Natural disaster rescue-oriented satellite communication terminal rapid deployment and directional locking method

Abstract

The invention discloses a natural disaster rescue-oriented satellite communication terminal rapid deployment and directional locking method, which comprises the steps of acquiring and preprocessing sensor data, performing magnetic field interference mode classification, attitude estimation and filtering-clustering double-loop feedback control, generating a compensation parameter set and feeding the compensation parameter set back to the magnetic field interference mode classification and attitude estimation. Circularly executing until a precision threshold value is met; and executing satellite direction positioning and locking based on the terminal attitude estimation value meeting the precision threshold and the satellite position information, and outputting a satellite pointing angle and a pointing control instruction. According to the method, the technical problems of large attitude estimation error, discontinuous parameter adjustment, lack of differential compensation and the like of a traditional method in a magnetic field interference environment are solved, and the deployment efficiency and the communication reliability of the satellite communication terminal in a disaster area environment are improved.

Description

Technical Field

[0001] The present invention belongs to the field of satellite communication technology and automatic control, and in particular to a method for rapid deployment and directional locking of satellite communication terminals for natural disaster rescue. Background Art

[0002] During natural disaster relief operations, communications infrastructure often suffers severe damage, making satellite communications a crucial means of connecting disaster-stricken areas with the outside world. The rapid deployment and accurate directional lock capabilities of satellite communication terminals are directly related to rescue efficiency and communication quality. Rescuers often face the challenge of establishing stable communication links in a short period of time in extreme environments. Therefore, developing lightweight satellite communication terminal attitude estimation and directional lock technologies that can adapt to complex magnetic fields is crucial for improving disaster response capabilities and rescue efficiency.

[0003] Currently, attitude estimation in satellite communication terminals primarily relies on methods such as complementary filtering or extended Kalman filtering to fuse multi-sensor data. Mainstream technologies typically use accelerometers to provide a gravity reference, gyroscopes to provide angular velocity, and an electronic compass to provide a magnetic north reference, achieving highly accurate attitude estimation in static environments. Many systems employ low-pass filters to process accelerometer data and high-pass filters to process gyroscope data to reduce the impact of noise in each. In commercial systems, the common AHRS (Attitude and Heading Reference System) fuses sensor data using a nonlinear extended Kalman filter to construct attitude state equations and observation equations, enabling highly accurate attitude estimation in static environments.

[0004] However, existing methods face multiple technical challenges in the complex magnetic field environments of disaster zones. First, traditional attitude estimation algorithms simply treat magnetic field interference as outliers and lack the ability to distinguish between different types of magnetic interference. This leads to significantly increased azimuth estimation errors in areas with wrecked reinforced concrete buildings or damaged power facilities. Second, existing filtering and clustering methods typically employ one-way dependencies, making it difficult to use filtering results as feedback to adjust the clustering process, potentially leading to inconsistent judgments in complex environments. Third, traditional parameter adjustments often employ step or linear changes, which can easily cause attitude estimation to jump when switching between magnetic field interference modes, affecting the terminal's stable pointing. Furthermore, fixed-weight sensor fusion methods are unable to dynamically adjust strategies based on different magnetic field interference modes, resulting in poor system performance under various conditions, such as persistent environmental drift, sudden interference, and periodic interference. These issues severely restrict the deployment efficiency and communication reliability of satellite communication terminals in disaster zones with volatile magnetic fields. Summary of the Invention

[0005] The purpose of the invention is to provide a method for rapid deployment and directional locking of satellite communication terminals for natural disaster relief, in order to solve at least one technical problem existing in the prior art.

[0006] The technical solution is a method for rapid deployment and directional locking of satellite communication terminals for natural disaster relief, including:

[0007] Acquire data from various sensors, pre-process them, and form data packets;

[0008] Receive data packets, classify magnetic field interference patterns, and obtain magnetic field interference pattern identifiers; perform attitude estimation based on the identifiers, and generate terminal attitude estimation values ​​and uncertainties;

[0009] Based on the magnetic field interference pattern identification, terminal attitude estimation value and uncertainty, filtering-clustering dual-loop feedback control is performed to generate a compensation parameter set including a filtering parameter adjustment set and a clustering parameter adjustment set; the clustering parameter adjustment set is fed back to the magnetic field interference pattern classification, and the filtering parameter adjustment set is fed back to the attitude estimation, and the cycle is executed until the accuracy threshold is met;

[0010] Based on the terminal attitude estimation value and satellite position information that meet the accuracy threshold, satellite direction positioning and locking are performed, and the satellite pointing angle and pointing control instructions are output.

[0011] Beneficial effects: The present invention solves the problems of large attitude estimation error, discontinuous parameter adjustment and lack of differentiated compensation in traditional methods under magnetic field interference environment, breaks the one-way dependency relationship between traditional filtering and clustering algorithms, establishes a two-way feedback mechanism, improves the deployment efficiency and communication reliability of satellite communication terminals in harsh environments, and provides reliable communication guarantee for disaster relief. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of the steps of a method for rapid deployment and directional locking of satellite communication terminals for natural disaster relief provided by an embodiment of the present invention.

[0013] Figure 2 A flowchart of the steps for classifying magnetic field interference patterns provided in an embodiment of the present invention.

[0014] Figure 3 A flowchart of the steps for processing enhanced magnetic field eigenvectors using a progressive incremental clustering algorithm with a core-satellite structure provided by an embodiment of the present invention.

[0015] Figure 4 A flowchart of the steps for determining the minimum distance and corresponding clustering provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0017] It should be noted that to clearly illustrate the steps of this application, serial numbers are assigned to each step in the specification. These serial numbers are for illustrative purposes only and do not limit the order in which the steps must be executed. In actual operation, depending on the technical requirements of the specific implementation scenario, the steps may be executed in a different order than shown in the specification, and in some cases, parallel processing between steps may be implemented.

[0018] like Figure 1 As shown, a method for rapid deployment and directional locking of satellite communication terminals for natural disaster relief includes the following steps:

[0019] S1, obtain the data of each sensor, pre-process it, and form a data packet;

[0020] Specifically, sensor data includes gyroscope raw data, accelerometer raw data, and electronic compass raw data. Gyroscope raw data can measure an object's angular velocity (i.e., rotational speed); accelerometer raw data can measure an object's acceleration (e.g., changes in direction and force); and electronic compass raw data can measure direction and orientation.

[0021] S2, receiving a data packet, classifying the magnetic field interference pattern, and obtaining a magnetic field interference pattern identifier;

[0022] Specifically, magnetic fields can be disturbed by the surrounding environment (such as electronic devices and magnetic materials), which can lead to inaccurate direction measurements. The system analyzes data packets to identify characteristic patterns of these interferences (for example, identifying the source or type of interference). Once the interference pattern is identified, it is assigned a "label" or "identifier" to indicate the specific type of interference (for example, whether it comes from electronic devices or is due to an abnormal environmental magnetic field).

[0023] S3. Perform attitude estimation based on the magnetic field interference pattern identification to generate the terminal attitude estimation value and uncertainty;

[0024] Specifically, based on information about magnetic field interference, the system begins calculating the terminal's current posture, such as its orientation (facing direction) and angle (tilt). The core of posture estimation is analyzing sensor data and compensating for the effects of magnetic field interference to produce the most accurate results possible. As the posture estimate is calculated, the system evaluates its accuracy. For example, it generates an "uncertainty" value to indicate the potential error in this estimate.

[0025] S4. Perform filtering-clustering dual-loop feedback control based on the magnetic field interference pattern identifier, the terminal attitude estimation value, and the uncertainty to generate a compensation parameter set including a filtering parameter adjustment set and a clustering parameter adjustment set; feed the clustering parameter adjustment set back to the magnetic field interference pattern classification, and feed the filtering parameter adjustment set back to the attitude estimation, and execute the loop until the accuracy threshold is met;

[0026] Specifically, a dual-loop filtering and clustering feedback control system is implemented. The filtering control focuses on removing interference or noise from the data, making the attitude estimation more accurate; the clustering control focuses on classifying and identifying patterns of magnetic field interference, making the classification more reliable. Based on this analysis, a set of adjustment parameters is generated: the filtering parameter adjustment set is used to optimize the attitude estimation process; the clustering parameter adjustment set is used to improve the accuracy of magnetic field interference classification. The filtering adjustment parameters are fed back to the attitude estimation module to continuously improve the accuracy of the direction estimation; the clustering adjustment parameters are fed back to the magnetic field interference classification module to further improve the accuracy of the classification. This process continues until the attitude estimation and interference classification results meet the preset accuracy threshold.

[0027] S5. Based on the terminal attitude estimation value that meets the accuracy threshold and the satellite position information, perform satellite direction positioning and locking, and output the satellite pointing angle and pointing control instructions.

[0028] Specifically, the system combines the device's attitude information and the satellite's position to calculate the direction the device needs to be adjusted to accurately align with the satellite, and obtains the required angle adjustment value and pointing control instructions to drive the device to complete the physical direction adjustment and lock onto the satellite.

[0029] This embodiment implements real-time data collection and processing, enabling the terminal to quickly adapt to complex magnetic field environments and improving system response speed. It also continuously improves the accuracy and reliability of terminal attitude estimation, ensuring the high precision of the final output satellite pointing angle, thereby enhancing communication quality. It also dynamically adjusts to different magnetic field interference patterns and attitude estimation errors, enhancing the system's adaptability and ensuring its flexibility in varying environmental conditions. This embodiment, with its real-time performance, adaptability, and high precision, is particularly suitable for satellite communication needs in complex environments such as natural disaster relief scenarios.

[0030] According to one aspect of the present application, the pre-processing step comprises:

[0031] S11. Collect multi-sensor data in parallel. Acquire gyroscope raw data at a 500Hz sampling rate and record triaxial angular velocity measurements; acquire accelerometer raw data at a 100Hz sampling rate and record triaxial acceleration measurements; acquire electronic compass raw data at a 20Hz sampling rate and record triaxial magnetic field strength measurements. Record the acquisition timestamp of each data set to ensure time correlation in subsequent processing.

[0032] S12. Perform sensor data calibration and noise removal. Read pre-calibrated sensor calibration parameters, including the bias vector and scale factor matrix; apply the temperature compensation model to the gyroscope raw data to generate temperature-compensated angular velocity data; apply the calibration parameters to the accelerometer raw data to generate calibrated acceleration data; apply hard-iron and soft-iron calibration to the electronic compass raw data to generate calibrated magnetic field data; apply a bandpass filter to all calibrated data to generate denoised sensor data.

[0033] S13. Perform multi-rate data synchronization and alignment. Design a time window alignment algorithm to establish a unified time base with a base sampling rate of 100 Hz. Downsample the temperature-compensated angular velocity data to generate 100 Hz angular velocity data. Interpolate the calibrated magnetic field data to generate 100 Hz magnetic field data. Integrate the sensor data streams to form a time-aligned multi-sensor data packet containing angular velocity, acceleration, and magnetic field data. Calculate quality assessment metrics for each time-aligned multi-sensor data packet and flag outliers.

[0034] like Figure 2 As shown, according to one aspect of the present application, the steps of classifying magnetic field interference patterns and obtaining magnetic field interference pattern identifiers include:

[0035] S21, receiving a data packet, an initial or feedback clustering parameter adjustment set, extracting time domain features of the magnetic field data, and constructing an enhanced magnetic field feature vector;

[0036] S22. Use the core-satellite structured progressive incremental clustering algorithm to process the enhanced magnetic field feature vector, and dynamically adjust the clustering threshold based on the clustering parameter adjustment set to obtain the magnetic field interference pattern identification and clustering credibility score.

[0037] S23. Analyze the magnetic field interference pattern identification and cluster credibility score, evaluate the current magnetic field environment state, and generate a magnetic field reliability index and interference compensation suggestion; wherein, the magnetic field interference pattern identification, magnetic field reliability index, and interference compensation suggestion are used for subsequent adaptive parameter adjustment posture estimation.

[0038] Specifically, 100Hz magnetic field data from time-aligned multi-sensor data packets are received, a 500ms sliding window is constructed, and time domain features, including mean, variance, maximum volatility, and frequency characteristics, are extracted from the magnetic field data within the window. Compared with the historical reference model, the magnetic field anomaly index is calculated to quantify the degree to which the current magnetic field deviates from the normal value. The magnetic field measurement values ​​and the extracted features are combined to construct an enhanced magnetic field feature vector.

[0039] The enhanced magnetic field eigenvector and the clustering state information from the previous cycle are read, and an adaptive distance metric is calculated between the current eigenvector and the known clustering pattern to identify the most matching clustering pattern. Depending on the degree of match, one of the following operations is performed: If the match is high, the core parameters of the corresponding cluster are updated, and the pattern matching identifier is recorded. If the match is moderate, the eigenvector is used as a satellite point for the cluster and the cluster boundaries are updated. If the match is low, the pattern is temporarily saved as a potential new pattern, and a new cluster may be created after sufficient samples are accumulated. The currently identified magnetic field interference pattern identifier and cluster confidence score are output.

[0040] Analyze the characteristics of identified magnetic field interference patterns, including duration, intensity trends, and directionality. Match the current pattern characteristics with predefined interference type templates to determine the type of interference (e.g., persistent environmental drift, sudden interference, etc.). Calculate magnetic field reliability indicators based on cluster credibility scores and pattern stability. Generate interference compensation recommendations, including adjustments to filter weights.

[0041] This embodiment constructs a core-satellite structured clustering model, which solves the problem that traditional clustering methods cannot adapt to the characteristics of time series data. It enables the clustering model to continuously adapt to the changing environment without the need to store and process all historical data, while also improving clustering accuracy.

[0042] like Figure 3 As shown, according to one aspect of the present application, the steps of using a core-satellite structured progressive incremental clustering algorithm to process enhanced magnetic field feature vectors to obtain magnetic field interference pattern identification and clustering credibility score include:

[0043] For the enhanced magnetic field eigenvector, the adaptive distance between the current eigenvector and the existing cluster is calculated, and the minimum distance and the corresponding cluster are determined, which is recorded as the best matching cluster;

[0044] According to the minimum distance and clustering parameters, the matching threshold in the set is adjusted, the core-satellite structure update is performed, the current sample is assigned as a core update, a satellite point or a potential new pattern, and the cluster core parameters, cluster boundary information or potential new pattern buffer are updated accordingly;

[0045] When the preset conditions are met, cluster maintenance and optimization are performed, the cluster set is updated according to the cluster activity and similarity, an updated cluster set is generated, and the magnetic field interference pattern identifier of the current most matching cluster is determined;

[0046] A cluster confidence score is calculated based on the status of the updated cluster set.

[0047] Specifically, the calculation process of the adaptive distance in progressive incremental clustering is: adaptive distance D(ct, ci) = ωd·Dd(ct, ci)+ωm·Dm(ct, ci) + ωt·Dt(ci); where Dd(ct, ci) = 1-cos(θ) is the direction-sensitive distance; θ is the angle between the direction of the current eigenvector ct and the direction of the cluster core ci; Dm(ct, ci) = |log(Mt / Mi)| is the intensity-sensitive distance; Mt is the magnetic field intensity of the current eigenvector; Mi is the magnetic field intensity of the cluster core; Dt(ci) = 1-exp(-λt·ΔT) is the time decay weight coefficient; ΔT is the difference between the current time and the last update time of cluster ci; λt is the time decay coefficient; ωd, ωm, ωt are the weights of the three components of direction, intensity and time, respectively, and ωd+ωm+ωt=1.

[0048] like Figure 4 As shown, according to one aspect of the present application, the steps of calculating the adaptive distance between the current feature vector and the existing clusters and determining the minimum distance and the corresponding clusters include:

[0049] performing normalization processing on the enhanced magnetic field eigenvector to generate a standardized eigenvector including a direction component and an intensity component;

[0050] The difference between the current time and the last update time of each cluster is calculated based on the standardized feature vector, and the exponential decay function is applied to generate the time decay weight coefficient;

[0051] Calculate the direction-sensitive distance based on the cosine value of the angle between the direction component and the core direction of each cluster;

[0052] The intensity-sensitive distance is calculated based on the logarithmic ratio of the intensity component to the core intensity of each cluster;

[0053] Use preset weight coefficients to combine direction-sensitive distance, intensity-sensitive distance and time-decay weight coefficients to generate an adaptive distance metric value;

[0054] Compare the adaptive distance metric values ​​of all clusters and determine the minimum distance and the corresponding cluster.

[0055] According to one aspect of the present application, the steps of performing a core-satellite structure update and correspondingly updating cluster core parameters, cluster boundary information, or a potential new pattern buffer zone include:

[0056] Compare the minimum distance with the matching threshold to determine whether to perform core update operation, satellite point update operation or potential new mode processing;

[0057] When performing a core update operation, the adaptive learning rate is calculated and the cluster core parameters, statistics, and active timestamps are updated;

[0058] When performing the satellite point update operation, the current sample is recorded as a satellite point, and the cluster boundary information and satellite point set are updated;

[0059] When executing potential new pattern processing, the current sample is added to the temporary buffer. When the sample in the temporary buffer meets the preset conditions for creating a new cluster, a new cluster core is generated and the potential new pattern buffer is updated.

[0060] Specifically, the core-satellite structure update mechanism is as follows: core update learning rate α(ci) = β·exp(-γ·N(ci))·(1-Sc(ci)); where α(ci) is the adaptive learning rate of cluster core ci; β is the basic learning rate constant; γ is the decay factor; N(ci) is the cumulative number of samples of cluster ci; Sc(ci) is the stability index of cluster ci, with a value range of [0, 1]; the core parameter update formula is: Pnew(ci) = (1-α(ci))·Pold(ci) + α(ci)·Pcur; Pnew(ci) is the updated cluster core parameter; Pold(ci) is the cluster core parameter before update; Pcur is the current sample feature parameter.

[0061] According to one aspect of the present application, the steps of performing cluster maintenance and optimization and determining the magnetic field interference pattern identifier that best matches the current cluster include:

[0062] Perform cluster aging detection, calculate the inactivity time of each cluster, and mark clusters whose inactivity time exceeds a predetermined threshold as dormant; perform cluster merging detection, calculate the cluster core distance based on the updated cluster core parameters, and when the distance between two cluster cores is less than the preset merging threshold, calculate the weighted average to generate a new core and merge the statistics and satellite point sets; perform cluster structure rebalancing, regularly optimize cluster boundaries and core distribution to reduce overlap and redundancy; and output the updated cluster set;

[0063] Based on the updated cluster set, the overall stability index of the current cluster model is calculated, and the magnetic field interference pattern identifier of the current most matching cluster is determined.

[0064] According to one aspect of the present application, the step of calculating the cluster credibility score based on the updated state of the cluster set includes:

[0065] Based on the updated cluster set, the number of samples, internal variance, and inactivity time are extracted, and the stability index of the matching cluster is calculated;

[0066] Based on the stability index and the minimum distance, a cluster credibility score is generated to quantify the reliability of the current cluster identification result, and the interference pattern characteristics (key feature parameter set) of the matching cluster are extracted, including duration, intensity change rate and directional stability, to generate a cluster feature parameter set.

[0067] Specifically, the process of calculating the cluster credibility score is: cluster credibility score Cfid(ci) = ws·Ss(ci)+ wv·(1-Sv(ci)) + wa·Sa(ci); where Cfid(ci) is the credibility score of cluster ci, with a value range of [0, 1]; Ss(ci) is the sample size standardized score of cluster ci; Sv(ci) is the internal variance standardized score of cluster ci; Sa(ci) is the activity score of cluster ci; ws, wv, and wa are the weights of the three components of sample size, internal variance, and activity, respectively, and ws+wv+wa=1.

[0068] In one embodiment of the present application, an adaptive distance metric is calculated. Read the current enhanced magnetic field feature vector F t and the existing cluster set C={C1, C2, ..., C k}; Apply normalization to each feature dimension to generate a standardized feature vector F t norm ; Calculate the time attenuation weight coefficient w t = e -λ(tcurrent-tlast) , giving higher weight to recent samples. Define direction-sensitive distance function: d dir (v_1, v_2) = 1 - |cos(v_1, v_2)|, captures the change in magnetic field direction. Define the intensity-sensitive distance function: d mag (m_1, m_2) = |log(||m_1|| / ||m_2||)|, capturing the change in magnetic field strength. Combining to construct an adaptive distance metric: d(F t , C i ) =α⋅d dir (F t dir , C i dir ) +β⋅d mag (F t mag , C i mag ) +γ⋅d temporal (F t temp , C i temp ) . Calculate the distance d(F) between the current sample and each existing cluster t , C i), find the minimum distance d min and the corresponding cluster C min Where e is the base of the natural logarithm, λ is the time decay parameter, and t current is the current time; t last is the previous time point; v_1 and v_2 are the magnetic field direction feature vectors; m_1 and m_2 are the magnetic field intensity feature vectors; Ci is the i-th cluster in the existing cluster set; α is the weight coefficient of the direction-sensitive distance function; F t dir is the directional characteristic part of the current enhanced magnetic field characteristic vector; C i dir is the directional feature part of the i-th cluster in the cluster set; β is the weight coefficient of the intensity sensitive distance function; F t mag is the intensity characteristic part of the current enhanced magnetic field characteristic vector; C i mag is the intensity feature of the i-th cluster in the cluster set; γ is the weight coefficient of the time-sensitive distance function; d temporal is the time-sensitive distance function; F t temp is the time characteristic part of the current enhanced magnetic field characteristic vector; C i temp is the time feature part of the i-th cluster in the cluster set.

[0069] Update the core-satellite structure. Determine the minimum distance d min and matching threshold θ match Relationship: If d min <θ match :Execute core update operation; if θ match ≤d min <θ satellite : Execute satellite point update operation; if d min ≥θ satellite : Perform potential new pattern processing, where θ satellite is the satellite point threshold. When performing the core update operation, the learning rate η=η0⋅(1−e −Ni / τ ), where η0 is the initial learning rate, τ is the time constant, and Ni is the cluster C min The number of samples; Update the clustering core parameter: C min core =(1−η)⋅C min core +η⋅F t norm ; Update cluster statistics: increase sample count Ni, update variance σ i 2 ; Update the last active timestamp tlast When performing satellite point update operation, record the current enhanced magnetic field characteristic vector F t As cluster C min Satellite points, update cluster boundary information: B min =max(B min , d min ); Update satellite point set: S min =S min ∪{F t}; Only the most recent K satellite points are retained to prevent the memory from growing indefinitely. When executing the potential new mode processing, the current enhanced magnetic field feature vector F t Add to temporary buffer T buffer , if the temporary buffer T buffer The average mutual distance of the midpoints is less than the new pattern threshold θ new And the number exceeds the minimum number threshold n min : Create a new cluster C new , the initial core is the temporary buffer T buffer The average of the midpoints; set the initial sample count, variance and timestamp; add the new cluster to the cluster set: C = C∪{C new}; Clear the temporary buffer T buffer .

[0070] Perform cluster maintenance and optimization. Perform cluster aging detection: calculate the inactive time t of each cluster inactive =t current −t last If t inactive >T max : Mark the cluster as dormant, not participating in regular matching but retained in the library, where T max Is the maximum threshold of the inactive time of the cluster. Perform cluster merging detection: If the distance between the two cluster cores is less than the merging threshold θ merge : Calculate the weighted average to get the new core; merge the statistics and satellite point sets; delete the original two clusters and add the merged cluster. Calculate the overall stability index S of the current clustering model model ; Output the magnetic field interference pattern identifier M of the current most matching cluster i .

[0071] Evaluate the incremental clustering results. Calculate the stability index of the matching cluster: Stab i =f(Ni,σ i 2 , t inactive ); Based on the stability index and matching distance, calculate the clustering credibility score: Conf i =g(Stab i , d min); analyze cluster features and extract key characteristic parameters of the interference pattern, such as duration, intensity change rate, and directional stability. This generates a cluster feature parameter set for subsequent interference type identification. The magnetic field interference pattern identification and cluster credibility score are passed to subsequent processing modules. Here, f is a function that matches the stability index of the cluster, and g is a function that matches the cluster credibility score.

[0072] In another embodiment of the present application, the core algorithm of incremental clustering takes a new sample x_t and an existing set of clusters C = {C_1, C_2, ..., C_k} as input, performs feature extraction, calculates the time window feature F_t, and constructs an enhanced data vector D_t = [x_t, F_t, ΔF_t], where ΔF_t is the change in the feature within the current time window. For each cluster C_i, an adaptive distance d(D_t, C_i) is calculated, and the minimum distance d_min and the corresponding cluster C_min are found. If the minimum distance d_min < the matching threshold θ_match: Assign the augmented data vector D_t to cluster C_min and incrementally update the core and statistics of cluster C_min. Otherwise, if the minimum distance d_min < the satellite threshold θ_satellite: Use the augmented data vector D_t as a satellite of cluster C_min and update the boundary information of cluster C_min. Otherwise, if sufficient resources are available and D_t is stable, create a new cluster C_new with the augmented data vector D_t as the core. The updated cluster set C' = C∪{C_new}. Otherwise, temporarily store the augmented data vector D_t as an unclassified sample. Perform cluster maintenance: Merge similar clusters; Remove aging clusters; Periodically rebalance the cluster structure; Output the updated cluster set C' and sample classification.

[0073] Perform cluster-compensation strategy mapping: For each identified cluster C_i, extract feature vectors (duration, intensity, directionality, rate of change, etc.); calculate similarity with predefined patterns; select the best-matching pattern type M_j; and apply the corresponding compensation strategy S_j. If the pattern type M_j is "persistent ambient magnetic field offset," recalibrate the magnetic field reference model and adjust the azimuth estimation bias. If the pattern type M_j is "sudden transient interference," temporarily increase the magnetometer noise covariance and increase the gyroscope integration weight. If the pattern type M_j is "periodic oscillation interference," apply a narrowband notch filter to suppress the corresponding frequency components in the frequency domain. If the pattern type M_j is "random noise enhancement," increase the smoothing filter strength and reduce the overall magnetometer weight.

[0074] According to one aspect of the present application, the steps of performing attitude estimation and generating a terminal attitude estimate and uncertainty include:

[0075] S31. Multi-sensor data fusion preprocessing. Receive denoised sensor data, magnetic field interference pattern identifiers, and magnetic field reliability indicators. Dynamically calculate sensor fusion weights based on the magnetic field reliability indicators, reducing magnetometer weights when the magnetic field is unreliable. Apply interference compensation recommendations to adjust filter parameters and generate adjusted filter configurations.

[0076] S32. Implement the adaptive extended Kalman filter. Perform a state prediction step based on the current attitude state estimate (quaternion representation) and 100 Hz angular velocity data. Calculate the predicted state uncertainty and update the system noise covariance matrix. Adjust the magnetometer-related components of the measurement noise covariance matrix based on the adjusted filter configuration. Perform a measurement update step using the calibrated acceleration data and 100 Hz magnetic field data. Output the updated attitude state estimate and estimated uncertainty.

[0077] S33. Perform attitude stability enhancement processing. Detect sudden changes in the attitude state estimate and trigger smooth transition processing if a sudden change is found. When magnetic field reliability is low, increase the gyroscope integration weight to reduce reliance on magnetic field measurements. Apply an exponentially weighted moving average to smooth short-term fluctuations while retaining rapid response capabilities. Convert the quaternion representation to Euler angles to generate terminal attitude angles (pitch, roll, and azimuth). Combined with the estimated uncertainty, generate an attitude quality score to provide a reliability indicator.

[0078] According to one aspect of the present application, the steps of performing filtering-clustering dual-loop feedback control and generating a compensation parameter set include:

[0079] S41. Calculate an anomaly detection statistic based on the filter residual, compare it with the clustering credibility score, and generate a consistency score that represents the degree of coordination between the filtering and clustering results;

[0080] S42: Determine the current state based on the consistency score. When the consistency score is lower than a preset threshold, trigger a depth adjustment and generate adjustment parameters, including a clustering threshold adjustment parameter and a filter weight adjustment parameter.

[0081] S43 . Based on the magnetic field interference pattern identifier and the adjustment parameters, a corresponding compensation strategy is selected, and the selected compensation strategy is parameterized to generate a compensation parameter set, including a filtering parameter adjustment set and a clustering parameter adjustment set.

[0082] According to one aspect of the present application, the steps of selecting a corresponding compensation strategy and generating a compensation parameter set include:

[0083] receiving a magnetic field interference pattern identifier, a clustering feature parameter set, and an adjustment parameter;

[0084] A feature matching algorithm is used to match the current interference pattern with a predefined interference type library, and a mapping relationship matrix including the correspondence between interference patterns and compensation strategies is established. The optimal compensation strategy is selected and a compensation parameter set is generated by combining the interference pattern characteristics and the preset strategy effect scores.

[0085] Specifically, the attitude state estimate, estimation uncertainty and magnetic field interference pattern identification are read, and the anomaly detection statistic based on the filter residual is calculated to evaluate the consistency between the measurement and prediction; the anomaly detection statistic is compared with the clustering credibility score to evaluate the consistency of the filtering and clustering results; and a consistency score is generated to quantify the degree of coordination between the filtering and clustering results.

[0086] Based on the consistency score, the system determines whether adjustments are needed. If consistency is high, the current parameter configuration is maintained; if consistency is moderate, minor parameter adjustments are made; if consistency is low, a deeper adjustment is triggered. When adjustments are necessary, clustering threshold adjustment parameters for clustering and filter weight adjustment parameters for filtering are generated. A smooth transition function is designed to ensure gradual parameter adjustments and avoid introducing new instabilities.

[0087] The system receives the magnetic field interference pattern identification and clustering feature parameters as input, uses the feature matching algorithm to map the current clustering pattern to the predefined interference type library, constructs a mapping relationship matrix, and establishes the correspondence between the interference pattern and the optimal compensation strategy; based on historical compensation effect data, it dynamically updates the strategy effect score; according to the interference pattern characteristics and strategy effect score, it selects the optimal compensation strategy and generates a detailed compensation strategy parameter set, including filter parameter adjustment, measurement weight correction and model update instructions.

[0088] Based on the magnetic field interference pattern identification, a corresponding compensation strategy is selected: for persistent environmental offsets, magnetic field reference model recalibration is applied; for sudden interference, a short-term magnetic field elimination strategy is applied; for periodic interference, frequency-domain selective filtering is applied. The selected compensation strategy is parameterized into a compensation parameter set, which is distributed to the attitude estimation module and clustering module to form a closed-loop control.

[0089] The clustering analysis process is decomposed into interruptible microtask units. The execution order and priority of microtasks are dynamically adjusted based on the posture quality score and system load. Time slices are allocated between the filtering main loop and the clustering auxiliary loop to ensure real-time performance. The execution time of each module is monitored and recorded to optimize the task scheduling strategy.

[0090] The consistency score is calculated as follows: consistency score Con = (1-|Anorm-Cfid|)·exp(-λc·σ 2est); Con is the consistency score of the filtering and clustering results, with a value range of [0, 1]; Anorm is the normalized anomaly detection statistic; Cfid is the credibility score of the current matching cluster; σ 2 est is the uncertainty of attitude estimation; λc is the adjustment coefficient, which controls the influence of uncertainty on consistency.

[0091] The compensation parameter set generation mechanism is as follows: filter parameter adjustment ΔRm = Rm0·(1-Con)·f(Mpat); where ΔRm is the magnetometer measurement noise covariance adjustment; Rm0 is the reference noise covariance matrix; Con is the consistency score; f(Mpat) is the adjustment function based on the magnetic field interference pattern Mpat; clustering parameter adjustment ΔTc = Tc0·(1-Con)·g(Anorm); where ΔTc is the cluster matching threshold adjustment; Tc0 is the reference matching threshold; and g(Anorm) is the adjustment function based on the anomaly detection statistic Anorm.

[0092] This embodiment breaks the one-way dependency between traditional filtering and clustering algorithms and establishes a bidirectional feedback mechanism. Through the consistency evaluation mechanism, mutual verification and compensation of filtering results and clustering results are achieved. The smooth transition function design ensures the continuity of attitude estimation when the interference mode is switched, avoiding the attitude jump in traditional methods.

[0093] According to one aspect of the present application, the steps of calculating anomaly detection statistics based on filter residuals and generating a consistency score include:

[0094] Calculate the prediction residual based on the terminal posture estimate and historical state estimate;

[0095] The Mahalanobis distance formula is used to calculate the standardized statistics of the prediction residuals and generate anomaly detection statistics;

[0096] The anomaly detection statistics and clustering credibility scores are weighted and fused to evaluate the consistency of filtering and clustering results; a consistency score is generated to quantify the degree of consistency.

[0097] According to one aspect of the present application, determining the current state according to the consistency score includes the following steps:

[0098] Compare the consistency score with the preset threshold to determine the current level of adjustment that needs to be performed;

[0099] Generate clustering threshold adjustment parameters for the clustering algorithm according to the adjustment level, including adjusting the matching threshold and satellite point threshold;

[0100] A filter weight adjustment parameter for the filtering algorithm is generated according to the adjustment level, including adjusting the magnetometer-related component in the measurement noise covariance matrix.

[0101] According to one aspect of the present application, the steps of performing satellite direction positioning and locking and outputting satellite pointing angles and pointing control instructions include:

[0102] S51. Calculate the satellite pointing angle. Receive the terminal attitude angle and satellite position information (azimuth and elevation), calculate the theoretical target pointing angle based on the current geographic location and satellite position; calculate the uncertainty of the pointing angle based on the attitude quality score; optimize the pointing strategy and generate antenna pointing control parameters.

[0103] S52: Perform directional mechanism control and feedback. Convert antenna pointing control parameters into servo control signals; monitor servo position feedback and calculate actual pointing error; implement closed-loop position control and fine-tune antenna pointing; and when the pointing is stable, lock the mechanical structure to reduce energy consumption.

[0104] S53: Perform signal quality monitoring and optimization. Monitor satellite signal strength and signal-to-noise ratio, calculate communication quality indicators; perform fine-tuning scans to find the optimal signal point and record the optimal pointing offset; feed the optimal pointing offset back to the attitude estimation module as a system calibration parameter; continuously assess communication stability and trigger the redirection process when necessary.

[0105] The following is a further explanation based on specific cases:

[0106] Case 1: In a rugged mountain environment, rescue teams need to quickly deploy satellite communication terminals to establish a communication connection with the command center. The complex magnetic field environment at the scene, including geomagnetic anomalies and interference from metal equipment, makes it difficult for traditional satellite orientation systems that rely on electronic compasses to accurately locate the satellite's direction. To meet the communication needs of the first rescue teams after an earthquake in a mountainous area, a method for rapid deployment and directional locking of satellite communication terminals for natural disaster rescue is proposed. Specifically:

[0107] Step 1: System composition and data acquisition.

[0108] The system of this embodiment includes: a three-axis gyroscope (model MPU-9250, sampling rate 500Hz); a three-axis accelerometer (model MPU-9250, sampling rate 100Hz); a three-axis magnetometer (model HMC5883L, sampling rate 20Hz); a single-board computer (equipped with an ARM Cortex-A72 processor); and a two-axis electrically adjustable antenna bracket (360° horizontal rotation and 0°-90° vertical adjustment).

[0109] After the system starts, it first executes the sensor initialization program to read the preset calibration parameters. Then it performs the following data acquisition and preprocessing steps:

[0110] Multi-sensor parallel data acquisition: The gyroscope acquires the three-axis angular velocity ω = [ωx, ωy, ωz] at a 500Hz sampling rate T , unit rad / s; the accelerometer acquires the three-axis acceleration a = [ax, ay, az] at a sampling rate of 100 Hz T , unit m / s 2 ; The magnetometer obtains the three-axis magnetic field strength m = [mx, my, mz] at a sampling rate of 20Hz T , unit is μT; each set of data is appended with the acquisition timestamp t, with the accuracy of milliseconds. T is transposed; ωx, ωy, ωz are the angular velocities around the X-axis, Y-axis, and Z-axis respectively; ax, ay, az are the accelerations along the X-axis, Y-axis, and Z-axis respectively; mx, my, mz are the magnetic field intensities along the X-axis, Y-axis, and Z-axis respectively.

[0111] Sensor Data Calibration and Noise Removal: Gyroscope Data Calibration: ω_cal = S_g (ω_raw - b_g); where S_g is the gyroscope scale factor matrix and b_g is the bias vector. Gyroscope Temperature Compensation: ω_comp = ω_cal - k_g (T - T_ref); where T is the current temperature, T_ref is the reference temperature, and k_g is the temperature coefficient. Accelerometer Data Calibration: a_cal = S_a (a_raw - b_a); where S_a is the accelerometer scale factor matrix and b_a is the bias vector. Magnetometer Hard-Iron Calibration: m_hard = m_raw - b_m, where b_m is the hard-iron bias vector and m_raw is the raw magnetometer data. Magnetometer Soft-Iron Calibration: m_cal = S_m m_hard, where S_m is the soft-iron correction matrix.

[0112] Perform multi-sampling rate data synchronization and alignment: establish a unified time base with a base sampling rate of 100 Hz, downsample the gyroscope data by taking the average of every 5 samples, and linearly interpolate the magnetometer data by interpolating the values ​​between two adjacent sampling points to form an aligned data packet: Data_packet(t) = {ω_comp(t), a_cal(t), m_cal(t), t}.

[0113] Step 2: Identification and classification of magnetic field interference patterns.

[0114] 2.1. Magnetic field data representation with enhanced temporal features.

[0115] Use the sliding window method to extract magnetic field data features: Construct a sliding window with a window length of 50 samples (500ms). Extract the time domain features of the magnetic field data within the window: Mean magnetic field intensity: M_mean = ||m|| avg= (Σ||m_i||) / 50; magnetic field direction unit vector: m_dir = m_mean / ||m_mean||; magnetic field strength standard deviation: M_std = sqrt[(Σ(||m_i|| - M_mean) 2 ) / 50]; magnetic field direction fluctuation: D_var = (Σ(1-cos(θ_i))) / 50, θ_i is the angle between m_i and m_dir; short-term change rate: ΔM_max = max(||m_i+1 - m_i||) / Δt. Construct the enhanced magnetic field feature vector: M_feat = [m_dir, M_mean, M_std, D_var, ΔM_max]. Where m_i is the three-dimensional vector of magnetic field data; ||m|| avg is the average modulus of the magnetic field data, that is, the average intensity; m_mean is the mean magnetic field vector within the sliding window; Δt is the time interval.

[0116] 2.2. Progressive incremental clustering algorithm for core-satellite structure.

[0117] This embodiment constructs a unique progressive incremental clustering algorithm with a "core-satellite" structure to adapt to dynamic changes in the magnetic field environment. The algorithm flow is as follows: During system initialization, three basic clustering templates are preset (stable environment, periodic perturbation, and strong magnetic field interference). For each new enhanced magnetic field feature vector M_feat, the adaptive distance from the existing cluster is calculated: adaptive distance D(M_feat, ci) = 0.6 (1-cos(θi)) + 0.3 |log(M_mean / M_i)| + 0.1 (1-exp(-0.05 ΔT_i)). Where θi is the angle between the direction of M_feat and the core direction of cluster ci; M_mean is the current mean magnetic field intensity, Mi is the core intensity of cluster ci; and ΔT_i is the difference (in seconds) between the current time and the last update time of cluster ci. Find the minimum distance D_min and the corresponding cluster c_match. Perform a core-satellite structure update based on the minimum distance D_min. If D_min < Th_core (core threshold, initially 0.15), classify the point as a core update and calculate the adaptive learning rate: α = 0.2 exp(-0.001 N_match) (1-S_match), where N_match is the number of cluster matches and S_match is the cluster stability metric. Update the cluster core parameters: c_match_new = (1-α) c_match + α M_feat. Update cluster statistics and timestamps. If Th_core ≤ D_min < Th_sat (satellite threshold, initially 0.3), classify the point as a satellite and record the enhanced magnetic field feature vector M_feat as a satellite point in cluster c_match. Update the cluster boundary information: radius_max = max(radius_max, D_min). Record the satellite point features for possible subsequent core splitting. If D_min ≥ Th_sat: classify it as a potential new pattern; add the enhanced magnetic field feature vector M_feat to the temporary buffer; if there are ≥5 similar samples in the buffer, create a new clustering core.

[0118] Perform cluster maintenance and optimization. Cluster aging detection: Calculate cluster inactivity time, marking any time exceeding 30 seconds as dormant. Cluster merging detection: If the core distance D(ci, cj) between two clusters is less than 0.1, merge them into one. Calculate the cluster confidence score: Cfid(c_match) = 0.4 min(1, N_match / 50) + 0.4 (1-min(1, V_match / 0.5)) + 0.2 exp(-0.1 ΔT_act), where N_match is the cumulative number of cluster samples; V_match is the internal variance of the cluster; and ΔT_act is the time interval between the last cluster activity (in seconds).

[0119] Output recognition results: magnetic field interference mode identification: Mode_ID = ID(c_match); clustering credibility score: Cfid(c_match); extracted pattern feature parameters: duration, intensity change rate, and directional stability.

[0120] Step 3: Posture estimation with adaptive parameter adjustment.

[0121] This embodiment uses an extended Kalman filter (EKF) for attitude estimation and dynamically adjusts the filter parameters according to the magnetic field interference pattern identifier.

[0122] 3.1. Construct filter state and observation model.

[0123] State vector: x = [q, b_ω] T ; q = [q0, q1, q2, q3] T is the attitude quaternion; b_ω = [bx, by, bz] T is the gyroscope bias estimate. The state transition equations are: q_k+1 = q_kΘq(ω_k·dt); b_ω_k+1 = b_ω_k; where Θ represents quaternion multiplication, and q(ω_k·dt) represents the quaternion increment obtained by integrating the angular velocity. Measurement updates use acceleration and magnetic field data: gravity direction observation: g_meas = a_cal / ||a_cal||; magnetic field direction observation: m_meas = m_cal / ||m_cal||. q_k is the current attitude quaternion, q_k+1 is the next attitude quaternion, b_ω_k is the current gyroscope bias estimate vector, and b_ω_k+1 is the next gyroscope bias estimate vector.

[0124] 3.2. Adaptive parameter adjustment mechanism.

[0125] The system dynamically adjusts EKF parameters based on the magnetic field interference mode identifier (Mode_ID) and the cluster confidence score (Cfid): Measurement noise covariance matrix adjustment: R_mag = R_mag0·(2 - Cfid)·f(Mode_ID); where R_mag0 is the magnetometer baseline noise covariance; and f(Mode_ID) is an adjustment function based on the interference mode: f = 1.0 for "stable environment" mode, f = 1.5 for "periodic disturbance" mode, and f = 3.0 for "strong magnetic field interference" mode. Process noise covariance matrix adjustment: When Cfid < 0.3, the gyroscope noise covariance is increased to reduce reliance on magnetic field measurements. When Mode_ID is "strong magnetic field interference" and the duration is > 10 seconds, the system switches to "gyroscope + gravity only" mode. Enhanced attitude stability: Detecting sudden attitude changes: If the difference between two consecutive estimated angles exceeds 20°, a smooth transition is triggered. Adaptive smoothing is applied: The output attitude q_out = SLERP(q_prev, q_curr, β), where SLERP is the spherical linear interpolation function and β is the smoothing coefficient, which is dynamically adjusted based on Mode_ID and Cfid. The terminal attitude estimate is output in Euler angles: [roll, pitch, yaw], where roll is roll, pitch is pitch, and yaw is yaw. Estimation uncertainty is calculated based on the diagonal elements of the state covariance matrix P.

[0126] Step 4: Filtering-clustering dual-loop feedback control.

[0127] This embodiment implements closed-loop control through a bidirectional feedback mechanism between filtering and clustering algorithms.

[0128] 4.1. Consistency evaluation of filtering-clustering results.

[0129] Calculate the filter residual statistics: S_res = (z - h(x)) T ·R -1 (z - h(x)); where z is the actual measurement value; h(x) is the predicted measurement value based on the state; and R is the measurement noise covariance matrix. Normalized residual statistic: Anorm = min(1, S_res / χ 2 α), where χ 2 α is the 95% critical value of the chi-square distribution with 5 degrees of freedom. The consistency score is calculated as: Con = (1 - |Anorm - Cfid|)·exp(-0.5·tr(P) / 3), where Cfid is the confidence score of the current matching cluster, and tr(P) is the trace of the state covariance matrix P, which represents the estimation uncertainty.

[0130] 4.2. Generate bidirectional feedback parameters.

[0131] Based on the consistency score Con, the system generates two sets of feedback parameters:

[0132] Filter parameter adjustment set: ΔR_mag = R_mag0·(1 - Con)·f(Mode_ID); ΔQ_gyro = Q_gyro0·(1 - Con)·(2 - Cfid); where R_mag0 is the magnetometer reference noise covariance; Q_gyro0 is the gyroscope reference process noise covariance; and f(Mode_ID) is the interference mode-based adjustment function. Clustering parameter adjustment set: ΔTh_core = Th_core0·(1 - Con)·(0.5 + 0.5·Anorm); ΔTh_sat = Th_sat0·(1 - Con)·(0.5 + 0.5·Anorm); where Th_core0 is the core threshold reference value of 0.15; and Th_sat0 is the satellite threshold reference value of 0.3. Parameter smoothing: A first-order low-pass filter is applied to smooth parameter changes; a parameter change rate limiter is used to prevent sudden changes.

[0133] 4.3. Implementation of dynamic compensation strategy.

[0134] The system implements specific compensation strategies for different magnetic field interference patterns: "Stable Environment" mode (Mode_ID=1): Normal weight configuration, trusting magnetic field measurements; R_mag = R_mag0, standard measurement update process. "Periodic Perturbation" mode (Mode_ID=2): Apply frequency-domain selective filtering to filter out periodic interference; adjust magnetic field measurement update weights: R_mag = 1.5·R_mag0; shorten the sliding window length to improve short-term responsiveness. "Strong Magnetic Field Interference" mode (Mode_ID=3): Implement a magnetic field rejection strategy, relying primarily on gyroscopes and gravity; significantly increase magnetic field measurement noise: R_mag = 3.0·R_mag0; use a historical magnetic field reference model to assist in orientation. "Unknown Interference" mode (New Clustering): A conservative strategy, increasing system uncertainty; slowing filter convergence to maintain stability.

[0135] Step 5: Satellite direction positioning and locking.

[0136] 5.1. Calculate the satellite pointing angle.

[0137] Based on the GPS position and target satellite parameters, the theoretical satellite azimuth angle Az_sat and elevation angle El_sat are calculated. Combined with the current terminal attitude estimate [roll, pitch, yaw], the azimuth and elevation angles that need to be adjusted for the antenna are calculated: Az_adjust = Az_sat - yaw + Az_offset; El_adjust = El_sat·cos(roll) - pitch·sin(Az_adjust) + El_offset; where Az_offset and El_offset are calibration offsets. The pointing angle uncertainty is calculated based on the attitude estimate uncertainty: σ_Az = sqrt(σ_yaw 2 + (σ_roll·sin(El_sat)) 2 );σ_El = sqrt(σ_pitch 2 +(σ_roll·cos(Az_adjust)) 2 ), where σ_Az is the azimuth uncertainty, σ_El is the elevation uncertainty, σ_yaw is the yaw uncertainty, σ_roll is the roll uncertainty, and σ_pitch is the pitch uncertainty.

[0138] 5.2. Control and lock the directional mechanism.

[0139] Convert the pointing angle into a servo control signal: horizontal servo angle = map(Az_adjust, -180°, 180°, 0°, 360°); vertical servo angle = map(El_adjust, 0°, 90°, 0°, 90°). Implement a two-stage locking strategy: coarse adjustment: Approach the target direction at a high speed; fine adjustment: Fine-tune to the optimal reception position at a low speed. Adaptive search algorithm: When the pointing uncertainty σ_Az > 10°, a scanning search is performed; the search range is an elliptical area of ​​±2σ; the step size is gradually reduced in proportion to the uncertainty. Signal peak lock: Monitor the received signal strength RSSI and record the peak position. When the RSSI exceeds the threshold and is relatively stable, the current position is locked. The mechanical lock is activated to fix the antenna pointing direction.

[0140] This embodiment was tested in a simulated disaster-affected magnetic field interference environment. Compared to traditional methods, the traditional method had an average satellite pointing accuracy error of 8.5° and a maximum error of 21.3°. This embodiment achieved an average satellite pointing accuracy error of 3.2° and a maximum error of 7.8°, a 62.4% reduction in average error. The traditional method had an average positioning time of 73 seconds, with a minimum of 126 seconds. This embodiment achieved an average positioning time of 31 seconds, with a minimum of 54 seconds, representing a 57.5% improvement in average time. In areas of strong magnetic field interference (magnetic field strength variation > 50 μT), the traditional method's magnetic field interference adaptability failed to reliably provide positioning (success rate < 30%), while this embodiment's magnetic field interference adaptability provided stable positioning (success rate > 85%). Computing resource usage of this embodiment: CPU peak usage < 40%; memory usage < 120 MB; battery life > 6 hours. This embodiment utilizes a progressive incremental clustering algorithm and a filtering-clustering dual-loop feedback control mechanism to improve the positioning accuracy and deployment speed of satellite terminals in harsh environments, providing reliable communication support for disaster relief efforts.

[0141] Case 2: This embodiment provides a method for rapid deployment and directional locking of satellite communication terminals for natural disaster relief. This method can achieve accurate attitude estimation and satellite directional locking of satellite communication terminals in complex magnetic field interference environments. The system uses a nine-axis sensor combination (three-axis accelerometer, three-axis gyroscope, and three-axis electronic compass) to collect data. Through a filtering-clustering dual-loop feedback control structure and a progressive magnetic field interference pattern incremental clustering algorithm, it solves the problem of attitude estimation errors caused by magnetic field interference in disaster areas. Specifically:

[0142] Step 1: Sensor data collection and preprocessing.

[0143] The terminal device uses a nine-axis sensor model MPU9250 to collect data according to the following sampling rate and accuracy requirements: gyroscope: sampling rate 500Hz, range ±250° / s, resolution 0.01° / s; accelerometer: sampling rate 100Hz, range ±2g, resolution 0.001g; electronic compass: sampling rate 20Hz, range ±4800μT, resolution 0.6μT.

[0144] The raw sensor data is calibrated using the following equation: Gyroscope calibration: ω_calibrated = (ω_raw - b_g)·S_g·T_comp; where ω_calibrated is the calibrated angular velocity data, ω_raw is the raw angular velocity data, b_g is the gyroscope bias vector, S_g is the scale factor matrix, and T_comp is the temperature compensation factor.

[0145] Accelerometer calibration: a_calibrated = (a_raw - b_a) · S_a; where a_calibrated is the calibrated acceleration data, a_raw is the raw acceleration data, b_a is the accelerometer bias vector, and S_a is the scale factor matrix.

[0146] Compass calibration: m_calibrated = (m_raw - b_m) S_m A; where m_calibrated is the calibrated magnetic field data, m_raw is the raw magnetic field data, b_m is the hard-iron calibration bias vector, S_m is the soft-iron calibration matrix, and A is the azimuth calibration matrix.

[0147] Sensor data with different sampling rates was unified to 100Hz, and a time window alignment algorithm was designed. A fourth-order Butterworth low-pass filter (cutoff frequency 45Hz) was applied to the gyroscope data to reduce noise, and cubic spline interpolation was used to increase the sampling rate of the electronic compass data. The three sensor data were aligned based on their timestamps, generating a time-aligned multi-sensor data packet D_aligned = {t, ω, a, m}, where t is the timestamp, ω is the angular velocity vector, a is the acceleration vector, and m is the magnetic field vector.

[0148] Step 2: Identification and classification of magnetic field interference patterns.

[0149] 2.1. Magnetic field data representation with enhanced temporal features.

[0150] Receive time-aligned multi-sensor data packets, build a 500ms sliding window, and extract the time domain features of the magnetic field data. The features are calculated as follows: Magnetic field intensity feature: M_mag = ||m|| = sqrt(m_x 2 + m_y 2 + m_z 2 ); Directional characteristics: M_dir = [arctan(m_y / m_x), arctan(m_z / sqrt(m_x 2 + m_y 2 ))]. Time domain statistical characteristics: mean: μ_m = (1 / N)·∑m_i; variance: σ 2 _m = (1 / N) ∑(m_i - μ_m) 2 ; Maximum rate of change: Δm_max = max(||m_i - m_{i-1}||) / Δt; Frequency domain energy: E_m = ∑|FFT(m)| 2Relative deviation characteristics: Δ_ref = ||m- m_ref|| / ||m_ref||; where m_ref is the reference magnetic field strength, determined according to the geographical location. The above characteristics are combined to construct the enhanced magnetic field feature vector F = [M_mag, M_dir, μ_m, σ 2 _m, Δm_max, E_m, Δ_ref]. Where m is the magnetic field vector; m_x, m_y, and m_z represent the components of the magnetic field vector on the x-axis, y-axis, and z-axis, respectively; N is the number of data points in the sliding window; m_i is the magnetic field vector data at the i-th moment in the current window; Δt is the time difference; and FFT stands for fast Fourier transform.

[0151] 2.2. Incremental clustering of progressive magnetic field interference patterns.

[0152] The following details the key calculation process of the progressive incremental clustering algorithm for the core-satellite structure:

[0153] Adaptive distance metric calculation: a) Feature normalization to generate a standardized feature vector: F_norm = (F - F_min) / (F_max - F_min). b) Time decay weight calculation: w_t = exp(-λ·(t_current - t_last)); where λ is the time decay coefficient (0.05 in this example), t_current is the current timestamp, and t_last is the last cluster update timestamp. c) Direction-sensitive distance calculation: d_dir(v1, v2) = 1 - |cos(v1, v2)| = 1 - |v1·v2| / (||v1||·||v2||); this distance measures the similarity between two magnetic field direction vectors and has a value range of [0, 1]. d) Intensity-sensitive distance calculation: d_mag(m1, m2) = |log(||m1|| / ||m2||)|; this distance captures the relative change in magnetic field strength, and the logarithmic function makes intensity changes of different sizes comparable. e) Time-sensitive distance calculation: d_temp(F1_temp, F2_temp) = sqrt(∑w_i)·(F1_temp_i - F2_temp_i) 2, where F_temp represents the time-domain statistical feature component, and w_i represents the weight of each feature. f) Comprehensive calculation of the adaptive distance metric function: d(F, C_i) = α·d_dir(F_dir, C_i_dir) + β·d_mag(F_mag, C_i_mag) + γ·d_temp(F_temp, C_i_temp); where α, β, and γ are weight coefficients. In this embodiment, their initial values ​​are set to 0.4, 0.3, and 0.3, respectively, and are dynamically adjusted with the degree of magnetic field interference. g) Calculate the distance between the current sample and each existing cluster, finding the minimum distance d_min and the corresponding cluster C_min. Where F is the feature vector; F_min is the minimum value of each feature in the feature vector; F_max is the maximum value of each feature in the feature vector; v1 and v2 are magnetic field direction vectors; m1 and m2 are magnetic field intensity vectors; F_dir is the direction feature of the sample; C_i_dir is the direction feature of the cluster center; F_mag is the magnetic field intensity feature of the sample; C_i_mag is the magnetic field intensity feature of the cluster center; F_temp is the time feature vector of the sample; C_i_temp is the time feature vector of the cluster center.

[0154] Core-Satellite Structure Update: a) Set two key thresholds: the matching threshold θ_match (initial value 0.15 in this example) and the satellite point threshold θ_satellite (initial value 0.3 in this example). b) Based on the relationship between the minimum distance d_min and the threshold, perform one of the following three operations: If d_min < θ_match: perform a core update; if θ_match ≤ d_min < θ_satellite: perform a satellite point update; if d_min ≥ θ_satellite: perform potential new pattern processing. c) Detailed calculation process for the core update operation: Learning rate calculation: η = η_0·(1 - exp(-N_i / τ)); where η_0 is the base learning rate (value 0.1), N_i is the number of samples in cluster C_min, and τ is the scaling parameter (value 50). Core parameter update: C_min_core = (1-η)·C_min_core + η·F_norm; Update cluster statistics: Increase sample count: N_i = N_i +1; Update variance: σ_i 2 = (1-η)·σ_i 2 + η·||F_norm - C_min_core|| 2; Update the most recent active timestamp: t_last = t_current. d) Detailed calculation process of satellite point update operation: Add the normalized feature vector F_norm as a satellite point to the cluster C_min: S_min = S_min ∪ {F_norm}; Update the cluster boundary information: B_min = max(B_min, d_min); Keep the size of the satellite point set limited: If |S_min| > K_max, remove the oldest satellite point. In this embodiment, the maximum capacity K_max of the satellite point set is set to 20. e) Detailed calculation process of potential new pattern processing: Add the normalized feature vector F_norm to the temporary buffer: T_buffer = T_buffer ∪ {F_norm}; Calculate the average pairwise distance of the points in the buffer: d_avg = (2 / n(n - 1))·∑∑d(F_i, F_j), where i < j; n is the number of points in the buffer; If the condition for creating a new cluster is met: d_avg < θ_new and |T_buffer| > n_min, create a new cluster C_new, with the initial core being the average of the points in the temporary buffer T_buffer: C_new_core = (1 / |T_buffer|)·∑F_i; Set the initial sample count N_new = |T_buffer|, the initial variance σ_new 2 = d_avg, and the initial timestamp t_new_last = t_current; Add the new cluster to the cluster set: C = C ∪ {C_new}; Clear the temporary buffer: T_buffer = ∅. In this embodiment, the distance threshold θ_new for the new cluster takes the value of 0.2, and the minimum point number threshold n_min for the new cluster takes the value of 5.

[0155] Cluster maintenance and optimization: The system performs the following operations every certain period of time (set to 30 seconds in this embodiment) or when a specific event is triggered: a) Cluster aging detection: Calculate the inactivity time of each cluster: t_inactive = t_current - t_last; If t_inactive > T_max (T_max is set to 300 seconds in this embodiment), mark the cluster as dormant. b) Cluster merge detection: Calculate the distance between the cores of any two active clusters: d(C_i_core, C_j_core); If d(C_i_core, C_j_core) < θ_merge (taking the value of 0.1 in this embodiment): Calculate the weighted average to obtain the new core: C_new_core = (N_i·C_i_core + N_j·C_j_core) / (N_i + N_j); Merge statistics: N_new = N_i + N_j σ_new 2= (N_i·σ_i 2 + N_j·σ_j 2 + N_i·N_j·||C_i_core - C_j_core|| 2 / (N_i + N_j)) / (N_i + N_j); Merge the satellite point sets: S_new = S_i ∪ S_j; If |S_new| > K_max, retain the closest K_max points; Delete the two original clusters and add the merged cluster. Where C_i_core is the core parameter of cluster C_i; θ_merge is the distance threshold for cluster merging.

[0156] Evaluation of incremental clustering results: a) Clustering stability index calculation: Stab_i = f(N_i, σ_i 2 , t_inactive) = N_i·exp(-k1·σ_i 2 )·exp(-k2·t_inactive / T_max); where k1 and k2 are weight parameters (in this embodiment, they are 2.0 and 0.5, respectively). b) Cluster credibility score calculation: Conf_i = g(Stab_i, d_min) = Stab_i·exp(-d_min / d_0); where d_0 is a scaling parameter (in this embodiment, it is 0.2). c) Extract interference pattern characteristic parameters: duration: T_duration = t_current - t_first_appear; intensity change rate: R_intensity = (||m_current|| - ||m_first||) / T_duration; directional stability: S_direction = 1 - σ(cos(m_i, μ_m)). d) Generate the magnetic field interference pattern identifier M_id and the cluster credibility score Conf_i. Where t_first_appear is the timestamp when the magnetic field interference pattern first appears; m_current is the magnetic field strength vector at the current moment; m_first is the magnetic field strength vector when the magnetic field interference pattern first appears; and σ is a statistic in directional stability.

[0157] Step 3: Posture Estimation with Adaptive Parameter Adjustment

[0158] 3.1. Multi-sensor data fusion preprocessing.

[0159] Based on the magnetic field interference pattern identifier M_id and the magnetic field reliability index Conf_i, the sensor fusion weight is dynamically adjusted: Sensor weight calculation: w_m = w_m0 exp(-k (1-Conf_i)); where w_m0 is the magnetometer base weight (0.3 in this example) and k is a tuning parameter (2.0). The measurement noise covariance matrix is ​​adjusted: R_m = R_m0 / w_m, where R_m0 is the magnetometer base noise covariance matrix. The covariance is increased when the confidence level is low.

[0160] 3.2. Implementation of adaptive extended Kalman filter.

[0161] State vector definition: x = [q, b_g, b_a, b_m] T ; where q is the attitude quaternion, b_g is the gyroscope bias, b_a is the accelerometer bias, and b_m is the magnetometer bias. State prediction steps: a) Quaternion integration: q*_k|k-1 =q_k-1|k-1 Θ q(Δt·(ω_k - b_g_k-1|k-1)) ; where Θ represents quaternion multiplication, q(·) represents the conversion from angular velocity to quaternion increment; q_k-1|k-1 is the attitude quaternion at the previous moment; ω_k is the gyroscope measurement at the current moment; and b_g_k-1|k-1 is the gyroscope bias at the previous moment. b) Bias prediction: b*_g_k|k-1 = b_g_k-1|k-1b*_a_k|k-1 = b_a_k-1|k-1 b*_m_k|k-1 = b_m_k-1|k-1; where b*_a_k|k-1 is the accelerometer bias predicted at the current moment; b_a_k-1|k-1 is the accelerometer bias at the previous moment; b*_m_k|k-1 is the magnetometer bias predicted at the current moment; b_m_k-1|k-1 is the magnetometer bias at the previous moment. c) Prediction state uncertainty update: P_k|k-1 = F_k·P_k-1|k-1·F_k T + Q_k; where P_k|k-1 is the uncertainty covariance matrix of the predicted state; F_k is the state transition matrix; and Q_k is the process noise covariance matrix. Measurement update steps: a) Set the measurement noise covariance matrix R_k based on the adjusted filter configuration. b) Calculate the Kalman gain: K_k = P_k|k-1·H_k T ·(H_k·P_k|k-1·H_k T + R_k) -1; where H_k is the measurement matrix. c) State update: x_k|k = x_k|k-1 + K_k·(z_k- h(x_k|k-1)); where x_k|k-1 is the predicted state vector at the current moment; z_k is the actual measurement value, and h(·) is the measurement function. d) Covariance update: P_k|k = (I - K_k·H_k)·P_k|k-1; where I is the identity matrix. Outputs the updated attitude state estimate (quaternion representation) and the estimated uncertainty P_k|k.

[0162] Perform attitude stability enhancement processing: Attitude mutation detection: If ||q_k|k Θ q_k-1|k-1-1|| > θ_jump (the value is 0.05 in this embodiment): trigger smooth transition processing. Smooth transition processing: q_smooth = slerp(q_k-1|k-1, q_k|k, s); where slerp is the spherical linear interpolation function, s is the smoothing coefficient, and the calculation formula is: s = 0.5 -0.5·cos(π·min(t / T_trans, 1)); T_trans is the transition time constant (the value is 1 second in this embodiment). Attitude angle calculation: Convert the quaternion q into the terminal attitude angle [φ, θ, ψ] expressed in Euler angles (representing the roll angle, pitch angle, and azimuth angle, respectively). Attitude quality score calculation: Q_attitude = exp(-tr(P_q) / σ_0 2 ), where P_q is the covariance matrix of the quaternion part, tr(·) represents the trace of the matrix, and σ_0 2 is a normalization parameter (in this embodiment, the value is 0.01).

[0163] Step 4: Filtering-clustering dual-loop feedback control.

[0164] 4.1. Consistency evaluation of filtering-clustering results.

[0165] Anomaly detection statistics calculation: a) Prediction residual calculation: r_k = z_k - h(x_k|k-1); where h() is the measurement function. b) Mahalanobis distance calculation: D_M 2 = r_k T ·S_k -1 ·r_k; where, S_k = H_k·P_k|k-1·H_k T + R_k is the residual covariance matrix. c) Abnormal detection threshold calculation: If D_M 2 > χ 2 n,α(χ 2 is the threshold of the chi-square distribution, n is the measurement dimension, α is the significance level, which is 0.05 in this embodiment), then an anomaly is detected.

[0166] Consistency score calculation: Consistency = w_filter exp(-D_M 2 / s_1) + w_cluster·Conf_i; where w_filter and w_cluster are weight coefficients (both are 0.5 in this embodiment), and s_1 is a scaling parameter (its value is the residual degree of freedom n).

[0167] 4.2. Generate bidirectional feedback parameters.

[0168] State adjustment judgment: If Consistency > θ_high (the value is 0.8 in this embodiment): maintain the current parameter configuration; if θ_low < Consistency ≤ θ_high (the value of θ_low is 0.5): slightly adjust the parameters; if Consistency ≤ θ_low: trigger deep adjustment.

[0169] Clustering threshold adjustment parameter calculation: a) Matching threshold adjustment: θ_match_new = θ_match (1 + k_adjust (θ_mid - Consistency)); where k_adjust is the adjustment strength parameter (valued at 0.5) and θ_mid is the mid-threshold (valued at 0.65). b) Satellite point threshold adjustment: θ_satellite_new = θ_satellite (1 + k_adjust (θ_mid - Consistency)). Filter weight adjustment parameter calculation: a) Measurement noise covariance scaling: scale_R = 1 + k_R max(0, θ_high - Consistency); where k_R is the amplification factor (valued at 4.0). b) Update the magnetometer-related components of the R matrix: R_m_new = scale_R R_m. Smooth transition function implementation: param_smooth = param_old + (param_new - param_old)·(0.5 - 0.5·cos(π·min(t / T_trans, 1))); where param_old is the parameter state before adjustment; param_new is the target new value of the parameter; and T_trans is the transition time constant (2 seconds in this embodiment).

[0170] 4.3. Clustering-compensation strategy mapping mechanism.

[0171] Feature matching algorithm: Calculate the similarity between the current clustering pattern and the pattern in the predefined interference type library: similarity(M_current, M_ref) = exp(-d(F_current, F_ref) / d_scale); where d(·, ·) is the aforementioned adaptive distance metric function, and d_scale is the scaling parameter (value is 0.3); M_current is the currently identified interference pattern; M_ref is the reference pattern in the predefined interference type library; F_current is the feature vector of the current pattern; and F_ref is the feature vector of the reference pattern.

[0172] Construct a mapping matrix: Build a mapping matrix based on the interference characteristics. Each row represents an interference type, each column represents a compensation strategy, and the element value indicates the degree of applicability. This embodiment supports the following interference types: persistent ambient magnetic field offset; sudden short-term interference; periodic oscillation interference; and random noise enhancement.

[0173] Policy effectiveness score update: Score_new(M_i, S_j) = λ·Score_old(M_i, S_j) + (1-λ)·ΔH; where Score_old is the historical effectiveness score; λ is the historical weight coefficient (set to 0.8), and ΔH is the change in state uncertainty before and after policy execution. Optimal compensation strategy selection: For the currently identified interference pattern Mi, select the strategy with the highest effectiveness score: S_opt = argmax_j(Map[M_i, S_j]·Score(M_i, S_j)). Compensation strategy parameterization: Based on the selected compensation strategy, the corresponding compensation parameter set is generated, including filter parameter adjustment instructions, measurement weight correction values, and model update instructions.

[0174] 4.4. Implementation of dynamic compensation strategy.

[0175] Differentiated processing is implemented for different types of magnetic field interference: Continuous ambient magnetic field offset compensation: a) Recalibrate the magnetic field reference model: m_ref_new = (1-α_ref)·m_ref_old + α_ref·m_current; where m_ref_old is the old magnetic field reference model and α_ref is the adaptation coefficient (set to 0.2). b) Adjust the azimuth estimation bias: ψ_bias = arctan2(m_ref_north×m_current, m_ref_north·m_current); where m_ref_north is the north component of the reference magnetic field; azimuth correction: ψ_corrected = ψ - ψ_bias.

[0176] Compensation for sudden, transient disturbances: a) Temporarily increase the magnetometer noise covariance: R_m_temp = β_temp · R_m, where β_temp is the temporary amplification factor (8.0). b) Increase the gyroscope integral weight: w_gyro_temp = min(1.0, w_gyro + Δw_gyro); where w_gyro is the gyroscope integral weight and Δw_gyro is the weight increment (0.3). c) Set the recovery time constant, T_recover, to 5 seconds, after which parameters gradually return to normal.

[0177] Periodic oscillation interference compensation: a) Spectral analysis identifies the main interference frequency f_disturb. b) Design a narrowband notch filter: H(z) = (1 - 2·cos(2πf_disturb / fs)·z -1 + z -2 ) / (1 - 2·r·cos(2πf_disturb / fs)·z -1 + r 2 ·z -2 ), where z is the variable in the filter, fs is the sampling frequency, and r is the attenuation coefficient (set to 0.9). c) Apply a notch filter to the magnetic field data.

[0178] This embodiment forms the core implementation of a method for rapid deployment and directional lock of satellite communication terminals for natural disaster relief. By using incremental clustering of progressive magnetic field interference patterns and filtering-clustering dual-loop feedback control, it effectively addresses the accuracy and stability issues of satellite communication terminal attitude estimation in magnetic field interference environments, enabling rapid deployment and reliable directional lock.

[0179] This invention designs a core-satellite clustering model, addressing the inability of traditional clustering methods to adapt to the characteristics of time-series data. It employs an incremental update mechanism, enabling the clustering model to continuously adapt to changing environments without the need to store and process all historical data. The adaptive distance metric function considers the temporal correlation and directional characteristics of magnetic field data, improving clustering accuracy. It breaks the one-way dependency between traditional filtering and clustering algorithms and establishes a bidirectional feedback mechanism. A consistency assessment mechanism enables mutual verification and compensation between filtering and clustering results. A smooth transition function design ensures the continuity of attitude estimation when switching between interference modes, avoiding the attitude jumps found in traditional methods. Differentiated compensation strategies are designed for different types of magnetic field interference, rather than the simple weight adjustment used in traditional methods. Different processing methods are used for persistent environmental offsets, sudden interference, and periodic interference, improving system adaptability. The parameterized design of the compensation strategy enables the system to flexibly respond to various complex interference environments. Decomposing computationally intensive cluster analysis into interruptible microtasks resolves the conflict between real-time performance and complex computations. Dynamically adjusting processing priorities based on posture quality ensures optimal system performance under resource-constrained conditions. A time-slice allocation mechanism for the filtering main loop and clustering auxiliary loop ensures that critical processing is not delayed.

[0180] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A method for rapid deployment and directional locking of satellite communication terminals for natural disaster relief, characterized in that: include: Acquire data from various sensors, pre-process them, and form data packets; receiving a data packet, classifying the magnetic field interference pattern, and obtaining a magnetic field interference pattern identifier; Based on this, attitude estimation is performed to generate the terminal attitude estimate and uncertainty; Based on the magnetic field interference pattern identification, terminal attitude estimation value and uncertainty, filtering-clustering dual-loop feedback control is performed to generate a compensation parameter set including a filtering parameter adjustment set and a clustering parameter adjustment set; the clustering parameter adjustment set is fed back to the magnetic field interference pattern classification, and the filtering parameter adjustment set is fed back to the attitude estimation, and the cycle is executed until the accuracy threshold is met; Based on the terminal attitude estimation value and satellite position information that meet the accuracy threshold, satellite direction positioning and locking are performed, and the satellite pointing angle and pointing control instructions are output.

2. The method according to claim 1, characterized in that The steps of classifying the magnetic field interference pattern and obtaining the magnetic field interference pattern identification include: receiving data packets, initial or feedback clustering parameter adjustment sets, extracting time domain features of magnetic field data, and constructing enhanced magnetic field feature vectors; The enhanced magnetic field feature vectors are processed using a progressive incremental clustering algorithm with a core-satellite structure, and the clustering threshold is dynamically adjusted based on the clustering parameter adjustment set to obtain the magnetic field interference pattern identification and clustering credibility score.

3. The method according to claim 2, characterized in that The steps of using the core-satellite structured progressive incremental clustering algorithm to process the enhanced magnetic field feature vectors and obtain the magnetic field interference pattern identification and clustering credibility score include: For the enhanced magnetic field eigenvector, the adaptive distance between the current eigenvector and the existing cluster is calculated, and the minimum distance and the corresponding cluster are determined, which is recorded as the best matching cluster; According to the minimum distance and clustering parameters, the matching threshold in the set is adjusted, the core-satellite structure update is performed, the current sample is assigned as a core update, a satellite point or a potential new pattern, and the cluster core parameters, cluster boundary information or potential new pattern buffer are updated accordingly; When the preset conditions are met, cluster maintenance and optimization are performed to generate an updated cluster set and determine the magnetic field interference pattern identifier that best matches the current cluster; A cluster confidence score is calculated based on the status of the updated cluster set.

4. The method according to claim 3, characterized in that The steps of calculating the adaptive distance between the current feature vector and the existing clusters and determining the minimum distance and the corresponding clusters include: performing normalization processing on the enhanced magnetic field eigenvector to generate a standardized eigenvector including a direction component and an intensity component; The difference between the current time and the last update time of each cluster is calculated based on the standardized feature vector, and the exponential decay function is applied to generate the time decay weight coefficient; Calculate the direction-sensitive distance based on the cosine value of the angle between the direction component and the core direction of each cluster; The intensity-sensitive distance is calculated based on the logarithmic ratio of the intensity component to the core intensity of each cluster; Use preset weight coefficients to combine direction-sensitive distance, intensity-sensitive distance and time-decay weight coefficients to generate an adaptive distance metric value; Compare the adaptive distance metric values ​​of all clusters and determine the minimum distance and the corresponding cluster.

5. The method according to claim 3, characterized in that The steps for performing a core-satellite structure update, and accordingly updating cluster core parameters, cluster boundary information, or potential new pattern buffers, include: Compare the minimum distance with the matching threshold to determine whether to perform core update operation, satellite point update operation or potential new mode processing; When performing a core update operation, the adaptive learning rate is calculated and the cluster core parameters, statistics, and active timestamps are updated; When performing the satellite point update operation, the current sample is recorded as a satellite point, and the cluster boundary information and satellite point set are updated; When executing potential new pattern processing, the current sample is added to the temporary buffer. When the sample in the temporary buffer meets the preset conditions for creating a new cluster, a new cluster core is generated and the potential new pattern buffer is updated.

6. The method according to claim 5, characterized in that The steps of cluster maintenance and optimization to determine the magnetic field interference pattern identifier that best matches the current cluster include: Perform cluster aging detection, calculate the inactivity time of each cluster, and mark clusters whose inactivity time exceeds a predetermined threshold as dormant; perform cluster merging detection, calculate the cluster core distance based on the updated cluster core parameters, and when the distance between two cluster cores is less than the preset merging threshold, calculate the weighted average to generate a new core and merge the statistics and satellite point sets; perform cluster structure rebalancing, regularly optimize cluster boundaries and core distribution to reduce overlap and redundancy; and output the updated cluster set; Based on the updated cluster set, the overall stability index of the current cluster model is calculated, and the magnetic field interference pattern identifier of the current most matching cluster is determined.

7. The method according to claim 3, characterized in that The steps of calculating the cluster credibility score based on the updated state of the cluster set include: Based on the updated cluster set, the number of samples, internal variance, and inactivity time are extracted to calculate the stability index of the matching cluster; Based on the stability index and the minimum distance, a cluster credibility score is generated to quantify the reliability of the current cluster recognition result. The interference pattern features of the matching clusters are extracted, including duration, intensity change rate and directional stability, and a cluster feature parameter set is generated.

8. The method according to claim 7, characterized in that The steps of performing filtering-clustering dual-loop feedback control and generating a compensation parameter set include: Calculate anomaly detection statistics based on filter residuals and compare them with cluster confidence scores to generate a consistency score that represents the degree of coordination between filtering and clustering results; When the consistency score is lower than the preset threshold, a depth adjustment is triggered and adjustment parameters are generated, including clustering threshold adjustment parameters and filter weight adjustment parameters; Based on the magnetic field interference pattern identification and adjustment parameters, a corresponding compensation strategy is selected, and the selected compensation strategy is parameterized to generate a compensation parameter set, including a filtering parameter adjustment set and a clustering parameter adjustment set.

9. The method according to claim 8, characterized in that Select the corresponding compensation strategy and generate the compensation parameter set. The steps include: receiving a magnetic field interference pattern identifier, a clustering feature parameter set, and an adjustment parameter; A feature matching algorithm is used to match the current interference pattern with a predefined interference type library, and a mapping relationship matrix including the correspondence between interference patterns and compensation strategies is established. The optimal compensation strategy is selected and a compensation parameter set is generated by combining the interference pattern characteristics and the preset strategy effect scores.

10. The method according to claim 8, characterized in that The steps to calculate the anomaly detection statistic based on the filter residuals and generate the consistency score include: Calculate the prediction residual based on the terminal posture estimate and historical state estimate; The Mahalanobis distance formula is used to calculate the standardized statistics of the prediction residuals and generate anomaly detection statistics; The anomaly detection statistics and clustering credibility scores are weighted and fused to evaluate the consistency of filtering and clustering results; a consistency score is generated to quantify the degree of consistency.

Images