Metal dust risk dynamic assessment and grading early warning method based on internet of things

Abstract

The application discloses a metal dust risk dynamic evaluation and grading early warning method based on an Internet of Things, and particularly relates to the technical field of metal dust monitoring; a plurality of types of sensor nodes are deployed in an industrial field to collect metal dust concentration, environmental parameters and equipment state information, after pre-processing by edge calculation, dynamic behavior characteristics of dust particles and abnormal characteristics of high-frequency signals are extracted, and based on the change trend and coupling relationship, a discrimination model is constructed to identify the reliability state of the sensor under different interference working conditions in real time; when potential misalignment risks of the sensor are detected, the system can automatically perform data correction or switch a redundant path to dynamically update the early warning result; the method effectively improves the stability of dust monitoring and the accuracy of early warning under complex working conditions, and significantly enhances the identification and prevention ability of the system to silent safety hazards.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metal dust monitoring, and particularly relates to a metal dust risk dynamic evaluation and hierarchical early warning method based on Internet of Things. BACKGROUND

[0002] The metal dust risk dynamic evaluation and hierarchical early warning refers to, for the metal dust generated in industrial production, continuously monitoring and analyzing risk factors such as generation amount, distribution, explosiveness, dynamically evaluating potential threats to safety and health, and dividing different early warning levels according to risk levels, so as to timely issue early warning signals, so that corresponding control and protection measures can be taken. This mechanism helps to realize early discovery and early intervention of dust risks, thereby improving the safety of the working environment and the accident prevention ability.

[0003] The prior art has the following deficiencies:

[0004] The existing technology generally ignores the interference problem of the charging effect of metal dust in a specific environment on the reliability of the sensor. In high-friction and high-speed processing scenarios, such as the grinding and conveying process of aluminum alloy or magnesium alloy, dust particles are easily charged due to frequent collision and peeling. The charged particles can form an electric field interference or micro-short circuit near the sensor probe, resulting in significant deviation of the dust concentration detection result, and even causing damage to the sensor module or signal loss. If the system lacks self-diagnosis and redundancy design, it is likely that no early warning will be issued during the high-risk period, thereby causing a "silent" safety hazard, which seriously threatens the safety and continuity of industrial production. SUMMARY

[0005] The purpose of the present application is to provide a metal dust risk dynamic evaluation and hierarchical early warning method based on Internet of Things to solve the problems in the background art.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a metal dust risk dynamic evaluation and hierarchical early warning method based on Internet of Things, comprising:

[0007] Deploying multiple types of sensor nodes in the target industrial area for real-time collection of metal dust concentration, environmental parameters and equipment running state;

[0008] Preprocessing the collected multi-source data through an edge computing device, and analyzing the preprocessed data using a risk evaluation model to output corresponding risk levels and hierarchical early warning information;

[0009] Extracting dust particle dynamic behavior related features and high-frequency signal abnormal features indirectly calculated from environmental charge disturbance from the preprocessed multi-source data;

[0010] Based on the dynamic behavior characteristics of dust particles and the changing trends and coupling relationships of high-frequency signal anomalies, a discriminant model is constructed to identify the reliability status of sensors under different interference conditions.

[0011] When the discrimination model identifies a potential risk of inaccuracy in the sensor output, it automatically performs dynamic correction on the multi-source data or activates redundant data paths, and updates the warning judgment results.

[0012] Preferably, anomalies in the instantaneous concentration change rate are generated after analyzing the anomalies in the dynamic behavior characteristics of dust particles. The method for obtaining these anomalies is as follows:

[0013] Suppose the sensor samples data at fixed intervals Δt, and the collected dust concentration sequence is C = {C1, C2, ..., C...} n}, where n is the total number of data points, and for each time point t i Calculate the rate of change of concentration before and after, the expression is: Among them: ICR i Let C be the instantaneous concentration change rate at the i-th sampling point. i Let be the dust concentration at the i-th sampling point;

[0014] Construct a statistical model of the rate of change within a sliding time window W, and calculate the average rate of change μ. ICR and standard deviation σ ICR Use the Z-score method to determine if the current rate of change is abnormal: Where: Z i The standardized score of the current rate of change; a preset Z-score anomaly detection threshold, if Z... i If the instantaneous concentration change rate is greater than or equal to the anomaly detection threshold, the corresponding instantaneous concentration change rate will be regarded as an anomaly value.

[0015] Preferably, after analyzing the number and amplitude changes of high-frequency transient spikes in the high-frequency signal anomaly characteristics indirectly inferred from environmental charge disturbances, high-frequency transient spike change values ​​are generated. The method for obtaining these values ​​is as follows:

[0016] Let the original time series acquired by the sensor be x(t). After normalizing the signal, the processed signal x(t) is subjected to improved empirical mode decomposition, which decomposes it into several intrinsic mode function sequences: Among them: IMF i r(t) is the i-th mode function, r(t) is the residual term, and m is the number of IMFs decomposed.

[0017] Based on the spectral energy distribution or experience, select the first k IMF components, denoted as: Applying the Hilbert transform to H(t) yields the corresponding analytic signal Z(t): in: A signal with the same imaginary part as H(t), where j is the imaginary unit, is defined with an amplitude threshold TA. A transient spike is considered to occur when Z(t) > TA. The change in the number of spikes ΔN is calculated. spike The expression is: ΔN spike =N cur -N ref N cur N represents the number of spikes detected within the current time period. ref To reference the average number of peaks over a historical period, calculate the average amplitude change ΔA. avg The expression is: Z ref This represents the reference average peak amplitude; the high-frequency transient peak change value is obtained by weighted average summation of the calculated changes in the number of peaks and the average amplitude.

[0018] Preferably, based on the changing trends and coupling relationships of the dynamic behavior characteristics of dust particles and the abnormal characteristics of high-frequency signals, a discriminant model is constructed to identify the reliability status of sensors under different interference conditions, specifically including:

[0019] Instantaneous concentration change anomalies and high-frequency transient peak change values ​​are converted into comprehensive feature vectors. These comprehensive feature vectors are used as input to a machine learning model. The machine learning model aims to predict the sensor reliability status analysis value label under different interference conditions for each set of comprehensive feature vectors. The training objective is to minimize the sum of prediction errors for the sensor reliability status analysis value labels under all different interference conditions. The machine learning model is trained until the sum of prediction errors converges, at which point the model training stops. The sensor reliability status analysis value under different interference conditions is determined based on the model output. The machine learning model is a multinomial regression model.

[0020] Preferably, the obtained sensor reliability status analysis values ​​under different interference conditions are compared with a predetermined threshold. If the sensor reliability status analysis value is greater than or equal to the predetermined threshold, the sensor is considered to be in a stable state, the signal is used as the input of the risk assessment model, and the system maintains normal operation. If the sensor reliability status analysis value is less than the predetermined threshold, it is determined that the sensor is currently affected by interference and there is a potential risk of inaccuracy.

[0021] Preferably, when the discrimination model identifies a potential risk of inaccuracy in the sensor output, it automatically performs dynamic correction on the multi-source data or activates redundant data paths, and updates the warning judgment result, specifically as follows:

[0022] When the sensor reliability state analysis value is less than a predetermined threshold value, it indicates that there is a potential misalignment risk in the sensor output, a conditional probability model is constructed for each data source, indicating its credibility to the true concentration C true under the current conditions: P(C true |D i )∝P(D i |C true )·P(C true );Wherein: D i is the concentration estimate from the first i data source, P(D i |C true ) is a sensor error model, and P(C true ) is the estimate or stationary prior of the last cycle;

[0023] The corrected concentration value C fused is calculated using the output of each data source and the weight: Wherein: is the Bayesian weight reflected based on the error variance of each data source, and q is the total number of data sources;

[0024] Replace the original data input of the main sensor with C fused ;

[0025] The corrected value is re-input into the risk assessment model to update the warning level of the current area.

[0026] In the above technical solution, the technical effects and advantages provided by the present application are as follows:

[0027] The present application overcomes the major safety hazards of sensor misalignment and system "silent" failure in existing technologies under specific working conditions such as high friction and high-speed machining, by introducing a metal dust charging effect interference recognition mechanism. By deploying multiple types of sensor nodes to collect metal dust and environmental data, combining edge computing and feature extraction technology, dynamic behavior features of dust particles and high-frequency signal anomaly features are extracted respectively, and a sensor reliability discrimination model is constructed, which significantly improves the system's perception ability and fault tolerance ability to abnormal working conditions, and realizes active identification and response to potential misalignment state.

[0028] Further, the present application uses a polynomial regression model to intelligently predict the reliability state of the sensor, and when a potential misalignment risk is identified, a Bayesian weighted data fusion algorithm is used to dynamically correct or switch redundant data paths for abnormal data, ensuring that the system can still output reliable warning results under the condition of data interference or local failure. This method has high adaptability, high stability and high accuracy, and is particularly suitable for industrial environments with high dust emission and high risk levels, significantly improving the robustness and safety control ability of the intelligent monitoring system. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0030] Figure 1 For the outline of the principle diagram of the present application.

[0031] Figure 2 For the mind map of the metal dust risk dynamic assessment and grading early warning method based on the Internet of Things. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0033] Embodiments, please refer to Figure 1 As shown in the drawings, the metal dust risk dynamic assessment and grading early warning method based on the Internet of Things described in the embodiments includes:

[0034] Deploying multiple types of sensor nodes inside the target industrial area for real-time collection of metal dust concentration, environmental parameters and equipment running state;

[0035] Preprocessing the collected multi-source data through an edge computing device, and analyzing the preprocessed data using a risk assessment model to output corresponding risk levels and grading early warning information;

[0036] Extracting dust particle dynamic behavior related features and high frequency signal abnormal features indirectly calculated from environmental charge disturbance from the preprocessed multi-source data;

[0037] Based on the change trend of the dust particle dynamic behavior related features and the high frequency signal abnormal features and their coupling relationship, a discriminant model for identifying the reliability state of the sensor under different interference conditions is constructed;

[0038] When the discriminant model identifies that there is a potential misalignment risk in the sensor output, automatically dynamically corrects the multi-source data or starts a redundant data path, and updates the early warning decision result.

[0039] In this embodiment, for a metal processing production workshop, a representative high-dust, high-temperature, and high-speed operation condition area is selected as the target industrial area, and a set of metal dust risk dynamic assessment and grading early warning system based on Internet of Things is deployed. The system mainly includes: a perception layer, an edge computing layer, a risk assessment layer, and an early warning response layer.

[0040] The following types of sensor nodes are laid out in the target area: metal dust concentration sensors (based on laser scattering method or charge induction type) for real-time detection of metal dust concentration in the air; temperature and humidity sensors for monitoring environmental temperature and relative humidity; wind speed and direction sensors for determining the influence of air flow on dust diffusion; oxygen concentration sensors for assessing explosion risk conditions; motor equipment state sensors such as vibration sensors and current sensors for determining whether the equipment is operating abnormally. All sensor nodes are connected to edge computing devices through wireless communication methods such as LoRa, Wi-Fi, or NB-IoT.

[0041] The edge computing device is deployed in the on-site electrical control cabinet and has basic data caching, processing, and analysis capabilities. The specific preprocessing process includes: data synchronization and timestamp correction to ensure consistency of multi-source data; noise filtering and data cleaning to eliminate short-term outliers, communication packet loss, and other interference; data format standardization to convert various sensor outputs into a unified format for subsequent analysis; and initial judgment of abnormal trends to locally mark short-term dust concentration fluctuations for early warning.

[0042] The processed data is uploaded to an industrial cloud platform or a local server and input into a trained risk assessment model. The model combines the following key parameters for risk judgment: current dust concentration value and its trend; current environmental humidity and temperature combination conditions; whether the oxygen concentration is close to the critical point of the three elements of dust explosion; equipment operating state and abnormal vibration amplitude; historical data comparison results and pattern deviation.

[0043] Based on the above inputs, the model outputs the risk level of the current work area, for example, divided into four levels (R1-R4), corresponding to normal, attention, alert, and severe states.

[0044] When the system determines that the current risk level reaches the preset threshold (such as R3 or R4), the grading early warning mechanism is automatically triggered, and the following operations are performed:

[0045] Start the audible and visual alarm device to issue a prompt on site; push alarm information to the management terminal APP and platform SMS module; automatically link the on-site dust removal system and spray dust suppression device to start; if the risk level reaches the severe level (R4), the system can be stopped and the risk area can be locked through the control interface.

[0046] The system deployed in this embodiment can realize real-time monitoring, dynamic assessment and graded response of metal dust risks, greatly improving the level of on-site safety management and control, and is particularly suitable for working environments with flammable metal dust such as aluminum, magnesium and titanium.

[0047] After preprocessing the multi-source data, the system further extracts key features to determine the reliability of the sensors under dust interference conditions.

[0048] Extraction of features related to the dynamic behavior of dust particles: Based on the concentration change curves of dust concentration sensors within a continuous time window, combined with wind speed sensor data, a multi-dimensional feature set describing the motion state of dust particles is constructed, including: instantaneous concentration change rate (ΔC / Δt); the offset angle between the particle drift direction and the wind direction vector; the coefficient of variation (CV) and local peak frequency (spike rate) of concentration change; and the standard deviation and autocorrelation of the concentration signal within a short period. These features can be used to identify whether dust behavior conforms to the laws of natural diffusion and airflow drive, and to assist in judging the physical rationality of the concentration signal.

[0049] High-frequency signal anomaly characteristics indirectly inferred from environmental charge disturbances: By analyzing the spectral characteristics of high-frequency components in the output signal of the dust concentration sensor, abnormal fluctuations caused by environmental electrostatic interference or charged dust clusters can be identified. Specifically, these include: the energy proportion of the signal power spectral density (PSD) in the high-frequency range; the number and amplitude of high-frequency transient spikes; non-periodic high-frequency oscillations that occur without obvious physical disturbances; and asynchronous disturbance characteristics that are independent of environmental humidity, temperature, and motor operating cycle.

[0050] Based on the aforementioned high-frequency characteristics, the system constructs a discrimination model by matching it with "static templates" under historical normal conditions, thereby indirectly inferring whether there is a risk of sensor error caused by metal dust charging in the current environment.

[0051] Anomalies in the instantaneous concentration change rate were generated by analyzing the relevant characteristics of the dynamic behavior of dust particles. The method for obtaining these anomalies is as follows:

[0052] Suppose the sensor samples data at fixed intervals Δt (in seconds), and the collected dust concentration sequence is C = {C1, C2, ..., C...}. n}, the unit is mg / m 3 , where n is the total number of data points. Basic filtering (such as moving average) is performed to remove high-frequency measurement noise.

[0053] For each time point t i Calculate the rate of change of concentration before and after, the expression is: Among them: ICR iThe instantaneous concentration change rate of the i-th sampling point, with the unit of mg / m 3 / s, C i The dust concentration of the i-th sampling point;

[0054] A change rate statistical model is constructed within a sliding time window W (such as 30 seconds or 60 seconds), and the average change rate μ ICR and the standard deviation σ ICR are calculated respectively; the Z-score method is used to determine whether the current change rate is abnormal: Where: Z i is the standardized score of the current change rate; the preset Z-score abnormality determination threshold (such as 2 or 2.5) can be dynamically adjusted according to the actual environment; if Z i is greater than or equal to the abnormality determination threshold, the corresponding instantaneous concentration change rate is taken as the instantaneous concentration change abnormal value.

[0055] If instantaneous concentration change abnormal values appear at a plurality of consecutive sampling points, the system can determine that the dust particle behavior is abnormal, such as sudden release or suspension cluster disturbance; this feature can be cross-verified with the high-frequency signal abnormality feature to determine whether there is charged dust interference; it can also be used to enhance the recognition ability of the dynamic risk model for explosive dust instantaneous accumulation.

[0056] After analyzing the number and amplitude changes of high-frequency transient spikes in the high-frequency signal abnormality feature indirectly calculated from environmental charge disturbance, a high-frequency transient spike change value is generated, and the acquisition method is:

[0057] Let the original time series collected by the sensor be x(t), which is a one-dimensional continuous signal; normalize the signal and apply a band-pass filter to remove the power frequency and low-frequency trend (such as wind speed slow change trend); the filtered signal is used as the input for subsequent analysis.

[0058] The processed signal x(t) is subjected to improved empirical mode decomposition, which is decomposed into a plurality of intrinsic mode function sequences: Where: IMF i (t) is the i-th mode function, r(t) is the residual term, and m is the number of decomposed IMFs.

[0059] According to the spectral energy distribution or experience, the first k IMFs are selected and denoted as: These components mainly reflect high-frequency spike disturbance signals.

[0060] The Hilbert transform is applied to H(t) to obtain the corresponding analytic signal Z(t): Where: The imaginary part signal with the same length as H(t), which suppresses the negative frequency part of the real part signal in the frequency domain, makes the result a single sideband spectrum complex signal, j is the imaginary unit, and is used to combine the Hilbert transform result as the imaginary part into the complex number space.

[0061] The amplitude threshold TA is defined, and when Z(t) > TA, it is considered that a transient spike occurs, and the spike number change ΔN is calculated spike , the expression is: ΔN spike = N cur -N ref ; N cur is the number of spikes detected in the current time period, and N ref is the average number of spikes in the reference historical time period; the average amplitude change ΔA avg is calculated, and the expression is: Z ref represents the reference average spike amplitude, which is obtained from the historical normal interval; the calculated spike number change ΔN spike and the average amplitude change ΔA avg are weighted and averaged to obtain the high-frequency transient spike change value.

[0062] The high-frequency transient spike change value as a key discriminant index can be used to judge whether the sensor has a misalignment or failure tendency under a specific working condition; when the high-frequency transient spike change value exceeds the preset threshold, a signal correction, self-diagnosis or alarm process can be triggered.

[0063] Based on the change trend of the related features of the dynamic behavior of dust particles and the abnormal features of high-frequency signals and their coupling relationship, a discriminant model for identifying the reliability state of the sensor under different interference working conditions is constructed, which specifically includes:

[0064] The instantaneous concentration change abnormal value and the high-frequency transient spike change value are converted into a comprehensive feature vector, the comprehensive feature vector is taken as the input of the machine learning model, the machine learning model takes the analysis value label of the sensor reliability state under different interference working conditions as the prediction target, and the minimum sum of the prediction error of the analysis value label of the sensor reliability state under all different interference working conditions is taken as the training target. The machine learning model is trained until the sum of the prediction error reaches convergence, and the model training is stopped. The sensor reliability state analysis value under different interference working conditions is determined according to the model output result, wherein the machine learning model is a polynomial regression model.

[0065] The sensor reliability state analysis value obtained under different interference conditions is compared with the predetermined threshold value. If the sensor reliability state analysis value is greater than or equal to the predetermined threshold value, it is considered that the sensor is currently in a stable state, and the signal can be used as the input of the risk assessment model, and the system maintains normal operation. If the sensor reliability state analysis value is less than the predetermined threshold value, it is judged that the sensor is currently affected by non-typical working condition interference such as charged dust, electromagnetic interference or dynamic drift, and there is a potential misalignment risk. The system will issue a diagnostic warning signal and trigger a backup sensor path or prompt maintenance operation.

[0066] When the discriminant model identifies that the sensor output has a potential misalignment risk, the multi-source data is automatically dynamically corrected or the redundant data path is started, and the warning judgment result is updated.

[0067] When the sensor reliability state analysis value is less than the predetermined threshold value, it indicates that the sensor output has a potential misalignment risk.

[0068] A conditional probability model is constructed for each data source to represent its credibility to the true concentration C true under the current condition: P(C true |D i ) ∝ P(D i |C true ) · P(C true ); Wherein: D i is the concentration estimate from the ith data source, P(D i |C true ) is the sensor error model, and P(C true ) is the estimate or stationary prior of the last period.

[0069] The output and weight of each data source are used to calculate the corrected concentration value C fused : Wherein: is the Bayesian weight based on the error variance of each data source, and q is the total number of data sources.

[0070] The original data input of the main sensor is replaced by C fused ;

[0071] The corrected value is re-input into the risk assessment model to update the warning level of the current area;

[0072] At the same time, record this event as a risk calculation corrected by the redundant path for subsequent backtracking or model retraining.

[0073] The above formulas are dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the latest true situation. The preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0074] The above-described embodiments can be implemented in part or in whole through software, hardware, firmware or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When loaded and executed by a computer, the computer instructions or computer programs cause the computer to perform all or part of the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0075] It should be understood that the term "and / or" used herein is only to describe an association relationship of associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after it, but it can also represent an "and / or" relationship, which can be understood in the context before and after it. Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0076] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for dynamic risk assessment and graded early warning of metal dust based on the Internet of Things, characterized in that: include: Deploy multiple types of sensor nodes within the target industrial area to collect real-time data on metal dust concentration, environmental parameters, and equipment operating status. The collected multi-source data is preprocessed by an edge computing device, and the preprocessed data is analyzed using a risk assessment model to output the corresponding risk level and graded early warning information. The dynamic behavior-related features of dust particles and the high-frequency signal anomaly features indirectly inferred from environmental charge disturbances were extracted from the preprocessed multi-source data. Based on the dynamic behavior characteristics of dust particles and the changing trends and coupling relationships of high-frequency signal anomalies, a discriminant model is constructed to identify the reliability status of sensors under different interference conditions. When the discrimination model identifies a potential risk of inaccuracy in the sensor output, it automatically performs dynamic correction on the multi-source data or activates redundant data paths, and updates the warning judgment results.

2. The method for dynamic risk assessment and graded early warning of metal dust based on the Internet of Things as described in claim 1, characterized in that: Anomalies in the instantaneous concentration change rate were generated by analyzing the relevant characteristics of the dynamic behavior of dust particles. The method for obtaining these anomalies is as follows: Suppose the sensor samples data at fixed intervals Δt, and the collected dust concentration sequence is C = {C1, C2, ..., C...} n }, where n is the total number of data points, and for each time point t i Calculate the rate of change of concentration before and after, the expression is: Among them: ICR i Let C be the instantaneous concentration change rate at the i-th sampling point. i Let be the dust concentration at the i-th sampling point; Construct a statistical model of the rate of change within a sliding time window W, and calculate the average rate of change μ. ICR and standard deviation σ ICR Use the Z-score method to determine if the current rate of change is abnormal: Among them: Z i The standardized score of the current rate of change; a preset Z-score anomaly detection threshold, if Z... i If the instantaneous concentration change rate is greater than or equal to the anomaly detection threshold, the corresponding instantaneous concentration change rate will be regarded as an anomaly value.

3. The method for dynamic risk assessment and graded early warning of metal dust based on the Internet of Things as described in claim 2, characterized in that: After analyzing the number and amplitude variations of high-frequency transient spikes in the high-frequency signal anomaly characteristics indirectly inferred from environmental charge disturbances, high-frequency transient spike variation values ​​are generated. The method for obtaining these values ​​is as follows: Let the original time series acquired by the sensor be x(t). After normalizing the signal, the processed signal x(t) is subjected to improved empirical mode decomposition, which decomposes it into several intrinsic mode function sequences: Among them: IMF i r(t) is the i-th mode function, r(t) is the residual term, and m is the number of IMFs decomposed. Based on the spectral energy distribution or experience, select the first k IMF components, denoted as: Applying the Hilbert transform to H(t) yields the corresponding analytic signal Z(t): in: A signal with the same imaginary part as H(t), where j is the imaginary unit, is defined with an amplitude threshold TA. A transient spike is considered to occur when Z(t) > TA. The change in the number of spikes ΔN is calculated. spike The expression is: ΔN spike =N cur -N ref N cur N represents the number of spikes detected within the current time period. ref To reference the average number of peaks over a historical period, calculate the average amplitude change ΔA. avg The expression is: Z ref This represents the reference average peak amplitude; the high-frequency transient peak change value is obtained by weighted average summation of the calculated changes in the number of peaks and the average amplitude.

4. The method for dynamic risk assessment and graded early warning of metal dust based on the Internet of Things as described in claim 3, characterized in that: Based on the dynamic behavior characteristics of dust particles and the changing trends and coupling relationships of high-frequency signal anomalies, a discriminant model is constructed to identify the reliability status of sensors under different interference conditions. Specifically, this includes: Instantaneous concentration change anomalies and high-frequency transient peak change values ​​are converted into comprehensive feature vectors. These comprehensive feature vectors are used as input to a machine learning model. The machine learning model aims to predict the sensor reliability status analysis value label under different interference conditions for each set of comprehensive feature vectors. The training objective is to minimize the sum of prediction errors for the sensor reliability status analysis value labels under all different interference conditions. The machine learning model is trained until the sum of prediction errors converges, at which point the model training stops. The sensor reliability status analysis value under different interference conditions is determined based on the model output. The machine learning model is a multinomial regression model.

5. The method for dynamic risk assessment and graded early warning of metal dust based on the Internet of Things according to claim 4, characterized in that: The obtained sensor reliability status analysis values ​​under different interference conditions are compared with a predetermined threshold. If the sensor reliability status analysis value is greater than or equal to the predetermined threshold, the sensor is considered to be in a stable state, the signal is used as the input of the risk assessment model, and the system maintains normal operation. If the sensor reliability status analysis value is less than the predetermined threshold, it is determined that the sensor is currently affected by interference and there is a potential risk of inaccuracy.

6. The method for dynamic risk assessment and graded early warning of metal dust based on the Internet of Things as described in claim 5, characterized in that: When the discrimination model identifies a potential risk of inaccuracy in the sensor output, it automatically performs dynamic correction on the multi-source data or activates redundant data paths, and updates the warning judgment result, specifically as follows: When the sensor reliability status analysis value is less than a predetermined threshold, it indicates a potential risk of inaccuracy in the sensor output. A conditional probability model is constructed for each data source to represent its accuracy relative to the true concentration C under current conditions. true Credibility: P(C) true |D i )∝P(D i |C true )·P(C true ); where: D i For the concentration estimate from the i-th data source, P(D) i |C true P(C) represents the sensor error model. true ) is an estimate or stationary prior of the previous period; The corrected concentration value C is calculated using the outputs and weights of each data source. fused : in: Based on the error variance of each data source The reflected Bayesian weights, where q is the total number of data sources; Use C fused Replace the raw data input of the main sensor; The corrected values ​​are re-entered into the risk assessment model to update the warning level for the current region.

Images