A Dataset Quality Analysis and Evaluation Method

Abstract

This invention belongs to the field of data quality analysis, specifically disclosing a dataset quality analysis and evaluation method. It first performs an integrity assessment on the current dataset, initiating data confidence analysis only after the number of data groups meets the required standard. This ensures that the confidence assessment is based on structurally complete data, avoiding misjudgments due to missing data and improving analysis reliability. Simultaneously, during data confidence analysis, it constructs adjacent datasets for each data group, implementing dynamic comparison between the current data and its local temporal neighborhood. Furthermore, it combines historical reference datasets for longitudinal comparison with historical fluctuation ranges, integrating local consistency and historical comparability for dual verification. This significantly improves the sensitivity and accuracy of anomaly detection. Finally, when determining usability based on dataset integrity and confidence assessment results, a closed-loop feedback path is established with the acquisition end, enabling adaptive adjustment of the acquisition strategy based on risk location results.

Description

Technical Field

[0001] This invention belongs to the field of data quality analysis, and specifically discloses a method for analyzing and evaluating the quality of datasets. Background Technology

[0002] In today's digital age, data has become a core basis for business processes and management decisions, widely used across various fields. Data sources have shifted from traditional manual recording to real-time acquisition by sensors, intelligent instruments, and other devices. However, even data acquired by high-precision instruments can be affected by various factors. Furthermore, modern data acquisition systems typically acquire multiple sets of time-series data continuously and in batches. In this context, an anomaly in a single data point can trigger a chain of misjudgments. Therefore, a systematic data quality assessment of the acquired dataset is essential after data acquisition is completed.

[0003] Existing technologies include technical solutions for dataset quality analysis. For example, Chinese invention patent CN116821203A proposes a dataset quality assessment method, device, and electronic device. This method involves acquiring the dataset, evaluating its quality using basic evaluation indicators, obtaining a basic evaluation result, and then, if the basic evaluation result indicates that the dataset's quality is undetermined, acquiring the dataset's data type. Finally, it performs a quality assessment on the dataset using specific evaluation indicators that match the data type, resulting in an in-depth evaluation result.

[0004] For example, Chinese invention patent CN113448955A proposes a dataset quality assessment method, apparatus, computer equipment, and storage medium. The method involves acquiring the data to be assessed from the dataset, statistically analyzing the attribute characteristics of the data to be assessed under multiple assessment dimensions, and performing a quality assessment on the data to be assessed based on the attribute characteristics under the multiple assessment dimensions to obtain the quality assessment results of the data to be assessed under the multiple assessment dimensions.

[0005] Although the two schemes mentioned above have constructed an analytical framework based on multi-dimensional indicators or hierarchical evaluation, they still have several key technical limitations: First, they have not conducted a systematic analysis of the structural integrity of the dataset. Data integrity is a prerequisite for quality assessment. Its absence will lead to subsequent analysis being based on incomplete samples, which is prone to evaluation bias or even misjudgment. Especially in scenarios where data is lost or collection is interrupted, the evaluation results lack representativeness and credibility.

[0006] Secondly, the quality assessment of datasets mainly focuses on the static attributes of individual data points, such as distribution, type, and missing rate, without fully considering the essential characteristics of datasets as time-series continuously acquired results. Modern data acquisition typically obtains multiple sets of related data in time series form, exhibiting inherent temporal dependencies and local smoothness. However, the aforementioned methods lack modeling and analysis of the dynamic correlations of data within local time windows, failing to identify short-term abrupt changes, step drifts, or local abnormal fluctuations. This results in insensitivity to local anomalies, difficulty in distinguishing between normal fluctuations and true data degradation, and consequently, insufficient anomaly detection sensitivity.

[0007] The re-evaluation results are only used for status labeling or alarm output, without establishing a feedback path with the data acquisition end. This results in a lack of proactive intervention and optimization capabilities for the acquisition process, limiting the adaptability and robustness in complex environments. Summary of the Invention

[0008] Therefore, one objective of this application is to provide a dataset quality analysis and evaluation method that effectively solves the problems existing in the prior art by evaluating dataset quality from two aspects: data integrity and data confidence.

[0009] The objective of this invention can be achieved through the following technical solution: A dataset quality analysis and evaluation method, comprising the following steps: Data integrity evaluation: performing data group quantity statistics on the current dataset acquired by the data acquisition terminal and comparing it with a preset quantity threshold;

[0010] If the number of data sets meets the requirement, proceed with the data confidence analysis step;

[0011] If the number of data sets does not meet the standard, the distribution characteristics of the missing data sets are identified to determine whether they are continuous or intermittent missing.

[0012] When a continuity is determined to be missing, an unusable conclusion is generated directly;

[0013] When a data point is determined to be intermittently missing, a partial confidence analysis is performed based on the existing dataset.

[0014] Data confidence analysis: Perform inter-group fluctuation analysis on the complete dataset that meets the criteria, and combine it with the comparison analysis with the fluctuation range of historical data to determine whether there is a confidence risk;

[0015] Data availability decision: Based on the data integrity assessment results and data confidence analysis results, an availability judgment conclusion is generated, which includes effective availability, partial availability, or unavailability;

[0016] When a dataset is determined to be valid and usable, it is retained; when it is partially usable, the valid dataset is retained and the invalid dataset is marked; when it is unusable, the current dataset is discarded.

[0017] Data collection optimization feedback: For conclusions regarding partial availability and unavailability, a re-collection command is sent to the data collection terminal.

[0018] Combining all the above technical solutions, the positive effects of this invention are as follows: 1. This invention first performs an integrity assessment on the current dataset collected by the data acquisition terminal, and only starts data confidence analysis after the number of data groups meets the standard. This ensures that the confidence assessment is based on structurally complete data, avoids misjudgment caused by missing data, improves the reliability of analysis, reduces invalid computational overhead, and ensures the accuracy and interpretability of the quality assessment results.

[0019] 2. When performing data confidence analysis on the current dataset, this invention constructs adjacent datasets for each data group to dynamically compare the current data with the local time-series neighborhood; at the same time, it combines historical reference datasets to conduct longitudinal comparison with historical fluctuation ranges, integrating dual verification of local consistency and historical comparability, effectively identifying local anomalies such as short-term mutations and drifts, and greatly improving the sensitivity and accuracy of anomaly detection.

[0020] 3. When determining availability based on dataset integrity and confidence assessment results, this invention establishes a closed-loop feedback path with the acquisition end, which can dynamically trigger supplementary acquisition or parameter tuning instructions based on risk location results, thereby achieving adaptive adjustment of the acquisition strategy and effectively improving the closed-loop control capability of data quality and the autonomous optimization level of the system. Attached Figure Description

[0021] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0022] Figure 1 This is a diagram illustrating the implementation steps of the method of the present invention.

[0023] Figure 2 This is a flowchart illustrating the process of generating availability determination conclusions based on data integrity assessment results and data confidence analysis results in this invention.

[0024] Figure 3 This is a flowchart illustrating the implementation of the present invention where valid data groups are retained and invalid data groups are marked when some data groups are available. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] See Figure 1 As shown, this invention proposes a dataset quality analysis and evaluation method, including the following steps: Data integrity evaluation: Count the number of data groups in the current dataset obtained by the data collection terminal and compare it with a preset number threshold;

[0027] If the number of data sets meets the requirement, proceed with the data confidence analysis step;

[0028] If the number of data sets does not meet the standard, the distribution characteristics of the missing data sets are identified to determine whether they are continuous or intermittent missing.

[0029] When a continuity is determined to be missing, an unusable conclusion is generated directly;

[0030] When intermittent missing data is identified, perform partial data confidence analysis based on the existing data set;

[0031] As a preferred implementation of the above scheme, the identification of the distribution characteristics of missing data groups to determine whether they are continuous or intermittent missing is as follows: extract the identifier numbers of all data groups from the current dataset, and perform a set difference operation with the preset complete number sequence range to determine the number set of missing data groups.

[0032] It's important to understand that the data acquisition process is essentially about generating an ordered sequence of data in chronological or logical order. Each data set is assigned a unique, incrementing identifier, and ideally, the identifier sequence should be continuous and uninterrupted. When the number of data sets in a dataset falls below a preset threshold, it indicates missing data. By extracting the identifier set from the current dataset and performing a set difference operation with the theoretically complete identifier range, the missing identifier set can be accurately identified.

[0033] The set of missing data group numbers is arranged in ascending order of numerical value. Then, the ordered sequence is traversed and the missing segments are divided according to the numerical interval between adjacent numbers. If the difference between adjacent missing numbers is greater than 1, it is considered as the boundary of different missing segments. Thus, the missing number sequence is divided into several consecutive numbered sub-intervals, and each sub-interval constitutes a missing segment.

[0034] The above process of sorting the missing numbers and segmenting them based on the difference between adjacent missing numbers is essentially a decomposition of the connected components of the missing set on the integer axis. Each missing segment represents a maximum continuous integer sub-interval, reflecting the clustering of data loss in the sequence space.

[0035] Count the total number of missing segments and the length of each missing segment, and identify the characteristics of the missing segment distribution according to the following criteria:

[0036] a) Continuous missing segments: The number of missing segments is 1, and the length of the missing segment is not less than 2;

[0037] b) Intermittent deletion: The number of missing segments is not less than 2, or the number of missing segments is 1 but the length is 1.

[0038] The above-mentioned criteria for continuous missing segments are set as the length of missing segments not less than 2, which eliminates the interference of accidental single-point loss. The criteria for discontinuity are set as the number of missing segments not less than 2, which reflects the random characteristics of multiple occurrences and dispersion.

[0039] It should be noted that when the number of data groups is found to be insufficient during the integrity assessment of the current dataset, the dataset is not immediately deemed invalid. Instead, the distribution characteristics of the missing data groups are further analyzed to distinguish the causes and impacts of the missing data. Continuous missing data usually reflects systemic failures, disrupting the temporal continuity of the data and severely affecting the reliability of the analysis; such datasets should be discarded entirely. Intermittent missing data, on the other hand, is mostly caused by random disturbances such as transient interference or packet loss; directly discarding such data would lead to a waste of valuable data. By performing local confidence analysis on such datasets, local anomalies can be identified and eliminated, while reliable data segments are retained, achieving efficient utilization of data resources and quality-level hierarchical control.

[0040] Data confidence analysis: Perform inter-group volatility analysis on the complete dataset that meets the criteria, and combine it with the comparative analysis of the volatility range of historical data to determine whether there is a confidence risk.

[0041] As an optional implementation of the above scheme, the data confidence analysis process is as follows: Based on the collection timestamps of each data group in the current dataset, construct the neighborhood data group set corresponding to each data group in chronological order. That is, take the target data group as the center, select one or more adjacent data groups before and after it to form a local time series window, and form a neighborhood data group set for comparative analysis.

[0042] For each monitoring indicator, the standard deviation of its numerical sequence is calculated within the neighborhood data set and compared with a preset indicator stability threshold. If the standard deviation of a monitoring indicator in the neighborhood data set is higher than the indicator stability threshold, the arithmetic mean is used as the neighborhood reference value for that indicator; otherwise, the median is used as the neighborhood reference value for that indicator.

[0043] The stability threshold of the aforementioned indicators reflects the acceptable degree of fluctuation of the monitoring indicators within the local time-series neighborhood. It is the decision boundary for distinguishing between normal fluctuations and abnormal disturbances. Specifically, it can be calculated by using historical data under normal operating conditions to calculate the standard deviation distribution of similar indicators within the neighborhood window, and then taking its quantile as the indicator stability threshold.

[0044] It should be noted that the standard deviation of a monitoring indicator within its neighborhood dataset reflects the dispersion of its numerical sequence. The smaller the standard deviation, the more stable the fluctuation of the indicator is within a local time series, and the more concentrated the data distribution. In this case, using the arithmetic mean as the neighborhood reference value can more accurately characterize its trend center and improve the sensitivity and representativeness of the reference value. The larger the standard deviation, the more significant the fluctuation or potential outliers, and the more dispersed the data distribution. In this case, using the median as the reference value has stronger robustness, can effectively suppress the influence of outliers, avoid reference benchmark shift, and ensure the stability and reliability of subsequent comparative analysis.

[0045] For each monitoring indicator in the current data set, calculate the absolute deviation between its value and the corresponding neighborhood reference value, and calculate the ratio of the absolute deviation to the neighborhood reference value as the volatility of each indicator in each data set relative to the neighborhood data set.

[0046] Furthermore, in the time-series acquisition of the dataset, the time intervals between adjacent data groups are short, exhibiting high temporal continuity and sequence correlation. Under normal operating conditions, each data group and its neighboring data should show a smooth trend with small fluctuations. Significant fluctuations indicate a confidence risk.

[0047] Based on the environmental parameters and collection time periods collected in the current dataset, a historical matching dataset is constructed by retrieving historical data group sequences with the same environmental configuration and time period characteristics from the historical reference database.

[0048] For example, environmental parameters used in the above operations can be temperature, humidity, etc.

[0049] It's important to understand that retrieving historical reference databases based on current environmental parameters and time-period characteristics aims to construct a historical dataset that matches the current context, ensuring data comparability under consistent operating conditions. Since monitoring indicator values ​​are significantly influenced by environmental and time factors, directly using all historical data as a reference would introduce operating condition biases, leading to inaccurate anomaly detection. Context matching ensures that historical benchmarks reflect typical behavior under current conditions, improving the accuracy and reliability of anomaly detection.

[0050] The historical fluctuation range of each monitoring indicator is statistically analyzed for the historical matching dataset.

[0051] In the specific implementation of the above scheme, the fluctuation range can be defined by the mean ± k times the standard deviation in the historical matching dataset.

[0052] For each data set, the volatility of each indicator and its original value are double-verified against the allowable volatility threshold and the historical volatility range.

[0053] If the volatility of a certain indicator exceeds a preset limit or the original value of the indicator exceeds the corresponding historical fluctuation range, then the data set is determined to have data confidence risk.

[0054] The volatility limit of the monitoring indicators mentioned above reflects the maximum relative change that the indicator is allowed in the local time-series neighborhood. It is a key threshold for judging whether data mutations exceed the normal fluctuation range. The limit can be determined based on the equipment operation specifications of the acquisition terminal and the dynamic response parameters such as the maximum rate of change provided by the manufacturer to meet the actual application requirements.

[0055] This invention employs a dual perspective of neighborhood fluctuation and historical fluctuation in data confidence analysis: neighborhood fluctuation reflects the dynamic consistency of data within a short time window, based on local temporal continuity; historical fluctuation reflects the long-term variation patterns of indicators under similar operating conditions, based on global statistical characteristics. The combination of these two approaches achieves a fusion of local anomaly sensitivity and global trend robustness, improving the accuracy and reliability of data quality assessment.

[0056] The following is the procedure for partial confidence analysis of the data applied to the above scheme:

[0057] Extract continuous, complete data subsequences from all missing data groups, that is, data segments that are numbered or timestamped consecutively and are preserved between missing intervals.

[0058] Understandably, local confidence analysis targets complete data subsequences rather than those containing missing segments because neighborhood comparisons rely on neighboring data to construct local reference benchmarks. Performing this analysis within missing intervals would result in incomplete neighborhood data, distorted reference values, and compromised analysis accuracy. Furthermore, complete data subsequences possess temporal or logical continuity, accurately reflecting the system's dynamic behavior within a short time window, which is a prerequisite for assessing local consistency. More importantly, even with missing data in the overall dataset, high-quality continuous subsequences may still exist. By performing confidence assessments only on complete subsequences, reliable data segments can be identified and preserved, enabling refined utilization of available data.

[0059] For each data group in the complete data subsequence, the comparison of the neighboring data group set and the historical matching dataset are performed in the same way. If the volatility of a certain indicator corresponding to a certain data group exceeds the preset allowable threshold or the original value of the indicator exceeds the corresponding historical volatility range, the local confidence is determined to be unsuccessful; otherwise, the local confidence is determined to be successful.

[0060] The above operations enable a re-evaluation of the credibility of valid segments in incomplete datasets, avoiding the complete rejection of usability due to missing local data. The data confidence analysis process follows the same logic as the confidence analysis under the condition that the number of data groups meets the standard, ensuring the consistency of evaluation criteria and supporting the subsequent retention and utilization of high-confidence subsequences in some usable datasets, thereby improving the utilization rate of data resources.

[0061] Data availability decisions: See Figure 2 As shown, based on the data integrity assessment results and data confidence analysis results, an availability determination conclusion is generated, which includes effective availability, partial availability, or unavailability.

[0062] Preferably, the data availability decision includes the following: when the number of data groups meets the standard and there is no confidence risk, the current dataset is determined to be valid and available.

[0063] If the number of data groups meets the requirement but there is a confidence risk, extract the identifiers of all risky data groups and sort them in ascending order.

[0064] Extract the minimum and maximum numbers from the sorted risk number sequence, and define the risk span as the difference between the maximum and minimum numbers.

[0065] It should be explained that the risk span represents the span of all risk data groups on the number axis, reflecting their distribution breadth.

[0066] Confidence risk clustering is constructed based on the ratio of the number of risk data sets to the risk span.

[0067] It's important to further explain that confidence risk clustering is essentially a density indicator of risk data along the number axis, i.e., the number of risk data groups within a unit span. When risk data groups appear consecutively, their numbers are consecutive, the risk span is small, and with a fixed number, the clustering value is high, reflecting a concentrated distribution of anomalies. When risk data groups are discretely distributed, the number intervals are large, the risk span is large, and with the same number, the clustering value is low, indicating dispersed anomalies. This indicator effectively distinguishes the spatial distribution characteristics of anomalies.

[0068] The confidence risk percentage is calculated based on the number of risky data groups and the total number of data groups in the current dataset.

[0069] The aforementioned proportion of confidence risk measures the relative size of risk data within the entire dataset. Even if the risk is diversified, if the overall proportion is too high, it may still affect the credibility of the overall analysis results, reflecting the quality control concept that quantitative change leads to qualitative change.

[0070] The confidence risk index is defined as the product of the confidence risk clustering degree and the confidence risk proportion. The confidence risk index is compared with the risk threshold. If the confidence risk index reaches the risk threshold, the current dataset is determined to be unusable; otherwise, the current dataset is determined to be partially usable.

[0071] It should be explained that the confidence risk index is constructed by multiplying the confidence risk clustering degree by the confidence risk proportion, aiming to comprehensively reflect the spatial concentration and overall impact scale of anomalies. This product form has a non-linear amplification effect: when the value of any one dimension is high, even if the other is moderate, it can still lead to an increase in the index; only when both are low will the index remain at a low level. This mechanism conforms to the quality judgment logic that high risk must simultaneously satisfy local concentration and global significance.

[0072] The aforementioned confidence risk threshold reflects the risk tolerance boundary for determining whether a dataset is unusable. Since the maximum value of confidence risk clustering is around 1, and the maximum value of confidence risk proportion is 1, this means that the values ​​of both are normally within a certain range. When the two are multiplied, the resulting confidence risk index has a non-linear decay characteristic. Only when both are large will the confidence risk index approach a high level. When a single indicator is too high, the product is still limited by the suppression of the other term, and the overall value is relatively moderate. Therefore, the confidence risk threshold should not be set too high, otherwise most risk scenarios will not be able to trigger the judgment, resulting in the risk of missed judgment.

[0073] If the number of data sets is insufficient and the partial confidence is passed, then the current dataset is considered partially usable.

[0074] If the number of data sets is insufficient and the local confidence test fails, the current dataset is deemed unusable.

[0075] Understandably, when generating availability assessment conclusions, this invention does not employ a simple binary judgment, such as directly classifying data as unusable or partially usable, when the number of data groups meets the standard but there is a confidence risk. Instead, it comprehensively considers the number and distribution characteristics of the risky data groups, constructing a multi-dimensional evaluation model for tiered decision-making. This mechanism can effectively distinguish between isolated random anomalies and systematic data degradation, avoiding excessive discarding of datasets due to minor local anomalies, improving data utilization efficiency, and ensuring the rigor of judgments in high-risk scenarios, thus achieving refined data availability assessment.

[0076] Furthermore, it can be understood that when generating an availability determination, if the number of data sets is insufficient but the partial confidence level is passed, it indicates that although there are missing data in the current dataset, there are consecutively numbered complete data subsequences within the missing intervals. Furthermore, verification through neighborhood and historical comparisons shows good internal data consistency and data reliability. Such subsequences contain valuable information; directly removing them entirely would waste useful data and reduce system data utilization. Therefore, they are determined to be partially usable, supporting the conditional use of reliable subsequences. Conversely, if the number of data sets is insufficient and the partial confidence level is not passed, it indicates not only structural missing data but also significant abnormal fluctuations or deviations from historical patterns detected in the existing complete subsequences. This indicates that the data quality problem combines completeness defects and content reliability failure, making it impossible to guarantee that any fragment meets basic analytical requirements. In this case, it is determined to be unusable to avoid drawing misleading conclusions based on low-quality data and to ensure the safety and reliability of data use.

[0077] When a dataset is determined to be valid and usable, it is retained. When it is partially usable, the valid dataset is retained and the invalid dataset is marked. When it is unusable, the current dataset is discarded.

[0078] For specific implementation details of the above operations, please refer to [link / reference]. Figure 3 As shown, when some data is available, the following operations are performed to retain valid data groups and mark invalid data groups: when the number of data groups meets the standard and it is determined that some data is available, risky data groups are extracted from the current dataset and removed as invalid data groups. The data groups that are retained are recorded as valid data groups.

[0079] When the number of data groups is insufficient and the dataset is deemed partially available, the data groups that did not participate in the local confidence analysis are extracted from the current dataset and removed as invalid data groups. The data groups that participated in the local confidence analysis are then considered as valid data groups.

[0080] It should be noted that when the number of data sets is insufficient but they are deemed partially usable, it indicates that no anomalies were detected in the complete data subsequence after partial confidence analysis, and these data sets can be retained as valid data sets. However, data sets that did not participate in confidence analysis are located in incomplete boundary regions with incomplete neighborhood information, making consistency verification impossible and compromising their reliability. Marking them as invalid and removing them avoids introducing unverified and uncertain data, ensuring the dual reliability of the output results in terms of temporal continuity and data credibility, and improving the accuracy and security of subsequent analyses.

[0081] Data collection optimization feedback: For conclusions regarding partial availability and unavailability, a re-collection command is sent to the data collection terminal.

[0082] To optimize the implementation of the above scheme, for some usable and unusable conclusions, a re-collection instruction is sent to the data acquisition terminal, including the following: when an unusable conclusion is generated, locate the collection time interval and associated acquisition device identifier of the data acquisition terminal when collecting the current dataset.

[0083] Based on the device identifier, retrieve the device's operation records during the aforementioned collection period from the device operation log database.

[0084] Time alignment is used to identify whether there are abnormal operations within the collection time interval from the operation records.

[0085] Specifically, actual operating parameters in the equipment's operation records, such as operating temperature, voltage, vibration amplitude, and communication status, can be compared with the normal parameter ranges determined based on the equipment model and specifications. These benchmark parameters can be extracted from the equipment's technical manual or manufacturer's specifications. If a parameter deviates from the normal operating range, it is determined to be abnormal operating behavior, and the parameter is marked as an abnormal feature.

[0086] Based on the identified anomalies, targeted instructions for adjusting sensor calibration parameters are generated.

[0087] The generated calibration parameter adjustment command is sent to the data acquisition terminal as a reacquisition command.

[0088] It should be noted that when a dataset is deemed unusable, it usually stems from missing data continuity or a loss of confidence, often caused by system-level anomalies in the data acquisition terminal. If the root cause is not identified before re-acquisition, the same fault may recur, resulting in wasted resources and compromised data quality. By analyzing equipment operating status, identifying and correcting abnormal operating parameters, and implementing targeted parameter optimization before re-acquisition, subsequent data acquisition can be based on a stable system and compliant configuration, effectively improving data integrity and reliability, preventing the recurrence of similar defects, and achieving closed-loop quality control.

[0089] Further optimization of the above scheme, for partially usable and unusable conclusions, sending a re-acquisition instruction to the data acquisition terminal also includes the following: when generating partially usable conclusions, extracting the spatial coordinates of the failed data group and the missing data group.

[0090] A supplementary data acquisition area map is generated based on the spatial location coordinates and sent to the data acquisition terminal.

[0091] Understandably, in spatial data acquisition, data quality is often closely related to its physical or logical spatial location. Missing or invalid data sets directly correspond to blind spots or low-quality areas in the space. Local data gaps can be estimated through interpolation of neighboring data, but the accuracy is limited. The most reliable completion method is to perform targeted supplementary acquisition near the missing location, driving the acquisition system to actively optimize data coverage.

[0092] It should be noted that when generating partially usable conclusions, it usually stems from intermittent data gaps or partial confidence risks. These problems are often caused by non-systematic factors such as transient interference, communication packet loss, or random environmental disturbances, reflecting sporadic anomalies during the data acquisition process. Since such faults generally do not involve equipment parameter drift or system-level failures, complex parameter tuning or mode reconstruction is unnecessary, avoiding excessive intervention that could lead to system state oscillations or configuration instability. In this case, by directly extracting the spatial coordinates of the failed and missing data groups, the data quality defect areas can be accurately located, and targeted supplementary acquisition instructions can be generated. This allows for selective re-acquisition of only the problematic areas while preserving existing high-quality data, effectively reducing redundant acquisition overhead, significantly improving data repair efficiency, optimizing resource utilization, and achieving rapid and robust data integrity recovery.

[0093] In the innovative implementation of the above solution, the collection and optimization feedback also includes equipment health correlation:

[0094] The cumulative number of re-acquisition commands triggered by the same acquisition device.

[0095] When the number of re-acquisition commands exceeds the preset alarm threshold, a device maintenance command is generated.

[0096] The preset alarm thresholds mentioned above reflect the maximum reasonable frequency at which the data acquisition device can trigger re-acquisition due to data quality issues within the statistical period. For specific settings, please refer to the device reliability indicators and maintenance cycle recommendations provided by the manufacturer.

[0097] The above operations, by associating abnormal behavior during data acquisition with equipment health monitoring, frequently trigger re-acquisition commands. This indicates that the acquisition device has failed to acquire high-quality data in multiple attempts, potentially suggesting a hardware failure or decreased environmental adaptability. Timely generation of equipment maintenance commands can prevent low-quality data from continuously flowing into the analysis system, reducing the risk of misjudgments due to erroneous data. Simultaneously, it avoids repeatedly executing invalid acquisition tasks on faulty equipment, saving manpower, time, and energy costs.

[0098] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0099] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0100] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0101] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0102] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for analyzing and evaluating the quality of a dataset, characterized in that, Includes the following steps: Data integrity assessment: Count the number of data groups in the current dataset acquired by the data acquisition terminal and compare it with a preset number threshold; If the number of data sets meets the requirement, proceed with the data confidence analysis step; If the number of data sets does not meet the standard, the distribution characteristics of the missing data sets are identified to determine whether they are continuous or intermittent missing. When a continuity is determined to be missing, an unusable conclusion is generated directly; When a data point is determined to be intermittently missing, a partial confidence analysis is performed based on the existing dataset. Data confidence analysis: Perform inter-group fluctuation analysis on the complete dataset that meets the criteria, and combine it with the comparison analysis with the fluctuation range of historical data to determine whether there is a confidence risk; Data availability decision: Based on the data integrity assessment results and data confidence analysis results, an availability judgment conclusion is generated, which includes effective availability, partial availability, or unavailability; When a dataset is determined to be valid and usable, it is retained; when it is partially usable, the valid dataset is retained and the invalid dataset is marked; when it is unusable, the current dataset is discarded. Data collection optimization feedback: Send a re-collection command to the data collection terminal based on the conclusions of partial availability and unavailability; The data confidence analysis is performed as follows: Based on the collection timestamps of each data group in the current dataset, a neighborhood data set is constructed in chronological order for each data group; for each monitoring indicator, the standard deviation of its numerical sequence is calculated within the neighborhood data set and compared with a preset indicator stability threshold. If the standard deviation of a monitoring indicator in the neighborhood data set is higher than the indicator stability threshold, the arithmetic mean is used as the neighborhood reference value for that indicator; otherwise, the median is used as the neighborhood reference value for that indicator; for each monitoring indicator in each data group, the absolute deviation between its value and the corresponding neighborhood reference value is calculated, and the ratio of the absolute deviation to the neighborhood reference value is calculated as the volatility of each data group relative to each indicator in the neighborhood data set; Based on the environmental parameters and collection period of the current dataset, historical data sets with the same environmental configuration and time period characteristics are retrieved from the historical reference database to construct a historical matching dataset. The historical fluctuation range of each monitoring indicator is calculated for the historical matching dataset. The volatility and original value of each indicator in each data set are double-verified with the allowable volatility threshold and the historical fluctuation range. If the volatility of an indicator in any data set exceeds the preset limit or the original value of the indicator exceeds the corresponding historical fluctuation range, the current dataset is determined to have confidence risk and the data set is marked as a risk data set. Otherwise, the current dataset is determined to have no confidence risk.

2. The dataset quality analysis and evaluation method as described in claim 1, characterized in that: The distribution characteristics of the identified missing data groups are determined as either continuous or intermittent missing, as detailed below: Extract the identifiers of all data groups from the current dataset, and perform a set difference operation with the preset complete number sequence range to determine the set of numbers for the missing data groups; The set of missing data group numbers is arranged in ascending order of numerical value. Then, the ordered sequence is traversed and the missing segments are divided according to the numerical interval between adjacent numbers. If the difference between adjacent missing numbers is greater than 1, it is regarded as the boundary of different missing segments. Thus, the missing number sequence is divided into several sub-intervals with consecutive numbers, and each sub-interval constitutes a missing segment. Count the total number of missing segments and the length of each missing segment, and identify the characteristics of the missing segment distribution according to the following criteria: a) Continuous missing segments: The number of missing segments is 1, and the length of the missing segment is not less than 2; b) Intermittent deletion: The number of missing segments is not less than 2, or the number of missing segments is 1 but the length is 1.

3. The dataset quality analysis and evaluation method as described in claim 2, characterized in that: The local confidence analysis is performed as follows: Extract consecutive, complete data subsequences from all missing data groups; For each data group in the complete data subsequence, the comparison of the neighboring data group set and the historical matching dataset are performed in the same way. If the volatility of a certain indicator corresponding to a certain data group exceeds the preset allowable threshold or the original value of the indicator exceeds the corresponding historical volatility range, the local confidence is determined to be unsuccessful; otherwise, the local confidence is determined to be successful.

4. The dataset quality analysis and evaluation method as described in claim 3, characterized in that: The process of generating a usability determination conclusion based on the data integrity assessment results and data confidence analysis results includes the following: The current dataset is considered valid and usable when the number of data sets meets the requirement and there is no confidence risk. If the number of data groups meets the standard but there is a confidence risk, extract the identifiers of all risky data groups and sort them in ascending order. Extract the minimum and maximum numbers from the sorted risk number sequence, and define the risk span as the difference between the maximum and minimum numbers; Confidence risk clustering is constructed based on the ratio of the number of risk data groups to the risk span; The confidence risk percentage is calculated based on the number of risk data groups and the total number of data groups in the current dataset. The confidence risk index is defined as the product of the confidence risk clustering degree and the confidence risk proportion. The confidence risk index is compared with the risk threshold. If the confidence risk index reaches the risk threshold, the current dataset is determined to be unusable; otherwise, the current dataset is determined to be partially usable.

5. The dataset quality analysis and evaluation method as described in claim 4, characterized in that: The process of generating a usability determination conclusion based on the data integrity assessment results and data confidence analysis results also includes the following: If the number of data sets is insufficient and the local confidence test is passed, then the current dataset is considered partially usable. If the number of data sets is insufficient and the local confidence test fails, the current dataset is deemed unusable.

6. The dataset quality analysis and evaluation method as described in claim 2, characterized in that: When the portion is available, retain the valid data group and mark the invalid data group as follows: When the number of data groups meets the standard and is determined to be partially available, risky data groups are extracted from the current dataset as invalid data groups and removed. The remaining data groups are recorded as valid data groups. When the number of data groups is insufficient and the dataset is deemed partially available, the data groups that did not participate in the local confidence analysis are extracted from the current dataset and removed as invalid data groups. The data groups that participated in the local confidence analysis are then considered as valid data groups.

7. The dataset quality analysis and evaluation method as described in claim 1, characterized in that: The data collection and optimization feedback includes the following: When an unusable conclusion is generated, locate the data acquisition time interval and associated acquisition device identifier when the data acquisition terminal acquires the current dataset. Based on the device identifier, retrieve the device's operation records during the aforementioned data collection period from the device operation log database; Time alignment is used to identify whether there are abnormal operations within the collection time interval from the operation records; Generate targeted sensor calibration parameter adjustment instructions based on the identified abnormal features; The generated calibration parameter adjustment command is sent to the data acquisition terminal as a reacquisition command.

8. The dataset quality analysis and evaluation method as described in claim 1, characterized in that: The data collection optimization feedback also includes the following: When generating partially usable conclusions, extract the spatial coordinates of the failed and missing data groups. A supplementary data acquisition area map is generated based on the spatial location coordinates and sent to the data acquisition terminal.

9. The dataset quality analysis and evaluation method as described in claim 1, characterized in that: The data collection and optimization feedback also includes device health association: The cumulative number of re-acquisition commands triggered by the same acquisition device; When the number of re-acquisition commands exceeds the preset alarm threshold, a device maintenance command is generated.

Images