Data-driven intelligent dust remover operation state sensing method

Abstract

The invention relates to the technical field of data processing, in particular to a method for sensing the running state of an intelligent dust remover driven by data, which comprises the following steps of: acquiring monitoring data of the dust remover in real time; preprocessing the operation parameter monitoring data stream; performing time window preliminary interception processing on the operation data stream of each dimension to obtain a mother data stream of each dimension; performing time window secondary interception processing on the mother data stream of each dimension to obtain a sub data stream of each dimension; based on the similarity between the sub-data stream of each dimension and the corresponding mother data stream, the operation fault possibility is obtained through calculation; and judging the running state of the dust remover based on the running fault possibility. According to the invention, the operation state is sensed according to the sub-data streams extracted according to the data change degree structure of the data stream and the sub-time window with the flexibly changed data fluctuation degree, so that the sensing accuracy can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for sensing the operating status of an intelligent dust collector driven by data. Background Art

[0002] As an important part of industrial environmental protection facilities, the operating status sensing technology of dust collectors has developed rapidly in recent years. With the continuous improvement of industrial emission standards and the increasingly stringent requirements for energy conservation and environmental protection, the operating status sensing technology of dust collectors has become a key technology to improve dust collector efficiency, reduce operating costs, and extend the service life of equipment. With the advancement of information technology and sensor technology, the operating status sensing technology of dust collectors has been significantly improved. Modern dust collectors generally use online monitoring systems to monitor the operating status of equipment in real time through sensors.

[0003] The dust collector will generate a large amount of streaming data during operation. Usually, the operating status perception is achieved by performing fixed-size window analysis on the streaming data. However, this fixed window size method cannot effectively adapt to the operating environment. The accuracy of the dust collector's operating status perception results is often low due to unreasonable window size division. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a data-driven intelligent dust collector operating status perception method. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows: In a first aspect, the present application provides a data-driven method for sensing the operating status of an intelligent dust collector, the method comprising: collecting monitoring data of the dust collector in real time and converting it into an operating parameter monitoring data stream, the monitoring data comprising operating parameter monitoring data of at least two dimensions; preprocessing the operating parameter monitoring data stream to obtain an operating data stream corresponding to each dimension; performing preliminary time window interception processing on the operating data stream of each dimension respectively to obtain a mother data stream of each dimension in a mother time window, the duration of the mother time window is calculated by the degree of data change in each operating data stream, and the end time of the mother time window is the current time; performing secondary time window interception processing on the mother data stream of each dimension respectively to obtain a child data stream of each dimension in a child time window, the duration of the child time window is calculated by the degree of data fluctuation in each mother data stream, and the end time of the child time window is the current time; calculating the possibility of operating failure based on the similarity between the child data stream of each dimension and the corresponding mother data stream; and determining the operating status of the dust collector based on the possibility of operating failure and a preset judgment threshold.

[0005] In combination with the first aspect, in a possible implementation method, the method for calculating the duration of the mother time window includes: obtaining the cycle duration of the dust removal cycle; performing data extraction processing in the operation data stream based on the cycle duration, extracting from the current moment forward through the cycle duration to obtain an initial operation data stream, and the duration of the initial operation data stream is an integer multiple of the cycle duration; respectively calculating the sum of the degree of change of each numerical value in each dust removal cycle in the initial operation data stream to obtain a cumulative value of the degree of change, and each degree of change of the numerical value is the difference between the data corresponding to one moment in a dust removal cycle and the corresponding data at the previous moment; calculating the initial operation data stream according to the cycle duration, all the cumulative values ​​of the degree of change and a preset first duration calculation function to obtain the duration of the mother time window.

[0006] In combination with the first aspect, in a possible implementation method, the method for calculating the duration of the sub-time window includes: obtaining the cycle duration of the dust removal cycle; calculating an initial coefficient based on the extreme value difference in the mother data stream of the target dimension, where the target dimension is any dimension; calculating a similarity change correction coefficient based on the similarity of changes between all mother data streams in the mother time window of the target dimension; calculating a noise correction coefficient based on the noise intensity in the mother data stream of the target dimension; and correcting the cycle duration based on the noise correction coefficient, the similarity change correction coefficient, and the initial coefficient to obtain the duration of the sub-time window.

[0007] In combination with the first aspect, in a possible implementation method, an initial coefficient is calculated based on the difference of extreme values ​​in the mother data stream of the target dimension, including: screening the mother data stream to obtain a maximum value; calculating a difference based on the maximum value and a preset indicator threshold; calculating the difference between adjacent extreme values ​​in the mother data stream one by one; and calculating the initial coefficient based on the difference between all adjacent extreme values ​​and the difference.

[0008] In combination with the first aspect, in a possible implementation method, a similarity change correction coefficient is calculated based on the change similarity between all the mother data streams in the mother time window of the target dimension, including: processing the mother data streams of all the dimensions based on a preset clustering algorithm to obtain a clustering result, the clustering object of the clustering algorithm is the data collected and preprocessed at each moment; extracting the clustering subclusters under the target dimension from the clustering results; calculating the Euclidean distance between each two of the clustering subclusters one by one, and summing them up to obtain the Euclidean distance cumulative value; respectively calculating the dynamic time warping distance of the mother data streams of the target dimension and other dimensions; and calculating the similarity change correction coefficient based on the Euclidean distance cumulative value and all the dynamic time warping distances.

[0009] In combination with the first aspect, in a possible implementation manner, the clustering algorithm is a k-means clustering algorithm.

[0010] In combination with the first aspect, in a possible implementation method, a noise correction coefficient is calculated based on the noise intensity in the mother data stream of the target dimension, including: performing data decomposition on the mother data stream of the target dimension based on a preset decomposition algorithm to obtain at least one intrinsic mode function component and a residual; respectively counting the maximum peak amplitude in each of the intrinsic mode function components; respectively calculating the variance of each of the intrinsic mode function components; and calculating the noise correction coefficient based on the variance and maximum peak amplitude of each of the intrinsic mode function components.

[0011] In combination with the first aspect, in a possible implementation manner, the decomposition algorithm is an intrinsic time scale decomposition algorithm.

[0012] In combination with the first aspect, in a possible implementation manner, the judgment threshold is 0.8.

[0013] In combination with the first aspect, in a possible implementation manner, the preprocessing includes data cleaning, data standardization and normalization.

[0014] The present invention has the following beneficial effects: In the present invention, a preliminary interception is first performed according to the data change degree of the running data stream to obtain a mother data stream. Then, a secondary interception is performed according to the data fluctuation state of the mother data stream to obtain a child data stream. Through the above method, the sub-data stream obtained by extracting the sub-time window that flexibly changes according to the structural characteristics of the data itself is used to perceive the running status, which can effectively improve the perception accuracy of the running status. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. 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 creative work.

[0016] Figure 1 A schematic diagram of a flow chart of a method for sensing the operating status of an intelligent dust collector driven by data provided by an embodiment of the present invention; Figure 2 A schematic diagram of a flow chart of step S3 provided in one embodiment of the present invention; Figure 3 A schematic flow chart of step S4 provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a data-driven intelligent dust collector operating state perception method proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0018] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0019] The following is a detailed description of a data-driven intelligent dust collector operating status perception method provided by the present invention in conjunction with the accompanying drawings.

[0020] See also Figure 1 , which shows a flow chart of a method for sensing the operating status of an intelligent dust collector using data-driven method provided by an embodiment of the present invention, specifically including steps S1 to S6.

[0021] S1. Collect monitoring data of the dust collector in real time and convert it into an operating parameter monitoring data stream, wherein the monitoring data includes operating parameter monitoring data of at least two dimensions.

[0022] In this step, a total of For example, the operating parameter monitoring data involved in this embodiment includes airflow pressure, dust concentration, temperature and airflow volume collected by different sensors. These parameters together constitute Data monitoring in multiple dimensions. That is, in this embodiment, The specific value of is 4. However, for those skilled in the art, some dimensions can be selected or other dimensions of operating data can be added according to actual needs and specific circumstances. For example, in addition to the above four dimensions, the monitoring of parameters such as motor operating current and motor operating voltage can be added, and a total of The operating parameter monitoring data of each dimension, that is, regarding the specific selection of the operating parameter monitoring data, the technicians in this field can flexibly decide according to the actual application requirements and specific circumstances, and no specific limitation is made in this embodiment.

[0023] S2. Preprocess the operation parameter monitoring data stream to obtain an operation data stream corresponding to each dimension.

[0024] The preprocessing involved in this stage includes but is not limited to data cleaning, data standardization and normalization. Among them, data cleaning aims to ensure the accuracy and completeness of the data by identifying and correcting errors, missing values ​​and outliers in the data. Data standardization refers to converting data according to established rules so that it conforms to a specific format or standard for subsequent processing and analysis. Normalization is to scale the data to a range of 0-1 to eliminate the influence of different dimensions and orders of magnitude, and ensure the fairness and consistency of the data during the analysis process. Through these preprocessing steps, the data quality and the reliability of the analysis results can be significantly improved. That is, the operating data obtained after the above processing is a set of one-dimensional time series data formed by the preprocessing of the collected data at each moment. It can also be called the time series curve of each dimension under the time series.

[0025] Furthermore, in this embodiment, it is considered that the dust collector will be cleaned regularly during operation, and the accumulation and regular cleaning of dust will affect the airflow pressure, dust concentration, temperature, and airflow volume. For example, during the normal operation of the dust collector, the airflow pressure increases with the increase of dust accumulation on the filter bag. The dust accumulation will increase the resistance of the airflow, resulting in an increase in pressure. The periodic cleaning operation will reduce the dust accumulation on the filter bag, reduce the airflow resistance, and cause the airflow pressure to drop, and the system pressure will return to a lower level. Therefore, correspondingly, the operating data flows of these dimensions will be in a certain change cycle when they fluctuate normally.

[0026] Therefore, in this embodiment, the duration of a dust removal cycle of the dust collector is recorded as , and analyze the correlation between the operating parameter monitoring data of each dimension and the dust cleaning operation one by one. The stronger the correlation, the stronger the periodicity of the change in the dimension. Therefore, this embodiment can first select a parent time window containing fewer dust collector cleaning cycles to ensure that the data streams in the parent time window are all data during the normal operation of the dust collector, and then compare the subsequent intercepted data with the normal operation data according to the corresponding time, so as to achieve fault identification. For detailed steps, please refer to steps S3-S5.

[0027] S3. Perform preliminary time window interception processing on the operating data stream of each dimension respectively to obtain the mother data stream of each dimension within the mother time window. The duration of the mother time window is calculated by the degree of data change in each operating data stream, and the end time of the mother time window is the current time.

[0028] Therefore, according to the above analysis, this embodiment believes that when analyzing the correlation between the operation data flow of each dimension and the change of the dust removal cycle, it should be observed whether the operation data flow of one dimension has a numerical change accompanying the dust cleaning operation, and the degree of numerical change in multiple dust cleaning operations is similar. If it meets the above change characteristics, it means that the data of this dimension is more correlated with the periodic change of the dust cleaning operation. Therefore, this feature is used to confirm the duration of the parent time window. See Figure 2 , Figure 2 It is shown in the figure that step S3 also includes steps S31 to S34.

[0029] S31. Obtaining the duration of the dust removal cycle .

[0030] S32. Perform data extraction processing in the running data stream based on the cycle duration, extract from the current moment forward by the cycle duration to obtain an initial running data stream, the duration of the initial running data stream being an integer multiple of the cycle duration.

[0031] Specifically, in this example, the duration of the initial running data flow is 5 times the duration of the cycle. At the same time, those skilled in the art may select other multiples, which are not specifically limited in this embodiment.

[0032] S33. Calculate the sum of the degree of change of each numerical value in each dust removal cycle in the initial operation data stream to obtain a cumulative value of the degree of change, wherein each degree of change of the numerical value is the difference between the data corresponding to one moment in a dust removal cycle and the corresponding data at the previous moment.

[0033] S34. Calculate the initial operation data stream according to the cycle duration, all the accumulated values ​​of the degree of change and a preset first duration calculation function to obtain the duration of the mother time window.

[0034] Specifically, the first duration calculation function mentioned in this step is as follows: ; in, Indicates The length of the parent time window of each dimension; Indicates the cycle duration of the dust removal cycle; Indicates the multiple of the initial operation data flow duration and the dust removal interval duration. In this embodiment ; represents the maximum and minimum normalized function; Indicates the total number of dimensions; Indicates that in the initial run data stream, The cumulative value of the change degree of the nth dimension within the dust removal interval; Represents the variance of the initial run data stream.

[0035] In the above calculation function, the variance It can express the discrete degree of data within a dust removal cycle or between two dust removal operations, which can show the degree of change of data from one aspect. At the same time, the cumulative value of the degree of change It can express the degree of change of data within a dust removal cycle or between two cleaning operations, which can show the degree of change of data from another aspect. Combining the data characteristics of the above two aspects, it can better reflect whether the operation data flow of one dimension has a numerical change accompanied by the cleaning operation, and the degree of numerical change in multiple cleaning operations is similar. If it meets the above change characteristics, it means that the data of this dimension is more correlated with the periodic change of the cleaning operation. The larger the value of , the smaller the correlation. The smaller the value of .

[0036] At the same time, in this embodiment, it is also considered that during the operation of the dust collector, the fault data is often expressed as a section of abnormal data. In order to perceive the operating status of the dust collector based on the sensor data obtained in real time, it is also necessary to further segment the parent data stream of each dimension using a sub-time window. Therefore, see step S4 for details.

[0037] S4. Perform secondary time window interception processing on the mother data stream of each dimension respectively to obtain a sub-data stream in a sub-time window of each dimension. The duration of the sub-time window is calculated by the degree of data fluctuation in each mother data stream, and the end time of the sub-time window is the current time.

[0038] In this embodiment, it is considered that too large a sub-time window will lead to untimely feedback of the operating status, and the same sub-time window will capture too much interference data and cannot accurately capture fault data. However, too small a sub-time window may cause excessive burden on computing resources and affect the real-time performance of the system. Therefore, in this embodiment, it is necessary to adjust the duration of the sub-time window in real time according to the changes in the data. Therefore, in this embodiment, one of the dimensions is used as an exemplary sub-time window duration calculation, see Figure 3 Steps S41 to S44 shown in FIG.

[0039] S41. Calculate initial coefficients based on extreme value differences within the mother data stream of a target dimension, where the target dimension is any dimension.

[0040] In this embodiment, it is considered that when the data changes relatively smoothly, the possibility that the data is fault data is small. In order to save computing resources, a larger sub-time window should be selected; when the data changes and fluctuates greatly, it may indicate that the dust collector has an operating failure. At this time, a smaller sub-time window needs to be selected to better obtain the fault status data of the dust collector. Specifically, in this step S4, the exemplary calculation of the initial duration shown in steps S411-S414 is also included.

[0041] S411, screening the mother data stream to obtain a maximum value.

[0042] S412: performing a difference calculation based on the maximum value and a preset indicator threshold to obtain a difference value.

[0043] S413, calculating the differences between adjacent extreme values ​​in the mother data stream one by one.

[0044] S414. Calculate an initial coefficient based on the difference between all adjacent extreme values ​​and the difference.

[0045] Specifically, the calculation function of the initial coefficient in this step is as follows: ; in, Indicates The initial coefficients of the sub-time windows of dimensions; Indicates The number of extreme values ​​in the parent data stream of each dimension; Indicates The first dimension in the parent data stream The extreme value and The difference between extreme values; Represents the absolute value calculation function; Indicates The difference between the maximum value in the parent data stream of the dimension and the preset indicator threshold; Represents a natural constant.

[0046] In this embodiment, it is considered that for the determination of the sub-time window, when a certain dimension is The greater the difference between the data at each moment, the greater the change in the data of the dimension, and the larger the sub-time window corresponding to the data dimension should be, so as to avoid the sub-window selected during fault identification being too small to capture important features. At the same time, in this embodiment, it is also considered that when a certain value of a mother data stream is closer to the preset indicator threshold of the dimension, the possibility of the dust collector failure is greater. It should also be noted that the preset indicator threshold of a dimension mentioned in this embodiment can usually be set to be greater than ten times the average value of the dimension.

[0047] S42. Calculate a similarity change correction coefficient based on the change similarity between all the mother data streams in the mother time window of the target dimension.

[0048] In this embodiment, when the data change curves of multiple dimensions are relatively similar, the information repetition of these dimensions may be high, and the sub-window corresponding to the corresponding data dimension should be smaller to avoid considering too much repeated information during fault identification. Therefore, the similarity between multiple dimensions is also considered in this application. Therefore, in order to clarify the calculation process of the similarity change correction coefficient, see steps S421-S425 for details.

[0049] S421. Processing the parent data streams of all the dimensions based on a preset clustering algorithm to obtain clustering results, wherein the clustering object of the clustering algorithm is the data collected and pre-processed at each moment.

[0050] Specifically, the clustering algorithm used in this embodiment is the k-means clustering algorithm, which is an algorithm widely used in the field of data mining and statistical analysis. It distributes data points to k clusters through an iterative process so that the sum of the squares of the distances between each data point and the center point of the cluster to which it belongs is minimized. However, those skilled in the art should understand that in addition to the k-means clustering algorithm, there are many other types of clustering algorithms to choose from, such as hierarchical clustering, DBSCAN, spectral clustering, etc. Those skilled in the art may also use other clustering algorithms, which are not specifically limited in this embodiment.

[0051] S422: extracting cluster subclusters under the target dimension from the clustering results.

[0052] S423, calculating the Euclidean distance between every two of the cluster sub-clusters one by one, and summing them up to obtain an accumulated Euclidean distance value.

[0053] S424, respectively calculating the dynamic time warping distances of the mother data streams of the target dimension and other dimensions.

[0054] In this step, the dynamic time warping distance is calculated by a dynamic time warping algorithm (DTW for short), which is an algorithm for measuring the similarity between two time series (data series with a time order). Since it is a prior art, the specific calculation process is not described in detail in this embodiment.

[0055] S425. Calculate a similarity change correction coefficient based on the Euclidean distance accumulation value and all the dynamic time warping distances.

[0056] Specifically, the calculation function of the similarity change correction coefficient in this step is as follows: ; in, Indicates Similarity change correction coefficient of each dimension; Represents the accumulated value of Euclidean distance; Indicates the total number of dimensions; Indicates The first dimension of the parent data flow and the Dynamic time warping distance of the dimensional mother data stream.

[0057] In the above calculation formula, when the data change similarity between the target dimension and other dimensions is similar, when the data change curves of the target dimension and other dimensions are relatively similar, the information repetition of the two dimensions may be high, and the sub-time window corresponding to the corresponding data dimension should be smaller. Therefore, It can better reflect the purpose that the higher the similarity, the smaller the similarity change correction coefficient, and the smaller the corresponding sub-time window should be. It represents the degree of influence of the target dimension on other dimensions. When the target dimension and other dimensions show a high degree of information repetition, the Euclidean distance between the corresponding cluster subclusters should be smaller, the corresponding quasi-change correction coefficient should be smaller, and the subwindow corresponding to the corresponding data dimension should be smaller.

[0058] S43. Calculate a noise correction coefficient according to the noise intensity in the mother data stream of the target dimension.

[0059] At the same time, considering that the data collected by the sensor may generate some noise signals due to the unique mode of the dust collector operation, for example, when the fan starts and stops, the airflow pressure may fluctuate in the short term, resulting in noise in the data. Therefore, the degree of noise signals contained in a certain dimension of data should also be monitored in real time. When a certain data contains more noise signals, the sub-window corresponding to the data dimension should be smaller. Therefore, in order to clarify the calculation process of the noise correction coefficient, see steps S431-S434 for details.

[0060] S431. Perform data decomposition on the mother data stream of the target dimension based on a preset decomposition algorithm to obtain at least one intrinsic mode function component and a residual.

[0061] The decomposition algorithm used in this embodiment is the intrinsic time-scale decomposition (ITD) algorithm. In this embodiment, the ITD algorithm can adaptively decompose the mother data stream into a series of intrinsic mode function components (IMFs) and residuals (Trend), and these IMF components can more intuitively reflect the data changes of the mother data stream at a certain frequency.

[0062] S432, respectively counting the maximum peak amplitude in each of the intrinsic mode function components.

[0063] S433, respectively calculating the variance of each of the intrinsic mode function components.

[0064] S434. Calculate a noise correction coefficient based on the variance and maximum peak amplitude of each of the intrinsic mode function components.

[0065] Specifically, the calculation function of the noise correction coefficient in this step is as follows: ; in, Indicates The noise correction coefficient of each dimension; total represents the total number of intrinsic mode function components; Indicates The maximum peak amplitude of the intrinsic mode function component in each dimension; Represents the average of the variances of all the IMF components.

[0066] In this embodiment, it is considered that noise usually appears as high-frequency, random fluctuations. Therefore, higher-order intrinsic mode function components are more likely to contain noise, and noise usually increases the volatility of the signal, causing the amplitude of these intrinsic mode function components to increase or the frequency to become irregular. At the same time, noise usually causes the variance of the signal to increase. When the variance of the intrinsic mode function component is larger, the intensity of the noise is higher. Therefore, in the above calculation formula It can better describe the variance change of the intrinsic mode function components. At the same time, Giving a larger weight to the high-order intrinsic mode function component and combining the maximum peak amplitude of the intrinsic mode function component can better reflect the noise intensity of the mother data stream.

[0067] S44. Correct the period duration based on the noise correction coefficient, the similarity change correction coefficient, and the initial coefficient to obtain the duration of the sub-time window.

[0068] Specifically, the calculation function of the duration of the sub-time window in this step is as follows: ; in, Indicates The length of the sub-time window corresponding to each dimension; Indicates the cycle duration of the dust removal cycle; represents the maximum-minimum normalized function; Indicates The initial coefficients of the sub-time windows of dimensions; Indicates Similarity change correction coefficient of each dimension; Indicates The noise correction factor in each dimension.

[0069] By combining the coefficients of noise, similarity between different dimensions and characteristics of the data itself, the length of the sub-data stream with appropriate amount of information can be better intercepted.

[0070] S5. Calculate the possibility of operation failure based on the similarity between the sub-data streams in each dimension and the corresponding parent data stream.

[0071] Further, in this embodiment, in order to determine the similarity between a sub-data stream and a parent data stream in a dimension. In this embodiment, the dynamic time warping algorithm (DTW) is used to compare their similarities. And in this embodiment, the length of the sub-time window of each dimension is used as the weight of the dimension, and when the sub-time window of the dimension is longer, it contains more useful information, and it should provide a higher reference when the similarity is compared. At the same time, in this embodiment, it is also considered that when the sub-data stream is compared with the parent data stream, the position of the sub-data stream in a dust cleaning cycle should be found, and the similarity should be compared with the data at the corresponding position of the parent data stream. The higher the similarity, the lower the possibility of dust collector operation failure at this time. That is, first, the parent data stream should be divided into several data stream segments according to the duration of the dust removal cycle, and then the similarity of the sub-data stream with other comparison data streams is compared respectively, and the sum of their similarities is accumulated to obtain the similarity between the sub-data stream and the parent data stream, wherein each comparison data stream is extracted according to the sub-data stream in the time period of a dust removal cycle in other data stream segments.

[0072] Therefore, based on the above analysis, the function formula for calculating the probability of failure in this step is as follows: ; in, Indicates the possibility of operating failure; represents the maximum-minimum normalized function; Indicates the total number of dimensions; Indicates The length of the sub-time window corresponding to each dimension; Indicates The length of the parent time window of each dimension; Indicates the dust removal interval of the dust collector; Indicates The sub-data stream of the dimension is the same as the parent data stream The DTW distance of the comparison data stream, among which, The comparison data stream is a comparison of the first and second sub-data streams in the main data stream according to the duration of the sub-data stream. The cut-off time of the comparison data stream is equal to the first The end time of a dust removal cycle.

[0073] S6. Determine the operating state of the dust collector based on the possibility of the operating failure.

[0074] Therefore, in this embodiment, the operating state of the dust collector can be judged whether it is a fault by the operating fault possibility value and the preset judgment threshold. For example, in this embodiment, the judgment threshold is 0.8, then when the operating fault possibility is greater than 0.8. It is considered that the dust collector is diagnosed as a fault. Once the fault is diagnosed, a fault-tolerant control strategy needs to be adopted to maintain the normal operation of the dust collector.

[0075] In this embodiment, for the large amount of streaming data generated by the dust collector during operation, the prior art usually adopts a method of fixed time window size to analyze the streaming data to realize the perception of the operating status. However, the method of fixed window size has an adverse effect on the accuracy of the operating perception results. Every time a judgment is made, the rationality of the window size cannot be guaranteed, which leads to a reduction in the accuracy of the perception results of the operating status of the dust collector. Therefore, in this embodiment, firstly, according to the degree of data change of the operating data stream, the mother data stream is initially intercepted. Subsequently, a secondary interception is performed according to the data fluctuation state of the mother data stream to obtain the child data stream. In the process of intercepting the child data stream, the characteristics of the operating data stream of each dimension, the similarity between different dimensions, and the inherent noise intensity of each dimension are also fully considered, and the window size is adjusted accordingly. Finally, the sub-data stream extracted by the above-mentioned flexible sub-time window is used to perceive the operating status, which can significantly improve the accuracy of perception.

[0076] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0077] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A data-driven intelligent dust collector operating status perception method, characterized in that: The method comprises: Collect monitoring data of the dust collector in real time and convert it into an operating parameter monitoring data stream, wherein the monitoring data includes operating parameter monitoring data of at least two dimensions; Preprocessing the operating parameter monitoring data stream to obtain an operating data stream corresponding to each dimension; Performing preliminary time window interception processing on the running data stream of each dimension respectively, to obtain the mother data stream of each dimension in the mother time window, the duration of the mother time window is calculated by the degree of data change in each running data stream, and the end time of the mother time window is the current time; Among them, the duration calculation method of the mother time window is as follows: obtain the cycle duration of the dust removal cycle; perform data extraction and processing in the operation data stream based on the cycle duration, extract from the current moment forward through the cycle duration, and obtain the initial operation data stream, and the duration of the initial operation data stream is an integer multiple of the cycle duration; respectively calculate the sum of the degree of change of each numerical value in each dust removal cycle in the initial operation data stream to obtain the cumulative value of the degree of change, and the degree of change of each numerical value is the difference between the data corresponding to one moment in a dust removal cycle and the corresponding data at the previous moment; calculate the initial operation data stream according to the cycle duration, all cumulative values ​​of the degree of change and the preset first duration calculation function to obtain the duration of the mother time window; Performing a secondary time window interception process on the mother data stream of each dimension respectively, to obtain a sub-data stream in a sub-time window of each dimension, wherein the duration of the sub-time window is calculated by the degree of data fluctuation in each mother data stream, and the end time of the sub-time window is the current time; The duration calculation method of the sub-time window is as follows: the initial coefficient is calculated based on the extreme value difference in the mother data stream of the target dimension, and the target dimension is any dimension; the similarity change correction coefficient is calculated based on the change similarity between all mother data streams in the mother time window of the target dimension; the noise correction coefficient is calculated based on the noise intensity in the mother data stream of the target dimension; the period duration is corrected based on the noise correction coefficient, the similarity change correction coefficient, and the initial coefficient to obtain the duration of the sub-time window; Obtaining the possibility of operation failure based on similarity calculation between the sub-data streams of each dimension and the corresponding parent data stream; The operating state of the dust collector is determined based on the operating failure possibility and a preset judgment threshold.

2. The data-driven intelligent dust collector operating status perception method according to claim 1 is characterized in that: The initial coefficients are calculated based on the extreme value differences in the mother data stream of the target dimension, including: Screening the mother data stream to obtain a maximum value; A difference is calculated based on the maximum value and a preset indicator threshold to obtain a difference; Calculating the differences between adjacent extreme values ​​in the mother data stream one by one; The initial coefficients are calculated based on the differences between all adjacent extreme values ​​and the difference.

3. The data-driven intelligent dust collector operating status perception method according to claim 1 is characterized in that: The similarity change correction coefficient is calculated based on the change similarity between all the mother data streams in the mother time window of the target dimension, including: Processing the parent data streams of all the dimensions based on a preset clustering algorithm to obtain clustering results, wherein the clustering object of the clustering algorithm is the data collected and preprocessed at each moment; Extracting cluster subclusters under the target dimension from the clustering results; Calculating the Euclidean distance between each two of the cluster subclusters one by one, and summing them up to obtain the accumulated Euclidean distance value; Calculate the dynamic time warping distance of the target dimension and the mother data stream of the other dimensions respectively; A similarity change correction coefficient is calculated based on the Euclidean distance accumulation value and all the dynamic time warping distances.

4. The data-driven intelligent dust collector operating status perception method according to claim 3 is characterized in that: The clustering algorithm is a k-means clustering algorithm.

5. The data-driven intelligent dust collector operating status perception method according to claim 1 is characterized in that: The noise correction coefficient is calculated according to the noise intensity in the mother data stream of the target dimension, including: Performing data decomposition on the mother data stream of the target dimension based on a preset decomposition algorithm to obtain at least one intrinsic mode function component and a residual; Counting the maximum peak amplitude of each intrinsic mode function component respectively; Calculating the variance of each of the intrinsic mode function components respectively; The noise correction coefficient is calculated based on the variance and the maximum peak amplitude of each of the intrinsic mode function components.

6. The data-driven intelligent dust collector operating status perception method according to claim 5 is characterized in that: The decomposition algorithm is an intrinsic time scale decomposition algorithm.

7. The data-driven intelligent dust collector operating status perception method according to claim 1 is characterized in that: The judgment threshold is 0.

8.

8. The data-driven intelligent dust collector operating status perception method according to claim 1 is characterized in that: The preprocessing includes data cleaning, data standardization and normalization.

Images