Method for supervising operation of filter bag dust removal system

Abstract

The invention relates to the technical field of dust removers, and provides a filter bag dust removal system operation supervision method which comprises the following steps: acquiring various operation parameters and parameter data of production parameters in multiple production-dust removal processes through a filter bag; carrying out correlation analysis on the operation parameters and the production parameters, and constructing a relation matrix of a production-ash removal process; analyzing similar production-ash removal processes to obtain a plurality of types of ash removal behaviors; analyzing the moment when the resistance change amplitude of the filter bag is large in the production-ash removal process and the final resistance information of the last production-ash removal process, and obtaining a plurality of short-time deviation time points and a plurality of long-time marking parameters of each production-ash removal process in combination with the moment when the parameter data change amplitude of each parameter is large; and scoring and pruning the neurons, constructing a filter bag service life prediction model, and realizing operation supervision of the filter bag dust removal system. The invention aims to solve the problem of low efficiency of a filter bag service life prediction model caused by redundant filter bag information and parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of dust collectors, and in particular to an operation monitoring method for a filter bag dust removal system. Background Art

[0002] Bag filter dust collectors (cloth bag dust collectors) have become one of the most effective treatment equipment for controlling particulate matter and emission standards in industrial production due to their high dust removal efficiency, strong adaptability, stable and reliable operation, etc. Therefore, the stability of their operation determines the emission effect; however, during the use of bag filter dust collectors, the filter bags will be continuously recycled and continuously affected by the blowing process. At the same time, it is difficult to detect the damage of the filter bags in time, so it is impossible to accurately evaluate the service life of the filter bags.

[0003] In the bag filter resistance operation and maintenance method of reference patent CN117654175A, various bag filter information during bag filter operation, such as bag filter resistance, cleaning cycle, injection pressure, and number of injections, is continuously analyzed. This information is then analyzed and comprehensively predicted to determine the bag filter lifespan, thereby enabling operational supervision of the bag filter system. In the actual prediction process, the bag filter information must be input into a trained prediction model. Because bag filter resistance growth has a lag and bag filter lifespan is affected by a long cycle, LSTM models are widely used for prediction. However, during the training of the LSTM prediction model based on bag filter information, the bag filter information and parameters are redundant, and the impact of some parameters on bag filter lifespan is not consistently reflected. This results in an unpruned model with slow inference speed and inability to meet the bag filter cleaning control response requirements. This results in inaccurate predictions of bag filter lifespan, which in turn affects the overall process supervision of the bag filter system. Summary of the Invention

[0004] The present invention provides a method for monitoring the operation of a filter bag dust removal system to solve the problem that the existing filter bag information and parameters are redundant and thus the filter bag life prediction model is inefficient. The technical solution adopted is as follows: The present invention proposes a method for supervising the operation of a filter bag dust removal system, which comprises the following steps: The filter bags arranged in the bag filter are used to obtain parameter data of various operating parameters and production parameters during multiple production and cleaning processes, and the pressure data of the filter bag inlet and outlet are recorded in real time; Based on the data change trends of the operating parameters in the production-cleaning process, the process is segmented, and the correlation between the operating parameters and the production parameters is analyzed in combination with the data changes of the production parameters. A relationship matrix of the production-cleaning process is constructed. Based on the relationship matrix constructed based on the correlation between the operating parameters and the production parameters, similar production-cleaning processes are analyzed to obtain several types of cleaning behaviors. Analyze the moments when the filter bag resistance changes significantly during the production-cleaning process, as well as the final resistance information from the previous production-cleaning process. Combined with the moments when the parameter data of each parameter changes significantly, several short-term deviation time points of each production-cleaning process are obtained. By analyzing the correlation changes at the same position in the relationship matrix of adjacent time sequences for similar cleaning behaviors, several long-term marker parameters of each production-cleaning process are obtained. Based on the short-term deviation time points and long-term marking parameters of each production-cleaning process in the training dataset, the neurons in the LSTM model are scored and pruned, and a filter bag life prediction model is constructed to realize the operation supervision of the filter bag dust removal system.

[0005] Optionally, segmenting based on the data change trend of the operating parameters in the production-cleaning process, performing correlation analysis between the operating parameters and the production parameters in combination with the data change of the production parameters, and constructing a relationship matrix of the production-cleaning process includes the following specific methods: Analyze the fluctuation of the slope of the parameter data of the operating parameters during the primary production-cleaning process, and segment the data change curve of the operating parameters; The correlation analysis between operating parameters and production parameters was conducted in combination with the data change curve of production parameters, and the relationship matrix of each production-cleaning process was constructed.

[0006] Optionally, the analysis of the fluctuation of the slope of the parameter data of the operating parameters during the primary production-cleaning process and segmenting the data change curve of the operating parameters includes the following specific methods: For any production-cleaning process, based on the time series parameter data of each operating parameter in the production-cleaning process and the time series parameter data of each production parameter in the production-cleaning process, obtain the data change curve of each operating parameter and the data change curve of each production parameter; For the The data change curve of the operating parameters is obtained, and the slope of each data point is obtained; for any data point, the variance of the slopes of the data point and all the data points before it is obtained as the variance of the slope before the data point, and the variance of the slopes of all the data points after the data point is obtained as the variance of the slope after the data point; Based on the front slope variance and the back slope variance of each data point in the data change curve, a first dividing point and a second dividing point of each operating parameter are obtained; The data curve between the first data point and the first division point is taken as the The initial data change curve of the operating parameters; the data curve between the first data point after the first division point and the second division point is taken as the first The mid-term data change curve of the operating parameters; the data curve composed of all data points after the second division point is used as the first The later data change curve of the operating parameters.

[0007] Optionally, the obtaining of the first dividing point and the second dividing point of each operating parameter includes the following specific methods: The first The front slope variance of each data point in the data change curve of the operating parameter is arranged in the time sequence of the data points to obtain the first The front slope variance sequence of the data change curve of the operating parameters is obtained, and several maximum values ​​are obtained for the front slope variance sequence. The judgment starts from the first maximum value, and the front slope variance threshold is set. If the first maximum value is greater than the front slope variance threshold, the data point corresponding to the first maximum value is taken as the first maximum value. The first dividing point of the data change curve of the operating parameter is selected. If the first maximum value is less than or equal to the previous slope variance threshold, the second maximum value is judged until the first dividing point is obtained; Arrange the posterior slope variances of all data points after the first partition point in the time sequence of the data points to obtain the The stable posterior slope variance sequence of the data change curve of the operating parameters is obtained, and several maximum values ​​are obtained for the stable posterior slope variance sequence. The judgment starts from the first maximum value, and the posterior slope observation threshold is set. If the first maximum value is greater than the posterior slope variance threshold, the data point corresponding to the first maximum value is taken as the first maximum value. The second dividing point of the data change curve of the operating parameter is selected. If the first maximum value is less than or equal to the previous slope variance threshold, the second maximum value is judged until the second dividing point is obtained.

[0008] Optionally, the specific method of constructing the relationship matrix of each production-cleaning process includes: According to the The first and second dividing points of the operating parameters are The data change curves of the production parameters are obtained in the early, mid and late stages respectively; operating parameters and The data change curves of each production parameter are used to obtain the Pearson correlation coefficients of the data change curves at the early, mid and late stages; The initial weight, mid-term weight and late weight are preset to correspond to the initial, mid-term and late data change curves respectively. The Pearson correlation coefficients of the three segments of data change curves are weighted and summed by the corresponding weights. The sum is used as the first weight in the production-cleaning process. The operating parameters and Correlation values ​​of production parameters; Obtain the correlation values ​​between each operating parameter and each production parameter in the production-cleaning process, and construct a matrix, in which the same row in the matrix represents the correlation value between the same operating parameter and each production parameter, and the same column represents the correlation value between each operating parameter and the same production parameter. The obtained matrix is ​​used as the relationship matrix of the production-cleaning process.

[0009] Optionally, the specific methods for obtaining several types of dust cleaning behaviors include: All production-cleaning processes are density clustered, and the distance measurement adopts the inverse proportional value of the cosine similarity between the relationship matrices of each production-cleaning process to obtain several clusters. The production-cleaning processes in the same cluster are regarded as the same type of cleaning behavior.

[0010] Optionally, the analysis of the moments when the filter bag resistance changes significantly during the production-cleaning process, as well as the final resistance information of the previous production-cleaning process, is combined with the moments when the parameter data of each parameter changes significantly to obtain several short-term deviation time points of each production-cleaning process, including the specific method of: Based on the pressure difference change at the inlet and outlet of the filter bag during one production-cleaning process, and the pressure difference at the end of the previous production-cleaning process, several pressure difference change moments are screened and obtained; Several changing moments of various parameters in a production-cleaning process are obtained, and difference analysis is performed in combination with the pressure difference changing moments to obtain several short-term deviation time points of each production-cleaning process.

[0011] Optionally, the screening to obtain a plurality of pressure difference change moments includes the following specific methods: For the training dataset During the production-cleaning process, the pressure difference between the inlet and outlet at each moment is obtained, and the difference obtained by subtracting the pressure data of the outlet from the pressure data of the inlet is used as the pressure difference between the inlet and outlet at each moment; The inlet and outlet pressure difference at the last moment of the production-cleaning process; the coordinate system is constructed with the order value corresponding to each moment as the horizontal axis and the pressure difference as the vertical axis. The inlet and outlet pressure differences at all moments in the production-cleaning process are mapped to the coordinate system according to the corresponding sequence values, and the pressure difference data points at each moment are obtained. The inlet and outlet pressure difference at the last moment of the production-cleaning process is used as the pressure difference corresponding to the origin of the horizontal coordinate. Connect each pressure difference data point to get the The pressure difference data curve of the production-cleaning process is used to calculate the slope of the pressure difference data points at each moment except the 0th moment in the pressure difference data curve, and the slope is used as the change slope of the next pressure difference data point; The slope threshold is preset, and the time corresponding to the pressure difference data point whose slope is greater than the slope threshold is taken as the first The moment when the pressure difference changes during the production-cleaning process.

[0012] Optionally, the specific method for obtaining several short-term deviation time points of each production-cleaning process includes: For the The data change curve of any operating parameter or production parameter in the production-cleaning process is obtained, the slope of each data point is obtained, the parameter slope threshold is preset, and the time corresponding to the data point with a slope greater than the parameter slope threshold is used as the first The moment when this parameter changes during the production-cleaning process; Based on the Several pressure difference change moments and several change moments of any parameter in the production-cleaning process. For any pressure difference change moment, obtain the absolute value of the difference between it and the change moments of the parameter, and use the minimum value as the time offset of the pressure difference change moment; Obtain the time offset of each pressure difference change moment. If the time offset of any pressure difference change moment is greater than the time offset threshold, take the pressure difference change moment as a short-term deviation time point and obtain the first Secondary cleaning - several short-term deviation points in the production process.

[0013] Optionally, a specific method for obtaining several long-term marking parameters of each production-cleaning process is as follows: For several production-cleaning processes in any type of cleaning behavior, sort the relationship matrices of all production-cleaning processes in chronological order. For adjacent relationship matrices, calculate the absolute value of the difference between the elements at the same position and use it as the element difference at the corresponding position. If the element difference is greater than the aging threshold, the operating parameters and production parameters corresponding to the elements at that position in the production-cleaning process corresponding to the two relationship matrices are marked as long-term marking parameters in the two production-cleaning processes; the aging of adjacent relationship matrices in the same type of cleaning behavior is judged to obtain several long-term marking parameters for each production-cleaning process.

[0014] The beneficial effects of the present invention are as follows: by performing correlation analysis on operating parameters and production parameters in the production-cleaning process, while considering the impact of the stability of the generated products at different stages on the parameter performance, the correlation coefficients are quantified in segments and correlation values ​​are obtained, and then a relationship matrix is ​​constructed to reflect the correlation performance of the operating parameters and production parameters. Similar correlation performance indicates that the overall parameter operation relationship of the production-cleaning process is similar, and is more likely to correspond to similar cleaning behaviors. By performing deviation analysis on the changes in filter bag resistance and parameter changes, short-term deviation time points are screened to reflect the impact of their unique parameter information on the filter bag life and use them in neuron pruning in the subsequent model. By analyzing the correlation changes in the relationship matrix of similar cleaning behaviors, aging information is analyzed to determine the long-term impact of periodic changes, and the corresponding parameters are marked to reflect the long-term aging information of the corresponding operating parameters or production parameters. The short-term deviation time points and long-term aging information are used to score and judge pruning of neurons, thereby improving the inference efficiency of the LSTM prediction model and constructing a filter bag life prediction model. In addition, the filter bag dust removal system in the bag dust collector can be visually monitored for life, thereby improving the operational supervision efficiency of the filter bag dust removal system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 A schematic flow chart of a method for supervising the operation of a bag filter dust removal system provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0018] See also Figure 1 , which shows a flow chart of a method for supervising the operation of a bag filter dust removal system provided by one embodiment of the present invention, the method comprising the following steps: Step S001: Obtain parameter data of various operating parameters and production parameters during multiple production-cleaning processes through filter bags arranged in a bag filter, and record pressure data of the filter bag inlet and outlet in real time.

[0019] The purpose of this embodiment is to analyze the short-term changes and long-term periodic effects of filter bag information and various parameters on the filter bag life when constructing a filter bag life prediction model through the LSTM prediction model, so as to prune the neurons in the model and improve the reasoning efficiency of the life prediction model. Then, the filter bag life prediction model is obtained by training the pruned LSTM prediction model to realize the operation supervision of the filter bag dust removal system.

[0020] Specifically, data is collected on the filter bag information arranged in the bag dust collector, several production parameters in the production process, and various operating parameters in the cleaning process. The filter bag information includes the resistance information of the filter bag. In this embodiment, the resistance information is reflected by recording the pressure data at the inlet and outlet of the filter bag, and the sampling time interval is set to 1 second; the production parameters of the production process include the smoke outlet temperature and humidity of the production process and other production parameters set according to the production goals, and the parameter data of each production parameter is collected at the same sampling time interval; the operating parameters of the cleaning process include the blowing pressure, blowing cycle and number of blowing times, etc., which are also recorded at the same sampling time interval to obtain the parameter data of each operating parameter.

[0021] Furthermore, a training data set is constructed by collecting parameter data of various operating parameters and production parameters during the large-scale production and cleaning process of filter bags of the same model of bag dust collector, as well as the recorded filter bag information, for the subsequent training of the filter bag life prediction model.

[0022] Step S002: Segment the process based on the data change trend of the operating parameters in the production-cleaning process, analyze the correlation between the operating parameters and the production parameters in combination with the data changes of the production parameters, and construct a relationship matrix for the production-cleaning process; based on the relationship matrix constructed based on the correlation between the operating parameters and the production parameters, analyze similar production-cleaning processes to obtain several types of cleaning behaviors.

[0023] It should be noted that there is a certain correlation between the operating parameters and the production parameters. By performing a correlation analysis on the data change curves of the operating parameters and the production parameters, the correlation between the operating parameters and the production parameters can be reflected. Based on this correlation, similar cleaning behaviors are analyzed, and the cleaning behaviors are classified by constructing a relationship matrix and performing clustering based on the relationship matrix.

[0024] As an example, the specific method included in this step is: Analyze the fluctuation of the slope of the parameter data of the operating parameters during the primary production-cleaning process, and segment the data change curve of the operating parameters; Combined with the data change curve of production parameters, the correlation analysis between operation parameters and production parameters is carried out to construct the relationship matrix of each production-cleaning process; Cluster analysis of the production-cleaning process was performed based on the relationship matrix, and several types of cleaning behaviors were obtained.

[0025] It should be noted that a production-cleaning process includes an initial rising unstable period, a mid-term stable period and a late stop period. When the production starts, the reaction begins and the changes in the generated products are unstable. Therefore, the changes in this stage are unstable, and there will be large deviations when calculating the correlation. For the late stop period, it changes from stable to unstable. Compared with the initial data changes, its data changes are more stable than in the early stage. It is necessary to conduct a segmented analysis based on the change slope of the parameter data of the operating parameters, that is, the change trend. At the same time, since the main monitoring is the operation of the dust removal system, the production parameters are segmented with the compared operating parameters. Among them, due to the initial unstable changes, its slope will show large fluctuations, while the mid-term stable slope will show smaller fluctuations, and the fluctuation of the slope change in the late stop is smaller than the initial fluctuations. In this way, the front slope variance and the back slope variance are constructed for analysis, and the threshold is used to judge the acquisition of the first and second dividing points.

[0026] Preferably, in one embodiment of the present invention, the fluctuation of the change slope of the parameter data of the operating parameters in the production-cleaning process is analyzed, and the data change curve of the operating parameters is segmented, including the specific method of: For any production-cleaning process, which includes several operating parameters and production parameters, based on the time series parameter data of each operating parameter in the production-cleaning process, and the time series parameter data of each production parameter in the production-cleaning process, the data change curve of each operating parameter and the data change curve of each production parameter are obtained; for the The data change curve of the operating parameters is obtained, and the slope of each data point is obtained. According to the difference and time interval between any data point and its adjacent previous data point, the slope of the first data point in the data change curve is set as the slope of the second data point; for any data point, the variance of the slopes of the data point and all the data points before it is obtained as the variance of the slope before the data point. Similarly, the variance of the slopes of all the data points after the data point is obtained as the variance of the slope after the data point.

[0027] Further, the The front slope variance of each data point in the data change curve of the operating parameter is arranged in the time sequence of the data points to obtain the first The front slope variance sequence of the data change curve of the operating parameters is obtained, and several maximum values ​​are obtained for the front slope variance sequence. The judgment starts from the first maximum value and the front slope variance threshold is set. In this embodiment, the front slope variance threshold adopts the front slope variance corresponding to the first 10% of all front slope variances sorted from large to small. If the first maximum value is greater than the front slope variance threshold, the data point corresponding to the first maximum value is taken as the first maximum value. The first dividing point of the data change curve of the operating parameter is selected. If the first maximum value is less than or equal to the previous slope variance threshold, the second maximum value is judged until the first dividing point is obtained.

[0028] Furthermore, the posterior slope variances of all data points after the first division point are arranged in the time sequence of the data points to obtain the The stable posterior slope variance sequence of the data change curve of the operating parameters is obtained, and several maximum values ​​are obtained for the stable posterior slope variance sequence. The judgment starts from the first maximum value and the posterior slope observation threshold is set. In this embodiment, the posterior slope variance threshold adopts the posterior slope variance corresponding to the first 30% of the posterior slope variances of all data points in descending order. If the first maximum value is greater than the posterior slope variance threshold, the data point corresponding to the first maximum value is taken as the first maximum value. The second dividing point of the data change curve of the operating parameter is selected. If the first maximum value is less than or equal to the previous slope variance threshold, the second maximum value is judged until the second dividing point is obtained.

[0029] Furthermore, the data curve between the first data point and the first dividing point (including the first dividing point) is taken as the first The initial data change curve of the operating parameters; the data curve between the first data point after the first division point and the second division point (including the second division point) is used as the first The mid-term data change curve of the operating parameters; the data curve composed of all data points after the second division point is used as the first The later data change curve of the operating parameters.

[0030] Preferably, in one embodiment of the present invention, the correlation analysis between the operating parameters and the production parameters is performed in combination with the data change curve of the production parameters, and a relationship matrix of each production-cleaning process is constructed, including the specific method of: According to the same first dividing point and second dividing point (corresponding time), The data change curves of the production parameters are obtained in the early, mid and late stages respectively; operating parameters and The data change curves of the production parameters are used to obtain the Pearson correlation coefficients of the initial, mid-term and late data change curves, that is, the data change curves of the same period are used to calculate the correlation coefficients. In this embodiment, the initial weight is set to 0.2, the mid-term weight is set to 0.5, and the late weight is set to 0.3, corresponding to the initial, mid-term and late data change curves, respectively. The Pearson correlation coefficients of the three data change curves are weighted and summed by the corresponding weights, and the sum is used as the first value in the production-cleaning process. The operating parameters and The correlation values ​​of production parameters.

[0031] It should be noted that the stable changes of the mid-term stabilizer are the most reliable for the correlation analysis between the operating parameters and the production parameters. The largest mid-term weight is set. The instability of the later stop period has a smaller fluctuation range than the initial rising unstable period. Therefore, the later weight needs to be set larger than the initial weight. Then, the correlation coefficient of the segmented data change curve is combined with the corresponding weight to quantify the correlation value between the operating parameters and the production parameters in this production-cleaning process.

[0032] Furthermore, the correlation values ​​between each operating parameter and each production parameter in the production-cleaning process are obtained according to the above method, and a matrix is ​​constructed, wherein the same row in the matrix represents the correlation value between the same operating parameter and each production parameter, and the same column represents the correlation value between each operating parameter and the same production parameter. The obtained matrix is ​​used as the relationship matrix of the production-cleaning process.

[0033] It should be further explained that the correlation between the operating parameters and the production parameters in similar production-cleaning processes is similar. By calculating the cosine similarity of the relationship matrix, the larger the cosine similarity, the smaller the corresponding relationship matrix distance in the clustering process, and then several clusters are obtained, and the same type of cleaning behaviors are divided based on the clusters.

[0034] Preferably, in one embodiment of the present invention, cluster analysis is performed on the production-cleaning process based on the relationship matrix to obtain several types of cleaning behaviors, including the following specific methods: A relationship matrix is ​​obtained for all production-cleaning processes in the training data set, and density clustering is performed on all production-cleaning processes. In this embodiment, DBSCAN clustering is adopted, and the distance metric adopts the inverse proportional value of the cosine similarity between the relationship matrices of each production-cleaning process (the difference obtained by subtracting the cosine similarity from 1) to obtain several clusters. The production-cleaning processes in the same cluster are regarded as the same type of cleaning behavior, and each type of cluster corresponds to a type of cleaning behavior.

[0035] At this point, by conducting a correlation analysis on the operating parameters and production parameters in the production-cleaning process, and considering the impact of the stability of the generated products at different stages on the parameter performance, the correlation coefficients are quantified in segments and the correlation values ​​are obtained, and then a relationship matrix is ​​constructed to reflect the correlation performance of the operating parameters and production parameters. Similar correlation performance indicates that the overall parameter operation relationship of the production-cleaning process is similar, and it is more likely to correspond to the same type of cleaning behavior.

[0036] Step S003: Analyze the moments when the resistance of the filter bag changes greatly during the production-cleaning process, as well as the final resistance information of the previous production-cleaning process, and combine the moments when the parameter data of each parameter changes greatly to obtain several short-term deviation time points of each production-cleaning process; obtain several long-term marking parameters of each production-cleaning process through the changes in the correlation of the same position in the time-adjacent relationship matrix of the same cleaning behavior.

[0037] It should be noted that it is expected that the retained neurons can retain long-term information and short-term change information. The long-term information represents the aging information of the filter bag during use (the process in the filter bag of the entire dust removal system is: dust accumulation-formation of dust accumulation layer-cleaning behavior-cleaning and stripping (but remaining) dust, so with the continuous repetition of production-cleaning work, there will be some factors that affect the lifespan, which corresponds to the long-term information difference), and the short-term difference represents the accumulation of deposition information in a production-cleaning process, that is, the information change from the initial dust accumulation.

[0038] As an example, the specific method included in this step is: Based on the pressure difference change at the inlet and outlet of the filter bag during one production-cleaning process, and the pressure difference at the end of the previous production-cleaning process, several pressure difference change moments are screened and obtained; Obtain several changing moments of various parameters during a production-cleaning process, perform difference analysis based on the changing moments of pressure difference, and obtain several short-term deviation time points of each production-cleaning process; The changes of elements at the same position in the relationship matrix of the production-cleaning processes that are adjacent in time sequence in the same type of cleaning behavior are analyzed to obtain several long-term marker parameters of each production-cleaning process.

[0039] Preferably, in one embodiment of the present invention, based on the pressure difference change at the inlet and outlet of the filter bag during one production-cleaning process and the final pressure difference during the previous production-cleaning process, a plurality of pressure difference change moments are screened and obtained, including the specific method of: For the training dataset Second production-cleaning process (the first in the chronological order) The pressure difference between the inlet and outlet at each moment is obtained, that is, the pressure difference between the inlet and outlet of the filter bag is used to reflect the resistance growth rate of the filter bag. Since the inlet pressure is greater than the outlet pressure, the difference between the inlet pressure data and the outlet pressure data is used as the inlet and outlet pressure difference at each moment. At the same time, since the resistance growth rate is affected by the previous production-cleaning process in the sequence, the first The inlet and outlet pressure difference at the last moment of the production-cleaning process; the coordinate system is constructed with the order value corresponding to each moment as the horizontal axis and the pressure difference as the vertical axis. The inlet and outlet pressure differences at all times during the production-cleaning process are mapped to the coordinate system according to the corresponding sequence values, and the pressure difference data points at each time are obtained. The inlet and outlet pressure difference at the last moment of the production-cleaning process is used as the pressure difference corresponding to the origin of the horizontal coordinate, that is, the pressure difference at the 0th moment, and the corresponding pressure difference data point is obtained. The pressure difference data points are connected to obtain the The pressure difference data curve of the production-cleaning process is used to calculate the slope of the pressure difference data points at each moment except the 0th moment in the pressure difference data curve. The slope is obtained by subtracting the inlet and outlet pressure difference at the previous moment from the inlet and outlet pressure difference at the next adjacent pressure difference data point, and the ratio of the difference to the sampling time interval (i.e., the interval between adjacent moments). The slope is used as the change slope of the next pressure difference data point. It is particularly noted that the change slope of the first moment is obtained from the inlet and outlet pressure difference at the 0th moment, and is also compared with the sampling time interval to obtain the change slope. The slope threshold is preset. In this embodiment, the slope threshold is used. The slope threshold is 1 / 5 of the ratio of the range of the inlet and outlet pressure difference at all times during the production-cleaning process to the sampling time interval. The time corresponding to the pressure difference data point whose slope is greater than the slope threshold is taken as the first The moment when the pressure difference changes during the production-cleaning process is the moment when the resistance information changes most.

[0040] It should be noted that the pressure difference data curve corresponds to the dust deposition situation during the production-cleaning process. Therefore, the pressure difference data curve should correspond to the changing moments of the operating parameters and production parameters of this process. If the moments do not correspond, it indicates that there is some unique short-term information that affects the life of the filter bag, and it needs to be used as a short-term deviation time point for subsequent pruning.

[0041] Preferably, in one embodiment of the present invention, several changing moments of various parameters in a production-cleaning process are obtained, and difference analysis is performed in combination with the pressure difference changing moment to obtain several short-term deviation time points of each production-cleaning process. The specific method includes: For the The data change curve of any operating parameter or production parameter in the production-cleaning process is obtained, and the slope of each data point is preset. The parameter slope threshold is used in this embodiment. The 1 / 5 of the ratio of the parameter data range of the operating parameter or production parameter to the sampling time interval at all times during the production-cleaning process is used as the parameter slope threshold, and the time corresponding to the data point with a slope greater than the parameter slope threshold is used as the first The moment when the parameter (operating parameter or production parameter) of the production-cleaning process changes.

[0042] Furthermore, based on Several pressure difference change moments of the production-cleaning process and several change moments of any parameter (operating parameter or production parameter), for any pressure difference change moment, obtain the absolute value of the difference between it and each change moment of the parameter, and use the minimum value as the time offset of the pressure difference change moment; obtain the time offset of each pressure difference change moment, preset the time offset threshold, and the time offset threshold of this embodiment is described as 10 seconds. If the time offset of any pressure difference change moment is greater than the time offset threshold, the pressure difference change moment is regarded as a short-term deviation time point, and the first pressure difference change moment is obtained according to the above method. Secondary cleaning - several short-term deviation points in the production process.

[0043] It should be further explained that in the same type of cleaning behavior, the change difference of the elements in the same position in the relationship matrix adjacent in time sequence is obtained. The change difference represents the loss caused by long-term accumulation, that is, aging information. In other words, the entire production-cleaning process should be analyzed as a whole during the comparison process.

[0044] Preferably, in one embodiment of the present invention, the changes in the elements at the same position in the relationship matrix of the production-cleaning processes that are adjacent in time sequence in the same type of cleaning behavior are analyzed to obtain several long-term marker parameters of each production-cleaning process, including the specific method as follows: For several production-cleaning processes in any type of cleaning behavior, the relationship matrices of all production-cleaning processes are sorted in time sequence. For adjacent relationship matrices, the absolute value of the difference between the elements (correlation values) at the same position is calculated as the element difference at the corresponding position. An aging threshold is preset. In this embodiment, the aging threshold is described as 0.3. If the element difference is greater than the aging threshold, the operating parameters and production parameters corresponding to the elements at that position in the production-cleaning processes corresponding to the two relationship matrices are marked as long-time marking parameters in the two production-cleaning processes. According to the above method, the aging of adjacent relationship matrices in the same type of cleaning behavior is judged to obtain several long-time marking parameters for each production-cleaning process.

[0045] At this point, by performing deviation analysis on the filter bag resistance changes and parameter changes, the short-term deviation time points are screened to reflect the impact of its unique parameter information on the filter bag life and used in the neuron pruning in the subsequent model; and through the correlation changes in the relationship matrix of similar cleaning behaviors, the aging information is analyzed to conduct long-term impact of periodic changes, and then the corresponding parameters are marked to reflect the long-term aging information of the corresponding operating parameters or production parameters.

[0046] Step S004: Based on the short-term deviation time points and long-term marking parameters of each production-cleaning process in the training data set, the neurons in the LSTM model are scored and pruned, and a filter bag life prediction model is constructed to realize the operation supervision of the filter bag dust removal system.

[0047] It should be noted that it is expected to screen the network branches in the LSTM network that can represent both short-term and long-term information. Therefore, during the LSTM training process, the neuron activation status at these time points can determine the importance score of the neuron, and then prune it according to the importance score; the neuron activation status is represented by the neuron's hidden state value and cell state value, where the neuron's hidden state value stores short-term information, while the cell state value stores long-term information.

[0048] Specifically, for any production-cleaning process, for any short-term deviation time point, the number of neighbors is preset. In this embodiment, the number of neighbors is described as 5. The moments of the number of neighbors before and after the short-term deviation time point, as well as the short-term deviation time point, together constitute a short-term time window of the short-term deviation time point, and the average of the hidden state values ​​of each neuron in the short-term time window is used as the short-term activation intensity of the neuron corresponding to the short-term deviation time point.

[0049] Furthermore, for any long-term marking parameter in this production-cleaning process, the mean (entire cycle) of the cell state values ​​of all neurons corresponding to the long-term marking parameter in this production-cleaning process is used as the long-term activation intensity of the corresponding neurons at each moment under this long-term marking parameter.

[0050] Furthermore, the sum of the short-term activation intensity and the long-term activation intensity of any neuron (a neuron corresponding to an operating parameter or a moment in a production parameter during a production-cleaning process) is used as the final score of the neuron, and a screening ratio is preset. In this embodiment, the screening ratio is described as 60%. All neurons are arranged from large to small according to the final score, and the neurons in the front screening ratio are retained. The remaining neurons are subjected to structured pruning, and the LSTM model is trained to obtain a trained LSTM model; wherein the input data of the LSTM model is the training data set, and the output data is the filter bag life judged by professional manual labor, and the loss function adopts the root mean square error function; in the parameter update process of the LSTM model, the set hyperparameters are determined according to the specific implementation situation, and this embodiment does not make specific limitations; the trained LSTM model is used as a filter bag life prediction model to predict the life of the filter bag in the bag dust collector, thereby realizing the operation supervision of the filter bag dust removal system.

[0051] At this point, the neurons are scored and pruned by using short-term deviation time points and long-term aging information, which improves the reasoning efficiency of the LSTM prediction model and builds a filter bag life prediction model. This enables visual monitoring of the life of the filter bag dust removal system in the bag dust collector, thereby improving the operation and supervision efficiency of the filter bag dust removal system.

[0052] 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 principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for supervising the operation of a bag filter dust removal system, characterized in that: The method comprises the following steps: The filter bags arranged in the bag filter are used to obtain parameter data of various operating parameters and production parameters during multiple production and cleaning processes, and the pressure data of the filter bag inlet and outlet are recorded in real time; Based on the data change trends of the operating parameters in the production-cleaning process, the process is segmented, and the correlation between the operating parameters and the production parameters is analyzed in combination with the data changes of the production parameters. A relationship matrix of the production-cleaning process is constructed. Based on the relationship matrix constructed based on the correlation between the operating parameters and the production parameters, similar production-cleaning processes are analyzed to obtain several types of cleaning behaviors. Analyze the moments when the filter bag resistance changes significantly during the production-cleaning process, as well as the final resistance information from the previous production-cleaning process. Combined with the moments when the parameter data of each parameter changes significantly, several short-term deviation time points of each production-cleaning process are obtained. By analyzing the correlation changes at the same position in the relationship matrix of adjacent time sequences for similar cleaning behaviors, several long-term marker parameters of each production-cleaning process are obtained. Based on the short-term deviation time points and long-term marking parameters of each production-cleaning process in the training dataset, the neurons in the LSTM model are scored and pruned, and a filter bag life prediction model is constructed to realize the operation supervision of the filter bag dust removal system.

2. A method for monitoring the operation of a bag filter dust removal system according to claim 1, characterized in that: The method of segmenting based on the data change trend of the operating parameters in the production-cleaning process, analyzing the correlation between the operating parameters and the production parameters in combination with the data change of the production parameters, and constructing the relationship matrix of the production-cleaning process includes the following specific methods: Analyze the fluctuation of the slope of the parameter data of the operating parameters during the primary production-cleaning process, and segment the data change curve of the operating parameters; The correlation analysis between operating parameters and production parameters was conducted in combination with the data change curve of production parameters, and the relationship matrix of each production-cleaning process was constructed.

3. A method for monitoring the operation of a bag filter dust removal system according to claim 2, characterized in that: The analysis of the fluctuation of the slope of the parameter data of the operating parameters in the primary production-cleaning process and segmenting the data change curve of the operating parameters include the following specific methods: For any production-cleaning process, based on the time series parameter data of each operating parameter in the production-cleaning process and the time series parameter data of each production parameter in the production-cleaning process, obtain the data change curve of each operating parameter and the data change curve of each production parameter; For the The data change curve of the operating parameters is obtained, and the slope of each data point is obtained; for any data point, the variance of the slopes of the data point and all the data points before it is obtained as the variance of the slope before the data point, and the variance of the slopes of all the data points after the data point is obtained as the variance of the slope after the data point; Based on the front slope variance and the back slope variance of each data point in the data change curve, a first dividing point and a second dividing point of each operating parameter are obtained; The data curve between the first data point and the first division point is taken as the The initial data change curve of the operating parameters; the data curve between the first data point after the first division point and the second division point is taken as the first The mid-term data change curve of the operating parameters; the data curve composed of all data points after the second division point is used as the first The later data change curve of the operating parameters.

4. A method for monitoring the operation of a bag filter dust removal system according to claim 3, characterized in that: The specific method of obtaining the first dividing point and the second dividing point of each operating parameter includes: The first The front slope variance of each data point in the data change curve of the operating parameter is arranged in the time sequence of the data points to obtain the first The front slope variance sequence of the data change curve of the operating parameters is obtained, and several maximum values ​​are obtained for the front slope variance sequence. The judgment starts from the first maximum value, and the front slope variance threshold is set. If the first maximum value is greater than the front slope variance threshold, the data point corresponding to the first maximum value is taken as the first maximum value. The first dividing point of the data change curve of the operating parameter is selected. If the first maximum value is less than or equal to the previous slope variance threshold, the second maximum value is judged until the first dividing point is obtained; Arrange the posterior slope variances of all data points after the first partition point in the time sequence of the data points to obtain the The stable posterior slope variance sequence of the data change curve of the operating parameters is obtained, and several maximum values ​​are obtained for the stable posterior slope variance sequence. The judgment starts from the first maximum value, and the posterior slope observation threshold is set. If the first maximum value is greater than the posterior slope variance threshold, the data point corresponding to the first maximum value is taken as the first maximum value. The second dividing point of the data change curve of the operating parameter is selected. If the first maximum value is less than or equal to the previous slope variance threshold, the second maximum value is judged until the second dividing point is obtained.

5. The method for monitoring the operation of a bag filter dust removal system according to claim 3, characterized in that: The specific method of constructing the relationship matrix of each production-cleaning process includes: According to the The first and second dividing points of the operating parameters are The data change curves of the production parameters are obtained in the early, mid and late stages respectively; operating parameters and The data change curves of each production parameter are used to obtain the Pearson correlation coefficients of the data change curves at the early, mid and late stages; The initial weight, mid-term weight and late weight are preset to correspond to the initial, mid-term and late data change curves respectively. The Pearson correlation coefficients of the three segments of data change curves are weighted and summed by the corresponding weights. The sum is used as the first weight in the production-cleaning process. The operating parameters and Correlation values ​​of production parameters; Obtain the correlation values ​​between each operating parameter and each production parameter in the production-cleaning process, and construct a matrix, in which the same row in the matrix represents the correlation value between the same operating parameter and each production parameter, and the same column represents the correlation value between each operating parameter and the same production parameter. The obtained matrix is ​​used as the relationship matrix of the production-cleaning process.

6. The method for monitoring the operation of a bag filter dust removal system according to claim 1, characterized in that: The specific methods of obtaining several types of dust removal behaviors include: All production-cleaning processes are density clustered, and the distance measurement adopts the inverse proportional value of the cosine similarity between the relationship matrices of each production-cleaning process to obtain several clusters. The production-cleaning processes in the same cluster are regarded as the same type of cleaning behavior.

7. The method for monitoring the operation of a bag filter dust removal system according to claim 1, characterized in that: The analysis of the moments when the filter bag resistance changes greatly during the production-cleaning process, as well as the final resistance information of the previous production-cleaning process, is combined with the moments when the parameter data of each parameter changes greatly to obtain several short-term deviation time points of each production-cleaning process. The specific method includes: Based on the pressure difference change at the inlet and outlet of the filter bag during one production-cleaning process, and the pressure difference at the end of the previous production-cleaning process, several pressure difference change moments are screened and obtained; Several changing moments of various parameters in a production-cleaning process are obtained, and difference analysis is performed in combination with the pressure difference changing moments to obtain several short-term deviation time points of each production-cleaning process.

8. The method for monitoring the operation of a bag filter dust removal system according to claim 7, characterized in that: The screening method for obtaining a plurality of pressure difference change moments includes: For the training dataset During the production-cleaning process, the pressure difference between the inlet and outlet at each moment is obtained, and the difference obtained by subtracting the pressure data of the outlet from the pressure data of the inlet is used as the pressure difference between the inlet and outlet at each moment; The inlet and outlet pressure difference at the last moment of the production-cleaning process; the coordinate system is constructed with the order value corresponding to each moment as the horizontal axis and the pressure difference as the vertical axis. The inlet and outlet pressure differences at all moments in the production-cleaning process are mapped to the coordinate system according to the corresponding sequence values, and the pressure difference data points at each moment are obtained. The inlet and outlet pressure difference at the last moment of the production-cleaning process is used as the pressure difference corresponding to the origin of the horizontal axis. Connect the pressure difference data points to get the The pressure difference data curve of the production-cleaning process is used to calculate the slope of the pressure difference data points at each moment except the 0th moment in the pressure difference data curve, and the slope is used as the change slope of the next pressure difference data point; The slope threshold is preset, and the time corresponding to the pressure difference data point whose slope is greater than the slope threshold is taken as the first The moment when the pressure difference changes during the production-cleaning process.

9. The method for monitoring the operation of a bag filter dust removal system according to claim 8, characterized in that: The specific method for obtaining several short-term deviation time points of each production-cleaning process includes: For the The data change curve of any operating parameter or production parameter in the production-cleaning process is obtained, the slope of each data point is obtained, the parameter slope threshold is preset, and the time corresponding to the data point with a slope greater than the parameter slope threshold is used as the first The moment when this parameter changes during the production-cleaning process; Based on the Several pressure difference change moments and several change moments of any parameter in the production-cleaning process. For any pressure difference change moment, obtain the absolute value of the difference between it and the change moments of the parameter, and use the minimum value as the time offset of the pressure difference change moment; Obtain the time offset of each pressure difference change moment. If the time offset of any pressure difference change moment is greater than the time offset threshold, take the pressure difference change moment as a short-term deviation time point and obtain the first Secondary cleaning - several short-term deviation points in the production process.

10. The method for monitoring the operation of a bag filter dust removal system according to claim 5, characterized in that: The specific method for obtaining the long-term marking parameters of each production-cleaning process is as follows: For several production-cleaning processes in any type of cleaning behavior, sort the relationship matrices of all production-cleaning processes in chronological order. For adjacent relationship matrices, calculate the absolute value of the difference between the elements at the same position and use it as the element difference at the corresponding position. If the element difference is greater than the aging threshold, the operating parameters and production parameters corresponding to the elements at that position in the production-cleaning process corresponding to the two relationship matrices are marked as long-term marking parameters in the two production-cleaning processes; the aging of adjacent relationship matrices in the same type of cleaning behavior is judged to obtain several long-term marking parameters for each production-cleaning process.

Images