in low SNR or medium multipath scenarios. In this scenario, the system switches to an Adaptive Matched Filter (AMF). The AMF introduces a noise whitening filter on top of the matched filter to suppress correlated noise and reverberation background. In strong multipath scenarios, a sparse reconstruction algorithm based on compressed sensing is employed. The multipath channel is modeled as a sparse representation in the time-delay-amplitude domain, and the sparse solution is solved using the OMP (Orthogonal Matching Pursuit) or FOCUSS algorithm. Experimental comparisons show that in strong multipath environments with more than five multipath paths, the arrival time estimation error of the traditional matched filter is 8.3 μs, while the sparse reconstruction algorithm can reduce the error to 2.1 μs, a 75% improvement. The multipath separation processor uses a CLEAN iterative deconvolution algorithm to separate overlapping multipath components. This algorithm first detects the strongest multipath component in the channel response. The peak value is subtracted from the ideal impulse response corresponding to that peak value, and then the second strongest peak value is detected in the residual signal. The above process is repeated until the residual energy is below the threshold. The shape of the ideal impulse response in each iteration is determined by the offline measured system impulse response or transmitted signal waveform. Key parameters of the CLEAN algorithm include a cycle gain γ set to 0.8, which means that 80% of the peak energy is subtracted each time, leaving 20% ​​to avoid over-subtraction. The iteration termination threshold is set to 3 times the noise power. Experimental results show that the CLEAN algorithm can effectively separate multipath components with a time interval greater than half a pulse width, i.e., 10ms. For denser multipath components, it can be combined with super-resolution spectral estimation methods such as the MUSIC algorithm to further improve the resolution. The feature extraction processor... After separation, the direct wave signal is used to perform accurate time of arrival estimation. Parabolic interpolation is used to fit a quadratic curve near the relevant peak to obtain the peak position with sub-sample point accuracy. At a sampling rate of 48kHz, the single sample interval is about 20.8μs. Parabolic interpolation can improve the time resolution to within 2μs. At the same time, phase calculation method is used to further improve the accuracy. The phase information of the LFM signal is used to convert the phase difference between the received signal and the local reference signal into time delay. The phase measurement accuracy can reach 0.1°, corresponding to a time delay accuracy of about 0.14μs for a 2kHz bandwidth signal. The combined estimation of amplitude and phase information can achieve a time of arrival measurement accuracy of 1μs, corresponding to a distance resolution of 15cm.

[0112] The hierarchical position estimation unit adopts a two-level processing architecture to balance real-time performance and accuracy. The coarse estimation processor performs fast batch processing positioning on the arrival time data of a single observation. It employs a closed-form solution method based on the Chan algorithm to first transform the nonlinear spherical ranging equation into a pseudo-linear system of equations. The closed-form solution for the position is obtained through two weighted least squares operations. The advantages of the Chan algorithm are that it does not require iteration to guarantee convergence and has a small computational load, with a single positioning time of approximately 2ms. Its disadvantage is that it is sensitive to measurement noise, especially when the geometric distribution of the reference sound source is poor, which can easily lead to large deviations. To improve the reliability of the coarse estimation, the input... Obvious outliers are removed from the input data. A measurement value is considered outlier and removed if its deviation from the median exceeds three times the standard deviation. The position output from the coarse estimate is used as the initial value for the fine estimate and is also used to detect abrupt events. The fine estimate processor establishes a state-space model of the array element positions and uses an extended Kalman filter (EKF) for recursive updates. The state vector includes the three-dimensional position coordinates and velocities of each element, totaling 192 dimensions. For a 32-element array, the state transition equation uses a uniform motion model, assuming the element velocities remain constant over short time scales. The observation equation is a nonlinear ranging equation, which is linearized using a Jacobian matrix. The noise covariance is set according to the motion characteristics of the array. The rigid section has lower process noise, while the flexible section has higher process noise, reflecting its stronger deformation capability. The observation noise covariance is dynamically adjusted based on the real-time assessment of measurement uncertainty. The observation weight increases when the signal-to-noise ratio is high and decreases when the signal-to-noise ratio is low. The advantage of EKF is its ability to smooth measurement noise and continuously track the slow changes in array element positions. Experiments show that compared to the coarse estimate of 12cm error, the fine estimate can reduce the error to 5cm, an improvement of 58%. The multi-hypothesis tracking processor handles multi-mode solutions caused by multipath propagation. In strong multipath environments, time-of-arrival measurements may have modalities. Ambiguity leads to multiple possible position solutions. The multi-hypothesis tracker maintains K hypothetical trajectories simultaneously. K is usually set to 3 to 5. Each hypothesis corresponds to a possible sequence of array element positions. As new observation data arrives, the probability weights of each hypothesis are updated according to the likelihood function. When the probability of a hypothesis exceeds the threshold of 0.9, it is confirmed as a correct solution and the remaining hypotheses are deleted. When the probabilities of all hypotheses are low, it indicates poor observation data quality and triggers additional disambiguation measurements, such as adding a reference sound source or introducing angle of arrival information. In actual testing, the multi-hypothesis tracking mechanism effectively avoids position jumps caused by multipath interference and ensures the continuity of tracking.

[0113] The uncertainty quantification and confidence propagation unit performs a full-chain analysis of the positioning error. The error source decomposition processor quantifies the contribution of each error source based on variance analysis. The total position estimation variance is decomposed into a weighted sum of measurement noise variance, sound velocity uncertainty variance, geometrical precision factor (GDOP), and algorithm estimation variance. The measurement noise variance is estimated using the variance of repeated measurements. The sound velocity uncertainty variance is calculated based on the confidence interval of the sound velocity profile and the propagation path length. GDOP reflects the influence of the geometric distribution of the reference sound source on the positioning accuracy. When the reference source is tetrahedral, GDOP is minimized to approximately 1.5. When the GDOP increases to over 5 in a coplanar distribution, the algorithm estimation variance originates from linearization error and iteration truncation error, obtained through Monte Carlo simulation statistics. Experimental data shows that under typical operating conditions, measurement noise contributes 45%, sound velocity uncertainty contributes 30%, GDOP affects 20%, and algorithm error accounts for 5%. Based on the error source decomposition results, targeted optimizations can be made. For example, when sound velocity uncertainty dominates, the CTD measurement frequency should be increased; when GDOP is poor, the placement position of the reference sound source should be optimized. The sensitivity analysis processor calculates the partial derivative matrix of the position estimate with respect to the input parameters, i.e., the Jacobian matrix, through a first-order Taylor expansion. The covariance matrix of the input parameters is propagated to the covariance matrix of the output position. The Jacobian matrix is ​​calculated using a numerical differentiation method, applying a small perturbation (e.g., 0.1% relative perturbation) to each input parameter, observing the output change, and calculating the partial derivatives. In actual calculations, it was found that position estimation is most sensitive to time of arrival measurements, with a sensitivity coefficient of approximately 300 m / s, meaning a 1 μs time error results in a 30 cm position error. Sensitivity to sound velocity parameters is the next highest, with a sensitivity coefficient of approximately 0.1, meaning a 1% sound velocity error results in a 0.1% position error in the baseline. The confidence score processor integrates multi-source information to generate a comprehensive confidence score. Inputs include... The chi-square test statistic of the measurement residuals, the signal-to-noise ratio of the observed data, the GDOP value of the geometric distribution of the reference sound source, the success rate of historical calibration, and the consistency index of multi-source information are used to map multidimensional features into a confidence score of 0-100 through a fuzzy inference system or neural network. The confidence score is strongly correlated with the actual error of the location estimation. Experimental statistics show that the estimation results with a confidence score of 90 or above have a 95% probability of error less than 5 cm, the estimation results with a confidence score of 60-90 have an 80% probability of error less than 10 cm, and the estimation results with a confidence score below 60 have uncontrollable error and should be rejected.

[0114] The closed-loop calibration execution and self-learning optimization module includes a multi-mode compensation strategy decision-making unit, a closed-loop execution and effect verification unit, and an online learning and knowledge base update unit. The multi-mode compensation strategy decision-making unit intelligently selects a compensation scheme based on task priority and system status. The task awareness processor receives task parameters from the task controller, including task type (e.g., target detection, communication relay, or navigation positioning), task priority (divided into high, medium, and low levels), allowable calibration time ranging from 10 seconds to 300 seconds, and allowable calibration energy consumption ranging from zero power consumption to full-power transmission of the reference signal. For high-priority emergency tasks such as underwater target search and rescue requiring rapid activation of the detection system, a lower initial calibration accuracy is acceptable, employing a fast digital domain approach. The compensation scheme takes 10 seconds with an accuracy of 10 cm. For medium-priority routine tasks such as marine environmental monitoring, a moderate calibration time can be allocated, using standard digital domain compensation combined with preliminary mechanical adjustment, taking 60 seconds with an accuracy of 5 cm. For low-priority scientific research tasks such as high-precision seabed topography mapping, sufficient calibration time is allowed, using complete mechanical geometric adjustment and multi-round iterative optimization, taking 300 seconds with an accuracy of 2 cm. The cost-benefit assessment processor quantitatively evaluates each candidate scheme. The evaluation dimensions include the expected accuracy improvement (predicted reduction in position error after compensation through simulation), execution cost (including energy consumption calculation of the power consumption of the reference sound source emission and data processing), time cost calculation of the total time from calibration initiation to verification completion, and execution risk assessment. The probability of new disturbances or failures introduced by mechanical actions is estimated. Digital domain compensation schemes correct the time delay and phase parameters in the signal processing of each array element through software, eliminating the need for mechanical actions and thus having the lowest risk and cost, but their compensation capability is limited and only applicable to small distortions within 5cm. Mechanical geometry adjustment schemes physically adjust the array element positions using servo motors or shape memory alloy SMA drivers, achieving large distortion corrections (over 20cm), but with long execution time, high energy consumption, and the risk of mechanical jamming. Redundant array element switching schemes remove an array element from the effective array element set participating in beamforming when a fault or excessive positional error is detected, and activate a backup redundant array element; this is fast but reduces array aperture and gain. The strategy optimization processor is based on... The multi-objective optimization algorithm weighs accuracy, cost, and risk across three dimensions. An improved TOPSIS approximation-ideal-solution ranking method is used to calculate the closeness of each solution to the ideal solution. The ideal solution is defined as having the highest accuracy, lowest cost, and lowest risk, but is usually unattainable. In practical applications, the weight coefficients of the three dimensions are dynamically adjusted according to task constraints. For tactical tasks, time and concealment are prioritized, with a time weight of 0.5, an accuracy weight of 0.3, and a risk weight of 0.2. For scientific research tasks, accuracy is prioritized, with an accuracy weight of 0.6, a time weight of 0.2, and a risk weight of 0.2. The optimal solution is selected through a weighted comprehensive score. Experiments show that intelligent decision-making based on multi-objective optimization can improve overall performance by 15% compared to human experience-based selection under the same constraints.

[0115] The closed-loop execution and effect verification unit achieves precise control and real-time monitoring of the compensation action. The execution instruction is sent to the processor to generate specific control instructions based on the selected compensation strategy. For the digital domain compensation scheme, the time delay compensation value and phase compensation value of each array element are generated and sent to the FPGA digital signal processing board. The FPGA uses Xilinx's Kintex-7 series XC7K325T chip, which internally implements a 32-channel parallel FIR digital filter with 128 taps per channel, achieving nanosecond-level time delay adjustment accuracy and 0.1-degree phase adjustment accuracy. Time delay compensation is achieved by adjusting the FIR filter coefficients, and fractional delay phase compensation is achieved through Hilbert transform or all-pass filter. For the mechanical geometry adjustment scheme, the position instructions of the servo motors are generated and sent to the motion controller. The array uses eight linear servo motors distributed at eight key nodes of the array. The motors are Maxon EC-max 30 brushless DC motors paired with GP 32... The C-planetary gear reducer and MR encoder have a stroke of 50mm, a positioning accuracy of 0.05mm, and a maximum thrust of 20N. By coordinating the extension and retraction of eight motors, the overall array can achieve bending, torsion, and tension compensation. The motion controller uses an EtherCAT real-time bus communication cycle of 1ms to ensure synchronous operation of each motor. For the redundant array element switching scheme, the mask vector of the effective array element is generated and sent to the beamforming processor to update the beamforming weight matrix, reset the weight of the failed array element to zero, and activate the weight of the backup array element. The real-time monitoring processor continuously collects feedback information during execution. For digital domain compensation, the FPGA's working status is monitored to confirm successful parameter loading. For mechanical adjustment, the encoder feedback of the motor is used to judge whether the deviation between the actual displacement and the target displacement has been achieved. The allowable error at the preset position is 0.1mm. Simultaneously, the motor current is monitored to detect overload or jamming. Normal current is below 0.5A; jamming current may exceed 1.5A, triggering protection and stopping. For redundant switching, the array response after switching is monitored to confirm the new array element is working normally. After compensation, the processor immediately initiates the verification process, controlling the reference sound source to re-emit the calibration signal, collecting the response data of the 32 array elements, and re-executing the position estimation algorithm to calculate the compensated array element position error. The error after compensation is compared with the error before compensation to calculate the improvement rate. Simultaneously, beamforming performance indicators, including main lobe pointing error, main lobe width, and side lobe level, are evaluated. The azimuth estimation deviation of the array is measured by transmitting a test signal with a known azimuth. Ideally, the deviation should be less than 0.If the verification results show that the improvement rate reaches the expected target (e.g., the position error decreases from the initial 25cm to below the target 5cm and the beam pointing error is less than 1 degree), the calibration is considered successful, the calibration parameters are recorded, and the system enters normal operating mode. If the improvement rate does not meet the target (e.g., the position error only decreases to 12cm and does not reach the 5cm target), the system is marked as failing due to reasons such as model mismatch or execution error, triggering iterative calibration or strategy switching. The system allows a maximum of 5 iterative calibrations. Experimental data shows that in 90% of cases, the target accuracy can be achieved within 3 iterations. For cases where repeated iterations still fail to meet the target, the system will issue an alarm indicating possible hardware failure or environmental conditions exceeding the system design range, suggesting manual intervention for inspection.

[0116] The online learning and knowledge base update unit enables the system to learn autonomously and continuously optimize. The case learning processor records and analyzes each successful or failed calibration case. The case data structure includes an input feature vector, a selected strategy, execution parameters, final effect, and environmental labels. The input feature vector contains 64-dimensional data such as initial position error, ocean current velocity, ocean current direction, thermocline depth, multipath density, signal-to-noise ratio, GDOP value, and other environmental and system state parameters. The selected strategy field records digital domain compensation, mechanical adjustment, or redundancy switching and their detailed parameters. The final effect field records the compensated position error, beamforming accuracy, and number of iterations. The environmental labels are clustered. Algorithms such as K-means categorize operating conditions into several types, such as deep-sea calm, shallow-sea turbulent, and moderate multipath, to facilitate the retrieval of similar cases. The case library uses a NoSQL database such as MongoDB for storage, supporting efficient vector similarity queries and time-series queries. After six months of offshore deployment, the system has accumulated over 5,000 calibration cases covering different seasons and sea conditions in multiple sea areas, including the South China Sea, East China Sea, and Yellow Sea. The knowledge graph maintenance processor mines the mapping patterns between environmental parameters and optimal policies based on the accumulated case data, using decision tree or random forest algorithms to learn the mapping function from the multidimensional feature space to policy selection. The splitting criteria for the decision tree employ Gini impurity or information gain. The maximum depth is limited to 8 layers to avoid overfitting. The trained decision tree model has strong interpretability and can extract rules, such as prioritizing sparse reconstruction algorithms for signal processing when the ocean current speed is greater than 1.2 knots and the multipath density is greater than 5. Random forest improves generalization ability by integrating 100 decision trees, achieving a policy selection accuracy of 84% on the test set. The knowledge graph stores the structure in the form of a semantic network, including environment nodes, policy nodes, and effect nodes, as well as directed edges between them to represent causal relationships and correlation strength. The graph traversal algorithm can quickly infer the recommended policy sequence in a given environment. The predictive maintenance processor predicts the remaining service life (RUL) of components based on time series analysis. Degradation models were established for key components such as hydrophones, cables, and servo motors. Degradation characteristics included the sensitivity drift of hydrophones, the rate of decrease in insulation resistance of cables, and the positioning repeatability deviation of motors. A Long Short-Term Memory (LSTM) network was used to model historical degradation curves and extrapolate to predict future trends. The LSTM network structure consisted of two hidden layers, each with 128 neurons. The input sequence length was 30 sampling points, and the output was a prediction of the next 10 sampling points. The training data came from land-based accelerated aging tests and actual sea-based usage data. Experimental verification showed that the median prediction error for RUL (Rapid Indicator Usage Limit) prediction of hydrophone sensitivity degradation was 1.2 days in advance and 3 days in advance.The 5-day prediction period is of practical value. When the predicted Recovery Time (RUL) of a component is less than 15 days, the system automatically generates a preventative maintenance recommendation and reports it to the shore-based command center via underwater acoustic or satellite communication links to request logistical support. In the absence of communication, the system displays maintenance prompts on the local interface to guide operators in performing necessary checks and replacements. Compared to traditional periodic maintenance or post-failure maintenance, this predictive maintenance strategy increases system availability from 89% to 96% and reduces maintenance costs by approximately 30%.

[0117] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. An acoustic array adaptive calibration and correction system for underwater moving targets, characterized in that, It includes a multi-source excitation and environmental perception module, an intelligent array perception and diagnosis module, an adaptive position inversion and uncertainty quantification module, and a closed-loop calibration execution and self-learning optimization module; The multi-source excitation and environmental perception module is responsible for providing the reference signals required for calibration and establishing a correlation model between the environment and array distortion. The intelligent array perception and diagnosis module realizes multi-modal data acquisition, array element health status monitoring and array geometry self-sensing. The adaptive position inversion and uncertainty quantization module completes signal processing, position estimation and error quantization propagation. The closed-loop calibration execution and self-learning optimization module executes compensation strategies, verifies calibration effects and continuously optimizes system performance through online learning. The multi-source excitation and environmental perception module includes a distributed reference sound source network unit, a multi-physics environment modeling unit, and an environmental array coupling analysis unit. The distributed reference sound source network unit is responsible for managing multiple mobile or fixed reference sound sources. The multi-physics environment modeling unit measures and models the spatiotemporal variation of sound velocity profiles, ocean current fields, thermocline distribution, and multipath channel characteristics in real time. The environmental array coupling analysis unit is used to establish a causal relationship model between environmental parameters and array changes. The intelligent array sensing and diagnosis module includes a multimodal array signal acquisition unit, an array element state intelligent diagnosis unit, and an array geometry self-sensing unit. The multimodal array signal acquisition unit simultaneously acquires multi-sensor fusion data of acoustic signals, vibration signals, strain data, and temperature distribution. The array element state intelligent diagnosis unit is used to identify gradual and sudden faults of array elements. The array geometry self-sensing unit fuses multi-source information such as inertial measurement unit data, acoustic self-cross-correlation measurement, and reference source reverse positioning to estimate the relative position of array elements and the array shape change trend. The adaptive position inversion and uncertainty quantization module includes a multi-path robust signal processing unit, a hierarchical position estimation unit, and an uncertainty quantization and confidence propagation unit. The multi-path robust signal processing unit performs filtering, matched filtering, and pulse compression processing on the original waveform. The hierarchical position estimation unit is used to provide initial solutions for rapid batch processing and accurately estimate the three-dimensional coordinates of the array elements. The uncertainty quantization and confidence propagation unit quantifies measurement uncertainty, environmental model uncertainty, and algorithm estimation uncertainty. The closed-loop calibration execution and self-learning optimization module includes a multi-mode compensation strategy decision-making unit, a closed-loop execution and effect verification unit, and an online learning and knowledge base update unit. The multi-mode compensation strategy decision-making unit is used to intelligently select compensation schemes and realize task-aware adaptive decision-making. The closed-loop execution and effect verification unit issues specific execution instructions and monitors the execution process in real time. The online learning and knowledge base update unit learns the mapping relationship between environmental patterns, array distortion characteristics and optimal compensation strategies in historical calibration cases, and establishes and continuously updates the knowledge graph. The uncertainty quantification and confidence propagation unit includes an error source decomposition processor, a sensitivity analysis processor, and a confidence synthesis processor. The error source decomposition processor decomposes the total positioning error into independent or related error components such as measurement noise, clock drift, sound velocity profile uncertainty, multipath residuals, array structure deformation, and algorithm approximation errors, and quantifies the contribution ratio of each error source to the final position estimation. The sensitivity analysis processor is used to propagate the input uncertainty to the covariance matrix of the output position and identify the key parameters that have the greatest impact on positioning accuracy. The confidence synthesis processor integrates the statistical test results of the measurement residuals, model matching evaluation, and historical performance statistics to generate a comprehensive confidence score for the position estimation of each array element. The confidence synthesis processor calculates the estimated position of the j-th element after fusion according to the following formula. Covariance Matrix : ； ； ； ； ； ； in, The fusion weight for the k-th measurement method, Let tr() be the conflict confidence factor for the k-th measurement method, and tr() be the trace function. Let be the covariance matrix of the k-th measurement method. Let k be the average degree of conflict between the k-th measurement and other measurements. For conflict sensitivity parameters, I is the covariance inflation factor, and I3 is the identity matrix. The average degree of conflict across all measurements; The confidence synthesis processor calculates the confidence level (Conf) of the estimated position of the j-th array element according to the following formula. j : ； in, This serves as a reference uncertainty threshold.

2. The acoustic array adaptive calibration and correction system for underwater moving targets as described in claim 1, characterized in that, The environmental array coupling analysis unit includes a fluid-structure interaction (FSI) computation processor, a structural response modeling processor, and a distortion source identification processor. The FSI computation processor combines array material properties and geometric parameters to predict the bending, torsional, and tensile deformations of the array. The structural response modeling processor integrates measurement data from strain sensors, fiber Bragg grating sensors, and accelerometers to establish a dynamic response model of the array support structure under environmental loads, estimating the offset and vibration characteristics of each array element relative to the design reference. The distortion source identification processor identifies the main environmental factors causing array distortion by comparing the correlation between changes in environmental parameters and array deviations. The fluid-structure interaction computation processor calculates the deviation vector of all array element positions according to the following formula. : ； ； Where M represents the number of dominant modes. For the m-th order distortion mode, Let be the excitation coefficient for the m-th mode. The residual term introduced for structural damping and nonlinear effects. This is the environment weight vector corresponding to the m-th mode. The magnitude of the ocean current velocity. The direction angle of the ocean current. For pressure gradient, This represents the amplitude of the temperature gradient. Let t be the Reynolds number of the underwater vehicle, and t be the current time.

3. The acoustic array adaptive calibration and correction system for underwater moving targets as described in claim 2, characterized in that, The array geometry self-sensing unit includes an inertial navigation fusion processor, an acoustic self-localization processor, and a geometric constraint optimization processor. The inertial navigation fusion processor is used to map the designed position of each array element in the body coordinate system to the global coordinate system to obtain a priori position estimates. The acoustic self-localization processor is used to calculate the relative geometric relationships between array elements. The geometric constraint optimization processor combines the physical constraints of the array, fuses inertial data and acoustic measurement data, and outputs the optimal estimate of the position of each array element and its covariance matrix. For the position p of the j-th array element j The geometric constraint optimization processor constructs the optimization objective according to the following formula: ； ； ； ； ； in, For the optimized estimated position of the j-th array element, p j The actual position of the j-th element. For the global estimation of the bias vector, L TOA For the arrival time measurement fitting term, The measured arrival time from reference sound source i to array element j is s i Let i be the position of the reference sound source, and c be the average speed of sound. For adaptive weights of arrival time measurements, L IMU For inertial navigation prior fitting terms, The estimated element positions for the inertial measurement unit. L is the weight of inertial navigation measurements. geom For rigid / semi-rigid connection constraint terms, For a set of rigidly connected array element pairs, To design the reference spacing vector, L is the regularization parameter for geometric constraints. smooth For the smoothing constraint term of the flexible segment, The regularization parameter for smoothing constraints.

4. The acoustic array adaptive calibration and correction system for underwater moving targets as described in claim 3, characterized in that, The multi-mode compensation strategy decision unit includes a task awareness processor, a cost-benefit evaluation processor, and a strategy optimization processor. The task awareness processor receives the task type and task priority parameters currently being performed by the underwater vehicle. The cost-benefit evaluation processor quantitatively evaluates the expected calibration effect, execution cost, and failure risk of each candidate scheme for digital domain compensation, mechanical geometry adjustment, and redundant array element switching. The strategy optimization processor selects the optimal compensation scheme or a multi-scheme hybrid strategy based on task constraints, confidence thresholds, and cost-benefit evaluation results. The preferred processor selects a compensation strategy according to the following formula: ； ； in, For the selected optimal compensation strategy, S1 is digital domain compensation, S2 is mechanical geometric adjustment, and S3 is redundant array element switching. Let S be the expected improvement in positional accuracy of strategy S, R(S) be the inherent risk coefficient of strategy S, C(S) be the execution cost of strategy S, and x be the value of x. t This is the current environment state vector. , and Weighting coefficients are assigned to the three tasks.

Images