Airflow sensor data processing method and system based on edge calculation

Abstract

The invention relates to the technical field of edge calculation and airflow state recognition, and discloses an airflow sensor data processing method and system. The method comprises the steps that original airflow signal flow is acquired and denoised, and a cleaning airflow signal set is obtained; dimension reduction processing is carried out on the cleaning airflow signal set, a feature vector group is separated, and key feature vectors are obtained; performing standardization processing on the key feature vector to obtain a standardized feature distribution range; and performing similarity calculation in combination with a historical feature library, performing classification identification according to a calculation result, determining a corresponding confidence score, and determining an airflow state classification result according to a label corresponding to the maximum confidence score. According to the method, the airflow state can be quickly and accurately judged, and the method is suitable for real-time airflow sensing and intelligent control in various edge computing scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of airflow state monitoring and edge computing, and particularly relates to an airflow sensor data processing method and system based on edge computing. BACKGROUND

[0002] At present, in the scenes of industrial pipeline operation, intelligent ventilation system, aviation wind tunnel experiment and environmental monitoring, real-time identification of airflow state is of great significance to ensure system safety, improve response speed and reduce energy consumption. Airflow state usually includes laminar flow, turbulent flow, vortex or unidentified state, and its change process is often accompanied by dramatic fluctuations in multidimensional characteristics such as flow rate, pressure and frequency.

[0003] In one prior art, raw data is collected by deploying airflow sensors in industrial sites, and characteristic parameters such as gas flow rate and temperature are extracted; then, a single characteristic change trend is used to realize preliminary identification of welding airflow disturbance based on a set threshold judgment rule.

[0004] However, the prior art mainly relies on fixed threshold judgment and centralized processing mechanism, and has problems such as low recognition accuracy, insufficient real-time performance, insufficient processing capacity of edge devices and lack of confidence evaluation mechanism, which cannot meet the requirements of efficient and accurate identification of airflow state in complex environments. Therefore, the prior art results in low accuracy of airflow sensor data processing. SUMMARY

[0005] The present application provides an airflow sensor data processing method and system based on edge computing to solve the problem of low accuracy of airflow sensor data processing caused by the prior art.

[0006] In a first aspect, to solve the above technical problems, the present application provides an airflow sensor data processing method and system based on edge computing to solve the problem of low accuracy of airflow sensor data processing caused by the prior art. obtaining a raw airflow signal stream and denoising to obtain a cleaned airflow signal set; performing dimensionality reduction processing on the cleaned airflow signal set and separating out a feature vector group to obtain a key feature vector; performing standardization processing on the key feature vector to obtain a standardized feature distribution range; comparing the standardized feature distribution range with a pre-established historical airflow state feature library to obtain a similarity calculation result; determining whether the similarity calculation result is lower than a preset laminar flow similarity threshold, and if so, determining a laminar flow state and generating a unique state identifier to obtain a laminar flow state label; If not, it is determined whether the similarity calculation result is lower than a preset turbulence similarity threshold value, if yes, it is determined as a turbulence state and a fluctuation frequency parameter of a feature vector is extracted, and a turbulence fluctuation feature set is generated, If not, according to the turbulence fluctuation feature set, it is determined whether the fluctuation frequency parameter meets a vortex condition, if yes, it is determined as a vortex state, and a corresponding vortex state classification identifier is generated, if not, the current state is marked as an unidentified state; A confidence score of the judgment result is calculated, and a gas flow state classification result is determined according to a label corresponding to a maximum value of the confidence score.

[0007] In an optional implementation, the original gas flow signal stream is acquired and denoised to obtain a cleaned gas flow signal set, including: The original gas flow signal stream is acquired in real time to obtain an original signal data set; The original signal data set is subjected to low-pass filtering to obtain the cleaned gas flow signal set.

[0008] In an optional implementation, the cleaned gas flow signal set is subjected to dimension reduction processing and a feature vector group is separated to obtain a key feature vector, including: The cleaned gas flow signal set is subjected to dimension reduction calculation to obtain a dimension reduction result data; The dimension reduction result data is subjected to feature decomposition to obtain the feature vector group; The variance contribution rate of the feature vector group is subjected to sorting processing to generate a feature vector set; For the feature vector set, a key feature vector is obtained by iteratively comparing the contribution rate of the remaining feature vectors.

[0009] In an optional implementation, the key feature vector is subjected to standardization processing to obtain a standardized feature distribution range, including: The key feature vector is integrated and unified to obtain a multi-dimensional data structure; The multi-dimensional data structure is subjected to distribution interval detection, and an abnormal value is filtered according to a preset abnormality detection threshold value to obtain a feature data structure; The dimensional data distribution of the feature data structure is subjected to interval limitation to generate a preliminary data distribution range; Data that exceeds a preset distribution range in the data distribution range is subjected to linear transformation adjustment to obtain the standardized feature distribution range.

[0010] In an optional implementation, the standardized feature distribution range is compared with a pre-established historical gas flow state feature library to obtain a similarity calculation result, including: Calculate the distance between the normalized feature distribution range and each feature vector in the historical airflow state feature library, and determine the candidate feature vector by combining the preset distance threshold range to obtain a candidate feature set; Extract the feature vector that meets the preset feature distribution interval threshold from the candidate feature set to form a main feature matching set; Perform matching quantization calculation on the feature vector in the main feature matching set and the normalized feature distribution range to obtain a final similarity calculation result.

[0011] In an optional implementation, the judgment whether the similarity calculation result is lower than a preset laminar flow similarity threshold includes: If the distance between the similarity calculation result and the laminar flow feature is lower than the preset laminar flow similarity threshold, the laminar flow state is determined. A unique state identifier is generated by a hash function according to a current timestamp; the state identifier is composed of the timestamp and the state type; The unique state identifier is associated with the laminar flow state, and the association information is recorded in a preset state database to obtain the laminar flow state label.

[0012] In an optional implementation, the judgment whether the similarity calculation result is lower than a preset turbulent flow similarity threshold includes: If the distance between the similarity calculation result and the turbulent flow feature is lower than the preset turbulent flow similarity threshold, the turbulent flow state is determined, and a corresponding turbulent flow state label is generated. According to the turbulent flow state label, the feature vector in the turbulent flow state is subjected to fluctuation frequency parameter extraction to generate a turbulent flow fluctuation feature set.

[0013] In an optional implementation, the judgment whether the fluctuation frequency parameter meets the vortex condition according to the turbulent flow fluctuation feature set includes: The fluctuation frequency of the turbulent flow fluctuation feature set is obtained and subjected to frequency analysis to obtain the distribution characteristic value. The distribution characteristic value that meets the preset periodic feature threshold is determined to have a periodic feature, and a corresponding periodic feature identifier is generated. If the periodic feature identifier is consistent with at least one condition in a pre-established vortex condition database, the vortex state is determined, and a vortex state label is generated. The vortex state label is subjected to standardization processing to obtain a vortex state classification identifier. In an alternative embodiment, the state of the airflow is classified according to a confidence score of the similarity of the judgment result, and the label corresponding to the maximum confidence score is determined as the classification result of the state of the airflow, comprising: obtaining the judgment result; the judgment result comprises the laminar flow state label, the turbulent flow state label, the vortex state label or the unidentified state; performing cosine similarity calculation on the judgment result and the feature vector in the preset template database to obtain a similarity score; performing weighted calculation on the similarity score combined with the preset weight of each label to obtain a corresponding confidence score; sorting the confidence scores, and determining the state label corresponding to the maximum value as the final classification result of the state of the airflow.

[0014] In a second aspect, the present application provides an airflow sensor data processing system based on edge computing, comprising: an original signal acquisition module for acquiring an original airflow signal stream and performing denoising processing to obtain a cleaned airflow signal set; a dimension reduction processing module for performing dimension reduction processing on the cleaned airflow signal set and separating out a feature vector group to obtain a key feature vector; a feature standardization module for performing standardization processing on the key feature vector to obtain a standardized feature distribution range; a similarity calculation module for comparing the standardized feature distribution range with a pre-established historical airflow state feature library to obtain a similarity calculation result; a laminar flow determination module for determining whether the similarity calculation result is lower than a preset laminar flow similarity threshold value, and if so, determining the state as a laminar flow state and generating a unique state identifier to obtain a laminar flow state label; a turbulent flow determination module for determining whether the similarity calculation result is lower than a preset turbulent flow similarity threshold value if not, and if so, determining the state as a turbulent flow state and extracting a fluctuation frequency parameter of the feature vector to generate a turbulent flow fluctuation feature set, a vortex determination module for determining whether the fluctuation frequency parameter meets a vortex condition according to the turbulent flow fluctuation feature set if not, and if so, determining the state as a vortex state and generating a corresponding vortex state classification identifier, and if not, marking the current state as an unidentified state; a state confirmation module for calculating a confidence score of the similarity of the judgment result, and determining the classification result of the state of the airflow according to the label corresponding to the maximum confidence score.

[0015] In a third aspect, the present application also provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the air flow sensor data processing method according to any one of the preceding aspects when executing the computer program.

[0016] In a fourth aspect, the present application also provides a computer readable storage medium comprising a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute the air flow sensor data processing method according to any one of the preceding aspects when the computer program runs.

[0017] Compared with the prior art, the present application has the following beneficial effects: (1) The present application is based on an edge computing architecture, which can complete signal preprocessing and feature extraction in real time at the data acquisition end, avoiding the problem of uploading all data to the cloud for processing, significantly reducing system delay and data transmission pressure, and improving overall response efficiency.

[0018] (2) The present application introduces multi-dimensional feature vector modeling and principal component analysis dimension reduction technology, and compares with the historical state feature library to realize accurate discrimination of typical air flow states such as laminar flow, turbulent flow, and vortex, and improve the adaptability and discrimination ability of the system to complex air flow patterns.

[0019] (3) The present application uses similarity matching and confidence score mechanism to output confidence scores for each type of state, improving the stability and interpretability of the recognition process. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a flowchart of an air flow sensor data processing method based on edge computing provided by the first embodiment of the present application; Figure 2 is a structural schematic diagram of an air flow sensor data processing system based on edge computing provided by the second embodiment of the present application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0022] With reference to Figure 1 The first embodiment of the present application provides an air flow sensor data processing method based on edge computing, comprising the following steps: S101, acquire an original airflow signal stream and denoise to obtain a cleaned airflow signal set; S102, perform dimensionality reduction processing on the cleaned airflow signal set and separate out a feature vector group to obtain a key feature vector; S103, perform standardization processing on the key feature vector to obtain a standardized feature distribution range; S104, compare the standardized feature distribution range with a pre-established historical airflow state feature library to obtain a similarity calculation result; S105, determine whether the similarity calculation result is lower than a preset laminar flow similarity threshold value, if yes, determine a laminar flow state and generate a unique state identifier to obtain a laminar flow state label; S106, if not, determine whether the similarity calculation result is lower than a preset turbulent flow similarity threshold value, if yes, determine a turbulent flow state and extract a fluctuation frequency parameter of the feature vector to generate a turbulent flow fluctuation feature set, S107, if not, determine whether the fluctuation frequency parameter meets a vortex condition according to the turbulent flow fluctuation feature set, if yes, determine a vortex state and generate a corresponding vortex state classification identifier, if not, mark the current state as an unidentified state; S108, calculate a confidence score of the similarity of the determination result, and determine an airflow state classification result according to the label corresponding to the maximum confidence score.

[0023] In step S101, an original airflow signal stream is acquired and denoised to obtain a cleaned airflow signal set, including: Step S1011, real-time acquisition of an original airflow signal stream to obtain an original signal data set; Step S1012, low-pass filtering of the original signal data set to obtain a cleaned airflow signal set.

[0024] In step S1011, an original airflow signal stream is acquired in real time to obtain an original signal data set.

[0025] It should be noted that the real-time acquisition of the original airflow signal stream includes using a lightweight acquisition module deployed on an edge node to obtain continuously changing airflow state signals in the measured area from a micro airflow sensor, and performing millisecond-level high-frequency sampling on the airflow signals. The airflow signal can be a one-dimensional or multi-dimensional time series signal stream with obvious frequency fluctuation characteristics. The edge node, as a local computing unit close to the airflow source, has preliminary processing and caching capabilities, and the lightweight acquisition module integrated therein can stably complete high-precision, low-latency data acquisition process, thereby constructing a complete structure of the original signal data set, providing accurate and reliable basic data source for subsequent data preprocessing and analysis.

[0026] In an implementation manner, the micro air flow sensor is constructed based on thermal, capacitive or MEMS technology, and can sense characteristic quantities such as air flow speed, direction or pressure in a target environment in real time; the lightweight collection module includes an analog-to-digital conversion unit, a buffer storage unit and a data forwarding module, and has a sampling capacity of not less than 100 Hz, and is suitable for edge deployment scenarios such as industrial pipeline monitoring and heating, ventilation and air conditioning systems.

[0027] In step S1012, the original signal data set is low-pass filtered to obtain a cleaned air flow signal set.

[0028] It should be noted that the low-pass filtering of the original signal data set includes setting a preset cutoff frequency for each frequency component in the original signal data set. When the frequency is higher than the cutoff frequency, the corresponding component is considered as high-frequency noise and is filtered out, and only the effective component below the cutoff frequency is retained, so as to enhance the stability and expression accuracy of the air flow signal. The processing result is the cleaned air flow signal set.

[0029] In an implementation manner, the cutoff frequency can be set to 10 Hz, that is, the frequency components higher than 10 Hz (such as fluctuations above 20 Hz caused by pipeline vibration and electromagnetic interference) are filtered as noise, and only the low-frequency air flow components in the interval of 0-10 Hz are retained, to finally form a smooth and high-fidelity cleaned air flow signal set.

[0030] In step S102, the cleaned air flow signal set is processed by dimension reduction, and a feature vector group is separated to obtain a key feature vector, including: Step S1021, performing dimension reduction calculation on the cleaned air flow signal set to obtain dimension reduction result data; Step S1022, performing feature decomposition on the dimension reduction result data to obtain a feature vector group; Step S1023, performing sorting processing on the variance contribution rate of the feature vector group to generate a feature vector set; Step S1024, for the feature vector set, the contribution rate of the remaining feature vectors is obtained by iterative comparison to obtain a key feature vector.

[0031] In step S1021, the cleaned air flow signal set is processed by principal component analysis dimension reduction to obtain dimension reduction result data.

[0032] It should be noted that the principal component analysis dimension reduction processing of the cleaning airflow signal set includes constructing a covariance matrix and calculating its eigenvalues and eigenvectors, performing linear transformation on the original signal set, mapping the data to a new coordinate system, so that the data variance of each principal component axis direction is maximum and orthogonal to each other, thereby compressing the data dimension while retaining as much information as possible. In order to improve the response efficiency, the time consumption of dimension reduction calculation is controlled in the millisecond level by controlling the data input size and adjusting the algorithm parameters, and then the dimension reduction result data meeting the real-time requirement is obtained. At the same time, according to the variance contribution rate, the principal components with cumulative variance contribution rate not less than 90% are selected, and the number of signal dimensions is determined.

[0033] In an implementation manner, the original cleaning airflow signal set contains about 100 feature dimensions. After principal component analysis processing, the system automatically selects the first several principal components to explain more than 90% of the cumulative variance information, for example, 10 principal component dimensions are finally retained, that is, the data dimension reduction process from 100 dimensions to 10 dimensions is completed. The dimension reduction result data provides compressed representation for subsequent feature vector extraction, and has higher representativeness and discriminability.

[0034] In step S1022, the dimension reduction result data is subjected to feature decomposition to obtain a feature vector group.

[0035] It should be noted that the feature decomposition of the dimension reduction result data includes performing vector space reconstruction and linear factor decomposition operation on the dimension reduction result data after principal component analysis processing, extracting a group of statistical characteristic parameters with maximum information in the principal component space, and forming a feature vector group. The feature vector group significantly reduces the original dimension redundancy while keeping the main trend of data change, which helps to improve the accuracy and efficiency of subsequent state recognition and comparison.

[0036] In an implementation manner, the variance contribution rate threshold is set to 90%, the first n principal component dimensions are retained from the dimension reduction result data, and a plurality of representative feature vectors are generated based on the physical meaning and statistical interpretation of each principal component. For example, the first principal component representing the flow rate change in the original signal can directly reflect the airflow disturbance intensity, and the second principal component can reflect the periodic change degree. The finally formed feature vector group contains not only the amplitude characteristics of the airflow signal, but also the frequency, fluctuation mode and other multi-dimensional attributes, which can provide a comprehensive feature basis for airflow state classification and model matching.

[0037] In step S1023, the variance contribution rate of the feature vector group is sorted to generate a feature vector set.

[0038] It should be noted that the variance contribution rate sorting process of the feature vector set includes calculating the contribution degree of each feature vector to the variance of the original airflow signal data, sorting according to the contribution rate from high to low to form a sorted feature vector set. In the sorting process, a preset threshold of the contribution rate can be set to eliminate noise vectors and low-efficiency features, thereby improving the accuracy and efficiency of subsequent modeling.

[0039] In an implementation manner, the preliminary feature vector set obtained after dimension reduction by principal component analysis is taken as a processing object. By statistically analyzing the variance contribution rate of each vector, vectors with higher contribution rates are preferentially retained. For example, when the contribution rate of a certain vector is 20% and the contribution rate of another vector is only 1%, the former is preferentially arranged. This sorting helps to focus on the dominant change mode in the signal, such as the periodic fluctuation of wind speed or air pressure, thereby generating a representative feature vector set.

[0040] In step S1024, for the feature vector set, the contribution rates of the remaining feature vectors are iteratively compared to obtain key feature vectors.

[0041] It should be noted that the comparison of the contribution rates of the remaining feature vectors includes: first, a preset threshold is set according to the variance contribution rate of each vector in the preliminary sorted feature vector set; when the contribution rate of a certain feature vector is lower than the threshold, it is considered that the vector has weak ability to explain the overall data, and it is removed from the set; then, the cumulative contribution rate of the remaining feature vectors is compared again, if the overall contribution rate does not reach the specified retention standard, the feature set is reconstructed by adjusting the number of principal components, updating the retention dimension, and the like, until the contribution rate threshold and the information retention requirement are met, thereby obtaining a key feature vector set.

[0042] In step S103, the key feature vectors are standardized to obtain a standardized feature distribution range, including: In step S1031, the key feature vectors are integrated and unified to obtain a multi-dimensional data structure. In step S1032, the multi-dimensional data structure is detected for distribution interval, and abnormal values are filtered according to a preset abnormality detection threshold to obtain a feature data structure. In step S1033, the interval of the data distribution of each dimension of the feature data structure is limited to generate a preliminary data distribution range. In step S1034, data outside the preset distribution range in the data distribution range is linearly transformed and adjusted to obtain a standardized feature distribution range.

[0043] In step S1031, the key feature vectors are integrated and unified to obtain a multi-dimensional data structure.

[0044] It should be noted that the key feature vector is integrated and unified, including reconstructing multiple key feature vectors in the dimension structure, structurally arranging according to the physical meaning of each feature component, and generating an array data framework with unified dimension mapping relationship. In this processing process, distribution interval detection is performed on each feature dimension, and a preset threshold range is set. If the feature value in a certain dimension exceeds the reasonable range, it is considered abnormal and is filtered out to eliminate invalid or noise data, so as to ensure that the integrated feature array structure meets the subsequent calculation requirements in terms of dimension consistency and numerical reasonableness.

[0045] In an implementation manner, assuming that the original feature vector group contains 10 dimensions, each dimension represents a typical feature of the airflow signal (such as wind speed change, direction fluctuation, etc.), the array conversion module can be used to uniformly construct a 10-dimensional structured array as the input basis for subsequent standardization and error verification.

[0046] In step S1032, the multi-dimensional data structure is subjected to distribution interval detection, and abnormal values are filtered according to a preset abnormality detection threshold to obtain a feature data structure.

[0047] It should be noted that the multi-dimensional data structure is subjected to distribution interval detection, including: for each feature dimension in the multi-dimensional data structure, a corresponding reasonable value interval is set as an abnormality detection threshold; when a data point in a certain dimension exceeds the corresponding preset threshold range, it is considered as an abnormal value and is removed. In this way, the stability and accuracy of the feature data structure can be effectively improved, providing a data basis for subsequent standardization processing.

[0048] In an implementation manner, a specific abnormality detection interval can be set for each feature dimension based on the physical characteristics of the airflow signal. For example, for the wind speed change dimension, the reasonable interval can be set to -10 m / s to +10 m / s. When a data point exceeds this interval range, it is determined as noise or abnormal input and is removed.

[0049] In step S1033, the interval of the dimension data distribution of the feature data structure is limited to generate a preliminary data distribution range.

[0050] It should be noted that the interval of the dimension data distribution of the feature data structure is limited, including: interval limitation judgment is performed on the data distribution of each dimension in the feature data structure. If it is found that the data distribution range of a certain dimension exceeds the preset threshold interval, the data of the dimension is adjusted by using a linear transformation method to be normalized to the target interval. Through the above processing, the feature data of each dimension has a consistent scale standard, forming a preliminary standardized data distribution range, which is convenient for subsequent reconstruction error calculation and distribution verification operations.

[0051] In an implementation, if the original data distribution range of a certain feature dimension is -20 to 30, and the set standardization target interval is -1 to 1, a linear transformation is used to adjust the original data to the target interval. In this process, the relative proportion of each data point is kept unchanged, avoiding data distortion caused by scaling, thereby generating a standardized feature data structure with consistent scale characteristics.

[0052] In step S1034, the data outside the preset distribution range in the data distribution range is adjusted by linear transformation to obtain a standardized feature distribution range.

[0053] It should be noted that adjusting the data outside the preset distribution range in the data distribution range by linear transformation includes: for data outside the preset interval, the distribution range is compressed or expanded by a linear function mapping method to fall within the set standardized interval; this transformation process realizes the data consistency between feature dimensions on the basis of maintaining the relative relationship of the original data, thereby improving the comparability and adaptability of the overall feature array.

[0054] In an implementation, assuming that the error between the reduced dimension data and the original normalized array is 5%, and the preset threshold is 3%, the dimension reduction parameters need to be fine-tuned, such as increasing the number of principal components retained from 8 to 9 to reduce the error. If verification finds that the distribution of a certain dimension deviates from the standard range by 10%, the deviation part needs to be recorded and marked, and the final verification result is obtained by comparison.

[0055] In step S104, the standardized feature distribution range is compared with the pre-established historical air flow state feature library to obtain a similarity calculation result, including: Step S1041, the distance between the standardized feature distribution range and each feature vector in the historical air flow state feature library is calculated, and a candidate feature vector is determined in combination with a preset distance threshold range to obtain a candidate feature set; Step S1042, a feature vector that meets a preset feature distribution interval threshold is extracted from the candidate feature set to form a main feature matching set; Step S1043, the feature vectors in the main feature matching set are matched with the standardized feature distribution range for quantification calculation to obtain a final similarity calculation result.

[0056] In step S1041, the distance between the standardized feature distribution range and each feature vector in the historical air flow state feature library is calculated, and a candidate feature vector is determined in combination with a preset distance threshold range to obtain a candidate feature set.

[0057] It should be noted that the distance calculation method adopts the Euclidean distance to measure the difference between the standardized features and the historical features, and the distance threshold is used to limit the similarity range of effective matching. If the distance between a certain feature vector and the standardized feature is less than the preset threshold, it is considered as a candidate matching object and is included in the candidate feature set.

[0058] In an implementation manner, for the distance calculation of the standardized feature distribution range and the airflow feature data set, the Euclidean distance calculation can be adopted for processing. Assuming that the value of a certain dimension of the standardized feature is 2.5, the value of a certain dimension of a feature vector in the data set is 3.0, and another is 1.5, it is found through the Euclidean distance calculation that the distance with the vector of 1.5 is closer. If the preset threshold is 1.0, the feature vector with a distance less than the value will be marked as a candidate matching feature and added to the candidate set. This method can quickly screen out historical data similar to the current airflow feature, providing a data basis for subsequent comparison and analysis.

[0059] In step S1042, a feature vector meeting a preset feature distribution interval threshold is extracted from the candidate feature set to form a main feature matching set. It should be noted that the extraction of the feature vector meeting the preset feature distribution interval threshold from the candidate feature set includes: performing distribution range comparison on each dimension of each feature vector in the candidate feature set. If the vector meets the preset interval limitation condition on all key dimensions, it is included in the main feature matching set, and the data points deviating from the interval are excluded to improve the matching accuracy.

[0060] In an implementation manner, a reasonable distribution range can be set. For example, if the value of a certain dimension in the candidate feature set is distributed between-5 and 5, and the preset threshold range is-3 to 3, only the feature vector falling within the range is retained. Through this screening method, abnormal or edge samples can be excluded, ensuring that the main feature matching set formed is closer to the actual distribution state of the target standardized feature, and improving the accuracy of the comparison and analysis.

[0061] In step S1043, the feature vectors in the main feature matching set are matched with the standardized feature distribution range for matching quantization calculation to obtain a final similarity calculation result.

[0062] It should be noted that the matching quantization calculation of the feature vectors in the main feature matching set with the standardized feature distribution range includes: according to the difference between the feature values of each dimension, the similarity degree reflected by the Euclidean distance is used to score the similarity degree between each feature vector and the standardized feature. The higher the score result is, the more similar it is to the current airflow feature, which can be used as a reference basis for the target state.

[0063] In an implementation, it is assumed that the main feature matching set contains 3 candidate feature vectors, and the matching degree values thereof are calculated by similarity to be 0.9, 0.6 and 0.5 respectively, wherein the vector with the score of 0.9 is selected as the final best matching vector because it is closest to the normalized feature distribution. The result can be used for subsequent airflow state judgment and classification.

[0064] In step S105, it is judged whether the similarity calculation result is lower than the preset laminar flow similarity threshold value, if yes, it is determined as the laminar flow state and a unique state identifier is generated, to obtain the laminar flow state label, including: In step S1051, if the similarity calculation result is lower than the preset laminar flow similarity threshold value, it is determined as the laminar flow state. In step S1052, a unique state identifier is generated by a hash function according to the current timestamp; the state identifier is composed of the timestamp and the state type; In step S1053, the unique state identifier is associated with the laminar flow state, and the association information is recorded into the preset state database to obtain the laminar flow state label.

[0065] In step S1051, if the similarity calculation result is lower than the preset laminar flow similarity threshold value, it is determined as the laminar flow state.

[0066] It should be noted that the similarity calculation result is lower than the preset laminar flow similarity threshold value, including comparing the current normalized feature vector with the laminar flow feature template constructed based on historical data, measuring the difference between the two by the Euclidean distance calculation method, if the distance value calculated is lower than the set threshold value (such as 0.15), it is determined that the current airflow state is highly similar to the laminar flow feature, so as to judge it as the laminar flow state and generate the laminar flow state flag, and the judgment result will be an important basis for subsequent state identifier generation and storage.

[0067] In an implementation, the laminar flow feature template is a multi-dimensional standard vector, for example, [0.60, 0.40, 0.55], and the current collected and normalized feature vector is [0.65, 0.45, 0.50]. By Euclidean distance calculation, the distance result of the two is 0.10, which is lower than the set threshold value 0.15, so it is determined as the laminar flow state, and the system generates the "laminar flow" state flag for subsequent state identifier construction and storage.

[0068] In step S1052, a unique state identifier is generated by a hash function according to the current timestamp; the state identifier is composed of the timestamp and the state type; It should be noted that the unique state identifier generated by the hash function includes: combining the current system timestamp with the aforementioned state type (such as "laminar flow"), using a hash function or string concatenation method to form a unique state identifier, ensuring that the same state at different times has distinguishability and traceability.

[0069] In step S1053, the unique state identifier is associated with the laminar flow state, and the association information is recorded in the preset state database to obtain the laminar flow state label.

[0070] It should be noted that the unique state identifier is associated with the laminar flow state and recorded in the state database, including: binding the unique state identifier generated by the hash function with the label information representing the laminar flow state, and writing it into the preset state database through the database storage module, to realize the persistent recording and subsequent calling of the state information. The state database is preconfigured with a field structure for storing airflow state monitoring records, including at least "state identifier" and "state label" fields.

[0071] In an implementation manner, the database is a relational database structure, and the state identifier such as "20250718121400_Laminar" and the corresponding "laminar flow" label form a key-value pair form written in the database, supporting fast query function based on timestamp or state type. For example, the system can retrieve all laminar flow state records in a specified time period through an SQL statement, for subsequent analysis of airflow stability or equipment running state.

[0072] In step S106, if no, whether the similarity calculation result is lower than the preset turbulent flow similarity threshold, if yes, determine as turbulent flow state and extract the fluctuation frequency parameter of the feature vector, generate turbulent flow fluctuation feature set, including: Step S1061, if the similarity calculation result and the distance of turbulent flow feature are lower than the preset turbulent flow similarity threshold, determine as turbulent flow state, and generate the corresponding turbulent flow state label; Step S1062, according to the turbulent flow state label, the fluctuation frequency parameter of the feature vector under the turbulent flow state is extracted, and the turbulent flow fluctuation feature set is generated.

[0073] In step S1061, if the similarity calculation result and the distance of turbulent flow feature are lower than the preset turbulent flow similarity threshold, determine as turbulent flow state, and generate the corresponding turbulent flow state label.

[0074] It should be noted that the similarity calculation result and the distance of the turbulence feature are lower than the preset turbulence similarity threshold, including comparing the current feature vector with the standard feature template constructed based on the historical turbulence state, calculating the distance value (such as cosine distance, Euclidean distance, etc.) between the two, and comparing the distance with the preset turbulence threshold. If the distance value is less than the threshold, it is considered that the current feature is similar to the turbulence template, thereby meeting the standard of determining the turbulence state. The comparison process usually combines multiple dimensional features for quantitative analysis to ensure the accuracy and reliability of the judgment.

[0075] In an implementation manner, the similarity adopts the calculation of cosine similarity to perform the feature vector comparison. Assuming that the current input feature vector is A = [0.5, 0.3, 0.2], and the preset turbulence feature template vector is B = [0.6, 0.4, 0.1], the similarity is calculated as 0.95 through the cosine similarity formula. In order to facilitate the judgment of whether it belongs to the turbulence state, the “distance” can be defined as 1 minus the similarity, that is, the distance is 0.05. If the preset turbulence similarity threshold is 0.1, because 0.05 is lower than the threshold, it is determined that the current airflow is in the turbulence state, and the corresponding turbulence state label is automatically generated by the system for identification.

[0076] In step S1062, according to the turbulence state label, the feature vector in the turbulence state is subjected to fluctuation frequency parameter extraction to generate a turbulence fluctuation feature set.

[0077] It should be noted that the feature vector in the turbulence state is subjected to fluctuation frequency parameter extraction, including the frequency domain feature parameters such as the main frequency, the secondary frequency, and the power spectrum energy proportion extracted based on the time series signal, which are used to depict the disturbance frequency structure and the energy distribution in the turbulence state, and improve the quantitative analysis capability of the airflow state.

[0078] In an implementation manner, for the feature vector in the turbulence state, the original fluid velocity time series can be subjected to frequency domain analysis through fast Fourier transform (FFT), the power spectrum density is calculated, and the main frequency component is extracted. Assuming that the main frequency is 5 Hz, the secondary frequency is 10 Hz, and the corresponding energy proportions are 60% and 30% respectively, the structured turbulence fluctuation feature set is integrated, for example: {main frequency: 5 Hz, energy proportion: 60%; secondary frequency: 10 Hz, energy proportion: 30%}. The set can be further used for comparison with the historical turbulence mode to assist the turbulence judgment and the fluid regulation.

[0079] In step S107, if no, according to the turbulence fluctuation feature set, it is judged whether the fluctuation frequency parameter meets the vortex condition, if yes, it is judged as the vortex state, and the corresponding vortex state classification label is generated, if no, the current state is marked as an unidentified state, including: Step S1071, obtain the fluctuation frequency of the turbulence fluctuation feature set, and perform frequency analysis to obtain the distribution feature value; Step S1072, determine that the distribution feature value meeting the preset periodic feature threshold has periodicity feature, and generate a corresponding periodic feature identifier; Step S1073, if the periodic feature identifier is consistent with at least one condition in the pre-established vortex condition database, it is determined as a vortex state, and a vortex state label is generated. Step S1074, the vortex state label is standardized to obtain a vortex state classification identifier. In step S1071, the fluctuation frequency of the turbulence fluctuation feature set is obtained, and frequency analysis is performed to obtain the distribution feature value.

[0080] It should be noted that the fluctuation frequency of the turbulence fluctuation feature set and the frequency analysis include frequency domain decomposition and recording of the frequency parameters in the turbulence fluctuation feature set, extraction of main frequency components, amplitude variation and other feature information, so as to construct complete frequency distribution features.

[0081] In an implementation manner, the frequency analysis can use fast Fourier transform (FFT) to perform frequency spectrum decomposition on the collected velocity time series data, obtain frequency characteristics such as a main frequency of 10 Hz and a secondary frequency of 20 Hz, and estimate the energy distribution through power spectral density, such as a main frequency of 60% and a secondary frequency of 30%. The above frequency and energy characteristics constitute the distribution feature value, which provides a basis for subsequent vortex state judgment.

[0082] In step S1072, it is determined that the distribution feature value meeting the preset periodic feature threshold has periodicity feature, and a corresponding periodic feature identifier is generated.

[0083] It should be noted that the determination of the periodicity feature includes: comparing the frequency distribution feature value with the preset periodic feature threshold, if the amplitude ratio of a certain frequency component exceeds the preset threshold, it can be determined that the fluctuation has periodicity feature.

[0084] In an implementation manner, the preset periodic feature threshold is that the amplitude ratio of a certain frequency component exceeds 30%, and if the analysis result shows that the energy ratio of the 10 Hz frequency component is 40%, it is determined that it has periodicity feature, and a periodic feature identifier "periodic fluctuation-10Hz" is generated for subsequent vortex state judgment process.

[0085] In step S1073, if the periodic feature identifier is consistent with at least one condition in the pre-established vortex condition database, it is determined as a vortex state, and a vortex state label is generated.

[0086] It should be noted that the vortex condition database includes a plurality of historical identification rules constituting a matching template, and the template generally includes frequency range, amplitude threshold, energy proportion, and other condition information for characterizing the periodic characteristics of typical vortex states.

[0087] In an implementation manner, the periodic characteristic is identified as a main frequency of 10 Hz, an amplitude of 0.6 m / s, and an energy proportion of 45%; if there is a condition item “frequency range 5 Hz-15 Hz and amplitude greater than 0.5 m / s” in the vortex condition database, the current periodic characteristic identification satisfies the condition. The system judges that the state is a vortex state according to this, and generates a corresponding vortex state label, such as “vortex_10Hz”.

[0088] In step S1074, the vortex state label is standardized to obtain a vortex state classification identification.

[0089] It should be noted that the standardization processing refers to unified format conversion of the original vortex state label through a preset classification identification generation rule, so as to ensure consistency, comparability, and identifiability of the vortex label under different sources or different algorithm determinations. The classification identification generally includes vortex type (such as Karman vortex, shear vortex), frequency range, disturbance level, and other structure fields.

[0090] In an implementation manner, assuming that the vortex state label is “vortex_10Hz”, the system generates a standardized identification of “Karman vortex (medium intensity)” through a general classification tool (for example, based on a rule tree or model reasoning) combined with input flow rate, Reynolds number, and frequency amplitude and other additional parameters, which is used to further support scenario-based application analysis, such as device regulation, abnormal alarm, or vortex model matching.

[0091] In step S108, a confidence score of the similarity of the judgment result is calculated, and a gas flow state classification result is determined according to a label corresponding to a maximum value of the confidence score, including: Step S1081, obtaining the judgment result; the judgment result includes the laminar flow state label, the turbulent flow state label, the vortex state label, or an unidentified state; Step S1082, performing cosine similarity calculation on the judgment result and a feature vector in a preset template database to obtain a similarity score; Step S1083, performing weighted calculation on the similarity score combined with a preset weight of each label to obtain a corresponding confidence score; Step S1084, sorting the confidence scores, and determining a state label corresponding to a maximum value as a final gas flow state classification result.

[0092] In step S1081, the judgment result is obtained; the judgment result includes the laminar flow state label, the turbulent flow state label, the vortex flow state label, or the unidentified state.

[0093] It should be noted that the judgment result represents a preliminary classification and identification result of the current airflow state, which can be generated based on similarity analysis, distribution characteristic judgment, template matching, or other algorithm rules, and has a clear state identification function.

[0094] In an implementation manner, the airflow sensor is deployed in a specific test environment, such as a wind tunnel experimental platform, to collect airflow state data in real time under different working conditions. The sensor can obtain physical parameter information such as the speed and pressure of the airflow, and the collected raw data can be used as the input basis for subsequent feature extraction and state identification. For example, the flow speed collected at a certain moment is 5.2 m / s, and the pressure is 101.3 kPa.

[0095] Further, a plurality of key features including flow speed, pressure, and vorticity can be extracted from the raw data. Among them, the flow speed feature reflects the motion trend of the airflow, the pressure feature represents the energy state of the airflow, and the vorticity feature is used to describe the rotation intensity of the airflow. In an embodiment, the algorithm analyzes the vorticity of a specified region, and if the detection result shows that the vorticity value of the region is significantly higher than that of the surrounding region, for example, reaches 0.8 s⁻¹, it can be preliminarily judged that there may be a vortex structure in the region. The extraction process of the above features can provide an important basis for subsequent similarity calculation, state classification, and anomaly detection.

[0096] In step S1082, the judgment result is compared with the feature vectors in the preset template database to calculate the cosine similarity, and a similarity score is obtained.

[0097] It should be noted that the calculation method of the similarity score is based on the angle relationship between the feature vectors, that is, by calculating the cosine similarity, the similarity between the current judgment result and each airflow state template in the preset template database is measured. This method has the advantages of anti-amplitude interference and high calculation efficiency, and is especially suitable for multi-dimensional airflow feature comparison and analysis, which can effectively improve the accuracy and robustness of state identification in practical applications.

[0098] In an implementation, when comparing the result with the template database, the matching degree of the current sample feature vector with each known air flow state template in the database can be quantified by calculating the cosine similarity. Specifically, the standard feature templates of various typical air flow states, such as the laminar flow template, the turbulent flow template, and the vortex flow template, can be pre-stored in the template database. The system obtains the corresponding similarity scores by calculating the cosine similarity of the current sample feature vector with each template. For example, if the similarity score of the current sample with the vortex flow template is 0.9, and the similarity scores with the turbulent flow template and the laminar flow template are 0.5 and 0.3 respectively, it is preliminarily indicated that the sample is closer to the vortex flow state. This calculation method can efficiently identify the air flow state category, and provide basic support for subsequent confidence assessment and classification decision.

[0099] In step S1083, the similarity scores are combined with the preset weights of each label for weighted calculation to obtain the corresponding confidence scores.

[0100] It should be noted that different feature dimensions have different degrees of influence in air flow state classification judgment, and the key features should be highlighted through a weighting mechanism to improve the accuracy of the judgment result. For example, the vorticity feature usually has stronger discriminative power for identifying the vortex state, and therefore can be given a higher weight in the weighted calculation.

[0101] In an implementation, the similarity scores are processed by using a support vector machine classification model or an equivalent confidence estimation method. Assuming that the weight of the vorticity feature in the model is set to 0.6, and the weights of the flow velocity and the pressure are 0.3 and 0.1 respectively, the total confidence score obtained after weighting is 0.85. If this value is higher than a set confidence threshold (such as 0.7), it indicates that the current judgment result has a high credibility, and the corresponding air flow state is preliminarily confirmed as the vortex state.

[0102] In step S1084, the confidence scores are sorted, and the state label corresponding to the maximum value is determined as the final air flow state classification result.

[0103] It should be noted that the state label includes but is not limited to the laminar flow state label, the turbulent flow state label, and the vortex flow state label. By comparing the confidence scores corresponding to different states, the most representative state label can be effectively selected, thereby improving the accuracy and stability of the classification result.

[0104] In an implementation, if the confidence scores of multiple state labels are similar, auxiliary features (such as the instantaneous Reynolds number, the flow velocity fluctuation rate, etc.) can be further introduced for secondary determination to improve the boundary state recognition ability and ensure that the output classification result has a high credibility and actual engineering applicability.

[0105] In summary, the application provides a kind of airflow sensor data processing method based on edge computing architecture, covers original signal acquisition and denoising, feature dimension reduction and extraction, standardization processing and error verification, feature comparison and state recognition, state identification generation and record etc.Key processes.The method can complete the rapid determination and intelligent classification of airflow state on the edge side, especially suitable for including intelligent ventilation regulation, industrial fluid detection, air power monitoring etc.The application scene with higher real-time and stability requirements.

[0106] It should be noted that the above embodiments are only specific exemplary descriptions of the present application, and those skilled in the art can make equivalent substitutions or modifications to the embodiments without departing from the spirit and essence of the present application, and these equivalent forms should be included in the protection scope of the present application.

[0107] Reference Figure 2 The second embodiment of the application provides a kind of airflow sensor data processing device based on edge computing, comprising: The original signal acquisition module is used to obtain the original airflow signal stream from the micro airflow sensor, and upload to the edge computing module through the lightweight acquisition module; The denoising processing module is used to filter and denoise the original airflow signal, remove high-frequency interference, and obtain a cleaned airflow signal set; The dimension reduction processing module is used to perform principal component analysis (PCA) or other dimension reduction method processing on the cleaned airflow signal set to obtain a feature vector group; The similarity calculation module is used to compare the feature vector with the template data in the historical airflow feature library and calculate the similarity result; The laminar flow determination module is used to determine the airflow as laminar flow state when the similarity result is lower than the preset laminar flow threshold, and generate the corresponding state label; The turbulent flow determination module is used to determine as turbulent flow state when the similarity result meets the turbulent flow characteristics, and extract frequency parameters to construct a turbulent flow fluctuation feature set; The vortex determination module is used to determine as vortex state when the turbulent flow fluctuation feature meets the periodic characteristic threshold condition, and generate the corresponding vortex state classification identification.

[0108] The state confirmation module is used to calculate the confidence score of the similarity result for the generated laminar flow, turbulent flow or vortex state classification identification, and confirm the state through the confidence of the comparison result, finally obtain the airflow state classification result.

[0109] It should be noted that the edge computing-based airflow sensor data processing apparatus provided by the embodiments of the present application is used to execute all process steps of the edge computing-based airflow sensor data processing method provided by the above embodiments, and the working principles and beneficial effects of the two are one-to-one correspondence, thus not being described again.

[0110] The embodiments of the present application further provide an electronic device. The electronic device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, for example, an edge computing-based airflow sensor data processing program. The processor implements the steps in the above various edge computing-based airflow sensor data processing method embodiments when executing the computer program, for example Figure 1 The steps S101 to S108 shown. Alternatively, the processor implements the functions of the modules / units in the above various device embodiments when executing the computer program, for example, the original signal acquisition module, the dimension reduction processing module, the similarity calculation module, the laminar flow determination module, the turbulent flow determination module, the vortex determination module, and the state confirmation module.

[0111] For example, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0112] The electronic device can be a desktop computer, a notebook, a palm computer, and a smart tablet, etc. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device, and can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, etc.

[0113] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the electronic device, and connects various parts of the electronic device through various interfaces and lines.

[0114] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.

[0115] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0116] It should be noted that the above-described device embodiments are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the connection relationship between the modules in the device embodiment provided by the present application indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0117] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are only for the specific embodiments of the present application and are not used to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. An edge computing-based air flow sensor data processing method, characterized by, The method comprises the following steps: acquiring an original airflow signal stream and denoising to obtain a cleaned airflow signal set; performing dimensionality reduction processing on the cleaned airflow signal set and separating a feature vector group to obtain a key feature vector; performing standardization processing on the key feature vector to obtain a standardized feature distribution range; comparing the standardized feature distribution range with a pre-established historical airflow state feature library to obtain a similarity calculation result; determining whether the similarity calculation result is lower than a preset laminar flow similarity threshold value, and if so, determining a laminar flow state and generating a unique state identifier to obtain a laminar flow state label; if not, determining whether the similarity calculation result is lower than a preset turbulent flow similarity threshold value, and if so, determining a turbulent flow state and extracting a fluctuation frequency parameter of the feature vector to generate a turbulent flow fluctuation feature set, if not, determining whether the fluctuation frequency parameter meets a vortex condition according to the turbulent flow fluctuation feature set, and if so, determining a vortex state and generating a corresponding vortex state classification identifier, and if not, marking the current state as an unidentified state; calculating a confidence score of the similarity of the determination result, and determining an airflow state classification result according to the label corresponding to the maximum confidence score. 2.The edge computing based air flow sensor data processing method according to claim 1, characterized in that, The method comprises the following steps: real-time acquisition of an original airflow signal stream to obtain an original signal data set; low-pass filtering of the original signal data set to obtain a cleaned airflow signal set. 3.The edge computing based air flow sensor data processing method and system of claim 1, wherein, The method comprises the following steps: dimensionality reduction calculation of the cleaned airflow signal set to obtain a dimensionality reduction result data; feature decomposition of the dimensionality reduction result data to obtain a feature vector group; sorting processing of the variance contribution rate of the feature vector group to generate a feature vector set; key feature vector extraction by iteratively comparing the contribution rates of the remaining feature vectors for the feature vector set. 4.The edge computing based air flow sensor data processing method of claim 1, wherein, The method comprises the following steps: integration and unification of the key feature vector to obtain a multi-dimensional data structure; distribution interval detection of the multi-dimensional data structure and filtering of abnormal values according to a preset abnormality detection threshold to obtain a feature data structure; interval limitation of the dimensional data distribution of the feature data structure to generate a preliminary data distribution range; linear transformation adjustment of data outside the preset distribution range in the data distribution range to obtain a standardized feature distribution range. 5.The edge computing based air flow sensor data processing method according to claim 1, wherein, The method comprises the following steps: distance calculation between the standardized feature distribution range and each feature vector in the historical airflow state feature library, combination with a preset distance threshold range to determine a candidate feature vector and obtain a candidate feature set; extraction of feature vectors meeting a preset feature distribution interval threshold from the candidate feature set to form a main feature matching set; The feature vector in the primary feature matching set is matched with the normalized feature distribution range for matching degree quantization calculation, and a final similarity calculation result is obtained. 6.The edge computing based air flow sensor data processing method according to claim 1, wherein, The similarity calculation result is judged whether it is lower than a preset laminar flow similarity threshold value, if yes, a laminar flow state is determined and a unique state identifier is generated, and a laminar flow state label is obtained, including: If the similarity calculation result is lower than the preset laminar flow similarity threshold value, the laminar flow state is determined. According to the current timestamp, a unique state identifier is generated by a hash function; the state identifier is composed of the timestamp and the state type; The unique state identifier is associated with the laminar flow state, and the association information is recorded in a preset state database, and the laminar flow state label is obtained. 7.The edge computing based air flow sensor data processing method according to claim 1, wherein, The similarity calculation result is judged whether it is lower than a preset turbulent flow similarity threshold value, if yes, a turbulent flow state is determined and a fluctuation frequency parameter of the feature vector is extracted, a turbulent flow fluctuation feature set is generated, including: If the similarity calculation result is lower than the preset turbulent flow similarity threshold value, the turbulent flow state is determined and a corresponding turbulent flow state label is generated; According to the turbulent flow state label, the fluctuation frequency parameter of the feature vector in the turbulent flow state is extracted, and a turbulent flow fluctuation feature set is generated. 8.The edge computing based air flow sensor data processing method of claim 1, wherein, According to the turbulent flow fluctuation feature set, it is judged whether the fluctuation frequency parameter meets the vortex condition, if yes, the vortex state is determined and a corresponding vortex state classification identifier is generated, including: The fluctuation frequency of the turbulent flow fluctuation feature set is obtained and frequency analysis is performed, and the distribution characteristic value is obtained; The distribution characteristic value meeting the preset periodic characteristic threshold value is determined to have periodic characteristics, and a corresponding periodic characteristic identifier is generated; If the periodic characteristic identifier is consistent with at least one condition in a pre-established vortex condition database, the vortex state is determined and a vortex state label is generated; The vortex state label is standardized to obtain a vortex state classification identifier. 9.The edge computing based air flow sensor data processing method of claim 1, wherein, The confidence score of the similarity calculation result is calculated, and the airflow state classification result is determined according to the label corresponding to the maximum value of the confidence score, including: The judgment result is obtained; the judgment result includes the laminar flow state label, the turbulent flow state label, the vortex state label or the unidentified state; The judgment result is calculated with the feature vector in the preset template database to obtain a similarity score; The similarity score is combined with the preset weight of each label for weighted calculation to obtain a corresponding confidence score; The confidence scores are sorted, and the state label corresponding to the maximum value is determined as the final airflow state classification result.

10. An edge computing based air flow sensor data processing system, characterized by, Including: An original signal acquisition module is configured to acquire an original airflow signal stream and perform denoising processing to obtain a cleaned airflow signal set; A dimension reduction processing module is configured to perform dimension reduction processing on the cleaned airflow signal set and separate out a feature vector group to obtain a key feature vector; A feature standardization module is configured to perform standardization processing on the key feature vector to obtain a normalized feature distribution range; The similarity calculation module is configured to compare the standardized feature distribution range with a pre-established historical airflow state feature library to obtain a similarity calculation result; The laminar flow determination module is configured to determine whether the similarity calculation result is lower than a pre-set laminar flow similarity threshold value, and if yes, determine a laminar flow state and generate a unique state identifier to obtain a laminar flow state label; The turbulent flow determination module is configured to determine whether the similarity calculation result is lower than a pre-set turbulent flow similarity threshold value if the similarity calculation result is not lower than the laminar flow similarity threshold value, and if yes, determine a turbulent flow state and extract a fluctuation frequency parameter of the feature vector to generate a turbulent flow fluctuation feature set, The vortex determination module is configured to determine whether the fluctuation frequency parameter meets a vortex condition according to the turbulent flow fluctuation feature set if the similarity calculation result is not lower than the turbulent flow similarity threshold value, and if yes, determine a vortex state and generate a corresponding vortex state classification identifier, and if not, mark the current state as an unidentified state; The state confirmation module is configured to calculate a confidence score of the determination result, and determine an airflow state classification result according to a label corresponding to a maximum value of the confidence score.

Images