Intelligent ultra-wideband positioning data correction method

Abstract

The invention discloses an intelligent ultra-wideband positioning data correction method. The method comprises the following steps: S1, generating a preprocessed ultra-wideband positioning data set; s2, training the improved self-supervised learning network model to obtain a positioning feature vector; s3, outputting a dynamic mask parameter pair; s4, outputting the corrected ultra-wideband positioning data; and S5, performing error evaluation on the corrected ultra-wideband positioning data and the original ultra-wideband positioning data, updating the improved self-supervised learning network model and the abnormal mode adaptive clustering mask mechanism according to an error evaluation result, and outputting the corrected ultra-wideband positioning data passing the error evaluation for downstream navigation, tracking or position service application. According to the invention, the data stability and application reliability of downstream navigation, tracking and position service systems are effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of data correction, and particularly to an intelligent ultra-wideband positioning data correction method. Background Art

[0002] With the continuous development of ultra-wideband positioning technology, the ultra-wideband positioning system relies on its characteristics of high precision and low latency. However, in the actual application process, due to environmental interference, signal occlusion, hardware fluctuations, and the influence of complex dynamic scenarios, the ultra-wideband positioning system often inevitably generates abnormal data, which will directly affect the accuracy and stability of system positioning, and further affect the reliability of downstream applications such as navigation, tracking, and location services.

[0003] Currently, for the processing of ultra-wideband positioning abnormal data, traditional methods mainly include static threshold screening and anomaly detection models based on manual annotation. On the one hand, the static threshold screening method relies on manual experience to set fixed anomaly determination rules, which are difficult to adapt to different environments and changes in anomaly types, and are prone to misjudgment or missed judgment. On the other hand, although the anomaly detection method based on manual annotation can improve the recognition accuracy by training machine learning models, it requires a large number of high-quality annotation samples, with high annotation costs and difficult to cover all potential anomaly patterns, restricting the popularization and application of the method.

[0004] In recent years, some studies have tried to introduce deep learning methods for ultra-wideband positioning anomaly detection. However, these methods generally have the following problems: First, traditional deep learning methods usually rely on supervised learning, still require a large amount of labeled data as support, and have insufficient generalization ability when the anomaly distribution changes. Second, the anomaly detection stage usually based on fixed feature extraction and static classifiers, lacking the ability to dynamically adapt to different anomaly types and changing environments, resulting in still large errors in the positioning data after system correction. Third, most of the abnormal data correction means are simple interpolation or direct deletion, failing to carry out targeted optimization processing in combination with the characteristics of anomaly patterns, affecting the overall continuity and spatial accuracy of the corrected data.

[0005] In summary, there is an urgent need for a new method that can automatically identify different anomaly patterns, dynamically optimize the masking strategy, and efficiently correct abnormal data to solve the above technical problems. Summary of the Invention

[0006] An object of the present invention is to propose an intelligent ultra-wideband positioning data correction method, which effectively guarantees the data stability and application reliability of downstream navigation, tracking, and location service systems.

[0007] An intelligent ultra-wideband positioning data correction method according to an embodiment of the present invention includes the following steps: S1. Collect the original ultra-wideband positioning data and perform preprocessing to generate a preprocessed ultra-wideband positioning data set; S2. Train an improved self-supervised learning network model based on the preprocessed ultra-wideband positioning data set to obtain a set of positioning feature vectors; S3. Use the positioning feature vectors to cluster the abnormal features of the preprocessed ultra-wideband positioning data set to obtain the results of abnormal pattern clusters, and dynamically generate mask rules, and output dynamic mask parameter pairs; S4. Input the real-time ultra-wideband positioning data into the improved self-supervised learning network model, perform anomaly detection in combination with the dynamic mask parameter pairs, generate an abnormal ultra-wideband positioning data mask and abnormal data, and perform local reconstruction or weight adjustment on the abnormal data according to the mask rules corresponding to the dynamic mask parameter pairs based on the abnormal ultra-wideband positioning data mask, and output the corrected ultra-wideband positioning data; S5. Perform error evaluation on the corrected ultra-wideband positioning data and the original ultra-wideband positioning data, update the improved self-supervised learning network model and the abnormal pattern adaptive clustering mask mechanism according to the error evaluation results, and output the corrected ultra-wideband positioning data that has passed the error evaluation for downstream navigation, tracking or location service applications.

[0008] Optionally, the S1 includes the following steps: S11. Collect the original ultra-wideband positioning data and construct an original ultra-wideband positioning data set. Each piece of original ultra-wideband positioning data in the original ultra-wideband positioning data set includes a timestamp, three-dimensional spatial coordinate information, and a signal quality index. The original ultra-wideband positioning data set contains the original ultra-wideband positioning data; S12. Resample and synchronize the original ultra-wideband positioning data based on the timestamps in the original ultra-wideband positioning data set, and then perform noise suppression processing to obtain a denoised ultra-wideband positioning data set; S13. Perform z-score standardization on the three-dimensional spatial coordinate information and the signal quality index in the denoised ultra-wideband positioning data set respectively, screen the standardized three-dimensional spatial coordinate information according to the three-sigma principle, and screen the standardized signal quality index according to the signal quality index threshold, and only retain the data with an absolute value not exceeding 3 and a signal quality index not lower than the threshold to form a preprocessed ultra-wideband positioning data set.

[0009] Optionally, the S2 includes the following steps: S21. Based on the preprocessed ultra-wideband positioning data set, continuously sample using a dynamic sliding window method to form an input sample set. Each input sample segment in the input sample set contains continuous preprocessed ultra-wideband positioning data with a length of The sliding window length According to the input sample segment The weighted combination calculation result of the standard deviation of the internal three-dimensional space coordinate information and the standard deviation of the signal quality index is dynamically determined, and the input sample set includes M input sample segments in total; S22. For each input sample segment in the input sample set, calculate the local anomaly score of each data point inside the input sample segment , and the local anomaly score is calculated based on the weighted sum of the change rate of the three-dimensional space coordinates of the data point and the change rate of the signal quality. Select the data point corresponding to the position with the largest local anomaly score as the pseudo label , and all pseudo labels form a pseudo label set ; ; S23. Pair the input sample set and the pseudo label set one by one to form a training sample pair set , and each pair of training samples in the training sample pair set consists of the input sample segment and the corresponding pseudo label ; the training sample pair set contains a total of pairs of training samples; S24. Based on the training sample pair set, train the improved self-supervised learning network model , and assign an anomaly sensitivity weight to each training sample pair during the training process , and the anomaly sensitivity weight is dynamically calculated according to the mean value of the local anomaly scores inside the input sample segment . The training optimization objective is to minimize the weighted loss function, and the weighted loss function is the weighted sum of the basic loss functions between the predicted outputs and the pseudo labels of all training sample pairs; S25. Input the preprocessed ultra-wideband positioning data set into the improved self-supervised learning network model after training and optimization , extract the high-dimensional positioning feature vectors of each piece of preprocessed ultra-wideband positioning data , and the high-dimensional positioning feature vectors form a positioning feature vector set .

[0010] Optionally, the S24 includes the following steps: S241. For the training sample pair set, calculate the local anomaly score according to each piece of preprocessed ultra-wideband positioning data inside each input sample segment , where ; S242. Based on the local anomaly score , calculate the anomaly sensitivity weight for each input sample segment , the anomaly sensitivity weight is defined as the mean of all local anomaly scores within the input sample segment; S243. Construct a weighted loss function , the weighted loss function is the weighted sum of the basic loss functions between the predicted output and the pseudo-labels for all training samples ; S244. Using the weighted loss function as the optimization objective, adopt the gradient descent method to iteratively update the parameters of the improved self-supervised learning network model, complete the training process, and obtain the optimized improved self-supervised learning network model .

[0011] Optionally, the S3 includes the following steps: S31. Jointly process each high-dimensional positioning feature vector in the positioning feature vector set with its corresponding three-dimensional space coordinate vector and the standardized signal quality index to construct a weighted adjacency matrix. The weighted adjacency matrix is calculated based on the three-dimensional space coordinate distance and the local bandwidth parameter between two preprocessed ultra-wideband positioning data. The local bandwidth parameter is obtained through a proportional mapping function according to the neighborhood coordinate variance centered on the preprocessed ultra-wideband positioning data. Use the adaptive spectral clustering method to cluster the positioning feature vectors based on the weighted adjacency matrix to form an anomaly pattern cluster set , in the anomaly pattern cluster set represents the th anomaly pattern cluster; S32. For each anomaly pattern cluster, perform linear prediction based on the space coordinate vector of each preprocessed ultra-wideband positioning data within the cluster and its neighboring data to obtain a reference coordinate vector. Calculate the Euclidean distance between the actual coordinate vector of the th data and the reference coordinate vector, and combine the standardized signal quality index to obtain a trajectory distortion factor , the trajectory distortion factor reflects the smoothness of the data trajectory and the degree of influence by noise within the anomaly pattern cluster ; S33. For each anomaly pattern cluster , calculate the covariance matrix of the anomaly pattern cluster based on the three-dimensional space coordinate vectors of all preprocessed ultra-wideband positioning data within the cluster , and define the square root of the trace of the covariance matrix as the spatial dispersion , the spatial dispersion is used to measure the spatial diffusion degree of the data within the anomaly pattern cluster; S34. Generate a dynamic mask parameter pair based on the trajectory distortion factor and the spatial dispersion of each anomaly pattern cluster , including the mask coverage ratio and the mask intensity weight , the mask coverage ratio The masking intensity weight is obtained by proportionally scaling the ratio of the trajectory distortion factor to the sum of the spatial dispersions. The dynamic masking parameter pair is obtained by mapping through the Sigmoid function based on the relationship between the average normalized signal quality index within the abnormal pattern cluster and the signal quality index threshold. It reflects the influence degree of each type of abnormal pattern on the subsequent masking process. S35. Based on each abnormal pattern cluster of the dynamic masking parameter pair , construct the corresponding dynamic masking rule function. The dynamic masking rule function controls the ratio of the masked feature dimensions with the masking coverage ratio and controls the substitution method of the masked features with the masking intensity weight. Combining with the time decay parameter construct a masking decay function based on the time distance, assign a small masking influence to the data samples far from the current moment, and form a time-sensitive masking rule.

[0012] Optionally, the S4 includes the following steps: S41. For each real-time ultra-wideband positioning data, perform noise removal processing and normalization processing in sequence according to the same processing flow as the preprocessed ultra-wideband positioning data to obtain the normalized real-time ultra-wideband positioning data. S42. Input the normalized real-time ultra-wideband positioning data into the optimized improved self-supervised learning network model, extract the corresponding real-time positioning feature vectors, assign each real-time positioning feature vector to the nearest abnormal pattern cluster, and obtain the dynamic masking parameter pair of the corresponding abnormal pattern cluster according to the abnormal pattern cluster-masking mapping relationship, including the masking coverage ratio and the masking intensity weight. S43. According to the dynamic masking parameter pair of the corresponding abnormal pattern cluster, combine the inference of the current moment and the time stamp of the real-time ultra-wideband positioning data, and calculate the abnormal detection masking response value. The abnormal detection masking response value represents the response of the abnormal degree of the real-time ultra-wideband positioning data. S44. According to the set abnormal determination threshold, judge whether the abnormal detection masking response value is greater than or equal to the abnormal determination threshold. If the condition is met, mark the corresponding real-time ultra-wideband positioning data as abnormal data and record it in the abnormal ultra-wideband positioning data masking set. S45. For the real-time ultra-wideband positioning data marked as abnormal data, perform local reconstruction or weight adjustment processing according to the corresponding dynamic masking parameter pair rule. For local reconstruction, linearly interpolate using the three-dimensional space coordinate vectors of the nearest normal data before and after to generate the corrected three-dimensional space coordinate vectors, which are used to replace the spatial position information of the abnormal data. Summarize the real-time ultra-wideband positioning data after completing the local reconstruction or weight adjustment processing to generate the corrected ultra-wideband positioning data set.

[0013] Optionally, S5 includes the following steps: S51. Correspondingly match the corrected ultra-wideband positioning data set with the original ultra-wideband positioning data set within the corresponding time period; S52. Calculate the positioning error based on the difference in three-dimensional space coordinate information between the corrected ultra-wideband positioning data and the original ultra-wideband positioning data, and define the positioning error of each piece of data as the Euclidean distance between the corrected coordinate and the original coordinate; S53. Statistically analyze all the positioning errors to calculate the overall corrected error mean and the standard deviation of the corrected error to form an error evaluation result; S54. Compare the overall corrected error mean with a preset error threshold . If the overall corrected error mean is less than or equal to the error threshold , it is determined that the correction effect meets the requirements of downstream applications. Otherwise, incrementally fine-tune the improved self-supervised learning network model and the anomaly pattern adaptive clustering mask mechanism according to the error evaluation result; S55. During the fine-tuning process, re-weight the training samples or adjust the anomaly pattern cluster mask parameters according to the error distribution characteristics, so that the improved self-supervised learning network model and the anomaly pattern adaptive clustering mask mechanism can better adapt to the changing trend of actual ultra-wideband positioning data; S56. Output the corrected ultra-wideband positioning data that has passed the error evaluation, and define the reference rules for downstream application scenarios based on the spatial trajectory characteristics and stability of the corrected ultra-wideband positioning data; S57. Classify and organize the corrected ultra-wideband positioning data that meets the reference rules for each downstream application scenario, and provide them for the navigation, tracking, and location service modules to call and use respectively.

[0014] Optionally, the reference rules for downstream application scenarios are: Navigation data: The trajectory of the corrected ultra-wideband positioning data is continuous and the mean positioning error is less than the preset navigation accuracy requirement; Tracking data: The trajectory of the corrected ultra-wideband positioning data changes smoothly, and the amplitude of the single-step position change is within the allowable tracking speed range; Location service data: The spatial distribution density of the corrected ultra-wideband positioning data meets the coverage integrity standard of the set service area.

[0015] The beneficial effects of the present invention are: (1) The present invention proposes an adaptive spectral clustering algorithm that fuses spatial relationships and signal quality. By constructing a weighted adjacency matrix, the three-dimensional spatial coordinate characteristics of ultra-wideband positioning data and signal quality indicators are jointly encoded into the clustering process, and the local bandwidth parameter is adaptively adjusted to ensure that the clustering can fully reflect the spatial local anomaly density and signal degradation characteristics. Compared with the traditional clustering method based on a single criterion of feature vector similarity, it can more accurately depict the spatial distribution of different types of anomaly patterns and the signal decline trend, improving the accuracy of anomaly pattern clusters and greatly enhancing the refinement and adaptability of anomaly pattern recognition.

[0016] (2) The present invention proposes a generation strategy for a pair of dynamic mask parameters of a trajectory distortion factor and a spatial dispersion degree. By simultaneously quantifying the smoothness and spatial diffusion characteristics of the positioning trajectory, the mask coverage ratio and the mask intensity weight are dynamically generated to achieve differential mask application for different anomaly pattern categories. The dual-index mechanism significantly improves the anomaly detection sensitivity and correction accuracy. In a complex dynamic environment, the mean value of the overall positioning error of the corrected ultra-wideband positioning data is reduced, showing higher stability and robustness in a multi-source interference environment.

[0017] (3) The present invention introduces an anomaly detection mask response mechanism in the anomaly correction process. Combining the time decay characteristics of the inference time and the data timestamp, it realizes the dynamic temporal weighting of anomaly detection decisions, effectively avoiding the false detection and missed detection problems caused by single-point anomalies. By introducing time decay to control the anomaly determination threshold, the temporal coherence of anomaly detection is improved, and the trajectory continuity index of the corrected trajectory is significantly better than the prior art, effectively ensuring the data stability and application reliability of downstream navigation, tracking, and location service systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings: Figure 1 It is a flowchart of an intelligent ultra-wideband positioning data correction method proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0020] Refer to Figure 1 , an intelligent ultra-wideband positioning data correction method, including the following steps: S1. Collect the original ultra-wideband positioning data and perform preprocessing to generate a preprocessed ultra-wideband positioning data set; S2. Construct self-supervised learning tasks based on the preprocessed ultra-wideband positioning dataset, and train the improved self-supervised learning network model to obtain positioning feature vectors; S3. Cluster the abnormal features of the preprocessed ultra-wideband positioning dataset using the positioning feature vectors to obtain the results of abnormal pattern clusters, and dynamically generate mask rules, and output dynamic mask parameter pairs; S4. Input the real-time ultra-wideband positioning data into the improved self-supervised learning network model, perform anomaly detection in combination with the dynamic mask parameter pairs, generate an abnormal ultra-wideband positioning data mask and abnormal data, and perform local reconstruction or weight adjustment on the abnormal data according to the mask rules corresponding to the dynamic mask parameter pairs of the abnormal ultra-wideband positioning data mask, and output the corrected ultra-wideband positioning data; S5. Perform error evaluation on the corrected ultra-wideband positioning data and the original ultra-wideband positioning data, update the improved self-supervised learning network model and the abnormal pattern adaptive clustering mask mechanism according to the error evaluation results, and output the corrected ultra-wideband positioning data that passes the error evaluation for downstream navigation, tracking or location service applications.

[0021] In this embodiment, S1 includes the following steps: S11. Collect the original ultra-wideband positioning data and construct an original ultra-wideband positioning dataset. Each piece of original ultra-wideband positioning data in the original ultra-wideband positioning dataset includes a timestamp, three-dimensional spatial coordinate information, and a signal quality index. The original ultra-wideband positioning dataset contains the original ultra-wideband positioning data; S12. Resample and synchronize the original ultra-wideband positioning data based on the timestamps in the original ultra-wideband positioning dataset, and then perform noise suppression processing to obtain a denoised ultra-wideband positioning dataset; The resampling synchronization in step S12 means that for the problems of non-uniform distribution of timestamps, inconsistent data intervals, and sampling frequency fluctuations that may exist in the original ultra-wideband positioning data, the implementer first sets a unified sampling time interval (for example, 100 ms), and constructs an equally spaced target time series based on the sampling period.

[0022] The system performs resampling operations on the original data according to the set time reference. For each standard time point, the system searches for several pieces of ultra-wideband positioning data within the time window before and after this time point from the original data, and selects the optimal data as the representative value based on the time proximity and signal quality index; if there is no valid data within this time period, the three-dimensional spatial coordinate value and signal quality index value at this time point are estimated through an interpolation strategy.

[0023] During the interpolation process, the system preferentially uses linear interpolation: that is, for any target time point , if If it is between two actual sampling points in the original data, the system interpolates the data at the target time point based on the three-dimensional spatial coordinates and signal quality values of these two points according to the time ratio. The interpolated coordinate points and signal quality values will be used to replace the original incomplete or missing data.

[0024] The noise suppression process cleans the mutation points and jump trajectories in the original ultra-wideband positioning data caused by factors such as signal occlusion, multipath reflection, and device errors, including sliding window median filtering and anomaly detection based on the change of positioning speed, and eliminates the abnormal points with short-term drastic changes to ensure the smoothness and continuity of the data sequence.

[0025] S13. Perform z-score standardization on the three-dimensional spatial coordinate information and signal quality indicators in the denoised ultra-wideband positioning dataset respectively, screen the standardized three-dimensional spatial coordinate information according to the three-standard-deviation principle, and screen the standardized signal quality indicators according to the signal quality indicator threshold, and only retain the data with an absolute value not exceeding 3 and a signal quality indicator not lower than the threshold to form a preprocessed ultra-wideband positioning dataset.

[0026] Perform z-score standardization processing on the data after noise suppression, calculate the mean and standard deviation of the three-dimensional spatial coordinates and signal quality indicators respectively, and normalize each piece of data so that its mean is 0 and the standard deviation is 1 to eliminate the influence of different dimensions. At the same time, combined with the three-standard-deviation principle and the signal quality threshold, initially screen the data points that deviate significantly from the normal distribution in the standardized data to construct a stable and reliable preprocessed ultra-wideband positioning dataset.

[0027] By introducing time synchronization, noise suppression, and z-score standardization processing, effectively improve the temporal consistency and feature stability of ultra-wideband positioning data, eliminate abnormal disturbance points, and significantly improve the accuracy and robustness of subsequent model training and anomaly detection.

[0028] In this embodiment, S2 includes the following steps: S21. Based on the preprocessed ultra-wideband positioning dataset, continuously sample using a dynamic sliding window method to form an input sample set, and each input sample segment in the input sample set contains continuous preprocessed ultra-wideband positioning data with a length of The sliding window length is dynamically determined according to the weighted combination calculation result of the standard deviation of the three-dimensional spatial coordinate information and the standard deviation of the signal quality indicator within the input sample segment The input sample set includes a total of M input sample segments; S22. For each input sample segment in the input sample set, calculate the local anomaly score of each data point inside the input sample segment , the local anomaly score is calculated based on the weighted sum of the change rate of the three-dimensional spatial coordinates of the data points and the change rate of the signal quality, and the data point corresponding to the position with the largest local anomaly score is selected as the th data point as the pseudo-label , and all pseudo-labels form a pseudo-label set ; S23. Pair the input sample set with the pseudo-label set one by one to form a training sample pair set , and each pair of training samples in the training sample pair set consists of an input sample segment and the corresponding pseudo-label . The training sample pair set contains a total of pairs of training samples; S24. Train the improved self-supervised learning network model based on the training sample pair set. During the training process, an anomaly sensitivity weight is assigned to each training sample pair. The anomaly sensitivity weight is dynamically calculated according to the mean of the local anomaly scores within the input sample segment. The training optimization objective is to minimize the weighted loss function, and the weighted loss function is the weighted sum of the basic loss functions between the predicted outputs and the pseudo-labels of all training sample pairs; S25. Input the preprocessed ultra-wideband positioning data set into the improved self-supervised learning network model after training and optimization, and extract the high-dimensional positioning feature vectors of each preprocessed ultra-wideband positioning data. The high-dimensional positioning feature vectors form a positioning feature vector set .

[0029] By constructing training sample pairs composed of input sample segments and pseudo-labels, and combining the self-supervised learning mechanism to train the improved self-supervised learning network model, a feature extraction method without manual annotation is effectively realized. In the process of positioning feature learning, this method introduces a sliding window modeling and a center point prediction mechanism, enabling the model to capture the temporal change law and spatial distribution characteristics of the trajectory, so as to extract more representative high-dimensional positioning feature vectors. Compared with the traditional supervised learning method relying on manual labels, this method significantly reduces the data annotation cost, and improves the sensitivity and generalization ability of the model to various types of anomalies, providing a more accurate feature basis for subsequent anomaly clustering and mask generation.

[0030] In this embodiment, S24 includes the following steps: S241. For the training sample pair set, calculate the local anomaly score according to each preprocessed ultra-wideband positioning data within each input sample segment​​ , where ; When the method of the present invention processes each piece of fragment data, it does not directly judge whether the whole segment is abnormal, but evaluates point by point which piece of data deviates from the preset data. The rule for judging the deviation from the preset data is to see whether it deviates from the path, changes suddenly, or has a very poor signal. The rules are as follows: Trajectory smoothing deviation term: Calculate the linear interpolation deviation between the current point and its adjacent points to reflect whether there is a sudden change in the trajectory; Spatial local density deviation term: Judge whether it deviates from the cluster center or the trajectory path according to the distance difference degree between the current point and the neighboring points; Signal quality change term: Compare the signal quality of the current point with that of its previous and next points. If there is an abnormal decrease, the score will increase.

[0031] S242. Based on the local anomaly score For each input sample segment Calculate the anomaly sensitivity weight , and the anomaly sensitivity weight is defined as the mean value of all local anomaly scores within the input sample segment: ; Where is the is the number of preprocessed ultra-wideband positioning data included in each input sample segment, is the th local anomaly score of the th preprocessed ultra-wideband positioning data in the

[0032] S243. Construct a weighted loss function , and the weighted loss function is the weighted sum of the basic loss function between the predicted output and the pseudo-label for all training samples:

[0033] Where represents the predicted output of the improved self-supervised learning network model for the input sample segment , represents the pseudo-label data corresponding to the input sample segment , represents the basic loss function; In S243, based on the anomaly sensitivity weights calculated in S242, the system weights the loss values of each training sample pair to construct a weighted loss function. Through this weighting mechanism, the training sample segments containing more anomaly information have a greater influence during the training process, thereby guiding the self-supervised learning network model to prioritize optimizing the perception and expression capabilities of anomaly features, and enhancing the model's recognition and adaptive capabilities for complex anomaly patterns.

[0034] S244. Using the weighted loss function as the optimization objective, the gradient descent method is adopted to iteratively update the parameters of the improved self-supervised learning network model, complete the training process, and obtain the optimized improved self-supervised learning network model. 。

[0035] During the network training process, an anomaly feature suppression mechanism is further introduced. For the slight anomaly patterns that may be contained in the training sample pairs, the contribution of the anomaly region in the loss calculation is weakened through an adaptive masking method, thereby avoiding the interference of anomaly noise on the network learning process and ensuring that the extracted localization feature vectors can more accurately represent the normal trajectory change law and local structure characteristics of the localization data. After the above training optimization, the improved self-supervised learning network model can automatically extract a set of high-dimensional, strongly anomaly-sensitive, and excellent dynamic adaptation ability localization feature vectors from the preprocessed ultra-wideband localization data after standardization, providing a reliable data basis for anomaly pattern adaptive clustering and dynamic mask parameter pair generation.

[0036] The present invention proposes an adaptive spectral clustering algorithm that fuses spatial relationships and signal quality. By constructing a weighted adjacency matrix, the three-dimensional spatial coordinate characteristics and signal quality indicators of ultra-wideband localization data are jointly encoded into the clustering process, and the local bandwidth parameter is adaptively adjusted to ensure that the clustering can fully reflect the spatial local anomaly density and signal degradation characteristics. Compared with the traditional clustering method based on a single criterion of feature vector similarity, it can more accurately depict the spatial distribution and signal decline trend of different types of anomaly patterns, improve the accuracy of anomaly pattern clusters, and greatly enhance the refinement and adaptability of anomaly pattern recognition.

[0037] In this embodiment, S3 includes the following steps: S31. Jointly process each high-dimensional localization feature vector in the localization feature vector set with its corresponding three-dimensional spatial coordinate vector and standardized signal quality indicator to construct a weighted adjacency matrix. The weighted adjacency matrix is calculated based on the three-dimensional spatial coordinate distance and local bandwidth parameter between two preprocessed ultra-wideband localization data. The local bandwidth parameter is obtained through a proportional mapping function according to the neighborhood coordinate variance centered on the preprocessed ultra-wideband localization data. The localization feature vectors are clustered using the adaptive spectral clustering method based on the weighted adjacency matrix to form a set of anomaly pattern clusters. , in the abnormal mode cluster set represents the - th abnormal mode cluster; The local bandwidth parameter is used to control the local neighborhood perception range of each data point when constructing the weighted adjacency matrix, which determines the sensitivity of the data point to neighboring samples. In the present invention, the local bandwidth parameter is centered on the spatial coordinates of the i - th pre - processed ultra - wideband positioning data, and neighboring points within a fixed number or radius range are selected. The local density feature is reflected by calculating the coordinate variance of these points. The calculation method is as follows: taking the i - th data as the center, counting the three - dimensional coordinate variance of its neighborhood points, and mapping it into a bandwidth value through a proportional function.

[0038] The classification of positioning feature vectors means that based on the local structural similarity between samples reflected by the weighted adjacency matrix, an adaptive spectral clustering algorithm is used to divide all feature vectors into several classes, and each class is a potential abnormal mode cluster. Each clustering result contains positioning segments with similar abnormal behaviors, and the corresponding data is assigned to the same abnormal mode cluster.

[0039] S32. For each abnormal mode cluster, based on the spatial coordinate vector of each pre - processed ultra - wideband positioning data in the cluster and its neighboring data, a linear prediction is performed to obtain a reference coordinate vector, and the Euclidean distance between the actual coordinate vector of the - th data and the reference coordinate vector is calculated, and the trajectory distortion factor is obtained by combining the standardized signal quality index , and the trajectory distortion factor reflects the smoothness of the data trajectory and the degree of influence by noise within the abnormal mode cluster ; Linear prediction means that for each pre - processed ultra - wideband positioning data, within the abnormal mode cluster where it is located, based on the three - dimensional coordinates of its adjacent data points before and after, a linear interpolation method is used to predict the reference coordinate of the current position. The specific method is as follows: if the data is a point to be predicted, take the coordinates of its previous and next normal data points, calculate their coordinate mean value, and use it as the reference coordinate vector. The reference value represents the position where the point should appear if the trajectory develops smoothly, and is used to evaluate the deviation degree between the current data point and its normal trajectory.

[0040] S33. For each abnormal mode cluster , based on the three - dimensional spatial coordinate vectors of all pre - processed ultra - wideband positioning data in the cluster, calculate the covariance matrix of the abnormal mode cluster , and define the square root of the trace of the covariance matrix as the spatial dispersion , and the spatial dispersion is used to measure the spatial diffusion degree of the data within the abnormal mode cluster; For each abnormal pattern cluster, all three-dimensional space coordinate vectors in the cluster are formed into a data set, and the corresponding spatial covariance matrix of the cluster is constructed to reflect the diffusion characteristics of the cluster in spatial distribution. The covariance matrix is a 3x3 matrix, representing the joint variation of the three-dimensional coordinates in each dimension. Further calculate the trace of this matrix (i.e., the sum of the diagonals) and take the square root to obtain the spatial dispersion, which is used to describe the spatial instability degree of this abnormal pattern. The larger the spatial dispersion, the more discrete the positioning data of this abnormal pattern cluster is distributed in space, and the greater the error fluctuation.

[0041] S34. According to each abnormal pattern cluster Generate a dynamic mask parameter pair based on the trajectory distortion factor and the spatial dispersion of the cluster, including the mask coverage ratio and the mask intensity weight , the mask coverage ratio is obtained by proportionally scaling the ratio of the sum of the trajectory distortion factor and the spatial dispersion. The mask intensity weight is obtained by mapping through the Sigmoid function based on the relationship between the average normalized signal quality index within the abnormal pattern cluster and the signal quality index threshold. The dynamic mask parameter pair reflects the influence degree of each type of abnormal pattern on the subsequent mask processing; In S34, the trajectory perturbation factor and the spatial dispersion are jointly used to generate the dynamic mask parameter pair. The trajectory perturbation factor reflects the degree of damage to the temporal continuity of the abnormality, and the spatial dispersion reflects the degree of spatial fluctuation of the abnormality. The two constitute the core control factors of the abnormal pattern mask rule.

[0042] S35. Based on the dynamic mask parameter pair of each abnormal pattern cluster , construct the corresponding dynamic mask rule function. The dynamic mask rule function controls the proportion of the masked feature dimensions with the mask coverage ratio and controls the substitution method of the masked features with the mask intensity weight. Combine the time decay parameter to construct a mask decay function based on the time distance, and give a small mask influence to the data samples far from the current moment to form a time-sensitive mask rule.

[0043] The present invention proposes a dual-index dynamic mask parameter pair generation strategy for the trajectory distortion factor and the spatial dispersion. By simultaneously quantifying the smoothness and spatial diffusion characteristics of the positioning trajectory, the mask coverage ratio and the mask intensity weight are dynamically generated to achieve differential mask applications for different abnormal pattern categories. The dual-index mechanism significantly improves the sensitivity of abnormal detection and the accuracy of correction. In a complex dynamic environment, the overall positioning error mean of the corrected ultra-wideband positioning data is reduced, and it shows higher stability and robustness in a multi-source interference environment.

[0044] In this embodiment, S4 includes the following steps: S41. For each piece of real-time ultra-wideband positioning data, perform denoising processing and normalization processing in sequence according to the same processing flow as the preprocessed ultra-wideband positioning data to obtain normalized real-time ultra-wideband positioning data; S42. Input the normalized real-time ultra-wideband positioning data into the optimized improved self-supervised learning network model to extract the corresponding positioning feature vectors, assign each real-time positioning feature vector to the anomaly pattern cluster with the closest distance, and obtain the dynamic mask parameter pairs of the corresponding anomaly pattern cluster according to the anomaly pattern cluster-mask mapping relationship, including the mask coverage ratio and the mask intensity weight; S43. According to the dynamic mask parameter pairs of the corresponding anomaly pattern cluster, combine the inference of the current time and the time stamp of the real-time ultra-wideband positioning data to calculate the anomaly detection mask response value, and the anomaly detection mask response value represents the response of the anomaly degree of the real-time ultra-wideband positioning data; S44. According to the set anomaly determination threshold, determine whether the anomaly detection mask response value is greater than or equal to the anomaly determination threshold. If the condition is satisfied, mark the corresponding real-time ultra-wideband positioning data as anomaly data and record it in the anomaly ultra-wideband positioning data mask set; S45. For the real-time ultra-wideband positioning data marked as anomaly data, perform local reconstruction or weight adjustment processing according to the corresponding dynamic mask parameter pair rules. For local reconstruction, linear interpolation is performed using the three-dimensional space coordinate vectors of the nearest normal data before and after to generate a corrected three-dimensional space coordinate vector, which is used to replace the spatial position information of the anomaly data. The real-time ultra-wideband positioning data after completing local reconstruction or weight adjustment processing is summarized to generate a corrected ultra-wideband positioning data set.

[0045] In the anomaly correction process of the present invention, an anomaly detection mask response mechanism is introduced, combined with the time decay characteristic of the inference time and the data time stamp, to achieve dynamic time series weighting of anomaly detection decisions, effectively avoiding false detection and missed detection problems caused by single-point anomalies. By introducing time decay to control the anomaly determination threshold, the time coherence of anomaly detection is improved, and the trajectory continuity index after correction is significantly better than the prior art, effectively ensuring the data stability and application reliability of downstream navigation, tracking, and location service systems.

[0046] In this embodiment, S5 includes the following steps: S51. Match the corrected ultra-wideband positioning data set with the original ultra-wideband positioning data set within the corresponding time period; S52. Calculate the positioning error according to the difference in three-dimensional space coordinate information between the corrected ultra-wideband positioning data and the original ultra-wideband positioning data, and define the positioning error of each piece of data as the Euclidean distance between the corrected coordinate and the original coordinate; The calculation of the positioning error in S52 is achieved by comparing the three-dimensional space coordinates of the corrected ultra-wideband positioning data and the original ultra-wideband positioning data at the same timestamp. The system first matches the two types of data in the time dimension to ensure that each corrected data corresponds one-to-one with the corresponding original data. Subsequently, the system extracts the three-dimensional space coordinates of the two data, calculates the coordinate differences in the x, y, and z directions between them, squares, sums, and then takes the square root of the coordinate differences to obtain the Euclidean space distance of the data point, which is used as the positioning error of the point. After summarizing the positioning errors of all data points, the system calculates their average value and standard deviation, which are used to comprehensively evaluate the correction effect and serve as the feedback basis for model optimization.

[0047] S53. Statistically analyze all positioning errors to calculate the mean value of the overall correction error and the standard deviation of the correction error to form the error evaluation result; S54. Compare the mean value of the overall correction error with the preset error threshold . If the mean value of the overall correction error is less than or equal to the error threshold , it is determined that the correction effect meets the requirements of downstream applications. Otherwise, incrementally fine-tune the improved self-supervised learning network model and the anomaly pattern adaptive clustering mask mechanism based on the error evaluation result; S55. During the fine-tuning process, re-weight the training samples or adjust the anomaly pattern cluster mask parameters according to the error distribution characteristics, so that the improved self-supervised learning network model and the anomaly pattern adaptive clustering mask mechanism can better adapt to the change trend of the actual ultra-wideband positioning data; S56. Output the corrected ultra-wideband positioning data that has passed the error evaluation, and define the reference rules for downstream application scenarios based on the spatial trajectory characteristics and stability of the corrected ultra-wideband positioning data; S57. Classify and organize the corrected ultra-wideband positioning data that meets the reference rules of each downstream application scenario, and provide them for the navigation, tracking, and location service modules to call and use respectively.

[0048] In this embodiment, the reference rules for downstream application scenarios are as follows: Navigation data: The trajectory of the corrected ultra-wideband positioning data is continuous and the mean value of the positioning error is less than the preset navigation accuracy requirement; Tracking data: The trajectory of the corrected ultra-wideband positioning data changes smoothly, and the magnitude of the single-step position change is within the allowable tracking speed range; Location service data: The spatial distribution density of the corrected ultra-wideband positioning data meets the coverage integrity standard of the set service area.

[0049] After the ultra-wideband positioning abnormal data is corrected, the present invention introduces the reference rules of downstream application scenarios, which can automatically match different application requirements according to the trajectory continuity, positioning accuracy, and spatial stability of the corrected data, and achieve intelligent adaptation in multiple scenarios. By constructing the determination rules for application scenarios such as navigation, tracking, and location services, the system can accurately judge its applicable range according to the error level, trajectory change characteristics, and data density of the corrected data, effectively avoiding the risk of blindly applying the data after abnormal correction to a high-precision system.

[0050] Compared with the traditional processing method that only outputs the correction result without distinguishing the usage scenario, the present invention can automatically classify and manage the corrected data, provide differentiated data support for different scenarios, improve the reliability and efficiency of data usage, significantly reduce the manual judgment cost, and enhance the automatic decision-making ability and practicality of the system.

[0051] Example 1: In the actual application process, an ultra-wideband positioning system is deployed in a certain warehousing area to monitor the real-time positions of multiple automatic guided vehicles. The system continuously collects a large amount of ultra-wideband positioning data, and each piece of data includes a timestamp, three-dimensional spatial coordinates, and a signal quality index. After the positioning system operates for a period of time, the implementer notices that there are breakpoints, drifts, and jumps in the trajectories of some robots, which affects the normal operation.

[0052] During a peak operation period of a day, the system detects a set of continuous data records, and the positioning data of a specific robot numbered AGV_23 appears abnormal at multiple time points. In the example, during the period from 10:15:23 to 10:15:30, the spatial coordinates (x, y, z) of AGV_23 suddenly deviate from the normal driving route, and the deviation distance reaches 1.2 meters, and the signal quality index drops to 65% of the historical average level during this period.

[0053] In response to the above situation, the implementer first performs noise removal processing on this batch of original ultra-wideband positioning data, removes obvious data jitters, and standardizes the spatial coordinates and signal quality index according to the z-score to form a preprocessed data set.

[0054] Subsequently, using the method of the present invention, a self-supervised learning task is constructed based on the preprocessed data set, pseudo-labels are automatically generated, and an improved self-supervised learning network model is trained. Taking the data of AGV_23 during the abnormal period as an example, the system extracts the feature vector of each piece of data, and at the same time combines its corresponding spatial coordinates and standardized signal quality index to construct a weighted adjacency matrix for adaptive spectral clustering.

[0055] The clustering results show that the abnormal data of AGV_23 is classified into the abnormal pattern cluster C4, and the characteristics of this cluster are short-term continuous drift accompanied by a sudden drop in signal quality. The system generates a dynamic mask parameter pair for C4. The mask coverage ratio is 0.85, and the mask intensity weight is 0.92, indicating that the degree of abnormality of this pattern is relatively high and should be corrected preferentially.

[0056] Entering the real-time data processing stage, the system extracts the feature vector of the new positioning data of AGV_23 in real time, and calculates the abnormal detection mask response value based on the inference time and the data timestamp. In the embodiment, at the moment of 10:15:26, the data mask response value of AGV_23 is calculated to be 0.88, which exceeds the abnormal determination threshold of 0.7 set by the system, so it is marked as abnormal data in real time.

[0057] The system corrects this data point in a local reconstruction manner according to the abnormal mark. Specifically, the normal positioning data at two time points of 10:15:25 and 10:15:27 are selected, and the mean value of their three-dimensional space coordinates is calculated as the corrected coordinate value to replace the original space coordinates of the abnormal data. After correction, the trajectory of AGV_23 resumes continuity without obvious mutation.

[0058] During this abnormal correction process, the system detected a total of 23 pieces of abnormal data, all of which were successfully corrected. To verify the effect, the implementer conducted a comparative analysis of the trajectories before and after correction, and the statistical results are as follows: The average positioning error of AGV_23 before correction was 0.38 meters, and it decreased to 0.21 meters after correction; The trajectory continuity index before correction was 0.76, and it increased to 0.89 after correction; The recall rate of abnormal detection reached 95.3%, which was much higher than 83.7% of the traditional static threshold method.

[0059] For further comparison, in this embodiment, the same data set is used, and the traditional static threshold method and the method of the present invention are respectively used for abnormal detection and correction, and the following comparison data is obtained (taking 1000 test data as an example): Table 1 Comparison data between the method of the present invention and the static threshold screening method

[0060] Furthermore, when tracking the data trajectory of AGV_23 continuously driving for 5 minutes, the corrected trajectory of the traditional method had 4 breakpoints, while the corrected trajectory of the method of the present invention had no breakpoints and the overall trajectory deviation was always less than 0.3 meters, effectively ensuring the continuity and stability of the robot navigation system.

[0061] In addition, the method of the present invention also performs well in terms of real-time processing efficiency. The processing delay of a single piece of data under high concurrency conditions is stably within 14 ms, fully meeting the requirements of industrial AGVs for the ultra-wideband positioning system in terms of low latency and high reliability.

[0062] As can be seen from Embodiment 1, in a complex dynamic environment, the method of the present invention can efficiently and accurately identify ultra-wideband positioning abnormal data, and intelligently correct the mechanism through an adaptive dynamic mask parameter, greatly improving the overall positioning accuracy and trajectory continuity of the system, fully verifying the feasibility and superiority of the technical solution of the present invention.

[0063] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes should be covered within the protection scope of the present invention.

Claims

1. An intelligent ultra-wideband positioning data correction method, characterized in that, It includes the following steps: S1. Collect the original ultra-wideband positioning data and perform preprocessing to generate a preprocessed ultra-wideband positioning data set; S2. Train an improved self-supervised learning network model based on the preprocessed ultra-wideband positioning data set to obtain a set of positioning feature vectors; S3. Cluster the abnormal features of the preprocessed ultra-wideband positioning data set using the positioning feature vectors to obtain the results of abnormal pattern clusters, and dynamically generate mask rules, and output dynamic mask parameter pairs; S4. Input the real-time ultra-wideband positioning data into the improved self-supervised learning network model, perform anomaly detection in combination with the dynamic mask parameter pairs, generate an abnormal ultra-wideband positioning data mask and abnormal data, and perform local reconstruction or weight adjustment on the abnormal data according to the mask rules corresponding to the dynamic mask parameter pairs of the abnormal ultra-wideband positioning data mask, and output the corrected ultra-wideband positioning data; S5. Evaluate the error between the corrected ultra-wideband positioning data and the original ultra-wideband positioning data, update the improved self-supervised learning network model and the abnormal pattern adaptive clustering mask mechanism according to the error evaluation results, and output the corrected ultra-wideband positioning data that has passed the error evaluation for downstream navigation, tracking or location service applications.

2. The intelligent ultra-wideband positioning data correction method according to claim 1, wherein, The S1 includes the following steps: S11. Collect the original ultra-wideband positioning data and construct an original ultra-wideband positioning data set; S12. Resample and synchronize the original ultra-wideband positioning data based on the timestamps in the original ultra-wideband positioning data set, and then perform noise suppression processing to obtain a denoised ultra-wideband positioning data set; S13. Perform z-score standardization on the three-dimensional spatial coordinate information and signal quality indicators in the denoised ultra-wideband positioning data set respectively, screen the standardized three-dimensional spatial coordinate information according to the three-standard deviation principle, and screen the standardized signal quality indicators according to the signal quality indicator threshold to form a preprocessed ultra-wideband positioning data set.

3. An intelligent ultra-wideband positioning data correction method according to claim 2, characterized in that, The S2 includes the following steps: S21. Based on the preprocessed ultra-wideband positioning dataset, continuously sample in a dynamic sliding window manner to form an input sample set, where each input sample segment in the input sample set contains continuous preprocessed ultra-wideband positioning data with a length of , and the sliding window length is dynamically determined according to the weighted combination calculation result of the standard deviation of the three-dimensional space coordinate information and the standard deviation of the signal quality index within the input sample segment . The input sample set includes a total of M input sample segments; S22. For each input sample segment in the input sample set, calculate the local anomaly scores of each data point within the input sample segment , where the local anomaly scores are calculated based on the weighted sum of the change rate of the three-dimensional spatial coordinates of the data points and the change rate of the signal quality. Select the data point corresponding to the th position with the largest local anomaly score as the pseudo label . All pseudo labels form a pseudo label set ; S23. Pair the input sample set with the pseudo-label set one by one to form a training sample pair set , and each pair of training samples in the training sample pair set consists of an input sample segment and the corresponding pseudo-label . The total number of training sample pairs in the training sample pair set is pairs; S24. Improve the self-supervised learning network model based on the set of training sample pairs for training, and assign an anomaly sensitivity weight to each training sample pair during the training process The anomaly sensitivity weight is dynamically calculated according to the mean of the local anomaly scores within the input sample segment . The training optimization objective is to minimize the weighted loss function, and the weighted loss function is the weighted sum of the basic loss functions between the predicted outputs and the pseudo-labels for all training sample pairs; S25. Input the preprocessed ultra-wideband positioning dataset into the improved self-supervised learning network model optimized through training , and extract each piece of preprocessed ultra-wideband positioning data 's high-dimensional positioning feature vector . The high-dimensional positioning feature vectors form a positioning feature vector set .

4. The intelligent ultra-wideband positioning data correction method according to claim 3, characterized in that The S24 includes the following steps: S241. For the set of training sample pairs, calculate the local anomaly score according to each piece of preprocessed ultra-wideband positioning data within each input sample segment ; S242. Based on local anomaly scores For each input sample segment Calculate the anomaly sensitivity weight , where the anomaly sensitivity weight is defined as the mean of all local anomaly scores within the input sample segment; S243. Construct a weighted loss function , where the weighted loss function is the weighted sum of the basic loss functions between the predicted output and the pseudo-labels for all training samples ; S244. Using the weighted loss function as the optimization objective, the gradient descent method is adopted to iteratively update the parameters of the improved self-supervised learning network model, complete the training process, and obtain the optimized improved self-supervised learning network model .

5. A method for correcting intelligent ultra-wideband positioning data according to claim 4, characterized in that The S3 includes the following steps: S31. Jointly process each high-dimensional positioning feature vector in the positioning feature vector set with its corresponding three-dimensional space coordinate vector and the standardized signal quality index to construct a weighted adjacency matrix. The weighted adjacency matrix is calculated based on the three-dimensional space coordinate distance and the local bandwidth parameter between two preprocessed ultra-wideband positioning data. The local bandwidth parameter is obtained through a proportional mapping function according to the neighborhood coordinate variance centered on the preprocessed ultra-wideband positioning data. Use the adaptive spectral clustering method to cluster the positioning feature vectors based on the weighted adjacency matrix to form a set of abnormal pattern clusters , in the set of abnormal pattern clusters represents the th abnormal pattern cluster; S32. For each abnormal pattern cluster, based on the spatial coordinate vectors of each piece of preprocessed ultra-wideband positioning data within the cluster and its neighboring data, perform linear prediction to obtain a reference coordinate vector, calculate the Euclidean distance between the actual coordinate vector of the th piece of data and the reference coordinate vector, and combine the standardized signal quality index to obtain a trajectory distortion factor . The trajectory distortion factor reflects the smoothness of the data trajectory within the abnormal pattern cluster and the degree of influence by noise; S33. For each abnormal pattern cluster , calculate the covariance matrix of the abnormal pattern cluster based on the three-dimensional space coordinate vectors of all preprocessed ultra-wideband positioning data within the cluster , and define the square root of the trace of the covariance matrix as the spatial dispersion , where the spatial dispersion is used to measure the spatial diffusion degree of the data within the abnormal pattern cluster; S34. Generate a pair of dynamic mask parameters based on the trajectory distortion factor and spatial dispersion of each abnormal mode cluster , including the mask coverage ratio and the mask intensity weight . The mask coverage ratio is obtained by proportionally scaling the ratio of the sum of the trajectory distortion factor and spatial dispersion. The mask intensity weight is obtained by mapping through the Sigmoid function based on the relationship between the average normalized signal quality index within the abnormal mode cluster and the signal quality index threshold. The pair of dynamic mask parameters reflects the influence degree of each type of abnormal mode on the subsequent mask processing; S35. Based on each abnormal pattern cluster of dynamic mask parameter pairs , construct the corresponding dynamic mask rule function. The dynamic mask rule function controls the proportion of masked feature dimensions with the mask coverage ratio, controls the replacement method of masked features with the mask intensity weight, and combines the time decay parameter to construct a mask decay function based on time distance, assign a small mask influence to data samples far from the current moment, and form a time-sensitive mask rule.

6. The intelligent ultra-wideband positioning data correction method according to claim 5, wherein The S4 includes the following steps: S41. For each real-time ultra-wideband positioning data, perform denoising processing and standardization processing in sequence according to the same processing flow as the preprocessed ultra-wideband positioning data to obtain standardized real-time ultra-wideband positioning data; S42. Input the standardized real-time ultra-wideband positioning data into the optimized improved self-supervised learning network model, extract the corresponding real-time positioning feature vectors, assign each real-time positioning feature vector to the nearest abnormal pattern cluster, and obtain the dynamic mask parameter pairs of the corresponding abnormal pattern cluster according to the abnormal pattern cluster-mask mapping relationship, including the mask coverage ratio and the mask intensity weight; S43. According to the dynamic mask parameter pairs of the corresponding abnormal pattern cluster, combine the inference of the current moment and the timestamp of the real-time ultra-wideband positioning data, and calculate the anomaly detection mask response value, which represents the response of the abnormal degree of the real-time ultra-wideband positioning data; S44. According to the set anomaly determination threshold, judge whether the anomaly detection mask response value is greater than or equal to the anomaly determination threshold. If the condition is met, mark the corresponding real-time ultra-wideband positioning data as abnormal data and record it in the abnormal ultra-wideband positioning data mask set; S45. For the real-time ultra-wideband positioning data marked as abnormal data, according to the corresponding masking rules, local reconstruction or weight adjustment processing is carried out. For local reconstruction, linear interpolation is performed using the three-dimensional space coordinate vectors of the nearest normal data before and after, to generate a corrected three-dimensional space coordinate vector, which is used to replace the spatial position information of the abnormal data. The real-time ultra-wideband positioning data after local reconstruction or weight adjustment processing is summarized to generate a corrected ultra-wideband positioning data set.

7. A method for correcting intelligent ultra-wideband positioning data according to claim 6, characterized in that, The S5 includes the following steps: S51. Corresponding matching is performed between the corrected ultra-wideband positioning data set and the original ultra-wideband positioning data set within the corresponding time period; S52. Calculate the positioning error based on the difference in the three-dimensional spatial coordinate information between the corrected ultra-wideband positioning data and the original ultra-wideband positioning data, and define the positioning error for each piece of data. It is the Euclidean distance between the corrected coordinate and the original coordinate. S53. Statistically analyze all positioning errors and calculate the mean of the overall correction error and the standard deviation of the correction error to form the error evaluation result; S54. Calculate the mean of the overall correction error and compare it with the preset error threshold . If the mean of the overall correction error is less than or equal to the error threshold , it is determined that the correction effect meets the requirements of downstream applications; otherwise, the improved self-supervised learning network model and the anomaly pattern adaptive clustering mask mechanism are incrementally fine-tuned based on the error evaluation results. S55. During the fine-tuning process, the training samples are re-weighted or the abnormal pattern cluster mask parameters are adjusted according to the error distribution characteristics, so that the improved self-supervised learning network model and the abnormal pattern adaptive clustering mask mechanism can better adapt to the actual change trend of the ultra-wideband positioning data; S56. Output the corrected ultra-wideband positioning data that has passed the error evaluation, and define the reference rules for downstream application scenarios based on the spatial trajectory characteristics and stability of the corrected ultra-wideband positioning data; S57. Classify and organize the corrected ultra-wideband positioning data that meets the reference rules for each downstream application scenario, and provide them for the navigation, tracking, and location service modules to call and use respectively.

8. A method for correcting intelligent ultra-wideband positioning data according to claim 7, characterized in that, The reference rules for the downstream application scenarios are as follows: Navigation data: The trajectory of the corrected ultra-wideband positioning data is continuous and the mean value of the positioning error is less than the preset navigation accuracy requirement; Tracking data: The trajectory of the corrected ultra-wideband positioning data changes smoothly, and the amplitude of the single-step position change is within the allowable tracking speed range; Location service data: The spatial distribution density of the corrected ultra-wideband positioning data meets the coverage integrity standard of the set service area.

Images