An intelligent management method for shared oxygen generator based on the Internet of Things

Abstract

The present invention relates to the technical field of data management, and proposes a shared oxygen concentrator intelligent management method based on the Internet of Things, including: collecting operating data of multiple dimensions of the oxygen concentrator; obtaining the comprehensive change degree of each time point and dividing the time axis to obtain several time periods; clustering the time periods to obtain time period clusters and form slices; obtaining the representative degree of each time period in each slice in each dimension, and obtaining the benchmark time period in each slice under each dimension; obtaining the benchmark change sequence of each slice under each dimension; obtaining the comprehensive difference of each time period; obtaining the initial tolerance parameter and the final tolerance parameter of each time period; performing adaptive revolving door compression on the operating data of the oxygen concentrator in multiple dimensions in each time period, obtaining the compressed operating data of each time period and storing it. The present invention aims to solve the problem that lossy compression causes excessive loss of multi-dimensional data information of the oxygen concentrator, thus affecting the intelligent management of the oxygen concentrator.

Description

Technical Field

[0001] The present invention relates to the technical field of data management, and in particular to an intelligent management method for a shared oxygen generator based on the Internet of Things. Background Art

[0002] The application of Internet of Things (IoT) in the medical field is becoming more and more mature, so as to provide better medical services for patients. As a kind of medical equipment, oxygen concentrator can be managed and used more intelligently and efficiently through the application of IoT technology; however, there are some problems with the traditional management methods of oxygen concentrators, such as low equipment utilization and difficulty in real-time monitoring of equipment status. IoT technology connects oxygen concentrators to the Internet to realize data interaction and real-time monitoring between devices; this makes it possible to manage oxygen concentrators intelligently, for example, sensors can be used to monitor oxygen concentration, equipment working status and other information in real time, and transmit these data to the cloud for analysis and storage.

[0003] However, when the oxygen concentrator transmits and stores multidimensional data through the Internet of Things, the amount of data generated by the multidimensional data is too large, and it is necessary to use a distributed storage system to store the multidimensional data with a large amount of data generated by the oxygen concentrator; and after slicing the multidimensional data, directly compressing and storing the slices through the traditional revolving door compression algorithm will cause excessive information loss of the multidimensional data, which in turn affects the analysis and adjustment of the oxygen concentrator operation data itself, resulting in errors in the intelligent management of the oxygen concentrator. Summary of the invention

[0004] The present invention provides an intelligent management method for a shared oxygen concentrator based on the Internet of Things to solve the problem that the existing lossy compression causes excessive loss of multi-dimensional data information of the oxygen concentrator, thus affecting the intelligent management of the oxygen concentrator. The technical solution adopted is as follows:

[0005] An embodiment of the present invention provides an intelligent management method for a shared oxygen concentrator based on the Internet of Things, the method comprising the following steps:

[0006] Collecting operation data of multiple dimensions of the oxygen concentrator, wherein the operation data of the multiple dimensions correspond to the same time axis, and the time axis includes a plurality of time points;

[0007] According to the changes of operating data at the same time point in different dimensions, the comprehensive change degree of each time point is obtained, and the time axis is divided into several time periods; according to the correlation of the changes of operating data in different dimensions in different time periods, the correlation distance between time periods is obtained, and several time period clusters are obtained by clustering; the operating data of multiple dimensions under each time period cluster is sliced;

[0008] According to the similarity of the change trend of the running data in different time periods in the same slice under the same dimension, the representative degree of each time period in each slice in each dimension is obtained, and the benchmark time period in each slice under each dimension is obtained; for the running data of several time periods and the benchmark time period in the same slice under each dimension, the benchmark change sequence of each slice under each dimension is obtained through data matching;

[0009] According to the similarity between the running data of each time period under each dimension and the benchmark change sequence of each slice under each dimension, the comprehensive difference of each time period is obtained; according to the comprehensive difference, the initial tolerance parameter of each time period is obtained, and the final tolerance parameter of each time period is obtained by the revolving door compression algorithm combined with the comprehensive change degree of the time point;

[0010] According to the final tolerance parameters of each time period, the operating data of the oxygen concentrator in multiple dimensions in each time period are adaptively compressed by a revolving door to obtain and store the compressed operating data of each time period.

[0011] Optionally, the obtaining of the representativeness of each time period in each slice in each dimension to obtain the reference time period in each slice under each dimension includes the following specific methods:

[0012] Combining any two time periods in the same slice into a group, we can get several time period combinations; Slice The time period combination of time periods is recorded as Slice The reference time period combination of time periods;

[0013] According to the similarity of data change trends in different time periods in the same slice under the same dimension, combined with the preset window scale, obtain the trend difference of each reference time period combination of each time period in each slice under each window scale combination of each dimension;

[0014] No. Slice The time period is The degree of representation of the dimensions The calculation method is:

[0015]

[0016] in, Indicates Slice The number of reference time period combinations for time periods, Indicates Slice The time period The first The number of time points in a time period, Indicates Slice The time period The reference time period combinations except The number of time points in another time period outside the time period, Indicates The maximum absolute value of the difference between the number of time points in two time periods in all time period combinations of slices, represents the number of window scale combinations, Indicates Slice The time period The reference time period is combined in Dimension The trend difference under the combination of window scales, It means to find the absolute value. represents an exponential function with a natural constant as base;

[0017] Get the Each time segment in the slice The degree of representation of the dimensions will be The time period corresponding to the maximum value of the representative degree of each dimension is taken as the The dimension in The base time period for each slice.

[0018] Optionally, the step of obtaining the trend difference of each reference time period combination of each time period in each slice under each window scale combination of each dimension includes the following specific methods:

[0019] For In several time periods of a slice, The dimension in The sequence of running data corresponding to each time period in the slice is recorded as Slice Several operating data series of dimensions; perform STL time series decomposition on each operating data series through each window scale to obtain the trend item of each operating data series at each window scale;

[0020] The preset window scales are combined in pairs to obtain several window scale combinations; the first window scale in the window scale combination corresponds to the first window scale in the reference time period combination. The running data sequence corresponding to the time period, the second window scale corresponds to the reference time period combination except the For any reference time period combination and any window scale combination, obtain the trend items of the operation data sequences corresponding to the two time periods in the reference time period combination under the window scale combination, calculate the DTW distance of the two trend items, and record the result as the reference time period combination in the first The trend difference under this window scale combination in each dimension.

[0021] Optionally, the reference change sequence of each slice in each dimension is specifically obtained by:

[0022] For the operating data of several time periods and the benchmark time period in the same slice under each dimension, the benchmark data sequence and several adjusted data sequences in the same slice under each dimension are obtained through DTW matching, interpolation and deletion, as well as the fusion weight of each operating data in the benchmark data sequence and each adjusted data sequence;

[0023] For Dimension The running data of the same order value in the baseline data sequence of the slice and all the adjusted data sequences are weighted averaged according to the fusion weight, and the average value is used as the element of the corresponding order value. For each order value, an element is obtained and arranged in ascending order according to the order value to obtain a new sequence, which is recorded as Dimension The baseline change sequence of slices.

[0024] Optionally, the obtaining of the reference data sequence and the plurality of adjusted data sequences in the same slice in each dimension, and the fusion weight of each running data in the reference data sequence and each adjusted data sequence, includes the following specific methods:

[0025] For Dimension Among several operation data sequences in a slice, the operation data sequence corresponding to the reference time period is recorded as the reference data sequence, and the sequence length of the reference data sequence is used as the reference time length; the operation data sequence whose sequence length is not equal to the reference time length is recorded as the data sequence to be adjusted;

[0026] For any data sequence to be adjusted, DTW matching is performed on the data sequence to be adjusted and the benchmark data sequence to obtain several matching point pairs. According to the matching point pairs, the data sequence to be adjusted is interpolated or deleted to obtain an adjusted data sequence with the same length as the benchmark data sequence; the adjusted data sequence of the corresponding time period of each data sequence to be adjusted is obtained; according to the comprehensive change degree of the benchmark data sequence length and the time point corresponding to each running data in the data sequence to be adjusted, the fusion weight of each running data in the benchmark data sequence and the adjusted data sequence is determined.

[0027] Optionally, the method of obtaining the comprehensive difference of each time period includes:

[0028] Get the centroid for each time period cluster, and take the time period corresponding to the centroid as the central time period of each time period cluster; time period, obtain the The dimension in The sequence of running data of time periods is recorded as The dimension in The running sequence of the time period is obtained The running sequence of the dimension in the central time period of each time period cluster; calculate the The dimension in The DTW distance between the running sequence of each time period and the running sequence of the central time period of each time period cluster is calculated, and the time period cluster corresponding to the central time period corresponding to the minimum DTW distance is taken as the first The time period is The belonging cluster of the dimension; get the The dimension in The running sequence of each time period is the same as that of the The inverse proportional normalized value of the DTW distance of the benchmark change sequence of the slice corresponding to the cluster in the dimension is recorded as The time period is The initial difference of the dimensions;

[0029] Get the The initial difference of each dimension in each time period is taken as the mean of all initial differences. The overall difference between the time periods.

[0030] Optionally, the initial tolerance parameter of each time period is obtained by:

[0031] Subtract 1 from The difference obtained by the comprehensive difference of the time period is recorded as The first difference of the time period; the product of the first difference and the tolerance hyperparameter is used as the first The initial tolerance parameter for each time period.

[0032] Optionally, the method of obtaining the final tolerance parameter for each time period includes:

[0033] Compression starts from the first time period, and revolving door compression is performed on the running sequence of each dimension in the first time period according to the initial tolerance parameter of the first time period to obtain the compressed data of each dimension in the first time period, which is recorded as the first compressed data of each dimension in the first time period; according to the first compressed data of each dimension, the comprehensive change degree is recalculated for each time point in the first time period, and the result is recorded as the first comprehensive change degree of each time point, and the absolute value of the difference between the first comprehensive change degree and the original comprehensive change degree at each time point is obtained, and the average of all the absolute values ​​of the difference is taken as the information loss amount of the first compression of the first time period; an information loss threshold is preset, and if the information loss amount is less than or equal to the information loss threshold, the initial tolerance parameter of the first time period is taken as the final tolerance parameter of the first time period; if the information loss amount is greater than the information loss threshold, the initial tolerance parameter is adjusted, and the adjusted tolerance parameter of the first time period is adjusted. The calculation method is:

[0034]

[0035] in, represents the initial tolerance parameter for the first time period, represents the tolerance hyperparameter, represents the information loss threshold, It indicates the amount of information loss in the first compression in the first time period;

[0036] According to the adjustment tolerance parameter, the running sequence of each dimension in the first time period is re-performed with the revolving door compression to obtain the second compression data of each dimension, and the second comprehensive change degree of each time point is recalculated based on the second compression data to obtain the information loss amount of the second compression of the first time period, and continue to compare with the information loss threshold. If it is less than or equal to the information loss threshold, the adjustment tolerance parameter is used as the final tolerance parameter; if it is still greater than the information loss threshold, the adjustment tolerance parameter is continued to be adjusted according to the information loss amount of the second compression, and the information loss amount of the third compression is recalculated, and so on, until the information loss amount is less than or equal to the information loss threshold, and the final tolerance parameter of the first time period is obtained; the final tolerance parameter is obtained for the second time period and each time period thereafter.

[0037] Optionally, the comprehensive change degree at each time point is obtained by:

[0038] For any dimension, obtain the slope of the operating data at each time point in the dimension, and obtain the initial degree of change at each time point in the dimension based on the difference in slopes between adjacent time points; obtain the comprehensive degree of change at each time point by averaging the initial degree of change at each time point in different dimensions.

[0039] Optionally, the obtaining of the correlation distance between the time periods includes the following specific methods:

[0040] For The time period and time period, obtain the The DTW distance between the sequences of the running data in the two time periods is recorded as The dimension in The time period and The time period distance of each time period; get the time period distance of each dimension in the The time period and The time period distance of each time period is taken as the average of all time period distances. The time period and The correlation distance of a time period.

[0041] The beneficial effects of the present invention are as follows: the present invention slices and obtains the operation data of multiple dimensions of the oxygen concentrator, obtains the final tolerance parameter for each time period in different slices, and performs adaptive revolving door compression according to the final tolerance parameter, thereby realizing the compression storage of the multi-dimensional operation data of the oxygen concentrator, and at the same time ensuring that the operation data will not be lost too much to affect the intelligent management of the oxygen concentrator; wherein by analyzing the changes in the operation data of different time periods under the same dimension and the same slice, the representative degree of the time period in each slice under each dimension is quantified, and the reference time period is obtained according to the representative degree, so that the reference time period can best reflect the change relationship of the operation data in the corresponding slice in time series; based on the reference time period, a reference change sequence is obtained for the slices under each dimension, and the comprehensive difference of each time period is obtained according to the reference change sequence, and the change difference of the time period is reflected according to the comprehensive difference, and then the initial tolerance parameter is determined, the final tolerance parameter is iteratively obtained, and the adaptive revolving door compression is performed on each time period based on the final tolerance parameter, thereby avoiding the reduction of the revolving door compression efficiency caused by unreasonable tolerance parameters; so that the compression efficiency of different slices without affecting information loss is greatly improved, and the effectiveness of compression of data of different slices is guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be 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 labor.

[0043] Figure 1 A schematic flow chart of an intelligent management method for a shared oxygen generator based on the Internet of Things provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0045] See also Figure 1 , which shows a flow chart of a shared oxygen concentrator intelligent management method based on the Internet of Things provided by an embodiment of the present invention, the method comprising the following steps:

[0046] Step S001: Collect operating data of multiple dimensions of the oxygen concentrator.

[0047] The purpose of this embodiment is to perform distributed storage on the operating data of multiple dimensions generated during the operation of the oxygen concentrator. Therefore, it is necessary to arrange several different types of sensors on the oxygen concentrator to collect the operating data. This embodiment collects data of eight dimensions, including the concentration, flow rate and pressure of the oxygen production parameters of the oxygen concentrator, the air compressor parameters, pressure dew point and dew temperature during the operation of the oxygen concentrator, and the temperature and humidity in the environment where the oxygen concentrator is located.

[0048] Specifically, in this embodiment, the oxygen concentrator oxygen production parameters, such as concentration, flow rate and pressure, as well as the air compressor parameters, pressure dew point and dew temperature during the operation of the oxygen concentrator, a total of six dimensions of data are directly acquired through the oxygen concentrator equipment; the temperature and humidity in the environment are collected through the temperature sensor and the humidity sensor respectively, and the sampling time interval of the data in each dimension is 1 second. Data collection starts after the oxygen concentrator starts to run. After the data in each dimension is collected, the data in each dimension is numerically processed respectively, and the numerically processed data is recorded as the operating data of each dimension. Then, each operating data in each dimension corresponds to a time point, and the overall time axis corresponding to all operating data in different dimensions is the same, and the operating data in all dimensions are constrained by numerical processing. Within the range.

[0049] So far, a number of operating data of the oxygen concentrator in multiple dimensions have been collected.

[0050] It should be noted that in the storage process of oxygen concentrator operating data in different dimensions, in order to ensure that the oxygen concentration can be quickly adjusted and the oxygen concentrator equipment can operate stably according to the changes in the operating data in different dimensions, it is necessary to compress and store the oxygen concentrator operating data of different dimensions with large data volume; data slicing is to divide the data set into smaller parts, called "shards" or "slices", so that each slice can be easier to manage and process. In a distributed storage system, each slice can be stored on a different server, thereby improving storage capacity and access speed.

[0051] Step S002: According to the changes in the operating data at the same time point in different dimensions, the comprehensive change degree at each time point is obtained, and the time axis is divided to obtain several time periods; according to the correlation of the changes in the operating data in different dimensions in different time periods, the correlation distance between the time periods is obtained, and several time period clusters are obtained by clustering; the operating data of multiple dimensions under each time period cluster are sliced.

[0052] Preferably, in one embodiment of the present invention, according to the changes of the running data at the same time point in different dimensions, the comprehensive change degree at each time point is obtained, and the time axis is divided into several time periods, including the specific method of:

[0053] It should be noted that the operating data of different dimensions of the oxygen concentrator have different impacts on the monitoring of the oxygen concentrator, and there is a certain response relationship between the operating data of different dimensions during the operation of the oxygen concentrator. For example, certain parameters or indicators of the oxygen concentrator may respond to changes in the external environment or operating conditions. When the ambient temperature rises, the oxygen production efficiency of the oxygen concentrator may decrease; when the oxygen demand increases, the oxygen concentrator may automatically increase the oxygen production flow rate. Therefore, there is a certain relationship between the operating data of different dimensions of the oxygen concentrator. If the difference between the change in the operating data for a certain period of time and the above relationship is large, the data difference in this time period is large, and the corresponding data needs to be sliced ​​according to the difference.

[0054] Specifically, for any dimension, the slope of the operating data at each time point in the dimension is obtained, and the difference between the slope of each time point and the adjacent previous time point is used as the initial change degree at each time point in the dimension; by averaging the initial change degrees of each time point in different dimensions, the comprehensive change degree of each time point is obtained, wherein the initial change degree and comprehensive change degree are not calculated for the first time point.

[0055] Furthermore, a change threshold is preset. In this embodiment, the change threshold is described as 0.58. The time point when the comprehensive change degree is greater than the change threshold is used as the segmentation point. The time axis composed of all time points is divided into several time periods through the segmentation points, where each segmentation point is used as the first time point in the corresponding time period, and the first time point is also used as a segmentation point, that is, the first time point in the first time period.

[0056] It should be noted that at a time point with a greater degree of comprehensive change, the operating data of each dimension changes greatly at that time point, and segmentation is required to ensure that the changes between the operating data of the same dimension within the time period are small, and then subsequent slice acquisition is performed.

[0057] Preferably, in one embodiment of the present invention, according to the correlation of the operation data changes in different dimensions in different time periods, the correlation distance between the time periods is obtained, and a number of time period clusters are obtained by clustering; the operation data of multiple dimensions under each time period cluster is sliced, and the specific method includes:

[0058] First The time period and Take the time period as an example, and get the The DTW distance between the sequences of the running data in the two time periods is recorded as The dimension in The time period and The time period distance of each time period; get the time period distance of each dimension in the The time period and The time period distance of each time period is taken as the average of all time period distances. The time period and The correlation distance of a time period.

[0059] It should be noted that after dividing the multi-dimensional operating data of the oxygen generator into time periods, it is necessary to obtain the relationship between dimensions in time periods with the same or similar distribution, and determine the similarity between time periods by analyzing the changing relationship between the time periods and the relationship between the obtained dimensions, and then slice them according to the similarity; the slicing process is based on cluster analysis, and the oxygen generator is a cyclic process during operation, including compression, filtration, separation, desorption and emission to ensure stable oxygen output, and other types of data such as the environment have different response relationships to different processes. Therefore, in order to obtain accurate response relationships, different time periods need to be clustered.

[0060] Specifically, this embodiment performs DBSCAN clustering on all time periods, and the distance measurement adopts the correlation distance between time periods, so that a number of clusters are obtained, which are recorded as a number of time period clusters; for any time period cluster, a number of operating data of each dimension in all time periods in the time period cluster together constitute a slice corresponding to the time period cluster, and a slice corresponding to each time period cluster is obtained; it should be noted that the slice contains a sequence composed of all operating data of each dimension in each time period under the time period cluster corresponding to the slice, that is, the slice is composed of a number of operating data sequences, and the time periods corresponding to the operating data sequences all belong to the time period cluster corresponding to the slice.

[0061] Optionally, in other embodiments, other clustering methods, such as K-means clustering, are used to cluster the time periods and obtain slices, wherein the distance metric uses the correlation distance between the time periods, and the specific clustering process is not described in detail in this embodiment.

[0062] At this point, by analyzing the changes in the correlation of operating data of different dimensions in time series, the comprehensive degree of change at the time point is obtained and the time period is divided; through the correlation of the operating data of each dimension between the time periods, the correlation distance between the time periods is quantified and clustered, and finally the multi-dimensional operating data of the oxygen concentrator is sliced ​​and obtained according to the time period.

[0063] Step S003: According to the similarity of the change trend of the running data in different time periods in the same slice under the same dimension, the representativeness of each time period in each slice in each dimension is obtained, and the benchmark time period in each slice under each dimension is obtained; for the running data of several time periods and the benchmark time period in the same slice under each dimension, the benchmark change sequence of each slice under each dimension is obtained through data matching.

[0064] Preferably, in one embodiment of the present invention, according to the similarity of the data change trend of different time periods in the same slice under the same dimension, the representative degree of each time period in each slice in each dimension is obtained, and the reference time period in each slice under each dimension is obtained, and the specific method includes:

[0065] First Take the dimension as an example, for the In several time periods of a slice, The dimension in The sequence of running data corresponding to each time period in the slice is recorded as Slice A number of operating data sequences of dimensions; preset an initial window scale and growth step size. In this embodiment, the initial window scale is described as 3, the growth step size is set to 2, and the maximum window scale is set to 9, so a total of four window scales are obtained; each operating data sequence is subjected to STL time series decomposition through each window scale to obtain the trend item of each operating data sequence at each window scale; Any two time periods in the slices form a combination, and several time period combinations are obtained, including the Slice The time period combination of time periods is recorded as Slice The reference time period combination of time periods; since the reference time period combination includes two time periods, each of the two time periods corresponds to an operating data sequence, and the operating data sequence has a trend item under each window scale, the window scales are combined in pairs, and the two window scales in the combination can be the same window scale, then a total of 16 window scale combinations of 4×4 are obtained, and the first window scale in the window scale combination corresponds to the first window scale in the reference time period combination. The running data sequence corresponding to the time period, the second window scale corresponds to the reference time period combination except the For any reference time period combination and any window scale combination, obtain the trend items of the operation data sequences corresponding to the two time periods in the reference time period combination under the window scale combination, calculate the DTW distance of the two trend items, and record the result as the reference time period combination in the first The trend difference of the window scale combination in the dimension; get the Slice Each reference time period of the time period is combined in the The trend difference under each window scale combination in each dimension.

[0066] Further, Slice The time period is The degree of representation of the dimensions The calculation method is:

[0067]

[0068] in, Indicates Slice The number of reference time period combinations for time periods, Indicates Slice The time period The first The number of time points in a time period, Indicates Slice The time period The reference time period combinations except The number of time points in another time period outside the time period, Indicates The maximum absolute value of the difference between the number of time points in two time periods in all time period combinations of slices, represents the number of window scale combinations, Indicates Slice The time period The reference time period is combined in Dimension The trend difference under the combination of window scales, It means to find the absolute value. represents an exponential function with a natural constant as the base. In this embodiment, The model shows an inverse proportional relationship. As the input of the model, the implementer can set the inverse proportional function according to the actual situation; according to the above method, the first Each time segment in the slice The degree of representation of the dimension.

[0069] It should be noted that the average of the inverse proportional values ​​of trend differences under different window scale combinations in the reference time period combination is used as the first time period and another time period in The trend similarity benchmark value of each dimension is used. The smaller the trend difference is, that is, the smaller the DTW distance of the corresponding trend item is, the more similar the trend items are, and the greater the trend similarity is. At the same time, the time length of the two time periods in the reference time period combination, that is, the difference in the number of time points, is used as the weight. The smaller the difference in the number of time points is, the greater the reference of the trend similarity of the two time periods is, and the greater the weight is. If the trend similarity in different reference time period combinations is large, then The more time periods can represent the The slice is in The greater the temporal changes in each dimension, the greater the degree of representation.

[0070] Further, in the The time period corresponding to the maximum value of the representative degree of each dimension is taken as the The dimension in The base time period of each slice is obtained according to the above method; the representativeness of each time period in each slice in each dimension is obtained, and then the base time period of each slice under each dimension is obtained.

[0071] Optionally, in other embodiments, obtaining the Slice There are several running data sequences in dimensions, each of which corresponds to the In a time period of each slice, obtain the mean DTW distance between each running data sequence and each other running data sequence, record it as the difference distance of each running data sequence, and take the time period corresponding to the running data sequence with the smallest difference distance as the first Dimension The base time period for each slice.

[0072] It should be noted that, since different slices need to be compressed to different degrees and the data distribution characteristics in different slices are different, different slices need to be compressed by revolving door to different degrees. At the same time, the benchmark change sequence of each slice needs to be analyzed. First, it is necessary to obtain the benchmark time period in different dimensions for each slice, and perform DTW matching on each time period in the same slice according to the benchmark time period to obtain the benchmark change sequence. The benchmark time period needs to be similar to other time periods in the slice to the greatest extent. The benchmark time period of each slice in each dimension is finally obtained through the representative degree.

[0073] Preferably, in one embodiment of the present invention, for the operation data of several time periods and the reference time period in the same slice in each dimension, the reference change sequence of each slice in each dimension is obtained by data matching, and the specific method includes:

[0074] For Dimension In a plurality of operation data sequences in a slice, the operation data sequence corresponding to the reference time period is recorded as the reference data sequence, and the sequence length of the reference data sequence is used as the reference time length; the operation data sequence whose sequence length is not equal to the reference time length is recorded as the data sequence to be adjusted; for any data sequence to be adjusted, the data sequence to be adjusted is matched with the reference data sequence by DTW to obtain a plurality of matching point pairs, and the matching point pairs are taken as the reference, and the matching point pairs include one-to-one, two-to-many and many-to-two. In the one-to-one case, the operation data in the reference data sequence and the data sequence to be adjusted do not need to be changed; in the two-to-many case, that is, two adjacent operation data in the reference data sequence correspond to a plurality of continuous operation data in the data sequence to be adjusted, then the operation data of the data sequence to be adjusted needs to be deleted, and the comprehensive change degree of each operation data corresponding to the time point is obtained for the plurality of continuous operation data, and the operation data are deleted in the order of the comprehensive change degree from large to small, until the number of remaining operation data is equal to the number of operation data in the reference data sequence corresponding to the matching point pair, that is, stop when there are two operation data left.

[0075] Furthermore, for the many-to-two situation, that is, a plurality of continuous operating data in the reference data sequence corresponds to two adjacent operating data in the data sequence to be adjusted, it is necessary to interpolate the data sequence to be adjusted. This embodiment interpolates between the two adjacent operating data in the data sequence to be adjusted by quadratic linear interpolation until the sum of the number of inserted operating data plus 2 is equal to the number of continuous operating data in the reference data sequence corresponding to the matching point pair and stops; then, the data sequence to be adjusted is interpolated and deleted based on the DTW matching result, and the processed data sequence to be adjusted is recorded as the adjusted data sequence for the corresponding time period of the data sequence to be adjusted. It is particularly noted that due to interpolation, there are a certain number of operating data in the adjusted data sequence that do not have corresponding time points, and the corresponding comprehensive change degree cannot be obtained. For the operating data inserted in the many-to-two situation, this embodiment takes the average of the comprehensive change degrees of the corresponding time points of the two adjacent operating data in the data sequence to be adjusted corresponding to "two" as the comprehensive change degree corresponding to each inserted operating data.

[0076] Furthermore, according to the above method, the adjustment data sequence of the corresponding time period of each data sequence to be adjusted is obtained, and the Dimension An adjusted data sequence is obtained for each time period in a slice, the benchmark time period is the benchmark data sequence, all adjusted data sequences and the benchmark data sequence have the same length, and all non-inserted operating data in the benchmark data sequence and the adjusted data sequence have corresponding time points, then the comprehensive change degree corresponding to the time point and the comprehensive change degree corresponding to the inserted operating data are used as the fusion weight of each operating data in the benchmark data sequence and the adjusted data sequence; the operating data with the same order value in the benchmark data sequence and all adjusted data sequences are weighted and averaged according to the fusion weight, and the obtained mean is used as the element of the corresponding order value, then an element is obtained for each order value, and they are arranged in ascending order according to the order value to obtain a new sequence, recorded as the first Dimension The baseline change sequence of each slice in each dimension is obtained according to the above method.

[0077] It should be noted that by obtaining the benchmark time period and the benchmark time length, the operating data sequences of different time periods in the same slice under the same dimension are interpolated or deleted, and finally a benchmark data sequence and several adjustment data sequences are obtained, and then a benchmark change sequence is obtained to reflect the overall time series distribution in the slice under the dimension, which provides a basis for the subsequent quantification of the comprehensive difference.

[0078] At this point, by performing correlation analysis on the operating data sequences of different time periods in the slices under the same dimension, the benchmark time period is obtained, and each operating data sequence is adjusted based on the benchmark time period to obtain the benchmark change sequence of each slice under each dimension.

[0079] Step S004: Obtain the comprehensive difference of each time period according to the similarity between the operating data of each time period under each dimension and the benchmark change sequence of each slice under each dimension; obtain the initial tolerance parameters of each time period according to the comprehensive difference, and obtain the final tolerance parameters of each time period through the revolving door compression algorithm combined with the comprehensive change degree of the time point.

[0080] Preferably, in one embodiment of the present invention, the comprehensive difference degree of each time period is obtained according to the similarity between the operation data of each time period under each dimension and the benchmark change sequence of each slice under each dimension, and the specific method includes:

[0081] Get the centroid for each time period cluster, and take the time period corresponding to the centroid as the central time period of each time period cluster; time period, obtain the The dimension in The sequence of running data of time periods is recorded as The dimension in The running sequence of the time period is obtained The running sequence of the dimension in the central time period of each time period cluster; calculate the The dimension in The DTW distance between the running sequence of each time period and the running sequence of the central time period of each time period cluster is calculated, and the time period cluster corresponding to the central time period corresponding to the minimum DTW distance is taken as the first The time period is The belonging cluster of the dimension; since the time period cluster corresponds to the slice, obtain the The dimension in The running sequence of each time period is the same as that of the The inverse proportional normalized value of the DTW distance of the benchmark change sequence of the slice corresponding to the cluster in the dimension is recorded as The time period is The initial difference of the dimensions, in which this embodiment adopts Model to present inverse proportional relationship and normalization processing, As the input of the model, the implementer can set the inverse proportional function and normalization function according to the actual situation.

[0082] Further, get the The initial difference of each dimension in each time period is taken as the mean of all initial differences. The comprehensive difference of each time period is obtained according to the above method.

[0083] Optionally, in other embodiments, obtaining the The dimension in The running sequence of the time period is calculated and the running sequence is compared with the Dimension The inverse proportional normalized value of the DTW distance of the benchmark change sequence of the slice belonging to the time period is recorded as The time period is The initial difference of the dimension; get the The initial difference of each dimension in each time period is averaged and the first The overall difference between the time periods.

[0084] It should be noted that during the revolving door compression process, the tolerance parameters need to be adapted to ensure that the compression does not cause excessive information loss. It is necessary to perform a difference analysis on the operating sequence of each dimension in each time period, and quantify the difference in combination with the benchmark change sequence. Finally, the comprehensive difference of each time period is obtained, and the comprehensive difference provides a basis for determining the subsequent tolerance parameters of each time period.

[0085] Preferably, in one embodiment of the present invention, the initial tolerance parameter of each time period is obtained according to the comprehensive difference, and the final tolerance parameter of each time period is obtained by the revolving door compression algorithm in combination with the comprehensive change degree of the time point, including the specific method of:

[0086] For time period, its initial tolerance parameter The calculation method is:

[0087]

[0088] in, Indicates The comprehensive difference between the time periods, Represents the preset tolerance hyperparameter. This embodiment adopts Describe; obtain the initial tolerance parameters for each time period according to the above method.

[0089] Further, compression is started from the first time period, and the operation sequence of each dimension in the first time period is subjected to revolving door compression according to the initial tolerance parameter of the first time period, and compressed data of each dimension in the first time period is obtained, which is recorded as the first compressed data of each dimension in the first time period; according to the first compressed data of each dimension, the comprehensive change degree is recalculated for each time point in the first time period, and the calculation method is to replace the original operation sequence of each dimension with the corresponding first compressed data. It should be noted that the initial change degree of each dimension at each time point changes, and then the comprehensive change degree is recalculated; the obtained result is recorded as the first comprehensive change degree of each time point, and the absolute value of the difference between the first comprehensive change degree and the original comprehensive change degree at each time point is obtained, and the average of all the absolute values ​​of the difference is taken as the information loss amount of the first compression of the first time period; a preset information loss threshold is set, and the information loss threshold in this embodiment is described as 0.3. If the information loss amount is less than or equal to the information loss threshold, the initial tolerance parameter of the first time period is taken as the final tolerance parameter of the first time period; if the information loss amount is greater than the information loss threshold, the initial tolerance parameter is adjusted, and the adjusted tolerance parameter of the first time period is adjusted. The calculation method is:

[0090]

[0091] in, represents the initial tolerance parameter for the first time period, represents the preset tolerance hyperparameter, represents the information loss threshold, Indicates the amount of information loss in the first compression of the first time period.

[0092] It should be noted that if the amount of information loss is large, it is necessary to further reduce the tolerance parameter to reduce the amount of information loss.

[0093] Furthermore, the revolving door compression is performed again on the running sequence of each dimension in the first time period according to the adjustment tolerance parameter to obtain the second compression data of each dimension, and the second comprehensive change degree of each time point is recalculated based on the second compression data to obtain the information loss amount of the second compression of the first time period, and then compared with the information loss threshold. If it is less than or equal to the information loss threshold, the adjustment tolerance parameter is used as the final tolerance parameter; if it is still greater than the information loss threshold, the adjustment tolerance parameter is continued to be adjusted according to the information loss amount of the second compression, and the information loss amount of the third compression is recalculated, and so on, until the information loss amount is less than or equal to the information loss threshold, and the final tolerance parameter of the first time period is obtained.

[0094] Furthermore, the final tolerance parameters are obtained for the second time period and each subsequent time period according to the above method. It should be noted that, taking the second time period as an example, in the process of obtaining the final tolerance parameters, in the process of recalculating the comprehensive change degree of each time point in the second time period, the comprehensive change degree of each time point in the first time period is the comprehensive change degree of each time point corresponding to the final tolerance parameters, that is, the comprehensive change degree has been updated. Then, in the process of obtaining the final tolerance parameters of each subsequent time period, the comprehensive change degree of each time point in the previous time period is the corresponding comprehensive change degree under the final tolerance parameters.

[0095] It should be noted that after obtaining the initial tolerance parameters, it is necessary to ensure that the operating data of each dimension will not be lost too much after the revolving door is compressed, and then the final tolerance parameters are obtained by continuously adjusting the tolerance parameters.

[0096] At this point, by analyzing the operating sequences of each dimension in each time period and combining the centroid of each time period cluster, the comprehensive difference of each time period is quantified, and the tolerance parameters are adjusted based on the comprehensive difference. At the same time, the information loss after revolving door compression is considered, and finally the final tolerance parameters of each time period are obtained, which ensures that the revolving door compression will not cause excessive loss of operating data in each dimension.

[0097] Step S005: According to the final tolerance parameters of each time period, the operating data of the oxygen concentrator in multiple dimensions in each time period are adaptively compressed by a revolving door, and the compressed operating data of each time period is obtained and stored, so as to realize the intelligent management of the multi-dimensional operating data of the oxygen concentrator.

[0098] Preferably, in one embodiment of the present invention, after obtaining the final tolerance parameter of each time period, for the operation sequence of any dimension of any time period, revolving door compression is performed on the operation sequence of that dimension of the time period, and the tolerance parameter is the final tolerance parameter of the time period, so as to obtain the compressed operation data of that dimension of the time period; the operation sequence of each dimension of the time period is subjected to revolving door compression according to the final tolerance parameter of the corresponding time period, so as to obtain the compressed operation data of each dimension of the time period, and the compressed operation data of all dimensions of the time period is used as the compressed operation data of the time period.

[0099] Furthermore, the compressed operating data of each time period in the time cluster to which the time period belongs is obtained, and all the compressed operating data obtained are stored according to the slices corresponding to the time cluster; similarly, the compressed operating data of each time period in other time clusters is obtained and stored based on the slices, so that the multi-dimensional operating data of the oxygen generator is distributedly stored; the stored data slices are managed and maintained through programming interfaces and command line tools, so as to realize the operational management of the large amount of operating data generated by the oxygen generator.

[0100] At this point, by analyzing the time series changes of the multi-dimensional operation data of the oxygen concentrator and slicing it, and performing revolving door compression on the data in the slices, distributed storage of the multi-dimensional operation data of the oxygen concentrator is achieved, thereby improving the adjustment, operation and management efficiency of the oxygen concentrator.

[0101] 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 protection scope of the present invention.

Claims

1. An intelligent management method for a shared oxygen concentrator based on the Internet of Things, characterized in that: The method comprises the following steps: Collecting operation data of multiple dimensions of the oxygen concentrator, wherein the operation data of the multiple dimensions correspond to the same time axis, and the time axis includes a plurality of time points; According to the changes of operating data at the same time point in different dimensions, the comprehensive change degree of each time point is obtained, and the time axis is divided into several time periods; according to the correlation of the changes of operating data in different dimensions in different time periods, the correlation distance between time periods is obtained, and several time period clusters are obtained by clustering; the operating data of multiple dimensions under each time period cluster is sliced; According to the similarity of the change trend of the running data in different time periods in the same slice under the same dimension, the representative degree of each time period in each slice in each dimension is obtained, and the benchmark time period in each slice under each dimension is obtained; for the running data of several time periods and the benchmark time period in the same slice under each dimension, the benchmark change sequence of each slice under each dimension is obtained through data matching; According to the similarity between the running data of each time period under each dimension and the benchmark change sequence of each slice under each dimension, the comprehensive difference of each time period is obtained; according to the comprehensive difference, the initial tolerance parameter of each time period is obtained, and the final tolerance parameter of each time period is obtained by the revolving door compression algorithm combined with the comprehensive change degree of the time point; According to the final tolerance parameters of each time period, the operating data of the oxygen concentrator in multiple dimensions in each time period are adaptively compressed by a revolving door to obtain and store the compressed operating data of each time period; The specific method of obtaining the comprehensive difference degree of each time period includes: Get the centroid for each time period cluster, and take the time period corresponding to the centroid as the central time period of each time period cluster; time period, obtain the The dimension in The sequence of running data of time periods is recorded as The dimension in The running sequence of the time period is obtained The running sequence of the dimension in the central time period of each time period cluster; calculate the The dimension in The DTW distance between the running sequence of each time period and the running sequence of the central time period of each time period cluster is calculated, and the time period cluster corresponding to the central time period corresponding to the minimum DTW distance is taken as the first The time period is The belonging cluster of the dimension; get the The dimension in The running sequence of each time period is the same as that of the The inverse proportional normalized value of the DTW distance of the benchmark change sequence of the slice corresponding to the cluster in the dimension is recorded as The time period is The initial difference of the dimensions; Get the The initial difference of each dimension in each time period is taken as the mean of all initial differences. The comprehensive difference between the time periods; The initial tolerance parameter of each time period is obtained by: Subtract 1 from The difference obtained by the comprehensive difference of the time period is recorded as The first difference of the time period; the product of the first difference and the tolerance hyperparameter is used as the first The initial tolerance parameter for each time period; The specific method of obtaining the final tolerance parameter for each time period includes: Compression starts from the first time period, and revolving door compression is performed on the running sequence of each dimension in the first time period according to the initial tolerance parameter of the first time period to obtain the compressed data of each dimension in the first time period, which is recorded as the first compressed data of each dimension in the first time period; according to the first compressed data of each dimension, the comprehensive change degree is recalculated for each time point in the first time period, and the result is recorded as the first comprehensive change degree of each time point, and the absolute value of the difference between the first comprehensive change degree and the original comprehensive change degree at each time point is obtained, and the average of all the absolute values ​​of the difference is taken as the information loss amount of the first compression of the first time period; an information loss threshold is preset, and if the information loss amount is less than or equal to the information loss threshold, the initial tolerance parameter of the first time period is taken as the final tolerance parameter of the first time period; if the information loss amount is greater than the information loss threshold, the initial tolerance parameter is adjusted, and the adjusted tolerance parameter of the first time period is adjusted. The calculation method is: in, represents the initial tolerance parameter for the first time period, represents the tolerance hyperparameter, represents the information loss threshold, It indicates the amount of information loss in the first compression in the first time period; According to the adjustment tolerance parameter, the running sequence of each dimension in the first time period is re-performed with the revolving door compression to obtain the second compression data of each dimension, and the second comprehensive change degree of each time point is recalculated based on the second compression data to obtain the information loss amount of the second compression of the first time period, and continue to compare with the information loss threshold. If it is less than or equal to the information loss threshold, the adjustment tolerance parameter is used as the final tolerance parameter; if it is still greater than the information loss threshold, the adjustment tolerance parameter is continued to be adjusted according to the information loss amount of the second compression, and the information loss amount of the third compression is recalculated, and so on, until the information loss amount is less than or equal to the information loss threshold, and the final tolerance parameter of the first time period is obtained; the final tolerance parameter is obtained for the second time period and each time period thereafter.

2. According to claim 1, a shared oxygen concentrator intelligent management method based on the Internet of Things is characterized in that: The specific method of obtaining the representativeness of each time period in each slice in each dimension and obtaining the reference time period in each slice under each dimension includes: Combining any two time periods in the same slice into a group, we can get several time period combinations; Slice The time period combination of time periods is recorded as Slice The reference time period combination of time periods; According to the similarity of data change trends in different time periods in the same slice under the same dimension, combined with the preset window scale, obtain the trend difference of each reference time period combination of each time period in each slice under each window scale combination of each dimension; No. Slice The time period is The degree of representation of the dimensions The calculation method is: in, Indicates Slice The number of reference time period combinations for time periods, Indicates Slice The time period The first The number of time points in a time period, Indicates Slice The time period The reference time period combinations except The number of time points in another time period outside the time period, Indicates The maximum absolute value of the difference between the number of time points in two time periods in all time period combinations of slices, represents the number of window scale combinations, Indicates Slice The time period The reference time period is combined in Dimension The trend difference under the combination of window scales, It means to find the absolute value. represents an exponential function with a natural constant as base; Get the Each time segment in the slice The degree of representation of the dimensions will be The time period corresponding to the maximum value of the representative degree of each dimension is taken as the The dimension in The base time period for each slice.

3. According to claim 2, a shared oxygen concentrator intelligent management method based on the Internet of Things is characterized in that: The specific method of obtaining the trend difference of each reference time period combination of each time period in each slice under each window scale combination of each dimension includes: For In several time periods of a slice, The dimension in The sequence of running data corresponding to each time period in the slice is recorded as Slice Several running data series of dimensions; Perform STL time series decomposition on each operating data series at each window scale to obtain the trend item of each operating data series at each window scale; The preset window scales are combined in pairs to obtain several window scale combinations; the first window scale in the window scale combination corresponds to the first window scale in the reference time period combination. The running data sequence corresponding to the time period, the second window scale corresponds to the reference time period combination except the For any reference time period combination and any window scale combination, obtain the trend items of the operation data sequences corresponding to the two time periods in the reference time period combination under the window scale combination, calculate the DTW distance of the two trend items, and record the result as the reference time period combination in the first The trend difference under this window scale combination in each dimension.

4. According to the method of intelligent management of shared oxygen concentrator based on Internet of Things in claim 1, it is characterized in that: The specific method for obtaining the benchmark change sequence of each slice in each dimension is as follows: For the operating data of several time periods and the benchmark time period in the same slice under each dimension, the benchmark data sequence and several adjusted data sequences in the same slice under each dimension are obtained through DTW matching, interpolation and deletion, as well as the fusion weight of each operating data in the benchmark data sequence and each adjusted data sequence; For Dimension The running data of the same order value in the baseline data sequence of the slice and all the adjusted data sequences are weighted and averaged according to the fusion weight, and the average value is used as the element of the corresponding order value. For each order value, an element is obtained and arranged in ascending order according to the order value to obtain a new sequence, which is recorded as Dimension The baseline change sequence of slices.

5. According to claim 4, a shared oxygen concentrator intelligent management method based on the Internet of Things is characterized in that: The specific method of obtaining the reference data sequence and several adjustment data sequences in the same slice in each dimension, and the fusion weight of each running data in the reference data sequence and each adjustment data sequence includes: For Dimension A number of running data sequences in a slice, the running data sequence corresponding to the benchmark time period is recorded as the benchmark data sequence, and the sequence length of the benchmark data sequence is used as the benchmark time length; The running data sequence whose sequence length is not equal to the reference time length is recorded as the data sequence to be adjusted; For any data sequence to be adjusted, DTW matching is performed on the data sequence to be adjusted and the benchmark data sequence to obtain several matching point pairs. According to the matching point pairs, the data sequence to be adjusted is interpolated or deleted to obtain an adjusted data sequence with the same length as the benchmark data sequence; the adjusted data sequence of the corresponding time period of each data sequence to be adjusted is obtained; according to the comprehensive change degree of the benchmark data sequence length and the time point corresponding to each running data in the data sequence to be adjusted, the fusion weight of each running data in the benchmark data sequence and the adjusted data sequence is determined.

6. According to the method of intelligent management of shared oxygen concentrator based on Internet of Things in claim 1, it is characterized in that: The comprehensive change degree at each time point is specifically obtained by: For any dimension, obtain the slope of the operating data at each time point in the dimension, and obtain the initial degree of change at each time point in the dimension based on the difference in slopes between adjacent time points; obtain the comprehensive degree of change at each time point by averaging the initial degree of change at each time point in different dimensions.

7. According to the method of intelligent management of shared oxygen concentrator based on Internet of Things in claim 1, it is characterized in that: The specific method of obtaining the correlation distance between the time periods includes: For The time period and time period, obtain the The DTW distance between the sequences of the running data in the two time periods is recorded as The dimension in The time period and The time period distance of each time period; get the time period distance of each dimension in the The time period and The time period distance of each time period is taken as the average of all time period distances. The time period and The correlation distance of a time period.

Images