Fault early warning method for back corona vibrating electric dust remover based on all-working-condition monitoring

Abstract

The invention discloses a back corona rapping electric dust remover fault early warning method based on all-condition monitoring, and relates to the technical field of fault early warning, and the method comprises the steps: obtaining a rapping signal, and extracting back corona features based on wavelet transform and spectral analysis; constructing a dynamic coupling model of rapping strength-dust layer thickness, and combining back corona characteristics to obtain working condition characteristic vectors; obtaining a working condition data set, and obtaining a health degree scoring sequence under each time window in combination with the working condition feature vectors; decomposing the health degree scoring sequence into a plurality of intrinsic mode components, identifying abnormal fluctuation features based on a Markov chain, and performing risk level judgment; through all-condition monitoring and multi-modal data fusion, a back corona feature extraction, dust layer thickness prediction and health degree scoring model is constructed, dynamic identification and risk early warning of the operation state of the electric dust remover are achieved, and the problems that a traditional method is single in monitoring dimension and lags in response are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault early warning, and more particularly to a fault early warning method for a back-corona rapping electrostatic precipitator based on full working condition monitoring. Background Art

[0002] With the continuous advancement of industrialization, atmospheric pollution is becoming increasingly serious. As a key device for industrial waste gas treatment, electrostatic precipitators (ESPs) are widely used in industries such as power, metallurgy, cement, and chemicals, fulfilling the critical task of removing smoke and reducing particulate matter emissions. The stable operation of ESPs is directly related to the achievement of environmental protection goals and the safe production of enterprises. However, due to the complex operating environment, large load fluctuations, and diverse dust composition, ESPs are prone to various faults during long-term operation, such as dust deposition, vibration system failure, and abnormal back-corona discharge. These faults seriously affect dust removal efficiency and equipment lifespan, and may even cause safety accidents. Therefore, the development of efficient and accurate ESP fault warning systems has become a research hotspot in the field of industrial intelligent monitoring and maintenance.

[0003] For example, patent publication number CN106861915A discloses an intelligent monitoring and diagnostic control system for electrostatic precipitators based on GA technology. This relates to the field of intelligent monitoring and diagnosis. This GA-based intelligent monitoring and diagnostic control system dynamically optimizes the BP neural network system, adjusts network structure parameters, improves convergence speed and accuracy, and enhances online prediction results. This maximizes the availability, utilization, and efficiency of electrostatic precipitators, demonstrating excellent practical value.

[0004] For example, the invention patent with publication number CN114632624A discloses an electrostatic precipitator operation optimization system and optimization method, including: a state monitoring system, a multi-level and multi-level operation mode and an intelligent optimization operation system; the state monitoring system includes a unit state monitoring system, an electrostatic precipitator state monitoring system and a state analysis and evaluation system, and the state monitoring system evaluates the dust load of the electrostatic precipitator according to the state of the unit, electrostatic precipitator and other equipment and the flue gas parameters; the specific mode when the multi-level and multi-level operation mode is running is determined according to the relevant parameters of the state analysis and evaluation system; the intelligent optimization operation system adopts a neural network algorithm, takes historical operation data as a training sample, establishes a relationship model between the operation data and the multi-level and multi-level operation mode, and predicts the flue gas parameters at the inlet and outlet of the electrostatic precipitator and the optimal operation mode of the electrostatic precipitator in real time; the overall system of this application runs back and forth, continuously optimizes the prediction results, and realizes energy saving and emission reduction of the electrostatic precipitator.

[0005] The above disclosed technical solutions have at least the following technical problems: The existing electrostatic precipitator fault detection method only relies on a single electrical parameter (such as secondary current) to judge back corona, which is easily affected by operating condition fluctuations and has a high false alarm rate. In addition, there is a lack of a quantitative model for the correlation between the vibration system and the back corona state, making it difficult to timely identify dust layer abnormalities caused by vibration failure. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a back-corona vibration electrostatic precipitator fault warning method based on full-operating-condition monitoring and multimodal data fusion, constructs a back-corona feature extraction, dust layer thickness prediction and health scoring model, realizes dynamic identification of the operating status of the electrostatic precipitator and risk warning, and solves the problems of single monitoring dimension and delayed response of traditional methods.

[0007] To achieve the above object, the present invention provides the following technical solutions: A back-corona vibration electrostatic precipitator fault warning method based on full-operating-condition monitoring includes: obtaining vibration signals and extracting back-corona features based on wavelet transform and spectrum analysis; constructing a dynamic coupling model of vibration intensity-dust layer thickness and obtaining operating condition feature vectors in combination with back-corona features; obtaining an operating condition data set and obtaining a health score sequence in each time window in combination with the operating condition feature vectors; decomposing the health score sequence into multiple intrinsic mode components, identifying abnormal fluctuation features based on Markov chains, and making risk level determinations.

[0008] In a preferred embodiment, the rapping signal is obtained and the back-corona feature is extracted based on wavelet transform and spectrum analysis, specifically as follows: the rapping signal is collected and the rapping signal is decomposed into multiple scales by wavelet basis functions to obtain several frequency band sub-signals; time domain features related to the back-corona pulse in the frequency band sub-signals are extracted, feature vectors are constructed according to the time domain features, and a multidimensional feature matrix composed of multiple frequency band sub-signals is obtained, wherein each row in the multidimensional feature matrix corresponds to a sample point in the feature space; the multidimensional feature matrix is ​​clustered based on the K-means cluster analysis method, the frequency band sub-signals are divided into several categories, and the distribution categories of the frequency band sub-signals in the feature space are identified; according to the distribution categories, the frequency bands that meet the constraint conditions are screened out as candidate sets of back-corona feature bands; the rapping signal is subjected to spectrum analysis, and the energy density changes corresponding to the frequencies in the candidate frequency bands are integrated to evaluate the characteristic stability and back-corona response strength of each frequency band; the optimal back-corona feature band is screened out from the candidate frequency bands according to the fluctuation of the spectrum energy changes and the clustering categories; and the back-corona feature is extracted according to the optimal back-corona feature band.

[0009] In a preferred embodiment, the rapping signal is subjected to spectral analysis, and the energy density changes corresponding to the frequencies in the candidate frequency bands are integrated to evaluate the characteristic stability and back-corona response strength of each frequency band, specifically as follows: the rapping signal is subjected to fast Fourier transform processing to obtain a spectrum diagram, and the energy density of the corresponding frequency band in the spectrum diagram is output; the energy density value of each candidate frequency band in different time windows is obtained to form a frequency band energy change sequence; according to the frequency band energy change sequence, the stability index and frequency band response strength of the corresponding frequency band are output.

[0010] In a preferred embodiment, the matching of spectral energy change fluctuations with cluster categories is performed as follows: based on the average Euclidean distance, the spectral energy change characteristics are fused and compared with the cluster categories to which they belong, and the energy performance differences of frequency bands in different categories are analyzed, and the spectral energy change characteristics include stability indicators and frequency band response strengths; the candidate frequency bands are sorted according to the energy performance differences, and the optimal anti-corona characteristic frequency band is screened out.

[0011] In a preferred embodiment, the dynamic coupling model of the rapping intensity and the dust layer thickness is constructed, and the back-corona characteristics are combined to obtain the operating condition characteristic vector, specifically as follows: the rapping intensity data during the operation of the electrostatic precipitator is obtained, and the rapping response characteristics after the rapping are simultaneously obtained; based on the rapping intensity data and the rapping response characteristics, a dynamic coupling model of the rapping intensity and the dust layer thickness is constructed based on the radial basis function, and the predicted dust layer thickness is output; the dust layer thickness is integrated with the back-corona characteristics to construct a complete operating condition characteristic vector.

[0012] In a preferred embodiment, the operating condition data set is obtained and combined with the operating condition feature vector to obtain a health score sequence in each time window, specifically as follows: the operating condition data is obtained and time-aligned through the data interface module; the operating condition data is segmented according to a preset sliding time window, and the key operating condition features in each time window are extracted to form an operating condition feature vector in the window; the operating condition feature vector in the window and the operating condition feature vector are fused to obtain an enhanced state representation vector; an operating state network represented by a graph structure is constructed based on the enhanced state representation vector to obtain the operating state vector at the current moment; the operating state vector is input into the trained health score model to output the health score of the equipment operation in the current time window; the health scores in all time windows are arranged in chronological order to form a health score sequence.

[0013] In a preferred embodiment, the operation status network represented by the graph structure is constructed as follows: according to the structural composition and functional division of the electrostatic precipitator, the operation modules in the equipment are abstracted as nodes in the graph structure; according to the physical connection relationship between the nodes, the data coupling strength and the historical operation collaborative characteristics, the edges between the nodes in the graph are constructed, and weights are assigned to each edge; the operating condition feature vector extracted within the time window is mapped to the corresponding node according to the module dimension to which it belongs, and used as the initial feature input of the node to complete the initialization of the operation status graph; the enhanced state representation vector is spliced ​​with the initial features of each node, and input into the pre-trained graph neural network model to perform multiple rounds of feature propagation and aggregation; after the graph neural network propagation is completed, the final features of all nodes in the graph are aggregated based on the global pooling method to form a unified graph embedding vector, i.e., the operation status vector.

[0014] In a preferred embodiment, the health score sequence is decomposed into multiple intrinsic mode components, and abnormal fluctuation characteristics are identified based on the Markov chain, and risk level determination is performed, specifically as follows: the health score sequence is subjected to intrinsic mode decomposition to obtain several intrinsic mode function components; the top k key mode components are screened out based on the frequency bandwidth and energy distribution of the modal components; each key mode component is divided into equal frequencies to construct a finite discrete state set, and the modal value at each moment is mapped to the corresponding state interval to obtain a discrete state sequence; the transition frequency between any two states is counted based on the discrete state sequence to construct a state transition probability matrix; a first-order discrete-time Markov chain model is established based on the state transition probability matrix, and abnormal fluctuation characteristics are extracted; the abnormal fluctuation characteristics are vectorized and encoded to generate an abnormal fluctuation feature vector; the abnormal fluctuation feature vector is input into the abnormal determination module for risk assessment to obtain a risk level; the risk level is compared with a preset threshold to determine the current device status.

[0015] In a preferred embodiment, the abnormal fluctuation feature vector is input into the abnormality determination module for risk assessment to obtain the risk level, specifically as follows: different feature weights are assigned to each dimension in the abnormal fluctuation feature vector, and a weighted scoring function is constructed; the score obtained by the weighted scoring function is normalized and mapped to a specified range to obtain a standardized risk value; the standardized risk value is compared with a preset segmentation threshold to determine the risk level of the current operating state.

[0016] The technical effects and advantages of the present invention's back-corona rapping electrostatic precipitator fault early warning method based on full working condition monitoring are as follows: 1. This invention effectively extracts back-corona response features from rapping signals by integrating wavelet transform and spectrum analysis techniques. Compared with traditional time-frequency domain analysis methods, this solution has stronger multi-scale signal decomposition capabilities and frequency band energy assessment capabilities. It can extract back-corona characteristic frequency bands with high stability and high responsiveness from complex background signals, improving the accuracy and robustness of feature extraction and ensuring the validity of subsequent model input data.

[0017] 2. This invention uses nonlinear regression methods such as radial basis functions to couple vibration parameters (such as vibration frequency, acceleration, and duration) with vibration responses (such as current disturbance amplitude and acceleration decay rate) to achieve dynamic prediction of dust layer thickness. Compared with models based on empirical thresholds or static estimates, the coupled model established in this invention has greater adaptability and generalization capabilities, accurately reflecting the dynamic changes in dust cleaning efficiency and dust accumulation within the electrostatic precipitator.

[0018] 3. This invention fuses back-corona characteristics with the predicted dust layer thickness to form a multidimensional operating condition feature vector reflecting the overall operating status of the equipment. Furthermore, it introduces an operating status network represented by a graph structure, mapping the complex modular system of the electrostatic precipitator into a graph neural network model, effectively preserving the coupling relationships between structures and operational coordination information. During multiple rounds of propagation and aggregation in the graph neural network, implicit patterns between different operating modules can be efficiently mined, forming a high-dimensional, unified operating status embedding vector, providing powerful characterization capabilities for health scoring and fault warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flow chart of a method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring. DETAILED DESCRIPTION

[0020] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0021] Example 1, Figure 1 The present invention provides a method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring, comprising the following steps: S1, obtain the rapping signal and extract the back corona characteristics based on wavelet transform and spectrum analysis.

[0022] In this embodiment, the rapping signal is obtained, and the back-corona characteristics are extracted based on wavelet transform and spectrum analysis, as follows: The current, voltage or acceleration sensors arranged on the rapping hammer, plate or rapping control circuit are used to collect the rapping signal, corona discharge waveform signal and electromagnetic interference signal generated during the rapping process in real time, and the sampling frequency is not less than 10kHz to ensure that the high-frequency discharge pulse components are captured; The vibration signal is decomposed at multiple scales using wavelet basis functions to obtain several frequency sub-signals. Each sub-signal corresponds to a time domain signal within a specific frequency range, reflecting the local time-frequency characteristics of that frequency band. The signal decomposition level is adaptively set based on the Nyquist frequency and the target frequency band characteristics, and 3 to 5 levels of decomposition are usually selected. Extracting time domain features related to back-corona pulses from frequency band sub-signals, wherein the time domain features include mean energy, pulse density, and skewness; Constructing a feature vector based on the time domain features of each frequency band sub-signal to obtain a multi-dimensional feature matrix composed of multiple frequency band sub-signals, wherein each row in the multi-dimensional feature matrix corresponds to a sample point in the feature space; The multi-dimensional feature matrix is ​​clustered based on the K-means clustering analysis method to divide the distribution categories of frequency sub-signals in the feature space. Each category represents a similar signal behavior pattern, which is convenient for identifying potential back-corona feature aggregation areas. According to the distribution category, the frequency bands that meet the constraint conditions are selected as the optimal candidate set of anti-corona characteristic frequency bands. The constraint conditions are that the mean kurtosis of the category is significantly higher than that of other categories and the average pulse density of the category is significantly greater than the overall average. The candidate frequency band set is used as a frequency band screening template in the spectrum analysis process to limit the frequency band range of the spectrum analysis; Perform spectrum analysis on the rapping signal and integrate the energy density changes corresponding to the frequencies in the candidate frequency bands to evaluate the characteristic stability and back-corona response intensity of each frequency band; According to the matching of spectrum energy fluctuation and clustering category, the optimal anti-corona characteristic frequency band is selected from the candidate frequency bands; The back-corona features are extracted according to the optimal back-corona characteristic frequency band. The back-corona features include time domain features and frequency domain features. The frequency domain features include spectrum energy distribution and significant spectrum peaks (manifested as suddenly appearing high-amplitude frequency points in the FFT spectrum).

[0023] In this embodiment, the rapping signal is subjected to spectrum analysis, and the energy density changes corresponding to the frequencies in the candidate frequency bands are integrated to evaluate the characteristic stability and back-corona response strength of each frequency band, as follows: Perform fast Fourier transform on the rapping signal to obtain a spectrum diagram, and output the energy density of the corresponding frequency band in the spectrum diagram; Obtain the energy density value of each candidate frequency band in different time windows to form a frequency band energy change sequence; According to the frequency band energy change sequence, the sliding standard deviation and average energy density are output as the stability index and frequency band response strength of the frequency band. The stability index reflects whether the characteristics of the frequency band are continuous and stable. At the same time, the frequency band with high and stable response strength is given priority. Based on the average Euclidean distance, the spectrum energy change characteristics are integrated and compared with the cluster categories to which they belong, and the energy performance differences of the frequency bands in different categories are analyzed. The spectrum energy change characteristics include stability index and frequency band response strength; The candidate frequency bands are sorted according to the differences in energy performance, and the frequency bands with stable energy and significant response intensity are selected and determined as the optimal anti-corona characteristic frequency bands.

[0024] S2, construct a dynamic coupling model of vibration intensity-dust layer thickness, and combine it with the back corona characteristics to obtain the working condition characteristic vector.

[0025] In this embodiment, a dynamic coupling model of rapping intensity and dust layer thickness is constructed, and combined with the back corona characteristics, the working condition characteristic vector is obtained, which is as follows: Acquire rapping intensity data during the operation of the electrostatic precipitator, and simultaneously acquire rapping response characteristics after rapping, wherein the rapping intensity data includes rapping frequency, rapping response acceleration, and rapping duration, and the rapping response characteristics include current disturbance amplitude and acceleration envelope attenuation rate; Based on the vibration intensity data and vibration response characteristics, the mapping relationship between them and the dust layer thickness is learned based on the radial basis function. A dynamic coupling model of vibration intensity and dust layer thickness is constructed, and the predicted dust layer thickness is output. The nonlinear model is trained using labeled historical data samples, and the model parameters are optimized by minimizing the error between the predicted thickness and the actual thickness. The dust layer thickness and back-corona characteristics are integrated to construct a complete operating condition feature vector, which serves as the input basis for subsequent health assessment and fault warning models.

[0026] The dynamic coupling model of vibration intensity and dust layer thickness is as follows: Where: is the predicted dust layer thickness, is the nonlinear function learned through training, is the vibration frequency, is the rapping response acceleration, is the rapping duration, is the current disturbance amplitude, is the acceleration envelope decay rate.

[0027] S3, obtain the working condition data set and combine it with the working condition feature vector to obtain the health score sequence in each time window.

[0028] In this embodiment, the operating condition data set is obtained and combined with the operating condition feature vector to obtain the health score sequence in each time window, as follows: Obtaining operating condition data during the operation of the electrostatic precipitator, including but not limited to: vibration intensity parameters (such as vibration frequency, vibration duration, current peak value, etc.), electric field operating parameters (such as high-voltage power supply current and voltage value), flue gas parameters (such as flow rate, temperature and humidity, dust concentration), back corona eigenvectors (such as harmonic energy, frequency domain peak value), and predicted dust layer thickness; Through the data interface module, the working condition data is synchronously organized into structured data input on a continuous time axis; The operating condition data is segmented according to the preset sliding time window, and the key operating condition features in each time window are extracted to form the operating condition feature vector within the window; The operating condition feature vector in the window is fused with the operating condition feature vector to obtain an enhanced state representation vector; Based on the enhanced state representation vector, a running state network represented by a graph structure is constructed to obtain the running state vector at the current moment; Input the running state vector into the trained health score model and output the health score of the device in the current time window. Arrange the health scores in all time windows in chronological order to form a health score sequence.

[0029] In this embodiment, a running status network represented by a graph structure is constructed as follows: Based on the structural composition and functional division of the electrostatic precipitator, the key operating modules in the equipment are abstracted into nodes in the graph structure. The nodes include the rapping mechanism node (representing the operating status of the rapping device), the power module node (representing the power supply characteristics of the high-voltage power supply), the electric field plate node (representing the working conditions of the settling plate and anode wire), and the ash hopper node (representing the dust accumulation and ash unloading status). Each node will participate in the subsequent state propagation as an independent computing unit in the graph neural network; Based on the physical connection relationship between nodes, data coupling strength and historical operation synergy characteristics, the edges between the nodes in the graph are constructed, and weights are assigned to each edge to represent its influence strength. For example, there is a direct action relationship between the vibration mechanism and the electric field plate, and the edge weight can be dynamically calculated based on the correlation coefficient between the historical vibration action and the plate current fluctuation. There is an energy transmission relationship between the power module and each plate node, and the edge weight can be adjusted according to the power change rate and discharge frequency. The feedback effect of the ash hopper state on the plate state can also be reflected through the change in the accumulated particle concentration. The operating condition feature vector extracted within the time window is mapped to the corresponding node according to the module dimension to which it belongs, and used as the initial feature input of the node to complete the initialization of the operating state diagram. The specific method is as follows: Embed rapping intensity characteristics (such as rapping frequency and duration) into rapping nodes; Embed indicators such as electric field voltage and current into power nodes and plate nodes; Embed dust deposition thickness, dust discharge cycle, etc. into the ash hopper node; Embed harmonic energy and spectrum characteristics into auxiliary monitoring nodes; The enhanced state representation vector is concatenated with the initial features of each node and input into a pre-trained graph neural network model. Multiple rounds of feature propagation and aggregation are performed. Each layer of graph convolution updates the node state based on the node neighborhood features and edge weight information, thereby extracting high-order coupling relationships between subsystems. After the graph neural network propagation is completed, the final features of all nodes in the graph are aggregated based on the global pooling method to form a unified graph embedding vector, namely the operating state vector, which serves as the embedded expression of the overall operating state of the electrostatic precipitator in the current time window.

[0030] S4, decomposes the health score sequence into multiple intrinsic mode components, identifies abnormal fluctuation characteristics based on Markov chain, and makes risk level determination.

[0031] In this embodiment, the health score sequence is decomposed into multiple intrinsic mode components, and the abnormal fluctuation characteristics are identified based on the Markov chain, and the risk level is determined as follows: Obtain the first-order difference and sliding standard deviation of the health score series to obtain the score change rate and volatility; A standardized indicator sequence is constructed based on the rate of change and volatility of the score, and the standardized anomaly score value is calculated based on the Z-score method; If the absolute value of the Z-score at a certain time point exceeds the preset abnormal threshold, the point is marked as an abnormal fluctuation point; Based on the abnormal fluctuation points in the marked scoring sequence, determine whether the trigger conditions for modal analysis are met; If the number of abnormal fluctuation points reaches the preset trigger condition, the intrinsic mode decomposition process is started; Perform intrinsic mode decomposition on the health score sequence to obtain several intrinsic mode function components; According to the frequency bandwidth and energy distribution of the modal components, the top k key modal components are screened out; Each key modal component is divided into equal frequencies to construct a finite discrete state set. The modal value at each moment is mapped to the corresponding state interval to obtain a discrete state sequence. Each state represents a certain degree of health fluctuation level. According to the discrete state sequence, the transition frequency between any two states is counted to construct the state transition probability matrix; A first-order discrete-time Markov chain model is established based on the state transition probability matrix to model the transition rules in the state sequence and extract abnormal fluctuation characteristics for anomaly identification. The abnormal fluctuation characteristics include steady-state distribution, state transition entropy, and dangerous state migration probability. Vectorize and encode the abnormal fluctuation characteristics to generate an abnormal fluctuation feature vector; The abnormal fluctuation feature vector is input into the abnormality determination module to evaluate the risk level and obtain the risk level; Compare the risk level with the preset threshold to determine whether the current device is in "normal", "warning" or "failure risk" status; If the current state is judged to be "warning" state, the adjustment suggestions will be output to the vibration cycle optimization module to automatically adjust the vibration frequency, intensity or delay parameters; If the current state is determined to be a "fault risk" state, the operation alarm module will be triggered, an alarm signal will be generated, and the manual inspection prompt process can be started.

[0032] The eigenmode decomposition is as follows: Where: It represents the health score value corresponding to time t, which is a component of the health score sequence and is used to reflect the health status of the electrostatic precipitator in this time window. Its value range is usually [0,1]. Represents the i-th order modal component, reflecting the fluctuation characteristics under a specific frequency component, It represents the long-term trend term, also known as the "trend term", which is the remaining signal after deducting all IMF components. It is usually regarded as the overall operating trend or long-term baseline of the health score series and is used to model the steady-state level of the equipment system. It represents the total number of intrinsic mode function components (IMF) obtained by intrinsic mode decomposition. It is dynamically determined according to the complexity of the signal and is usually related to the sequence length and signal fluctuation frequency.

[0033] In this embodiment, the abnormal fluctuation feature vector is input into the abnormality determination module for risk level assessment to obtain the risk level, which is as follows: Assign different feature weights to each dimension in the abnormal fluctuation feature vector and construct a weighted scoring function; The scores obtained by the weighted scoring function are normalized and mapped to the range of [0,1] to obtain the standardized risk value, thereby eliminating the impact of dimensional differences in each feature dimension; Compare the standardized risk value with the preset segmentation threshold to determine the risk level of the current operating status. The specific determination rules are as follows: when When , the device is judged to be in normal state, indicating that there is no obvious abnormal fluctuation; when When the device is in the warning state, it is judged that there is a slight fluctuation and continuous monitoring is required; when When the device is judged to be in a state of risk of failure, it means that the fluctuation is abnormal and significant, and a control or alarm strategy should be triggered; in, is the standardized risk value, the threshold parameter and It can be set based on historical data training or manual experience, or dynamically updated through online learning to adapt to changes in different device types and operating environments. Typical setting examples are: =0.35, =0.65.

[0034] The weighted scoring function is as follows: Where: is the weighted score of the current device in a given time window, is the number of abnormal fluctuation features, is the risk weight coefficient corresponding to the abnormal volatility characteristics, represents the value of the i-th abnormal fluctuation feature, is the standardized risk value, is the lower limit of the score in historical data, is the upper limit of the score in historical data.

[0035] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0036] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0037] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0038] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0039] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A fault warning method for a back-corona rapping electrostatic precipitator based on full-operating-condition monitoring is characterized in that: include: Obtain rapping signals and extract back-corona features based on wavelet transform and spectrum analysis; A dynamic coupling model of vibration intensity and dust layer thickness is constructed, and the operating condition characteristic vector is obtained by combining the back-corona characteristics. Obtain the operating condition data set and combine it with the operating condition feature vector to obtain the health score sequence in each time window; The health score sequence is decomposed into multiple intrinsic mode components, and the abnormal fluctuation characteristics are identified based on the Markov chain, and the risk level is determined.

2. The method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring according to claim 1, characterized in that: The rapping signal is obtained and the back corona characteristics are extracted based on wavelet transform and spectrum analysis, as follows: The vibration signal is collected and decomposed into multiple scales using wavelet basis functions to obtain several frequency band sub-signals; Extracting time domain features related to the back-corona pulse from the frequency band sub-signals, constructing feature vectors based on the time domain features, and obtaining a multidimensional feature matrix composed of multiple frequency band sub-signals, wherein each row in the multidimensional feature matrix corresponds to a sample point in the feature space; Clustering the multidimensional feature matrix based on the K-means clustering analysis method, dividing the frequency band sub-signals into several categories, and identifying the distribution categories of the frequency band sub-signals in the feature space; According to the distribution category, the frequency bands that meet the constraints are selected as the candidate set of anti-corona characteristic frequency bands; Perform spectrum analysis on the rapping signal and integrate the energy density changes corresponding to the frequencies in the candidate frequency bands to evaluate the characteristic stability and back-corona response intensity of each frequency band; According to the matching of spectrum energy fluctuation and clustering category, the optimal anti-corona characteristic frequency band is selected from the candidate frequency bands; The back-corona features are extracted according to the optimal back-corona feature frequency band.

3. The method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring according to claim 2, wherein: The spectrum analysis of the rapping signal is performed, and the energy density changes corresponding to the frequencies in the candidate frequency bands are integrated to evaluate the characteristic stability and back-corona response intensity of each frequency band, as follows: Perform fast Fourier transform on the rapping signal to obtain a spectrum diagram, and output the energy density of the corresponding frequency band in the spectrum diagram; Obtain the energy density value of each candidate frequency band in different time windows to form a frequency band energy change sequence; According to the frequency band energy change sequence, the stability index and frequency band response strength of the corresponding frequency band are output.

4. The method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring according to claim 3, characterized in that: The matching of the spectrum energy fluctuations with the cluster categories is as follows: Based on the average Euclidean distance, the spectrum energy change characteristics are integrated and compared with the cluster categories to which they belong, and the energy performance differences of the frequency bands in different categories are analyzed. The spectrum energy change characteristics include stability index and frequency band response strength; The candidate frequency bands are sorted according to the differences in energy performance, and the optimal anti-corona characteristic frequency band is selected.

5. The method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring according to claim 4, characterized in that: The dynamic coupling model of vibration intensity and dust layer thickness is constructed, and combined with the back corona characteristics, the working condition characteristic vector is obtained, which is as follows: Acquire the vibration intensity data during the operation of the electrostatic precipitator and simultaneously obtain the vibration response characteristics after vibration; Based on the vibration intensity data and vibration response characteristics and the radial basis function, a dynamic coupling model of vibration intensity and dust layer thickness is constructed to output the predicted dust layer thickness; The dust layer thickness and back-corona characteristics are integrated to construct a complete working condition feature vector.

6. The back-corona rapping electrostatic precipitator fault early warning method based on full working condition monitoring according to claim 5 is characterized in that: The working condition data set is obtained and combined with the working condition feature vector to obtain the health score sequence in each time window, as follows: Obtain working condition data and perform time alignment on the working condition data through the data interface module; The operating condition data is segmented according to the preset sliding time window, and the key operating condition features in each time window are extracted to form the operating condition feature vector within the window; The operating condition feature vector in the window is fused with the operating condition feature vector to obtain an enhanced state representation vector; Based on the enhanced state representation vector, a running state network represented by a graph structure is constructed to obtain the running state vector at the current moment; Input the running state vector into the trained health score model and output the health score of the device in the current time window; Arrange the health scores in all time windows in chronological order to form a health score sequence.

7. The method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring according to claim 6, characterized in that: The operation state network represented by the graph structure constructed based on the enhanced state representation vector is specifically as follows: According to the structural composition and functional division of the electrostatic precipitator, the operating modules in the equipment are abstracted as nodes in the graph structure; Based on the physical connection relationship between nodes, data coupling strength and historical operation synergy characteristics, the edges between nodes in the graph are constructed and weights are assigned to each edge. The operating condition feature vector extracted within the time window is mapped to the corresponding node according to the module dimension to which it belongs, as the initial feature input of the node; The enhanced state representation vector is concatenated with the initial features of each node and fed into a pre-trained graph neural network model to perform multiple rounds of feature propagation and aggregation. After the graph neural network propagation is completed, the final features of all nodes in the graph are aggregated based on the global pooling method to form a unified graph embedding vector, namely the running state vector.

8. The method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring according to claim 7, characterized in that: The health score sequence is decomposed into multiple intrinsic mode components, and abnormal fluctuation characteristics are identified based on the Markov chain, and risk level determination is performed, as follows: Perform intrinsic mode decomposition on the health score sequence to obtain several intrinsic mode function components; According to the frequency bandwidth and energy distribution of the modal components, the top k key modal components are screened out; Perform equal frequency division on each key modal component, construct a finite discrete state set, and map the modal value at each moment to the corresponding state interval to obtain a discrete state sequence; According to the discrete state sequence, the transition frequency between any two states is counted to construct the state transition probability matrix; A first-order discrete-time Markov chain model is established based on the state transition probability matrix, and abnormal fluctuation characteristics are extracted; Vectorize and encode the abnormal fluctuation characteristics to generate an abnormal fluctuation feature vector; The abnormal fluctuation feature vector is input into the abnormality judgment module for risk assessment to obtain the risk level; Compare the risk level with the preset threshold to determine the current device status.

9. The method for early warning of a back-corona rapping electrostatic precipitator fault based on full-operating-condition monitoring according to claim 8, characterized in that: The abnormal fluctuation feature vector is input into the abnormality determination module for risk assessment to obtain the risk level, which is as follows: Assign different feature weights to each dimension in the abnormal fluctuation feature vector and construct a weighted scoring function; The scores obtained by the weighted scoring function are normalized and mapped to a specified range to obtain a standardized risk value; Compare the standardized risk value with the preset segmentation threshold to determine the risk level of the current operating status.

Images