A gas sensor array timing data validity scoring method and system

Abstract

The application discloses a kind of gas sensor array time series data validity scoring method and system, it is related to gas sensor signal processing technical field, including: combining gas sensor array time series data, sample attribute information, event marker information and environmental parameter time series data form to be scored data set, from it extract including data integrity features, signal quality features, event correlation features, data diversity features and multi-channel cooperativity features Scoring feature group, and the score of corresponding evaluation dimension is generated, the score of each evaluation dimension is weighted summation based on the most right weight determined after difference judgment to obtain comprehensive quality score, again with preset score threshold comparison to determine sample use label, it solves the problem that gas sensor array time series data lacks unified quantitative score and sample use label determining basis.

Description

Technical Field

[0001] This invention relates to the field of gas sensor signal processing technology, and in particular to a method and system for evaluating the validity of time-series data from a gas sensor array. Background Technology

[0002] Time-series data from gas sensor arrays are typically acquired by the same multi-channel gas sensor array during target gas detection. The time-series data of the gas response signals from each channel can reflect the dynamic response process of the sensor array under the influence of the target gas. This type of data has high application value in target gas identification, exhaled gas analysis, and detection model training. It usually needs to be processed in conjunction with information such as the acquisition scene, event markers, and environmental parameters, and then preprocessed, feature extracted, and sample screened to form data records that can be used for subsequent analysis.

[0003] In the process of selecting samples for time series data of gas sensor arrays, conventional methods often determine sample usability based on local data quality or single signal state, without adequately considering the comprehensive evaluation relationship between data integrity, signal quality, event correlation, data diversity, and multi-channel synergy. At the same time, the impact of acquisition scenarios and environmental parameters on the scoring and fusion process has not been processed with a unified logic, resulting in a lack of consistent quantitative determination basis for the use of labels for samples from different gas sensor array time series data. Summary of the Invention

[0004] In view of the aforementioned existing problems, the inventors have proposed this invention.

[0005] Therefore, this invention provides a method and system for evaluating the validity of time-series data from gas sensor arrays, solving the problems of quantifying the validity of time-series data from gas sensor arrays and determining the labels used for samples.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for evaluating the validity of time-series data from a gas sensor array, including: For the gas sensor array time series data newly included in the gas sensor array time series database, the sample attribute information, event marker information and environmental parameter time series data under its acquisition scenario are combined to form a dataset to be scored. The gas sensor array time series data consists of the gas response signal time series data of each channel acquired by the same multi-channel gas sensor array. The scoring feature set is extracted from the dataset to be scored, which includes data integrity features, signal quality features, event correlation features, data diversity features and multi-channel synergy features. Based on this, scores for five evaluation dimensions are generated: data integrity score, signal quality score, event correlation score, data diversity score and multi-channel synergy score. Based on the scores, dynamic adjustment factors for each evaluation dimension are determined. The initial fusion weights for each evaluation dimension are adjusted based on the dynamic adjustment factors to generate fusion weights. The fusion weights are scaled based on time-series environmental parameter data to generate weight scaling values. The final weights are determined based on the difference between the weight scaling values ​​and the initial fusion weights. The scores are then weighted and summed based on the final weights to obtain the comprehensive quality score. The overall quality score is compared with a preset score threshold to determine the labeling of the gas sensor array time series data samples.

[0007] As a preferred embodiment of the gas sensor array time-series data validity scoring method of the present invention, the sample attribute information includes the age category, gender category, and inhalation stage category of the sampled object in the collection scenario; the event annotation information includes the actual time when the sampled object inhaled the target gas in the collection scenario, and the start time when the target gas was detected in the gas sensor array time-series data; the environmental parameters include the ambient temperature and relative humidity in the collection scenario; and the timestamps of the gas response signal time-series data of each channel in the gas sensor array time-series data are aligned.

[0008] As a preferred embodiment of the gas sensor array time-series data validity scoring method of the present invention, the data integrity features include the missing rate, glitch index, and temporal continuity of the gas response signal time-series data of each channel. The calculation logic of the data integrity score is as follows: after calculating the data integrity score of the gas response signal time-series data of each channel based on the missing rate, glitch index, and temporal continuity, the average value is taken as the data integrity score of the gas sensor array time-series data in the dataset to be scored, and the data integrity score is negatively correlated with the missing rate, glitch index, and temporal continuity.

[0009] As a preferred embodiment of the gas sensor array time-series data validity scoring method of the present invention, the signal quality characteristics include the signal-to-noise ratio, baseline drift rate, and noise level of the gas response signal time-series data of each channel. The calculation logic of the signal quality score is as follows: after calculating the signal quality score of the gas response signal time-series data of each channel based on the signal-to-noise ratio, baseline drift rate, and noise level, the average value is taken as the signal quality score of the gas sensor array time-series data in the dataset to be scored. The signal quality score is positively correlated with the signal-to-noise ratio and negatively correlated with the baseline drift rate and noise level.

[0010] As a preferred embodiment of the gas sensor array time series data validity scoring method of the present invention, the event correlation features include event marker synchronization, false detection rate, and waveform matching degree between the gas sensor array time series data and the standard time series data. The calculation logic of the event correlation score is as follows: the event marker synchronization, the difference between 1 and the false detection rate, and the waveform matching degree are weighted and summed to obtain the event correlation score of the gas sensor array time series data in the dataset to be scored. The waveform matching degree calculation logic between the gas sensor array time series data and the standard time series data is as follows: For any channel gas response signal time series data in the gas sensor array time series data, calculate its DTW distance and cosine similarity of the morphological feature vectors with the gas response signal time series data of the same type in the standard time series data. Then, process and weighted sum the results to obtain the waveform matching degree of the gas response signal time series data of that channel. Take the average value of the waveform matching degree of the gas response signal time series data of each channel as the waveform matching degree between the gas sensor array time series data and the standard time series data. The morphological feature vectors include the rise slope, peak time and decay constant. The calculation logic for the false detection rate is as follows: extract the target gas detection results output by the multi-channel gas sensor array at each time stamp, and based on the target gas detection results and the actual target gas results, determine the number of timestamps in which the target gas was mistakenly identified, and use the ratio of this number to the total number of timestamps in the time series data of the gas sensor array as the false detection rate. The event tag synchronization is determined based on the time interval between the actual time and the start time, and the time interval is inversely proportional to the event tag synchronization.

[0011] As a preferred embodiment of the gas sensor array time-series data validity scoring method of the present invention, the data diversity feature includes the category proportion of gas sensor array time-series data in the dataset to be scored. The determination logic is as follows: based on sample attribute information and environmental parameter time-series data, the category to which the gas sensor array time-series data in the dataset to be scored belongs is determined as the target category. The number of gas sensor array time-series data belonging to the target category in the gas sensor array time-series database is counted, and the ratio of the target category to the total number of gas sensor array time-series data in the gas sensor array time-series database is calculated as the category proportion. Among them, there are four categories of environmental status, three categories of age of the sampling subjects, two categories of gender, and three categories of the stage of consumption; The environmental status categories include four types: abnormal temperature and humidity, abnormal temperature, abnormal humidity, and normal temperature and humidity. These categories are determined based on time-series data of environmental parameters and preset environmental temperature and relative humidity thresholds. The data diversity score is calculated based on the category percentage and is positively correlated with the category percentage, and a low percentage compensation score mechanism is introduced.

[0012] As a preferred embodiment of the gas sensor array time series data validity scoring method of the present invention, the multi-channel coordination feature includes channel synchronization and cross-interference coefficient. The calculation logic of the multi-channel coordination score is as follows: the difference between channel synchronization, 1 and cross-interference coefficient is weighted and summed to obtain the multi-channel coordination score of the gas sensor array time series data in the dataset to be scored.

[0013] As a preferred embodiment of the time-series data validity scoring method for the gas sensor array described in this invention, the scores of each evaluation dimension are normalized to obtain the dynamic adjustment factor of each evaluation dimension, and the dynamic adjustment factor of each evaluation dimension is multiplied by the corresponding initial fusion weight and then normalized to obtain the fusion weight of each evaluation dimension. The dimensional reliability parameter is generated based on the environmental parameter time series data. Specifically, the environmental state category to which the gas sensor array time series data belongs is determined based on the environmental parameter time series data. If the temperature and humidity are normal, the dimensional reliability parameter is set to 1. If the temperature or humidity is abnormal, the dimensional reliability parameter is set to a positive decimal less than 1. If the temperature and humidity are abnormal, the dimensional reliability parameter is set to a positive decimal less than 1, and is less than the dimensional reliability parameter corresponding to the environmental state category of abnormal temperature. Multiply the dimensional reliability parameter by the fusion weight of each evaluation dimension to obtain the weight scaling value of each evaluation dimension. Calculate the relative error between the weight scaling value and the initial fusion weight under each evaluation dimension and take the average value as the weight difference index. If the weight difference index is less than the preset threshold, the initial fusion weight is used as the final weight; otherwise, the weight scaling value is used as the final weight. The data integrity score, signal quality score, event relevance score, data diversity score, and multi-channel coordination score are weighted and summed according to their final weights to obtain the comprehensive quality score. The overall quality score is compared with a preset score threshold to determine the label to be used for the sample, which is a high-quality sample label, a qualified sample label, or a low-quality sample label. The output includes a comprehensive quality score, data integrity score, signal quality score, event relevance score, data diversity score, multi-channel synergy score, dimensional reliability parameter, final weights, and sample score records with sample usage labels.

[0014] As a preferred embodiment of the gas sensor array time series data validity scoring method of the present invention, wherein: for any gas sensor array time series data included in the gas sensor array time series database, the previous gas sensor array time series data included in the gas sensor array time series database is used as its reference data, and the final weight of its reference data in each evaluation dimension is used as its initial fusion weight in the corresponding evaluation dimension. For the first gas sensor array time series data included in the gas sensor array time series database, its initial fusion weights in each evaluation dimension are preset values, and satisfy the following: the initial fusion weight preset value for event correlation is the largest, the initial fusion weight preset value for signal quality is the second largest, the initial fusion weight preset value for data integrity and multi-channel synergy is the smallest, and the initial fusion weight preset value for data integrity and multi-channel synergy is the same, while the initial fusion weight preset value for data diversity is the smallest. The time series data of the gas sensor array included in the gas sensor array time series database is continuously monitored. When the latest gas sensor array time series data included in the gas sensor array time series database and the N gas sensor array time series data before it satisfy the condition that the final weight remains unchanged, the fusion weight of the gas sensor array time series data in each evaluation dimension is used as the initial fusion weight of the next gas sensor array time series data included in the gas sensor array time series database in the corresponding evaluation dimension. N is a preset quantity threshold and N is a positive integer greater than 1.

[0015] As described in this invention, a gas sensor array time-series data validity scoring system includes: The dataset to be scored module is used to form a dataset to be scored by combining the latest gas sensor array time series data included in the gas sensor array time series database with sample attribute information, event marker information and environmental parameter time series data in the acquisition scenario. The gas sensor array time series data consists of the gas response signal time series data of each channel acquired by the same multi-channel gas sensor array. The evaluation dimension score generation module is used to extract a set of scoring features from the dataset to be scored. The set of scoring features includes data integrity features, signal quality features, event correlation features, data diversity features, and multi-channel synergy features. Based on these features, scores for five evaluation dimensions are generated: data integrity score, signal quality score, event correlation score, data diversity score, and multi-channel synergy score. The comprehensive quality score generation module is used to determine the dynamic adjustment factors for each evaluation dimension based on the score, adjust the initial fusion weights of each evaluation dimension based on the dynamic adjustment factors to generate fusion weights, scale the fusion weights based on environmental parameter time series data to generate weight scaling values, determine the final weights based on the difference between the weight scaling values ​​and the initial fusion weights, and perform a weighted summation of each score based on the final weights to obtain the comprehensive quality score. The sample label generation module is used to compare the overall quality score with a preset score threshold to determine the sample labeling for the gas sensor array time series data.

[0016] The beneficial effects of this invention are as follows: by establishing a correspondence between the time-series data of the gas sensor array and the sample attribute information, event label information and environmental parameter time-series data, and generating scores from multiple evaluation dimensions such as data integrity, signal quality, event correlation, data diversity and multi-channel synergy, and then combining the final weights to obtain a comprehensive quality score, the usability of the gas sensor array time-series data can be determined in a unified quantitative way, thereby providing a consistent data evaluation basis for sample screening, model training data organization and sample label generation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the scoring method in this invention; Figure 2 This is a block diagram of the scoring system in this invention; Figure 3 This is a schematic diagram illustrating the scoring feature groups and the generation of scores for each evaluation dimension in this invention; Figure 4 This is a schematic diagram of the final weight generation in this invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Example 1, referring to Figure 1 , Figure 3 and Figure 4 This embodiment provides a method for evaluating the validity of time-series data from a gas sensor array. This method is suitable for assessing the quality of time-series data from gas sensor arrays and generating sample usage labels based on the evaluation results. The time-series data from a gas sensor array consists of time-series data of gas response signals from each channel collected by the same multi-channel gas sensor array. This embodiment uses an 8-channel gas sensor array collecting exhaled gas signals as an example. The method in this embodiment includes the following steps: S1 Multi-channel data synchronous acquisition and formation of the dataset to be scored For the latest gas sensor array time series data included in the gas sensor array time series database, the sample usage labels are not yet known. Therefore, this data is used as gas sensor array time series data to be scored to determine the sample usage labels. The data is combined with the sample attribute information, event tag information and environmental parameter time series data in the acquisition scenario to form a dataset to be scored.

[0023] The first The first channel in the The gas response signal at each sampling point is denoted as . ; in, Channel number, In this embodiment, the number of channels in the multi-channel gas sensor array is... =8, the gas response signal time series data is a data sequence formed by arranging the gas response signals from multiple sampling points in chronological order of sampling time. This is the sampling point number.

[0024] While simultaneously acquiring gas response signals from each channel, the timestamp, ambient temperature, and relative humidity corresponding to each sampling point are recorded simultaneously, and denoted as follows: ; in, Indicates the first The timestamp corresponding to each sampling point Indicates the first The ambient temperature corresponding to each sampling point Indicates the first The relative humidity corresponding to each sampling point.

[0025] The sample attribute information includes the age category, gender category, and inhalation stage category of the sampled object; the event marker information includes the actual time when the sampled object inhaled the target gas, and the start time of detection of the target gas in the time series data of the gas sensor array. In this embodiment, an 8-channel gas sensor array is used for nitrous oxide detection, so the target gas is set as nitrous oxide. The 8-channel gas sensor array is a highly integrated sensor module. It not only collects the gas response signals of each channel at each time stamp, but also analyzes the gas response signals of each channel at each time stamp through its own integrated microcontroller or digital signal processing chip to determine whether the target gas (i.e., nitrous oxide) is detected at each time stamp. The specific working principle is common knowledge to those skilled in the art and will not be described in detail here.

[0026] Establish a correspondence between the time-series data of the gas sensor array, sample attribute information, event tag information, and environmental parameter time-series data under the same acquisition scenario to form a dataset to be scored. The dataset to be scored serves as the basic data object for subsequent multi-dimensional feature extraction, five-dimensional score calculation, comprehensive quality score calculation, and sample label generation.

[0027] S2 Multidimensional Feature Extraction and Scoring The scoring feature set is extracted from the dataset to be scored. The scoring feature set includes data integrity features, signal quality features, event correlation features, data diversity features, and multi-channel synergy features. Based on these features, data integrity scores, signal quality scores, event correlation scores, data diversity scores, and multi-channel synergy scores are generated.

[0028] S2.1 Data Integrity Dimension Score Data integrity features include missing rate, glitch index, and temporal continuity; for the first... Each channel has a missing rate denoted as . The burr index is denoted as Time continuity is denoted as In this embodiment Expressed as timestamp breakpoint rate, all three are standard metrics known in the field, and their calculation methods are as follows: in, , and All are negative proportional indicators, with a value range of [0,1]. The larger the value, the more severe the data loss, glitches, or timestamp breakpoints in the corresponding channel.

[0029] When identifying spurs, first, the signal values ​​within the range of three sampling points before and after the current sampling point are taken as the neighborhood signals, and the median of these neighborhood signals is calculated. Then, the signal value of the current sampling point is compared with the median. When the absolute value of the difference between the two is greater than three times the standard deviation of the baseline segment, the sampling point is determined to be a spur. This determination relationship is expressed as follows: in, For the first The standard deviation of the baseline segment signal of each channel. The baseline segment is the stable sampling segment before the occurrence of the target gas event. The time interval of the target gas event is the time interval during which the multi-channel gas sensor array determines that the target gas has been detected. The baseline segment is taken from the first 20% of the sampling points before the occurrence of the target gas event. 20% is the preset baseline segment selection ratio in this embodiment.

[0030] After obtaining the missing rate, glitch index, and temporal continuity of each channel, the missing rate of the ch-th channel is used as the basis for further analysis. Spike Index and temporal continuity Calculate the first The data integrity score of the first channel will be used to determine the data integrity score of the second channel. The data integrity score for each channel is denoted as follows: The calculation formula is as follows: Among them, the missing rate has the greatest impact on data integrity, followed by the glitch index and temporal continuity, which have roughly the same impact. Therefore, in the calculation of data integrity score, the missing rate will be considered as a factor. The weighting coefficient is set to the maximum, while... and The weighting coefficients are set to be the same and less than 1. The weighting coefficients, as in this embodiment, are... , and The weighting coefficients are set to 0.4, 0.3, and 0.3 respectively, but other reasonable values ​​can also be used; no restrictions are imposed here. , and These are all negative indicators that are detrimental to data integrity. Higher values ​​indicate a greater degree of overall missing data, overall jagged edges, and overall timestamp breakpoints, respectively. Therefore, this embodiment first uses... , and Convert the anomaly ratio into an integrity component, and then calculate the data integrity score. .

[0031] The data integrity scores of each channel are averaged to obtain the data integrity scores of the gas sensor array time-series data in the dataset to be scored. The calculation formula is as follows: S2.2 Signal Quality Dimension Scoring Signal quality characteristics include signal-to-noise ratio, baseline drift rate, and noise level. For the first... There are 1 channel, and the signal-to-noise ratio is denoted as . The calculation formula is as follows: in, For the first Variance within the response window of each channel For the first The baseline variance of each channel is calculated based on the baseline segment signal before the occurrence of the target gas event, with the response window being the time interval corresponding to the target gas event.

[0032] The baseline drift rate is determined by the linear trend of the baseline segment signal, specifically for the first... The baseline segment signal of the nth channel is subjected to univariate linear fitting to obtain the nth channel. Baseline segment fitting slope of each channel Based on the first The maximum and minimum absolute values ​​of the baseline fitting slopes of each channel in history are subjected to max-min normalization to obtain the nth channel. Baseline drift rate of each channel The specific mathematical expression is: in, and They represent the first The maximum and minimum absolute values ​​of the baseline fitting slope for each channel in history, after max-min normalization. The value range is [0,1]. As a negative indicator of baseline drift, a larger value indicates a higher baseline drift. The more severe the baseline drift of each channel.

[0033] Noise level is used to characterize the degree of high-frequency fluctuations in the baseline signal, the first The noise level of each channel is denoted as The calculation formula is as follows: Among them, the high-frequency components are composed of the first... The time-series data of the gas response signals from each channel were obtained by extracting high-frequency components. The dynamic range was used to scale the high-frequency components, specifically the difference between the maximum and minimum values ​​of the channel range. Before participating in subsequent scoring, the data is first processed to the range [0,1] based on max-min normalization. Specifically, this can be based on the first... The maximum and minimum noise levels of each channel in history are used for max-min normalization, in the same way as described above; the relevant mathematical expressions will not be repeated here. As a negative indicator of noise, a larger value indicates the... The higher the noise level of each channel.

[0034] After obtaining the signal-to-noise ratio of each channel Baseline drift rate and noise level Then, based on the first The signal-to-noise ratio, baseline drift rate, and noise level of each channel are calculated. Signal quality score for each channel.

[0035] No. The signal quality score of each channel is denoted as follows: The calculation formula is as follows: Among these factors, the signal-to-noise ratio (SNR) has the greatest impact on signal quality, followed by the baseline drift rate, and then the noise level. Therefore, in the calculation of signal quality score, the signal-to-noise ratio (SNR) is considered the most important factor. The weighting coefficient is set to the maximum, while... The weighting coefficient is set to the second highest, and... The weighting coefficient is set to a higher value, such as in this embodiment, where... , and The weighting coefficients are set to 0.5, 0.3, and 0.2 respectively, but other reasonable values ​​can also be used; no restrictions are imposed here. and All are negative normalized indicators, therefore... and It is then converted into a positive rating component and used in the calculation.

[0036] For the signal-to-noise ratio segmental mapping function: In this embodiment, 20dB and 10dB are preset signal-to-noise ratio segment thresholds. A signal-to-noise ratio higher than 20dB is considered to have excellent signal quality, while a signal-to-noise ratio lower than 10dB is considered to have poor signal quality. 1.0 and 0.2 are the mapping values ​​for the corresponding segments, and can be adjusted by the staff according to the actual situation. When 10dB≤ When <20dB, It varies linearly between 0.2 and 1.0.

[0037] The signal quality scores of each channel are averaged to obtain the signal quality scores of the gas sensor array time-series data in the dataset to be scored. The calculation formula is as follows: S2.3 Event Relevance Dimension Scoring Event relevance features include waveform matching degree, event tag synchronization, and false detection rate. Waveform matching degree is denoted as... The response waveform is determined based on the typical response waveform of the target gas. This embodiment uses nitrous oxide detection as an example; the typical response waveform of the target gas includes a rapid rise, a plateau, and an exponential decay.

[0038] When calculating waveform matching degree, for the first... The gas response signal time-series data of each channel is dynamically time-warped (DTW) with the gas response signal time-series data of the same type of channel in the standard time-series data. The larger the DTW value, the greater the difference between the gas sensor array time-series data and the standard time-series data for that channel. The standard time-series data refers to the gas sensor array time-series data detected under typical target gas events. Typical target gas events can be selected from the most frequently occurring target gas events in history, or calibrated by relevant experts; no restrictions are imposed here. Meanwhile, from the... The rise slope, peak time, and decay constant are extracted from the gas response signal time-series data of each channel. These are well-known indicators to those skilled in the art, and the specific calculation process will not be elaborated here. These parameters constitute the morphological feature vector of the channel. The cosine similarity between the morphological feature vector of this channel and the morphological feature vector of the same type of channel in the standard time-series data is calculated. The higher the cosine similarity, the smaller the difference between the gas sensor array time-series data and the standard time-series data for that channel. The dynamic time warping distance corresponding to this channel is then subjected to maximum-minimum normalization to obtain a distance normalized value. This distance normalized value is subtracted from 1 to obtain the dynamic time warping similarity. Finally, the dynamic time warping similarity is weighted and fused with the cosine similarity of the morphological feature vector to obtain the ... The waveform matching degree of each channel is denoted as . The mathematical expression is as follows: .

[0039] In the formula, For the first The distance normalization value of the i-th channel, specifically, is the distance normalization value through the i-th channel. The maximum and minimum values ​​of the dynamic time-warped distance that have occurred historically for the i-th channel, for the i-th The dynamic time warp distances corresponding to each channel are obtained by max-min normalization. For the first The cosine similarity between the morphological feature vectors of each channel and the morphological feature vectors of the same type of channel in standard time series data, because It is a positive indicator, while Since it's a negative indicator, we'll pass it here first. In this way, The dynamic time-warped similarity, converted into a positive indicator, is then compared with... Perform weighted fusion. This refers to dynamic time warping similarity. Furthermore, both dynamic time warping similarity and cosine similarity are same-dimensional indicators reflecting waveform similarity; therefore, we set... and The weights for both waveform matching in the calculation are consistent, both being 0.5; The average waveform matching degree of each channel is taken to obtain the waveform matching degree between the gas sensor array time series data and the standard time series data. The calculation formula is as follows: Event tag synchronization is denoted as The calculation formula is as follows: in, The actual time it takes for the sampled object to inhale the target gas. This represents the start time of target gas detection in the timing data of the gas sensor array. This refers to the time interval between the two, in seconds. 0.5 is the preset synchronization time normalization coefficient in this embodiment. This is a positive synchronization indicator, with a value range of (0,1]. The smaller the time interval, the more timely the detection of the collected object, and the higher the event tag synchronization. The closer it is to 1.

[0040] False positive rate is denoted as When calculating the false detection rate, the target gas detection results output by the multi-channel gas sensor array at each time stamp are extracted. Based on the target gas detection results and the actual target gas results, the number of time stamps where the target gas was mistakenly detected is determined. The ratio of this number to the total number of time stamps in the time-series data of the gas sensor array is used as the false detection rate. The target gas detection results are based on the output of the multi-channel gas sensor array, and the actual target gas results are obtained by querying the data acquisition object or querying relevant monitoring. The calculation formula is as follows: in, It is a negative proportional indicator with a value range of [0,1]. The larger the value, the higher the proportion of timestamps that are misjudged as having the target gas. This means that the multi-channel gas sensor array has a lower detection accuracy for the target gas, and the lower the confidence level of the output gas response signals of each channel.

[0041] The event correlation score of the time series data of the gas sensor array in the dataset to be scored is denoted as: The calculation formula is as follows: Among them, waveform matching degree reflects the morphological consistency between the time-series data of the gas response signal and the typical response waveform of the target gas, and is the main basis for judging event correlation, so it is set with a large weight; event labeling synchronization reflects the consistency between the actual time of detection of the target gas and the time of target gas event labeling, and has an auxiliary role in judging event correlation, so it is set with the second highest weight; false detection rate is used to constrain the situation where the target gas is misjudged when no target gas event has occurred, and serves as a supplementary correction factor for event correlation scoring, so it is set with the lowest weight, as in this embodiment, , , The weighting coefficients are set to 0.6, 0.3, and 0.1 respectively, but other reasonable values ​​can also be used; no restrictions are imposed here. Because... The larger the value, the higher the degree of misjudgment of the event. Therefore, by... It is then converted into a positive rating component and used in the calculation.

[0042] S2.4 Data Diversity Dimension Scoring The data diversity characteristics include the proportion of categories in the time series data of the gas sensor array in the dataset to be scored, the age category, gender category and inhalation stage category in the sample attribute information, and the environmental state category determined by combining the environmental parameter time series data, so as to determine the category to which the dataset to be scored belongs, and the category to which it belongs is used as the target category.

[0043] The age categories include three categories: minors, young adults, and the elderly. If the subject is under 18 years old, the age category is set to minors; if the subject is between 18 and 60 years old, the age category is set to young adults; and if the subject is over 60 years old, the age category is set to the elderly. The gender category includes two categories: male and female. If the subject is male, the gender category is set to male; otherwise, the gender category is set to female. The inhalation stage is categorized into three types: saturation release stage, rapid decay stage, and trace residue stage. If the target gas is detected within 0-5 minutes after the subject inhales the target gas, and the concentration of nitrous oxide in the alveoli and blood is at its peak during this stage, the inhalation stage is classified as the saturation release stage. If the target gas is detected within 5-30 minutes after the subject inhales the target gas, and the subject's body rapidly detoxifies through respiratory movements during this stage, with the concentration of the target gas decreasing exponentially, the inhalation stage is classified as the rapid decay stage. If the target gas is detected within 30 minutes-1 hour after the subject inhales the target gas, and most of the nitrous oxide in the subject's body has been emptied through the lungs during this stage, with only a very small amount remaining in deep tissues or slowly precipitating from the blood into the alveoli, the inhalation stage is classified as the long-term residue stage. The environmental status categories include four types: abnormal temperature and humidity, abnormal temperature, abnormal humidity, and normal temperature and humidity. These categories are determined based on time-series environmental parameter data and preset environmental temperature and relative humidity thresholds. First, the environmental temperature and relative humidity time-series data are averaged to obtain the average environmental temperature and relative humidity values. Then, the suitable temperature and humidity ranges for the multi-channel gas sensor array are determined based on the product specifications or instruction manual. For example, the suitable temperature range is set to -10°C to 55°C, and the suitable humidity range is set to 15% to 95%. If both the average environmental temperature and average relative humidity are within their respective suitable ranges, the environmental status category is set to normal temperature and humidity. If neither is within their respective suitable ranges, the environmental status category is set to abnormal temperature and humidity. If the average environmental temperature is within the suitable temperature range but the average relative humidity is not, the environmental status is set to abnormal humidity. If the average environmental temperature is not within the suitable temperature range but the average relative humidity is within the suitable humidity range, the environmental status is set to abnormal temperature. As one implementation method, if the average ambient temperature in the collection scenario is 25 degrees Celsius, the average relative humidity is 40%, the age of the subject is 25 years old, and the gender is male, and the gas sensor array time series data is collected 20 minutes after the subject inhales the target gas, then the category of the gas sensor array time series data in the dataset to be scored is: normal temperature and humidity ∩ young adult ∩ male ∩ rapid decay period. After determining the category to which the dataset to be scored belongs, the data diversity coverage index is calculated, denoted as . The calculation formula is as follows: in, The percentage of different categories in the time-series data of the gas sensor array in the dataset to be scored. The theoretical number of categories is obtained by combining age category, gender category, stage of drug use category, and environmental state category.

[0044] In this embodiment: Wherein, 3, 2, 4, and 3 represent the preset number of age categories, gender categories, environmental state categories, and drug use stage categories in this embodiment, respectively. pass Normalization is performed; a larger value indicates that the sample is more fully covered in the theoretical class space.

[0045] For the time-series data of the gas sensor array in the dataset to be scored, if its category proportion meets the following conditions... If the score is less than 5%, it indicates that the category it belongs to is relatively scarce in the gas sensor array time series database. Therefore, a compensation score is introduced to improve its data diversity score in subsequent calculations, making it more likely to be regarded as a high-quality sample in subsequent calculations and improving the comprehensiveness of the model training data.

[0046] The data diversity score of the time series data of the gas sensor array in the dataset to be scored is denoted as: The calculation formula is as follows: 5% is the preset scarcity category judgment threshold in this embodiment. This is a preset scarcity category compensation coefficient in this embodiment, and its value is determined by the following rule: when... When the scarcity rate is less than 2%, it is considered highly scarce, and the scarcity category compensation coefficient is set to its maximum value, such as letting... =0.3, when 2%≤ When the scarcity rate is less than 5%, it is considered generally scarce, and the compensation coefficient for the scarcity category is taken as the median value, such as letting... =0.1, when When the percentage is ≥5%, it is considered not scarce, and the scarcity category compensation coefficient is set to its minimum value, such as letting... =0, where 2% is the preset threshold for judging highly scarce categories in this embodiment, and 0.3 and 0.1 are the compensation coefficients corresponding to highly scarce categories and general scarce categories, respectively.

[0047] S2.5 Multi-channel synergy dimension score The multi-channel cooperative characteristics include channel synchronization and cross-interference coefficient. The response time is the time point when the response signal rises to 80% of the peak amplitude after the channel sensor comes into contact with the target gas, in seconds. Here, 80% is a preset response time determination ratio in this embodiment.

[0048] After obtaining the response time of each channel, the standard deviation and mean of the response time of each channel are calculated, and the channel synchronicity is determined accordingly. Channel synchronicity is denoted as... The calculation formula is as follows: in, The standard deviation of the response time for each channel. This represents the average response time of each channel.

[0049] The higher the relative dispersion of response time, the greater the difference in response time between channels; after the above formula transformation, This is a positive channel synchronization indicator; the larger the value, the higher the synchronization of multi-channel response. The cross-interference coefficient was obtained by performing dimensionality reduction analysis on the channel response matrix using partial least squares regression. Specifically, the channel response matrix was formed based on the time-series data of the gas response signals for each channel, and the condition number of the regression model was calculated. Then normalize it to obtain the cross-interference normalization coefficient. : in, and These are the maximum and minimum condition numbers of sensors in the same batch under standard conditions, after normalization. The value range is [0,1]. As a negative indicator of cross-interference, the larger the value, the larger the condition number of the channel response matrix, and the higher the degree of cross-interference of the corresponding channel.

[0050] The multi-channel coherence score of the time series data of the gas sensor array in the dataset to be scored is denoted as: The calculation formula is as follows: Among them, the channel synchronization and cross-interference normalization coefficients are indicators of the same dimension reflecting the multi-channel coordination, therefore, here we set... and The weights for each channel in the multi-channel synergy score calculation are consistent, both being 0.5. A higher value indicates a higher degree of channel crosstalk, therefore... It is then converted into a positive rating component and used in the calculation.

[0051] S3 Dynamic Weight Allocation and Overall Scoring Based on the scores, dynamic adjustment factors for each evaluation dimension are determined. The initial fusion weights for each evaluation dimension are adjusted based on the dynamic adjustment factors to generate fusion weights. The fusion weights are scaled based on time-series environmental parameter data to generate weight scaling values. The final weights are determined based on the difference between the weight scaling values ​​and the initial fusion weights. The scores are then weighted and summed based on the final weights to obtain the comprehensive quality score.

[0052] After obtaining the scores for each evaluation dimension, the scores are normalized and used as dynamic adjustment factors for the corresponding evaluation dimensions. in, , , , and These are dynamic adjustment factors for the time series data of the gas sensor array in the dataset to be evaluated, in terms of data integrity, signal quality, event correlation, data diversity, and multi-channel coordination. Data integrity, signal quality, event correlation, data diversity, and multi-channel coordination are all evaluation dimensions. The formulas for calculating the fusion weights of each evaluation dimension are as follows: In the formula, For the time-series data of the gas sensor array in the dataset to be scored, its first... The fusion weights of the evaluation dimensions, For the time-series data of the gas sensor array in the dataset to be scored, its first... Dynamic adjustment factors for each evaluation dimension For the time-series data of the gas sensor array in the dataset to be scored, its first... Initial fusion weights for each evaluation dimension, For the index of evaluation dimensions, These correspond to data integrity, signal quality, event correlation, data diversity, and multi-channel synergy, respectively. Here, the initial fusion weights are adjusted based on a dynamic adjustment factor to generate the fusion weight settings, so that the fusion weights of the evaluation dimensions with higher scores are higher. This achieves an adaptive result of increasing the weights of high-quality evaluation dimensions and decreasing the weights of low-quality evaluation dimensions. This makes it easier to focus on high-quality evaluation dimensions in the subsequent scoring process, thereby improving the scoring results of gas sensor array time series data that perform well in at least one evaluation dimension. This also makes it easier to mine and screen gas sensor array time series data that perform well in at least one evaluation dimension. Furthermore, for any gas sensor array time series data included in the gas sensor array time series database, the previous gas sensor array time series data included in the gas sensor array time series database is used as its reference data, and the final weight of its reference data in each evaluation dimension is used as its initial fusion weight in the corresponding evaluation dimension, so as to ensure the continuity of weight updates. For the first gas sensor array time-series data included in the gas sensor array time-series database, its initial fusion weights for each evaluation dimension are preset values, satisfying the following conditions: the initial fusion weight for event correlation is the largest, followed by the initial fusion weight for signal quality, and then the initial fusion weights for data integrity and multi-channel synergy are the smallest, with the initial fusion weights for data integrity and multi-channel synergy being the same. The reason for this is that the gas sensor array time-series data is used for target gas identification and detection, so event correlation is the most important, hence its initial fusion weight is set to the largest. Signal quality directly reflects the confidence level of the gas sensor array time-series data, therefore signal quality is of secondary importance, hence its initial fusion weight is set to the second largest. Data integrity and multi-channel synergy are used to reflect the acquisition quality of time-series data from the gas sensor array. They are less important than the previous two, so the initial fusion weight preset values ​​for data integrity and multi-channel synergy are set to a lower value, and the initial fusion weight preset values ​​for data integrity and multi-channel synergy are set to be the same. Data diversity can only reflect the scarcity and typicality of time-series data from the gas sensor array, so the initial fusion weight preset value for data diversity is set to the minimum. For example, in this technical solution, the initial fusion weight preset values ​​for data integrity, signal quality, event correlation, data diversity, and multi-channel synergy are set to 0.15, 0.25, 0.35, 0.10, and 0.15, respectively. Of course, other reasonable values ​​can also be set, and the specific settings are determined by the staff according to the actual situation. No restrictions are imposed here. The system continuously monitors the gas sensor array time-series data included in the gas sensor array time-series database. When the latest gas sensor array time-series data included in the database, and the N gas sensor array time-series data preceding it, satisfy the condition that the final weight remains unchanged, the fusion weight of the data in each evaluation dimension is used as the initial fusion weight of the next gas sensor array time-series data included in the database in the corresponding evaluation dimension. N is a preset quantity threshold and is a positive integer greater than 1. This introduces a setting to force the update of the initial fusion weight, taking into account both adaptability and computational efficiency. N can generally be between 30 and 50. For example, when N is 49, it is considered that when the final weight of 50 consecutive gas sensor array time-series data remains unchanged, the setting to force the update of the initial fusion weight is introduced, so that the initial fusion weight is more in line with the calculation of the comprehensive quality score of the current time-series data. Dimensional reliability parameters are generated based on time-series environmental parameter data, as follows: Based on the environmental parameter time series data, the environmental state category to which the gas sensor array time series data belongs is determined. If the temperature and humidity are normal, it indicates that the multi-channel gas sensor array is in a suitable working state and the quality of the collected data is high. The dimension reliability parameter is set to 1, that is, no adjustment is made to the fusion weights, and the fusion weights are directly used as the weight scaling value. If the temperature or humidity is abnormal, it indicates that the multi-channel gas sensor array is in a poor working state and the quality of the collected data is low. The dimension reliability parameter is set to a positive decimal less than 1, such as 0.95, that is, each fusion weight is reduced proportionally, so that when the weight scaling value is used as the final weight to calculate the comprehensive quality score, the comprehensive quality score is adaptively reduced. If the temperature and humidity are abnormal, it indicates that the multi-channel gas sensor array is in the worst working state and the quality of the collected data is the lowest. The dimension reliability parameter is set to a positive decimal less than 1, and less than the dimension reliability parameter when the environmental state category is abnormal temperature, such as 0.9, that is, each fusion weight is reduced proportionally, so that when the weight scaling value is used as the final weight to calculate the comprehensive quality score, the comprehensive quality score is adaptively reduced. The reliability parameter of each dimension is multiplied by the fusion weight of each evaluation dimension to obtain the weight scaling value of each evaluation dimension. The relative error between the weight scaling value and the initial fusion weight under each evaluation dimension is calculated and the average is taken as the weight difference index. If the weight difference index is less than the preset threshold, it means that the weight scaling value and the initial fusion weight are not significantly different. That is, it means that the evaluation dimension of the gas sensor array time series data has not changed significantly compared with its reference data. In this case, the initial fusion weight is used as the final weight, that is, the final weight of its reference data in each evaluation dimension is used as its final weight in the corresponding evaluation dimension to maintain the consistency of the evaluation benchmark. Conversely, it means that the evaluation dimension of the gas sensor array time series data has changed significantly compared with its reference data. In this case, the weight scaling value is used as the final weight, that is, the weight scaling value of its evaluation dimension is used as its final weight in the corresponding evaluation dimension. The evaluation benchmark that fits it is used to score it. The preset threshold can generally be set between 5% and 20%. For example, it can be set to 10%. The weight difference index is the average relative error. When it is less than 10%, it means that the evaluation dimension has changed little before and after. Conversely, it means that the evaluation dimension has changed much before and after. The data integrity score, signal quality score, event relevance score, data diversity score, and multi-channel coordination score are weighted and summed according to their final weights to obtain the comprehensive quality score, which is denoted as [missing value]. The calculation formula is as follows: In the formula, These are the final weights of the time-series data of the gas sensor array in the dataset to be scored in terms of data integrity, signal quality, event correlation, data diversity, and multi-channel synergy. S4 Sample Grading Output The overall quality score is compared with a preset score threshold to determine the label to be used for the sample, which is a high-quality sample label, a qualified sample label, or a low-quality sample label. The output includes a comprehensive quality score, data integrity score, signal quality score, event relevance score, data diversity score, multi-channel synergy score, dimensional reliability parameter, final weights, and sample score records with sample labels. Specifically, the overall quality score is compared with preset high and low score thresholds. When high-quality sample labels are generated, they can be directly used for model training; when At that time, qualified sample labels are generated, which are then reviewed and adjusted by experts before being used for model training; when When a low-quality sample label is generated, it is directly removed. Here, 85 and 60 are the preset high score threshold and low score threshold in this embodiment, respectively. The specific values ​​can be adjusted by the staff according to the actual situation, and there is no restriction here.

[0053] Example 2, please refer to Figure 2 This embodiment provides a gas sensor array time-series data validity scoring system for executing the gas sensor array time-series data validity scoring method in Embodiment 1 above, including: The dataset to be scored module is used to form a dataset to be scored by combining the latest gas sensor array time series data included in the gas sensor array time series database with sample attribute information, event marker information and environmental parameter time series data in the acquisition scenario. The gas sensor array time series data consists of the gas response signal time series data of each channel acquired by the same multi-channel gas sensor array. The evaluation dimension score generation module is used to extract a set of scoring features from the dataset to be scored. The set of scoring features includes data integrity features, signal quality features, event correlation features, data diversity features, and multi-channel synergy features. Based on these features, scores for five evaluation dimensions are generated: data integrity score, signal quality score, event correlation score, data diversity score, and multi-channel synergy score. The comprehensive quality score generation module is used to determine the dynamic adjustment factors for each evaluation dimension based on the score, adjust the initial fusion weights of each evaluation dimension based on the dynamic adjustment factors to generate fusion weights, scale the fusion weights based on environmental parameter time series data to generate weight scaling values, determine the final weights based on the difference between the weight scaling values ​​and the initial fusion weights, and perform a weighted summation of each score based on the final weights to obtain the comprehensive quality score. The sample label generation module is used to compare the overall quality score with a preset score threshold to determine the sample labeling for the gas sensor array time series data.

[0054] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for evaluating the validity of time-series data from a gas sensor array, characterized in that, include: For the gas sensor array time series data newly included in the gas sensor array time series database, the sample attribute information, event marker information and environmental parameter time series data under its acquisition scenario are combined to form a dataset to be scored. The gas sensor array time series data consists of the gas response signal time series data of each channel acquired by the same multi-channel gas sensor array. The scoring feature set is extracted from the dataset to be scored, which includes data integrity features, signal quality features, event correlation features, data diversity features and multi-channel synergy features. Based on this, scores for five evaluation dimensions are generated: data integrity score, signal quality score, event correlation score, data diversity score and multi-channel synergy score. Based on the scores, dynamic adjustment factors for each evaluation dimension are determined. The initial fusion weights for each evaluation dimension are adjusted based on the dynamic adjustment factors to generate fusion weights. The fusion weights are scaled based on time-series environmental parameter data to generate weight scaling values. The final weights are determined based on the difference between the weight scaling values ​​and the initial fusion weights. The scores are then weighted and summed based on the final weights to obtain the comprehensive quality score. The overall quality score is compared with a preset score threshold to determine the labeling of the gas sensor array time series data samples.

2. The method for evaluating the validity of time-series data of a gas sensor array according to claim 1, characterized in that: The sample attribute information includes the age category, gender category, and inhalation stage category of the sampled object in the collection scenario. The event annotation information includes the actual time when the sampled object inhaled the target gas in the collection scenario, as well as the start time of detecting the target gas in the gas sensor array time series data. The environmental parameters include the ambient temperature and relative humidity in the collection scenario, and the timestamps of the gas response signal time series data of each channel in the gas sensor array time series data are aligned.

3. The method for evaluating the validity of time-series data of a gas sensor array according to claim 1, characterized in that: The data integrity features include the missing rate, glitch index, and temporal continuity of the time-series data of the gas response signals of each channel. The calculation logic of the data integrity score is as follows: after calculating the data integrity score of the time-series data of the gas response signals of each channel based on the missing rate, glitch index, and temporal continuity, the average value is taken as the data integrity score of the time-series data of the gas sensor array in the dataset to be scored. The data integrity score is negatively correlated with the missing rate, glitch index, and temporal continuity.

4. The method for evaluating the validity of time-series data of a gas sensor array according to claim 1, characterized in that: The signal quality characteristics include the signal-to-noise ratio, baseline drift rate, and noise level of the time-series data of the gas response signals of each channel. The calculation logic of the signal quality score is as follows: after calculating the signal quality score of the time-series data of the gas response signals of each channel based on the signal-to-noise ratio, baseline drift rate, and noise level, the average value is taken as the signal quality score of the time-series data of the gas sensor array in the dataset to be scored. The signal quality score is positively correlated with the signal-to-noise ratio and negatively correlated with the baseline drift rate and noise level.

5. The method for evaluating the validity of time-series data of a gas sensor array according to claim 2, characterized in that: The event correlation features include event marker synchronization, false detection rate, and waveform matching degree between gas sensor array time series data and standard time series data. The calculation logic of the event correlation score is as follows: the event marker synchronization, the difference between 1 and the false detection rate, and the waveform matching degree are weighted and summed to obtain the event correlation score of the gas sensor array time series data in the dataset to be scored. The waveform matching degree calculation logic between the gas sensor array time series data and the standard time series data is as follows: For any channel gas response signal time series data in the gas sensor array time series data, calculate its DTW distance and cosine similarity of the morphological feature vectors with the gas response signal time series data of the same type in the standard time series data. Then, process and weighted sum the results to obtain the waveform matching degree of the gas response signal time series data of that channel. Take the average value of the waveform matching degree of the gas response signal time series data of each channel as the waveform matching degree between the gas sensor array time series data and the standard time series data. The morphological feature vectors include the rise slope, peak time and decay constant. The calculation logic for the false detection rate is as follows: extract the target gas detection results output by the multi-channel gas sensor array at each time stamp, and based on the target gas detection results and the actual target gas results, determine the number of timestamps in which the target gas was mistakenly identified, and use the ratio of this number to the total number of timestamps in the time series data of the gas sensor array as the false detection rate. The event tag synchronization is determined based on the time interval between the actual time and the start time, and the time interval is inversely proportional to the event tag synchronization.

6. The method for evaluating the validity of time-series data of a gas sensor array according to claim 2, characterized in that: The data diversity characteristics include the category proportion of gas sensor array time-series data in the dataset to be scored. The determination logic is as follows: based on sample attribute information and environmental parameter time-series data, determine the category to which the gas sensor array time-series data in the dataset to be scored belongs, and use it as the target category. Count the number of gas sensor array time-series data belonging to the target category in the gas sensor array time-series database, and calculate the ratio of this ratio to the total number of gas sensor array time-series data in the gas sensor array time-series database, which is used as the category proportion. Among them, there are four categories of environmental status, three categories of age of the sampling subjects, two categories of gender, and three categories of the stage of consumption; The environmental status categories include four types: abnormal temperature and humidity, abnormal temperature, abnormal humidity, and normal temperature and humidity. These categories are determined based on time-series data of environmental parameters and preset environmental temperature and relative humidity thresholds. The data diversity score is calculated based on the category percentage and is positively correlated with the category percentage, and a low percentage compensation score mechanism is introduced.

7. The method for evaluating the validity of time-series data of a gas sensor array according to claim 1, characterized in that: The multi-channel coordination features include channel synchronization and cross-interference coefficient. The calculation logic of the multi-channel coordination score is as follows: the differences between channel synchronization, 1 and cross-interference coefficient are weighted and summed to obtain the multi-channel coordination score of the gas sensor array time series data in the dataset to be scored.

8. The method for evaluating the validity of time-series data of a gas sensor array according to claim 6, characterized in that: The scores of each evaluation dimension are normalized to obtain the dynamic adjustment factor of each evaluation dimension. The dynamic adjustment factor of each evaluation dimension is multiplied by the corresponding initial fusion weight and then normalized to obtain the fusion weight of each evaluation dimension. The dimensional reliability parameter is generated based on the environmental parameter time series data. Specifically, the environmental state category to which the gas sensor array time series data belongs is determined based on the environmental parameter time series data. If the temperature and humidity are normal, the dimensional reliability parameter is set to 1. If the temperature or humidity is abnormal, the dimensional reliability parameter is set to a positive decimal less than 1. If the temperature and humidity are abnormal, the dimensional reliability parameter is set to a positive decimal less than 1, and is less than the dimensional reliability parameter corresponding to the environmental state category of abnormal temperature. Multiply the dimensional reliability parameter by the fusion weight of each evaluation dimension to obtain the weight scaling value of each evaluation dimension. Calculate the relative error between the weight scaling value and the initial fusion weight under each evaluation dimension and take the average value as the weight difference index. If the weight difference index is less than the preset threshold, the initial fusion weight is used as the final weight; otherwise, the weight scaling value is used as the final weight. The data integrity score, signal quality score, event relevance score, data diversity score, and multi-channel coordination score are weighted and summed according to their final weights to obtain the comprehensive quality score. The overall quality score is compared with a preset score threshold to determine the label to be used for the sample, which is a high-quality sample label, a qualified sample label, or a low-quality sample label. The output includes a comprehensive quality score, data integrity score, signal quality score, event relevance score, data diversity score, multi-channel synergy score, dimensional reliability parameter, final weights, and sample score records with sample usage labels.

9. The method for evaluating the validity of time-series data of a gas sensor array according to claim 8, characterized in that, For any gas sensor array time series data included in the gas sensor array time series database, the previous gas sensor array time series data included in the gas sensor array time series database is used as its reference data, and the final weight of its reference data in each evaluation dimension is used as its initial fusion weight in the corresponding evaluation dimension. For the first gas sensor array time series data included in the gas sensor array time series database, its initial fusion weights in each evaluation dimension are preset values, and satisfy the following: the initial fusion weight preset value for event correlation is the largest, the initial fusion weight preset value for signal quality is the second largest, the initial fusion weight preset value for data integrity and multi-channel synergy is the smallest, and the initial fusion weight preset value for data integrity and multi-channel synergy is the same, while the initial fusion weight preset value for data diversity is the smallest. The time series data of the gas sensor array included in the gas sensor array time series database is continuously monitored. When the latest gas sensor array time series data included in the gas sensor array time series database and the N gas sensor array time series data before it satisfy the condition that the final weight remains unchanged, the fusion weight of the gas sensor array time series data in each evaluation dimension is used as the initial fusion weight of the next gas sensor array time series data included in the gas sensor array time series database in the corresponding evaluation dimension. N is a preset quantity threshold and N is a positive integer greater than 1.

10. A gas sensor array time-series data validity scoring system, used to execute the gas sensor array time-series data validity scoring method according to any one of claims 1-9, characterized in that, include: The dataset to be scored module is used to form a dataset to be scored by combining the latest gas sensor array time series data included in the gas sensor array time series database with sample attribute information, event marker information and environmental parameter time series data in the acquisition scenario. The gas sensor array time series data consists of the gas response signal time series data of each channel acquired by the same multi-channel gas sensor array. The evaluation dimension score generation module is used to extract a set of scoring features from the dataset to be scored. The set of scoring features includes data integrity features, signal quality features, event correlation features, data diversity features, and multi-channel synergy features. Based on these features, scores for five evaluation dimensions are generated: data integrity score, signal quality score, event correlation score, data diversity score, and multi-channel synergy score. The comprehensive quality score generation module is used to determine the dynamic adjustment factors for each evaluation dimension based on the score, adjust the initial fusion weights of each evaluation dimension based on the dynamic adjustment factors to generate fusion weights, scale the fusion weights based on environmental parameter time series data to generate weight scaling values, determine the final weights based on the difference between the weight scaling values ​​and the initial fusion weights, and perform a weighted summation of each score based on the final weights to obtain the comprehensive quality score. The sample label generation module is used to compare the overall quality score with a preset score threshold to determine the sample labeling for the gas sensor array time series data.

Images