Method and system for detecting and classifying multi-component toxic gases in a stamping plant

Abstract

The present application relates to the technical field of gas detection classification, and discloses a stamping workshop multi-component toxic gas detection and classification method and system. The method comprises obtaining multi-component toxic gas concentration data sequences of multiple detection points in the stamping workshop space, and then screening matching gas data segments from historical toxic gas data, which match the gas concentration distribution characteristics in the current detection period. Subsequently, the concentration change of the matching gas data segments in time sequence is analyzed, the concentration mutation point is determined and the time delay is calculated, and the concentration gradient data sequences and gas component data sequence changes before and after the concentration mutation point are compared to determine the component matching degree. Then, the spatial weight coefficient is determined by combining the time delay, the component matching degree and the concentration distribution difference, the component feature vector is weighted and fused by the coefficient to generate a feature vector set, and finally the feature vector set is input into the classifier to output the multi-component toxic gas classification result.

Description

Technical Field

[0001] This invention relates to the field of gas detection and classification technology, specifically to a method and system for detecting and classifying multi-component toxic gases in stamping workshops. Background Technology

[0002] In industrial production, stamping workshops, as key production sites in the machinery manufacturing field, continuously generate multi-component toxic gases due to the unique nature of their production processes. These toxic gases originate from complex sources, including volatile organic compounds generated during the reaction of metal materials and lubricants in the stamping process, as well as harmful gases produced during equipment operation. With the continuous expansion of industrial production scale and increasingly stringent requirements for the production environment, the need for accurate detection and classification of multi-component toxic gases in stamping workshops is becoming increasingly urgent.

[0003] Currently, methods for detecting multi-component toxic gases in stamping workshops primarily rely on analyzing gas concentration data from a single detection point. These methods typically collect gas concentration information only at a fixed location, ignoring the spatial variations in gas distribution within the workshop. Due to the complex spatial layout of stamping workshops, the amount of gas generated, the diffusion rate, and the concentration distribution vary significantly across different areas. Data from a single detection point cannot comprehensively reflect the gas conditions of the entire workshop, leading to substantial biases in the detection results and failing to provide an accurate basis for subsequent gas classification.

[0004] Existing detection methods often lack effective utilization of historical data when processing gas concentration data. In actual production processes, the operating conditions in stamping workshops exhibit a certain periodicity, and the gas concentration distribution characteristics may show similarities across different production cycles. However, most existing methods only analyze data independently within the current detection cycle, failing to extract information from historical data that is similar to the current operating conditions. They cannot leverage historical patterns in gas concentration changes to assist current detection and classification work, thus limiting the accuracy and reliability of the detection and classification process.

[0005] Traditional gas classification methods typically rely solely on absolute gas concentration values ​​or simple concentration trends, neglecting the temporal patterns of gas component changes and the interrelationships between different components. The concentration changes of multi-component toxic gases often exhibit temporal characteristics; for example, the concentrations of certain components may abruptly change at specific points in time, and these abrupt changes are closely related to the gas's source and properties. Furthermore, the concentration changes of different components may exhibit synergistic or restrictive relationships. The neglect of these factors by traditional methods makes it difficult to accurately identify the specific gas category, failing to meet the demands of precise classification of multi-component toxic gases in industrial production. This can potentially impact the safety management of the production environment and the health and safety of workers. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for detecting and classifying multi-component toxic gases in stamping workshops, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for detecting and classifying multi-component toxic gases in a stamping workshop, the method comprising:

[0008] Acquire multi-component toxic gas concentration data sequences from multiple detection points within the stamping workshop space;

[0009] Select matching gas data segments from historical toxic gas data that match the gas concentration distribution characteristics within the current detection period;

[0010] Analyze the concentration changes of the matched gas data segments over time to determine the concentration abrupt change points of each matched gas data segment;

[0011] The time difference between the start time of each matched gas data segment and the concentration mutation point is used as the time delay of that matched gas data segment;

[0012] By comparing the changes in the concentration gradient data sequence before the concentration abrupt change point with the changes in the gas component data sequence after the concentration abrupt change point, the component matching degree of each matching gas data segment is determined.

[0013] By combining the time delay, the component matching degree, and the difference in gas concentration distribution between the matched gas data segment and the current detection cycle, the spatial weight coefficient of each matched gas data segment is determined.

[0014] The component feature vectors of all matching gas data segments corresponding to the current detection period are weighted and fused using the spatial weight coefficients to generate a feature vector set of multi-component toxic gases.

[0015] The set of feature vectors is input into the classifier, which outputs the classification results of the multi-component toxic gases.

[0016] Preferably, the step of selecting matching gas data segments from historical toxic gas data that match the gas concentration distribution characteristics within the current detection period includes:

[0017] Identify abnormal concentration fluctuations in historical toxic gas data sequences;

[0018] The historical toxic gas data sequence is divided into multiple continuous gas data segments, and the concentration distribution characteristic value of each gas data segment is calculated.

[0019] From the abnormal concentration fluctuations in historical toxic gas data, select matching gas data segments that match the concentration distribution characteristics of the current detection period.

[0020] Preferably, identifying the abnormal concentration fluctuation portion in the historical toxic gas data sequence includes:

[0021] Calculate the rate of change of each concentration data point in the historical toxic gas data series;

[0022] The concentration data point where the maximum rate of change first appears is used as the dividing point;

[0023] The data sequence is divided into a preceding part and a subsequent part based on the aforementioned dividing point, and the average rate of change of the preceding part and the subsequent part are calculated respectively.

[0024] The portion with a large average rate of change is marked as the part with abnormal concentration fluctuations.

[0025] Preferably, the step of selecting matching gas data segments that match the concentration distribution characteristics of the current detection period includes:

[0026] Calculate the difference in concentration distribution characteristic values ​​between each gas data segment in the abnormal concentration fluctuation section of historical toxic gas data and the current detection period;

[0027] The concentration distribution feature value differences are negatively correlated to obtain the concentration distribution similarity.

[0028] Matching gas data segments are determined based on the similarity of the concentration distribution.

[0029] Preferably, the step of analyzing the concentration changes of the matched gas data segments over time and determining the concentration abrupt change points of each matched gas data segment includes:

[0030] Detect the extreme points of the concentration gradient in the matched gas data segment;

[0031] The extreme points of the concentration gradient are taken as the concentration abrupt change points of the matched gas data segment.

[0032] Preferably, the step of comparing the changes in the concentration gradient data sequence before the concentration abrupt change point with the changes in the gas component data sequence after the concentration abrupt change point to determine the component matching degree of each matching gas data segment includes:

[0033] Calculate the average difference in concentration gradient data sequences prior to the concentration abrupt change point;

[0034] Calculate the average variation difference of gas component data sequences after the concentration abrupt change point;

[0035] The difference between the average change in concentration gradient and the average change in gas components is negatively correlated to obtain the component matching degree.

[0036] Preferably, determining the spatial weight coefficient for each matching gas data segment by combining the time delay, the component matching degree, and the difference in gas concentration distribution between the matching gas data segment and the current detection cycle includes:

[0037] A set of coordinate points is formed by using the location of the concentration abrupt change point of each matched gas data segment as the x-axis and the time delay of the matched gas data segment as the y-axis.

[0038] Perform curve fitting on the set of coordinate points to generate a fitted curve;

[0039] Calculate the residual between each coordinate point and the fitted curve;

[0040] The validity index of each matched gas data segment is calculated based on the component matching degree, time delay, and residual.

[0041] The spatial weighting coefficient is determined based on the effectiveness index and the gas concentration distribution difference, wherein the effectiveness index is positively correlated with the spatial weighting coefficient, and the gas concentration distribution difference is negatively correlated with the spatial weighting coefficient.

[0042] Preferably, the calculation of the validity index for each matched gas data segment based on the component matching degree, time delay, and residual includes:

[0043] The ratio of component matching degree to the product of time delay and residual is used as an effectiveness indicator.

[0044] Preferably, the step of weighting and fusing the component feature vectors of all matching gas data segments corresponding to the current detection period using the spatial weighting coefficient includes:

[0045] The spatial weighting coefficients of all matched gas data segments are normalized.

[0046] The corresponding component feature vectors are weighted and summed using the normalized spatial weight coefficients.

[0047] Preferably, the present invention also includes a multi-component toxic gas detection and classification system for a stamping workshop, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the multi-component toxic gas detection and classification method for a stamping workshop as described above.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] By acquiring multi-component toxic gas concentration data sequences from multiple detection points within the stamping workshop, the spatial distribution information of gas concentrations within the workshop can be comprehensively captured. Compared to traditional methods relying on data from a single detection point, the multi-detection-point data acquisition approach can effectively cover different areas of the workshop, fully reflecting the actual situation of gas generation and diffusion in each area, avoiding detection biases caused by spatial differences, and providing a more comprehensive and reliable data foundation for subsequent gas analysis and classification.

[0050] During data processing, this method filters out matching gas data segments from historical toxic gas data that correspond to the gas concentration distribution characteristics within the current detection period, thus achieving an effective correlation between historical and current data. By leveraging the gas concentration variation patterns in historical data that are similar to those under current operating conditions, the data analysis dimensions for the current detection period can be further enriched, reducing the analytical limitations that may arise from relying solely on current data. This makes the gas detection and classification process more referential and more closely aligned with the periodic characteristics of actual production conditions.

[0051] By analyzing the concentration changes of matched gas data segments over time to identify concentration abrupt change points, and using the time difference from the start time to the concentration abrupt change point as a time delay, key nodes of gas concentration change over time can be accurately captured. Sudden changes in gas concentration are often closely related to the gas source and generation mechanism. Determining the time delay clearly reflects the temporal characteristics of gas concentration changes in different matched data segments, providing an important temporal reference for subsequent analysis of gas component variation patterns and helping to gain a deeper understanding of the underlying logic of gas concentration changes.

[0052] By comparing the concentration gradient data sequences before and after concentration abrupt changes with the gas component data sequences to determine the component matching degree, we can delve deeper into the changing patterns of gas components before and after key time points and the correlations between different components. This analytical method can accurately identify the changing characteristics of gas components in each matching data segment, understand the synergistic or restrictive relationships of concentration changes of different components, and thus provide more detailed component-level information for determining gas categories, improving our understanding of gas component characteristics.

[0053] By combining time delay, component matching degree, and the difference in gas concentration distribution between the matched gas data segment and the current detection cycle, spatial weighting coefficients can be determined, enabling differentiated consideration of different matched data segments. Since different matched data segments differ in their correlation with the current detection cycle, temporal characteristics, and component characteristics, spatial weighting coefficients can highlight the role of matched data segments that are more closely aligned with the current detection cycle, while reasonably weakening the influence of data segments with weaker correlation. This makes the subsequent feature vector weighting and fusion process more targeted and more consistent with the actual gas conditions of the current detection cycle.

[0054] By weighting and fusing the component feature vectors of all matching gas data segments corresponding to the current detection period using spatial weighting coefficients, a feature vector set of multi-component toxic gases is generated, which can integrate the advantageous information of each matching data segment. This fusion method is not a simple data superposition, but a precise integration based on weights. It can make full use of the information in each matching data segment that is valuable to the current classification, avoid the limitations of a single data segment, and form a more comprehensive and representative feature vector set, providing high-quality input data for the classifier.

[0055] The feature vector set is input into a classifier to output classification results. The entire process, through multi-dimensional data collection, effective utilization of historical data, in-depth analysis of time series and component features, and precise weight fusion, forms a complete and scientific detection and classification system. This system can effectively improve the accuracy and reliability of multi-component toxic gas detection and classification in stamping workshops, better adapt to the complex gas environment and cyclical production conditions of the workshop, provide strong support for the safety management of the workshop production environment and the health protection of workers, and meet the actual needs of industrial production for precise management of multi-component toxic gases. Attached Figure Description

[0056] Figure 1 This is a schematic diagram illustrating the working principle of the multi-component toxic gas detection and classification method in the stamping workshop described in this invention.

[0057] Figure 2 A flowchart for filtering matching gas data segments;

[0058] Figure 3 A flowchart for identifying abnormal concentration fluctuations;

[0059] Figure 4 A flowchart for determining the component matching degree;

[0060] Figure 5 A flowchart for determining spatial weighting coefficients. Detailed Implementation

[0061] 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.

[0062] Please see Figure 1 This invention provides a method for detecting and classifying multi-component toxic gases in a stamping workshop, the method comprising:

[0063] A multi-component toxic gas concentration data sequence is acquired from multiple detection points within the stamping workshop. This data sequence is collected in real-time by a gas sensor array deployed within the workshop, containing concentration measurements of various toxic gas components at different time points. Matching gas data segments that match the gas concentration distribution characteristics within the current detection period are selected from historical toxic gas data. The concentration changes of the matching gas data segments over time are analyzed to determine the concentration abrupt change points for each matching gas data segment. The time difference between the start time and the concentration abrupt change point of each matching gas data segment is used as the time delay of that matching gas data segment. The component matching degree of each matching gas data segment is determined by comparing the changes in the concentration gradient data sequence before the concentration abrupt change point with the changes in the gas component data sequence after the concentration abrupt change point. The spatial weight coefficient of each matching gas data segment is determined by combining the time delay, component matching degree, and the difference in gas concentration distribution between the matching gas data segment and the current detection period. The component feature vectors of all matching gas data segments corresponding to the current detection period are weighted and fused using the spatial weight coefficient to generate a feature vector set of multi-component toxic gases. The feature vector set is input into a classifier, and the classification results of the multi-component toxic gases are output.

[0064] Example 1: See Figure 2 In the actual operating environment of a stamping workshop, the monitoring of multi-component toxic gases relies on a pre-deployed sensor network. These sensors are placed in different locations within the workshop, such as near stamping machines, raw material storage areas, ventilation vents, and areas of high human activity, to capture representative gas concentration information within the space. Each sensor node periodically collects concentration readings of multiple target toxic gases, such as benzene, toluene, xylene, formaldehyde, and the total amount of volatile organic compounds, forming a multi-dimensional time-series data stream. This real-time data is transmitted to a central processing unit, constituting the gas concentration data sequence for the current detection period. Simultaneously, a vast historical database is maintained, storing the raw concentration data sequences recorded by all sensor nodes over long historical spans; this historical data forms the foundation for analysis and matching.

[0065] Identifying anomalous concentration fluctuations in historical toxic gas data sequences is the starting point of the entire process. Historical data is massive and continuous, making direct searching inefficient and susceptible to interference from numerous stable baselines. Therefore, it's crucial to first locate the intervals within the historical data where significant concentration changes have occurred—the anomalous fluctuation portions. This process doesn't simply involve setting a global threshold; instead, it relies on the local characteristics of the data itself. Specifically, the system traverses the historical data sequence, calculating the rate of change between each concentration data point and its predecessor. This rate of change reflects the speed and direction of concentration increase or decrease per unit time. The system identifies the point in the entire sequence with the largest absolute value of the rate of change, marking the moment of most dramatic concentration change, and designates it as the initial segmentation point. Using this segmentation point, the original long sequence is divided into two parts: all data before the segmentation point constitutes the preceding part, and all data after the segmentation point constitutes the following part. Next, the average absolute value of the rate of change for all data points within each of these two parts is calculated—the average rate of change. Comparing these two average rates of change, the portion with the larger value is marked as the anomalous concentration fluctuation portion. This is because abnormal fluctuations usually indicate a sudden change in the activity of the gas release source or in the environmental ventilation conditions, which manifests as a sustained high rate of change over a relatively concentrated period. Conversely, the portion with a smaller average rate of change typically represents a relatively stable background concentration or a slow diffusion process.

[0066] After identifying all potential anomalous fluctuations in historical data, the next step is to further subdivide these fluctuations into more refined, matchable units. The system employs a fixed-length sliding window, such as a five-minute window, to divide each anomalous fluctuation into multiple consecutive, potentially overlapping gas data segments. Each such segment contains concentration readings of multiple gas components within the corresponding time period. For each segment, its concentration distribution characteristic value is calculated. This characteristic value is a comprehensive statistic designed to characterize the overall distribution of multiple gas concentrations within that time period. It not only calculates the mean and variance of each gas concentration to describe its central tendency and dispersion but may also include higher-order moment features such as skewness and kurtosis to capture the asymmetry and leptokurticity of the concentration distribution. All these statistics are combined into a multidimensional feature vector to uniquely characterize the concentration distribution features of that data segment.

[0067] The current detection period, such as real-time data collected in the past ten minutes, is also processed in the same way, calculating a concentration distribution feature vector with the same dimensions. The goal of the screening is to find the data segments most similar to the characteristics of the current detection period from all gas data segments divided from historical abnormal fluctuations. This process is achieved by calculating the distance between feature vectors. The system calculates the Euclidean distance between the feature vector of each historical gas data segment and the feature vector of the current detection period. The smaller the Euclidean distance, the more similar the two data segments are in terms of concentration distribution. The system presets a similarity distance threshold. All historical gas data segments with calculated distance values ​​below this threshold are screened out and identified as matching gas data segments that match the gas concentration distribution characteristics within the current detection period. These matching data segments come from different historical periods, but they have all experienced concentration distribution states similar to the current moment. Therefore, their subsequent change patterns have important reference value for predicting or classifying the current gas situation. Thus, the system has successfully extracted a set of matching samples that are highly relevant to the current situation and have undergone preliminary screening from the vast historical data.

[0068] Example 2: See Figure 3 In the practical application of gas monitoring in stamping workshops, meticulous processing of historical data sequences is crucial for identifying meaningful patterns. Historical databases store continuous monitoring data from previous months or even years, recording concentration changes of multiple gas components within the workshop under various operating conditions, different production rhythms, and unexpected events. To locate segments potentially relevant to the current situation from such a massive dataset, an effective method is first needed to preliminarily identify those noteworthy intervals—that is, areas of abnormal concentration fluctuations.

[0069] The system reads stored historical concentration data, typically recorded at fixed time intervals, such as one record per minute. Each record contains concentration values ​​for multiple gases, including benzene, toluene, and xylene. Calculating the rate of change for each data point is the fundamental step. For each point in the sequence (except the first), the system calculates the difference in concentration for each gas between that point and the previous time point, using the absolute value or sum of squares of these differences as the overall rate of change for that time point. This process iterates through the entire historical sequence, generating a corresponding rate of change sequence. The system then searches for a point in this rate of change sequence where the global maximum rate of change first occurs. This selection is practically significant because it typically marks the beginning of a significant emission event or a sudden change in environmental conditions. For example, in the historical record at 2:15 PM on a certain day, the rate of change for multiple gas concentrations simultaneously and sharply increases, reaching a peak; this time point is marked as the dividing point. Using this dividing point as the boundary, the original historical concentration sequence is divided into two parts: all data before the dividing point constitutes the presequence part, which usually represents the relatively stable state before the event; all data after the dividing point constitutes the subsequent part, which includes the changes during and after the event.

[0070] To objectively determine which part is more noteworthy, the system calculates the average rate of change of all data points in both the preceding and subsequent parts. This average rate of change reflects the average severity of concentration changes within that time period. Comparing these two averages, the part with the significantly larger value is marked as the concentration anomaly fluctuation section. For example, if the average rate of change in the subsequent part is much higher than that in the preceding part, then the entire subsequent part is identified as a potential anomaly fluctuation range. Historical data may contain multiple such dividing points and fluctuation ranges, which the system processes one by one to identify all possible anomaly fluctuation sections. After successfully identifying each concentration anomaly fluctuation section in the historical data, the screening process enters a more refined stage. The system further divides each anomaly fluctuation section into multiple continuous gas data segments of fixed length. For example, an anomaly fluctuation section lasting three hours is divided into six continuous data segments with a 30-minute time window. Adjacent data segments may or may not overlap, depending on the specific configuration.

[0071] Each gas data segment needs a quantifiable indicator to characterize its concentration distribution features, i.e., a concentration distribution feature value. This feature value is not a single numerical value, but a comprehensive vector calculated based on the concentration readings of all gas components at all time points within the data segment. It may include the average concentration, standard deviation, and higher-order statistics reflecting the shape of the distribution for each component. These calculations ensure that the feature value captures the overall shape and internal differences in the gas concentration distribution within the data segment. Simultaneously, real-time monitoring data from the current monitoring period (e.g., the past thirty minutes) undergoes the exact same processing flow, calculating a concentration distribution feature value with the exact same dimensions and meaning. The core of the selection process lies in calculating the difference between the feature value of each gas data segment in the historical anomaly fluctuation section and the current feature value. This difference is typically calculated using distance metrics, such as the sum of the absolute differences between the two feature vectors in each dimension, or their Euclidean distance. The larger the distance, the less similar the historical data segment is to the current concentration distribution pattern; the smaller the distance, the higher the similarity.

[0072] Directly using distance differences is not conducive to intuitive understanding and subsequent calculations. Therefore, the system performs a negative correlation transformation, mapping the difference value to a similarity index. For example, a large difference value is mapped to a similarity value close to zero through the transformation function; while a small difference value is mapped to a similarity value close to one. This transformed value is called the concentration distribution similarity.

[0073] Matching gas data segments are determined based on the similarity of their concentration distributions. The system sets a similarity threshold; among all historical gas data segments, those with a concentration distribution similarity higher than this threshold are selected and formally identified as matching gas data segments. These matching segments come from different historical dates and different event contexts, but they all exhibit a high degree of similarity in gas concentration distribution patterns to the current detection period.

[0074] Example 3: See Figure 4 The core step is to perform in-depth time-series analysis on the selected matching gas data segments. Each matching gas data segment represents a record of gas concentration changes over a historical period, and the dynamic patterns contained within it are valuable for understanding the current situation. Identifying the key turning points in these data segments—concentration abrupt change points—and quantifying the characteristics of changes in their preceding and following sequences are important bases for assessing the fit between this historical segment and the current context.

[0075] When processing each matched gas data segment, the system first needs to accurately identify its internal concentration abrupt change points. This process is based on the analysis of the rate of concentration change within the data segment. The system calculates the concentration difference between adjacent time points in the data segment, constructing a concentration gradient sequence. This gradient sequence reflects the instantaneous rate of concentration change over time. Traversing the entire gradient sequence, the system finds the time point corresponding to the gradient value with the largest absolute value. This point represents the moment of most drastic concentration change within the time span of the data segment, typically corresponding to events such as a sudden opening of a gas leak source, a change in the ventilation system's status, or a violent chemical reaction. The system marks this time point as the concentration abrupt change point for that matched gas data segment. For example, in a matched data segment lasting thirty minutes, the system finds that at the tenth minute, the concentration gradient of benzene reaches its negative maximum value, and the gradient of toluene reaches its positive maximum value; this moment is thus determined to be the concentration abrupt change point.

[0076] Once the concentration abrupt change point is identified, the data segment is naturally divided into two subsequences: the portion before the abrupt change point is called the concentration gradient data sequence (although it contains the original concentration values, the focus is on their trend), and the portion after the abrupt change point is called the gas component data sequence (the focus shifts to the relative changes in the concentrations of each component). The system needs to analyze the characteristics of these two subsequences separately and compare the differences in their change patterns.

[0077] For concentration gradient data sequences preceding abrupt concentration changes, the system focuses on the consistency or volatility of these changes. It calculates the average difference in change for this subsequence. Specifically, it extracts the concentration gradient values ​​of this subsequence (i.e., the previously calculated concentration differences between adjacent points) and calculates the standard deviation of these gradient values. A larger standard deviation indicates more drastic fluctuations in concentration changes and greater inconsistency in the direction of change before the abrupt change; a smaller standard deviation indicates a relatively stable or consistent trend in concentration changes before the abrupt change. This standard deviation is the average difference in concentration gradient change. :

[0078] in: This indicates the number of gradient data points in the subsequence preceding the concentration abrupt change point. Indicates the first Gradient values, This represents the average of these gradient values. The dispersion of concentration changes before the mutation was quantified.

[0079] For gaseous component data sequences following concentration abrupt changes, the system focuses on the synergy or heterogeneity of concentration changes among multiple gaseous components. The average variation difference of this subsequence is calculated. Specifically, for each gaseous component (e.g., benzene, toluene, xylene), the standard deviation of its concentration value over time in the subsequence following the abrupt change is calculated. Then, the average of the standard deviations for all gaseous components is calculated. This average value represents the average variation difference of the gaseous components. . A larger value indicates that after a sudden change, the concentration changes of different gas components are not synchronized and fluctuate differently; a smaller value indicates that the changes of each component are relatively synchronous or stable. This reflects the coordination of the evolution of multi-component systems after mutation.

[0080] To assess the degree to which the matched gas data segment matches the current potential context in describing the patterns before and after the abrupt event, the system needs to comprehensively consider... and The system calculates the absolute difference between the two, representing the characteristics of change. This difference The size has a specific meaning: if A small value implies that the fluctuation characteristics of concentration changes before and after the mutation point (before the mutation) are relatively close to the coordinated characteristics of multi-component evolution (after the mutation), possibly indicating that this historical event has some inherent consistency or pattern; if A large value indicates a significant difference in the characteristics of the sequence before and after the mutation, suggesting that the pattern may be unusual or unstable.

[0081] To obtain a standardized and easily comparable matching index, the system... Perform a negative correlation transformation to map it to component matching degree. The goal of the transformation is: The smaller the value (indicating a high degree of similarity in characteristics before and after), the better. The closer it is to the maximum value of 1; The larger the value (indicating a greater difference in characteristics before and after), the better. The closer it gets to the minimum value of 0. This transformation can be achieved through various mathematical functions, such as linear scaling or non-linear functions. The final result is... This is a value between 0 and 1, representing the relative consistency of concentration change patterns before and after abrupt changes in historical events recorded within a specific matched gas data segment. This component matching degree... This information will be used in subsequent spatial weighting coefficient calculations, serving as one of the important dimensions for measuring the reference value of this historical segment. In this way, the system not only identifies key event points (mutation points) in historical data, but also deeply analyzes the correlation between dynamic patterns before and after the events, providing a quantitative basis for selecting the most relevant and reliable historical references.

[0082] Example 4: See Figure 5 The spatial weighting coefficient comprehensively reflects the reference value of the historical data segment in terms of time delay characteristics, component change patterns, and similarity to the current situation. The process of determining the spatial weighting coefficient involves geometric relationship modeling, consistency assessment, and multi-factor fusion calculation.

[0083] The system processes all matching gas data segments corresponding to the current detection cycle. For each matching segment, the system has determined two key parameters through the aforementioned steps: the location of the concentration abrupt change and the time delay of the segment. The location of the concentration abrupt change is usually represented by the time index of that point within the data segment. For example, in a data segment containing 30 time points (one point per minute), if the abrupt change is located at the 15th minute, its location is recorded as 15. The time delay is the length of time elapsed from the start of the data segment to the concentration abrupt change, for example, 15 minutes.

[0084] The system constructs a two-dimensional coordinate point for each matched gas data segment. The x-axis of this point uses the location of the concentration abrupt change point, and the y-axis uses the time delay value. Assuming there are currently five matched gas data segments, their coordinate point set is shown in Table 1.

[0085] Table 1: Coordinates of the matching gas data segment.

[0086]

[0087] The system performs curve fitting on this set of coordinate points. The goal of the fitting is to find a curve that best represents the distribution trend of these points. Typically, a least squares method is used for polynomial fitting, such as choosing a quadratic polynomial model. The fitting process calculates an optimal curve that minimizes the sum of the squares of the vertical distances (residuals) from all coordinate points to this curve. This generated curve is called the fitted curve, and it reveals a general statistical relationship that may exist between the location of concentration abrupt changes and the amount of time delay in historical data. The system evaluates the deviation of the coordinate points of each matched data segment from the fitted curve, i.e., it calculates the residual. The residual is defined as the absolute difference between the actual ordinate value (time delay) of the point and the function value corresponding to the abscissa (location of the abrupt change) of the fitted curve. The magnitude of the residual reflects the degree to which the time delay characteristics of the data segment conform to the general historical trend. A small residual indicates that the time delay pattern of the segment conforms to common historical patterns; a large residual indicates that the time delay characteristics of the segment are relatively special or abnormal.

[0088] The system has previously calculated the component matching degree for each matched data segment, which quantifies the relative consistency of the concentration change pattern before and after abrupt concentration changes recorded in that segment. Now, combining the newly calculated time delay and residual, the system needs to calculate a comprehensive indicator reflecting the effectiveness of the data segment. The calculation rule for the effectiveness indicator is: divide the component matching degree by the product of the time delay and the residual. The inherent logic of this calculation method is that a high component matching degree improves effectiveness, while a large time delay or a large residual (meaning a deviation from the general trend) reduces its effectiveness. Therefore, the effectiveness indicator tends to select data segments with consistent component change patterns, moderate time delays, and conformity to historical trends. Simultaneously, the system also needs to consider the difference between each matched gas data segment and the current detection period in the overall gas concentration distribution. This difference is measured by calculating the cosine similarity between the concentration distribution feature vectors of the two data segments. A cosine similarity value close to 1 indicates a high degree of similarity, while a value close to 0 indicates a significant difference. To correspond to the negative correlation of the weighting coefficients, the cosine similarity is usually converted into a difference value, for example, by subtracting the cosine similarity from 1.

[0089] The determination of spatial weighting coefficients integrates two factors: effectiveness indicators and concentration distribution differences. The spatial weighting coefficients are positively correlated with effectiveness indicators; that is, the higher the effectiveness indicator, the greater the weight assigned. Concentration distribution differences are negatively correlated; that is, the greater the difference in concentration distribution between the current segment and historical matching segments, the smaller the weight assigned to that historical segment. The specific fusion calculation can be a weighted combination or proportional relationship between the two factors, with the ultimate goal of giving higher weights to historical data segments that are effective in terms of temporal patterns and similar in distribution to the current situation.

[0090] Example 5: Processing and Feature Fusion Stage of Spatial Weighting Coefficients of Matching Gas Data Segments Receives the output from the preceding steps: a set of screened matching gas data segments, each of which has been assigned an initial spatial weighting coefficient. This coefficient comprehensively reflects its time delay characteristics, component matching degree, residual amount with the fitted curve, and difference from the gas concentration distribution of the current detection period.

[0091] The system first processes the set of spatial weight coefficients for all matched gas data segments. These coefficients typically have unequal values ​​and a wide range, and directly using them for weighting might lead to over-amplification or neglect of certain features. Therefore, normalization is necessary to transform all weight coefficients to a uniform, comparable scale, and to ensure that their sum is a fixed value. The normalization process is implemented using a mathematical function. This function takes the original weight coefficient sequence as input, performs an exponential operation on each original weight coefficient, and then sums all the exponential results to obtain the denominator. Dividing each exponential result by this denominator yields the corresponding normalized weight coefficient. This transformation ensures that the normalized weight coefficients all fall between zero and one, and that the sum of all coefficients is exactly one. The normalized weight coefficients represent the relative importance of each matched data segment in the final fused feature. For example, a data segment with a larger original weight coefficient will also have a relatively larger normalized weight coefficient, indicating a higher contribution to the final fused feature; conversely, the same applies.

[0092] Each matched gas data segment is associated with a component feature vector, a multi-dimensional array designed to comprehensively characterize the core properties of the multi-component toxic gases within that segment. Its composition typically includes three main aspects: concentration statistics, spectral characteristics, and time-series pattern characteristics. Concentration statistics include the average, maximum, minimum, standard deviation, skewness, and kurtosis of each gas concentration within the data segment, describing the central trend, dispersion, and shape of the concentration distribution. Spectral characteristics are obtained by performing Fast Fourier Transform or Wavelet Transform on the gas concentration time series, extracting the energy distribution characteristics of concentration changes in the frequency domain and identifying the dominant fluctuation period or frequency component. Time-series pattern characteristics may include autocorrelation coefficients, trend slopes within a sliding window, or implicit patterns extracted through specific models, used to capture the inherent laws and dynamic behavior of concentration evolution over time. All these features are combined into a high-dimensional feature vector, serving as a digital representation of the historical gas conditions represented by that matched data segment.

[0093] For each matched gas data segment, its associated component feature vectors are weighted using its corresponding normalized spatial weight coefficient. The weighting operation involves multiplying each element of the feature vector by the normalized weight coefficient. Then, the system sums the weighted feature vectors of all matched data segments. This summation is performed element-wise: the first element of each weighted vector is added to form the first element of the new vector, the second element is added to form the second element, and so on. This element-wise summation generates a single, fused feature vector. Although named a set, this fused feature vector set typically refers to this single new feature vector that integrates all historical matching information. It integrates key information from multiple relevant historical segments, and its dimensions are the same as the feature vector of a single matched data segment, but the value of each dimension is a weighted average of the features of all matched segments in that dimension. This fused vector contains a comprehensive feature pattern of multi-component toxic gases related to the current detection cycle, extracted based on historical similarity patterns.

[0094] The resulting set of fused feature vectors is input into a pre-trained classifier model. The classifier model employs a Support Vector Machine (SVM) architecture. SVM is a supervised learning model whose core idea is to find an optimal hyperplane that can clearly separate samples of different classes in the feature space. During training, the model learns using a large amount of historical gas feature vector data with known class labels, adjusting its internal parameters to construct the optimal classification decision boundary. In the current classification phase, the trained SVM model receives the fused feature vectors as input and performs calculations using its internal decision function. This decision function determines which class's decision region a feature vector falls into based on its position in the feature space. The classifier's output is a class label representing the system's final classification judgment of the multi-component toxic gas status within the current detection period, such as being identified as "normal ventilation," "minor local leak," "moderate multi-source release," or "serious accident emission," among other pre-defined categories. Thus, through the normalization of spatial weight coefficients, the weighted fusion of feature vectors, and the classifier's decision, the entire process from historical data matching to current gas status classification is completed.

[0095] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0096] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting and classifying multi-component toxic gases in a stamping workshop, characterized in that, The method includes the following steps: Acquire multi-component toxic gas concentration data sequences from multiple detection points within the stamping workshop space; Select matching gas data segments from historical toxic gas data that match the gas concentration distribution characteristics within the current detection period; Analyze the concentration changes of the matched gas data segments over time to determine the concentration abrupt change points of each matched gas data segment; The time difference between the start time of each matched gas data segment and the concentration mutation point is used as the time delay of that matched gas data segment; By comparing the changes in the concentration gradient data sequence before the concentration abrupt change point with the changes in the gas component data sequence after the concentration abrupt change point, the component matching degree of each matching gas data segment is determined. By combining the time delay, the component matching degree, and the difference in gas concentration distribution between the matched gas data segment and the current detection cycle, the spatial weight coefficient of each matched gas data segment is determined. The component feature vectors of all matching gas data segments corresponding to the current detection period are weighted and fused using the spatial weight coefficients to generate a feature vector set of multi-component toxic gases. The set of feature vectors is input into the classifier, which outputs the classification results of the multi-component toxic gases.

2. The method for detecting and classifying multi-component toxic gases in a stamping workshop according to claim 1, characterized in that, The process of selecting matching gas data segments from historical toxic gas data that match the gas concentration distribution characteristics within the current detection period includes: Identify abnormal concentration fluctuations in historical toxic gas data sequences; The historical toxic gas data sequence is divided into multiple continuous gas data segments, and the concentration distribution characteristic value of each gas data segment is calculated. From the abnormal concentration fluctuations in historical toxic gas data, select matching gas data segments that match the concentration distribution characteristics of the current detection period.

3. The method for detecting and classifying multi-component toxic gases in a stamping workshop according to claim 2, characterized in that, The identification of abnormal concentration fluctuations in historical toxic gas data sequences includes: Calculate the rate of change of each concentration data point in the historical toxic gas data series; The concentration data point where the maximum rate of change first appears is used as the dividing point; The data sequence is divided into a preceding part and a subsequent part based on the aforementioned dividing point, and the average rate of change of the preceding part and the subsequent part are calculated respectively. The portion with a large average rate of change is marked as the part with abnormal concentration fluctuations.

4. The method for detecting and classifying multi-component toxic gases in a stamping workshop according to claim 2, characterized in that, The process of selecting matching gas data segments that match the concentration distribution characteristics of the current detection period includes: Calculate the difference in concentration distribution characteristic values ​​between each gas data segment in the abnormal concentration fluctuation section of historical toxic gas data and the current detection period; The concentration distribution feature value differences are negatively correlated to obtain the concentration distribution similarity. Matching gas data segments are determined based on the similarity of the concentration distribution.

5. The method for detecting and classifying multi-component toxic gases in a stamping workshop according to claim 1, characterized in that, The analysis of the concentration changes of the matched gas data segments over time, and the determination of the concentration abrupt change points of each matched gas data segment, includes: Detect the extreme points of the concentration gradient in the matched gas data segment; The extreme points of the concentration gradient are taken as the concentration abrupt change points of the matched gas data segment.

6. The method for detecting and classifying multi-component toxic gases in a stamping workshop according to claim 1, characterized in that, The step of comparing the changes in the concentration gradient data sequence before and after the concentration abrupt change point to determine the component matching degree of each matching gas data segment includes: Calculate the average difference in concentration gradient data sequences prior to the concentration abrupt change point; Calculate the average variation difference of gas component data sequences after the concentration abrupt change point; The difference between the average change in concentration gradient and the average change in gas components is negatively correlated to obtain the component matching degree.

7. The method for detecting and classifying multi-component toxic gases in a stamping workshop according to claim 1, characterized in that, The spatial weighting coefficient for each matching gas data segment is determined by combining the time delay, the component matching degree, and the difference in gas concentration distribution between the matching gas data segment and the current detection cycle, including: A set of coordinate points is formed by using the location of the concentration abrupt change point of each matched gas data segment as the x-axis and the time delay of the matched gas data segment as the y-axis. Perform curve fitting on the set of coordinate points to generate a fitted curve; Calculate the residual between each coordinate point and the fitted curve; The validity index of each matched gas data segment is calculated based on the component matching degree, time delay, and residual. The spatial weighting coefficient is determined based on the effectiveness index and the gas concentration distribution difference, wherein the effectiveness index is positively correlated with the spatial weighting coefficient, and the gas concentration distribution difference is negatively correlated with the spatial weighting coefficient.

8. The method for detecting and classifying multi-component toxic gases in a stamping workshop according to claim 7, characterized in that, The validity index for calculating each matched gas data segment based on the component matching degree, time delay, and residual includes: The ratio of component matching degree to the product of time delay and residual is used as an effectiveness indicator.

9. The method for detecting and classifying multi-component toxic gases in a stamping workshop according to claim 1, characterized in that, The step of weighting and fusing the component feature vectors of all matching gas data segments corresponding to the current detection period using the spatial weighting coefficient includes: The spatial weighting coefficients of all matched gas data segments are normalized. The corresponding component feature vectors are weighted and summed using the normalized spatial weight coefficients.

10. A multi-component toxic gas detection and classification system for a stamping workshop, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-component toxic gas detection and classification method for stamping workshops as described in any one of claims 1 to 9.

Images