A data intelligent analysis method for intelligent chemical product supply chain management system
By analyzing the timing and fluctuation characteristics of chemical production data in the chemical supply chain management, adaptively adjusting the branch factor L1 of the BIRCH algorithm, the problems of clustering results accuracy and abnormality calculation deviation are solved, and the accuracy of abnormality detection is improved and safety of chemical product production process.
Patent Information
- Application Number
- CN202410307114.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-18
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-03-18
AI Technical Summary
In the existing BIRCH clustering algorithm, in the chemical supply chain management, the small value of branch factor L leads to a rapid increase in the depth of the tree, affecting the accuracy and readability of clustering results. The traditional methods ignore the timing characteristics of timing data. and fluctuation characteristics, resulting in deviations in the abnormality calculation.
Using intelligent data analysis method, by obtaining the fluctuation curve of chemical production data, dividing molecular curve segments, analyzing the matching degree and abnormal scores of sub-curve segments, adaptively adjusting the branch factor L1, and updating the cluster feature tree in real time to ensure the correlation of data and diversity, avoid improper growth of cluster feature trees.
The accuracy of cluster feature tree update results is improved, the accuracy of abnormal detection is enhanced, and the data in the production process of chemical products is adjusted in real time to ensure the safety of the production process of chemical products.
Smart Images

Figure CN118333547B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of data intelligent analysis, and specifically to a data intelligent analysis method for an intelligent chemical product supply chain management system. Background Art
[0002] The chemical supply chain refers to the entire process from raw material procurement, manufacturing, distribution and sales to after-sales service, involving multiple links and participants, including suppliers, manufacturers, distributors, retailers, etc. With the acceleration of the globalization process, the chemical supply chain is also constantly developing and changing. At present, the management of the chemical supply chain is increasingly dependent on information technology and data analysis. These technologies can help companies better manage inventory, logistics, production and other links, improve efficiency and reduce costs. In addition, with the increasing awareness of sustainable development and environmental protection, the chemical supply chain is also gradually developing towards a green, low-carbon, and circular economy.
[0003] By clustering the acquired supply chain data, we can analyze which products or suppliers have higher risks, and then take preemptive measures to reduce potential risks. BIRCH is a commonly used clustering method. The algorithm has three preset parameters, namely branching factor B, branching factor L, and maximum sample radius threshold T. Among them, the branching factor L is used to limit the maximum number of CFs allowed in the leaf node. Usually, the branching factor L is a fixed value. When new data is added later, if the value of the branching factor L is small, the depth of the tree will increase rapidly, affecting the accuracy and readability of the clustering results. Summary of the invention
[0004] In order to solve the above technical problems, the present invention provides a data intelligent analysis method for an intelligent chemical product supply chain management system to solve the existing problems.
[0005] The data intelligent analysis method of an intelligent chemical product supply chain management system of the present invention adopts the following technical solution:
[0006] An embodiment of the present invention provides a data intelligent analysis method for an intelligent chemical product supply chain management system, the method comprising the following steps:
[0007] Collect historical chemical production data; cluster all chemical production data using the BIRCH algorithm to obtain an initial clustering feature tree; each node of the initial clustering feature tree is composed of various CFs containing chemical production data;
[0008] The data of daily chemical production data in each dimension are used to form a fluctuation curve of the data in each dimension of each day; each sub-curve segment of the fluctuation curve is obtained; for each sub-curve segment where each data on the fluctuation curve is located, the fluctuation characteristic value of each sub-curve segment is determined according to the difference between each sub-curve segment and the adjacent sub-curve segment; the matching degree between any two sub-curve segments in the same dimension is determined according to the difference in the fluctuation characteristic value between any two sub-curve segments in the same dimension;
[0009] Based on the matching degree, the matching sub-curve segments of each sub-curve segment are obtained; according to the matching sub-curve segments of each sub-curve segment where each data point is located on the fluctuation curve of each dimension data every day, the abnormality of each chemical production data of each day is determined; for each leaf node in the initial clustering feature tree, the data stability of each leaf node is determined according to the abnormality of all chemical production data in each leaf node; based on the data stability of each leaf node and the hierarchical distribution in the initial clustering feature tree, the adaptive coefficient of each leaf node is determined; according to the adaptive coefficient of each leaf node, the adaptive branching factor L of each leaf node is determined 1 ;
[0010] Combined with adaptive branching factor L 1 , the initial clustering feature tree will be updated with the newly added chemical production data, and abnormal data will be adjusted in real time to ensure the safety of the chemical product production process.
[0011] Preferably, the obtaining of each sub-curve segment of the fluctuation curve comprises: obtaining all extreme value points in the fluctuation curve by using Newton's method; and forming each sub-curve segment of the fluctuation curve from two adjacent extreme value points and data therebetween.
[0012] Preferably, determining the fluctuation characteristic value of each sub-curve segment according to the difference between each sub-curve segment and an adjacent sub-curve segment includes:
[0013] The mean value of the slope of all data on the sub-curve segment is taken as the slope value of the corresponding sub-curve segment;
[0014] The difference in slope value between each sub-curve segment and its adjacent sub-curve segments on its left and right sides is calculated respectively; the absolute value of the sum of the two differences is used as the exponent of an exponential function with a natural constant as the base, and the calculation result of the exponential function is used as the fluctuation characteristic value of each sub-curve segment.
[0015] Preferably, determining the matching degree between any two sub-curve segments in the same dimension according to the difference in the fluctuation characteristic values between any two sub-curve segments in the same dimension includes:
[0016] Obtain the absolute value of the difference between the fluctuation characteristic values of any two sub-curve segments in the same dimension; take the reciprocal of the absolute value of the difference as the exponent and the exponential function obtained by taking the natural constant as the base as the first exponential function;
[0017] Obtain a normalized value of the absolute value of the difference between the sub-curve segment numbers on the corresponding fluctuation curves of any two sub-curve segments in the same dimension; calculate the sum of the normalized value and the number 1;
[0018] The ratio of the calculation result of the first exponential function to the sum value is used as the matching degree between any two sub-curve segments in the same dimension.
[0019] Preferably, obtaining the matching sub-curve segments of each sub-curve segment based on the matching degree includes: obtaining the sub-curve segments with the maximum matching degree in different days under the same dimension with each sub-curve segment, and respectively using them as the matching sub-curve segments of each sub-curve segment.
[0020] Preferably, the determination of the abnormality of each chemical production data of each day according to the matching sub-curve segments of each sub-curve segment where each data point on the fluctuation curve of each dimension data of each day is located includes:
[0021] For each matching sub-curve segment of each data point on the fluctuation curve of each dimension of data every day, the LOF algorithm is used to obtain the abnormal score of each data point added to each matching sub-curve segment;
[0022] Among them, the expression of the abnormality of the i-th chemical production data on the m-th day is:
[0023]
[0024]
[0025] In the formula, ρ mni represents the abnormality of the ith data on the fluctuation curve of the nth dimension data on the mth day, M represents the number of days of collection, α mni,p Represents the abnormal score of the p-th matching sub-curve segment of the i-th data on the fluctuation curve of the n-th dimension data on the m-th day. When p = 0, α mni,0 Represents the abnormal score of the sub-curve segment mni where the ith data is located on the fluctuation curve of the nth dimension data on the mth day; ρ mi represents the abnormality of the i-th chemical production data on the m-th day, and N represents the number of dimensions; the i-th data on the fluctuation curve of the n-th dimension data on the m-th day corresponds to the i-th chemical production data on the m-th day.
[0026] Preferably, determining the data stability of each leaf node according to the abnormality of all chemical production data in each leaf node includes:
[0027] Get the average value of the abnormality of all chemical production data in each leaf node;
[0028] For each chemical production data in each leaf node, the difference between the mean and the abnormality of each chemical production data is calculated, the normalized value of the difference is calculated, and the mean of the normalized values of all chemical production data in each leaf node is used as the data stability of each leaf node.
[0029] Preferably, the adaptive coefficient of each leaf node is determined based on the data stability of each leaf node and the hierarchical distribution in the initial clustering feature tree, and the expression is:
[0030]
[0031] In the formula, ε b Represents the adaptive coefficient of the bth leaf node, μ b , μ c represents the average abnormality of chemical production data in the bth and cth leaf nodes, respectively. C is the number of all leaf nodes at the level of the initial clustering feature tree where leaf node b is located. H b Indicates the level of the initial clustering feature tree where the leaf node b is located, ω b Indicates the data stability of the b-th leaf node.
[0032] Preferably, the adaptive branching factor L of each leaf node is determined according to the adaptive coefficient of each leaf node. 1 The method comprises: calculating the sum of the adaptive coefficient of each leaf node and the preset adjustment factor; taking the product of the sum and the preset reference value of the branching factor L as the adaptive branching factor L of each leaf node. 1 .
[0033] Preferably, the combination of the adaptive branching factor L 1 , update the initial clustering feature tree with the newly added chemical production data, and adjust the abnormal data in real time, including:
[0034] For the newly added chemical production data, the adaptive branching factor of each leaf node is Instead of the original branching factor L, the initial clustering feature tree is updated;
[0035] The CF of chemical production data in the leaf nodes of the initial clustering feature tree that is less than 5 is recorded as abnormal CF; when the newly added chemical production data is added to the abnormal CF, or the newly added chemical production data causes the number of CFs of the initial clustering feature tree to increase during the addition process, the newly added chemical production data is judged to be abnormal data; otherwise, it is normal data;
[0036] For each dimension data in the abnormal data, when the dimension data is greater than the average threshold of the corresponding dimension data, the corresponding dimension data will be reduced; otherwise, the corresponding dimension data will be increased; the average threshold is the numerical mean of all chemical production data collected historically in the corresponding dimension data.
[0037] The present invention has at least the following beneficial effects:
[0038] The present invention analyzes the characteristics of abnormal time series fluctuation of chemical production data in the supply chain management data of chemical products, realizes the division of sub-curve segments based on the distribution of extreme points in the fluctuation curve, so that the division result of the sub-curve segments is more convenient for analyzing the abnormal fluctuation situation between adjacent sub-curve segments; by analyzing the matching situation between any two sub-curve segments, it is possible to more accurately identify the similarity of fluctuations under different fluctuation amplitudes, construct sub-curve segments matching them, and mine the abnormal situation of data points under similar fluctuation curve conditions; based on the abnormal scores of data points in different matching sub-curve segments, it is easier to highlight abnormal values and distinguish abnormal data points; based on the distribution of each chemical production data in the time series data of different days, the abnormality of each chemical production data is calculated, and the abnormality of the chemical production data is mined at two levels of the time series characteristics and fluctuation characteristics of the time series data; according to the abnormality of the chemical production data in each leaf node in the initial clustering feature tree, the data stability in the leaf node is evaluated, and the adaptive branching factor L of each leaf node is adaptively determined. 1 , that is, the adaptive coefficient of the leaf node is adjusted accordingly, so that the data added to the leaf node has a stronger correlation while being able to accommodate diverse data. This can effectively avoid deviations in the update results of the clustering feature tree when new data is added, and at the same time avoid the situation where the amount of data inserted into the data point node is too large or too small, and constrain the growth of the clustering feature tree under the node characteristics; improve the accuracy of the update results of the clustering feature tree, and then improve the accuracy of subsequent anomaly detection results, adjust the data in the chemical product production process in real time, and ensure the safety of the chemical product production process. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] 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.
[0040] Figure 1 A flow chart of a data intelligent analysis method for an intelligent chemical product supply chain management system provided by the present invention;
[0041] Figure 2 is the adaptive branching factor L of the leaf node 1 The indicator construction flow chart. DETAILED DESCRIPTION
[0042] 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 data intelligent analysis method of an intelligent chemical product supply chain management system proposed by the present invention, its specific implementation method, structure, characteristics and effects, 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.
[0043] 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.
[0044] The specific scheme of the data intelligent analysis method for an intelligent chemical product supply chain management system provided by the present invention is described in detail below with reference to the accompanying drawings.
[0045] An embodiment of the present invention provides a data intelligent analysis method for an intelligent chemical product supply chain management system.
[0046] Specifically, the following data intelligent analysis method for intelligent chemical product supply chain management system is provided. Figure 1 , the method comprises the following steps:
[0047] Step S001, obtaining chemical production data from supply chain management data of chemical products.
[0048] The supply chain data of chemical products covers the entire process from raw material procurement to finished product sales. It is mainly divided into upstream links, intermediate product links, and downstream links. In the intermediate product link, the supply chain data mainly includes various data such as the production process.
[0049] This embodiment obtains chemical production data in the production process of chemical products. Specifically, the chemical production data of the past three months is obtained from the supply chain management system. Among them, the data in the daily production process includes multiple process parameters such as temperature, pressure, and flow. By analyzing these data and analyzing the real-time chemical production data in the subsequent supply chain data, the stability of the production process can be ensured.
[0050] In this embodiment, a total of M days of chemical production data are obtained, and U chemical production data are collected every day. Each chemical production data has N dimensions of data. The chemical production data collected every day are arranged according to the collection time to form a daily chemical production data sequence. Among them, in this embodiment, the number of days M for collection, the number of daily collections U, and the number of dimensions N are set to 30, 60, and 5 respectively, and the implementer can set them according to the actual situation.
[0051] At this point, historical chemical production data can be collected according to the above method.
[0052] Step S002: construct an initial clustering feature tree of chemical production data based on the BIRCH clustering algorithm, and analyze the adaptive branching factor L of each leaf node. 1 .
[0053] All chemical production data collected in the historical M days are clustered using the BIRCH method to obtain an initial clustering feature tree. Assume that there are P leaf nodes in the obtained initial clustering feature tree. The BIRCH method is a formula technology and will not be described in detail in this embodiment.
[0054] Since the abnormal degree of each chemical production data in the chemical production data sequence of each day is different, this embodiment calculates the abnormal degree of each chemical production data in the chemical production data sequence according to the fluctuation characteristics of each chemical production data. Then, the adaptive coefficient of each leaf node is calculated according to the distribution of the abnormal degree of the chemical production data in each leaf node. Then, the adaptive branching factor L of each leaf node is calculated according to the adaptive coefficient of each leaf node. 1 .
[0055] The traditional method of calculating the abnormality of the chemical production data sequence is usually calculated based on the abnormality of each chemical production data in all the data. For the chemical production data sequence of this embodiment, it ignores the temporal characteristics and fluctuation characteristics of the time series data. Therefore, when the abnormality of each chemical production data is calculated using the traditional method, the calculated abnormality will cause deviations. Therefore, this embodiment first calculates the abnormality of each chemical production data in the chemical production data sequence, and then calculates the abnormality of its similar position in other chemical production data sequences, and integrates all the calculated abnormalities to obtain a more accurate abnormality of each chemical production data.
[0056] For the i-th chemical production data in the chemical production data sequence of the m-th day. First, obtain the fluctuation curve of the chemical production data of the m-th day under each dimension data. Taking the fluctuation curve of the n-th dimension data of the m-th day as an example, first use Newton's method to calculate all the extreme points in the fluctuation curve, and then form each sub-curve segment by combining the two adjacent extreme points and the data between them. Newton's method is a well-known technology and will not be described in detail. Similarly, obtain each sub-curve segment of the fluctuation curve of the chemical production data sequence of other days in each dimension data.
[0057] First, the sub-curve segment where the i-th data is located on the fluctuation curve of the n-th dimension data of the m-th day is recorded as the curve segment mni, and then the curve segment mni is matched with the sub-curve segments on the fluctuation curves of the n-th dimension data of other days, and the successfully matched ones are recorded as similar curve segments of the curve segment mni. The specific matching method is as follows:
[0058] σ mni =exp|2l mni -(l mni-1 +l mni+1 )|
[0059] In the formula, σ mni represents the fluctuation characteristic value of the sub-curve segment mni, exp|| represents the exponential function with the natural constant e as the base, l mni Indicates the slope value of the sub-curve segment mni, l mni-1 , l mni+1 They respectively represent the slope values of the previous curve segment and the next curve segment of the fluctuation curve where the sub-curve segment mni is located. The slope value is obtained by the average value of the slopes of all data on the sub-curve segment.
[0060] It should be noted that the greater the difference in slope between the sub-curve segment mni and its adjacent sub-curve segment, the greater the fluctuation amplitude of the sub-curve segment mni, that is, σ mni The larger the value is, the more abnormal fluctuations may occur near the sub-curve segment mni during the production process of chemical products.
[0061] Since the fluctuation amplitude of the nth dimension data of different days is not the same, the fluctuation characteristic value of the sub-curve segment mni cannot be accurately represented by the slope value of the adjacent sub-curve segment mni. Therefore, it is necessary to analyze the fluctuation difference between the sub-curve segment mni and any sub-curve segment to obtain a more accurate fluctuation characteristic. This embodiment takes the matching degree between the sub-curve segment mni and the sub-curve segment anj as an example:
[0062]
[0063] In the formula, θ mni,anjrepresents the matching degree between the sub-curve segment mni and the sub-curve segment anj, where the sub-curve segment anj represents the sub-curve segment where the j-th data on the fluctuation curve of the n-th dimension data on the a-th day is located, norm() represents the normalization function, exp() represents the exponential function with the natural constant e as the base, and w mni 、w anj They represent the sub-curve segment numbers of sub-curve segment mni and sub-curve segment anj in their fluctuation curves, σ mni , σ anj They represent the fluctuation characteristic values of sub-curve segment mni and sub-curve segment anj respectively, and +1 is used to avoid the situation where the denominator is 0.
[0064] It should be noted that norm(|w mni -w anj The smaller the value of |), the closer the corresponding number values of sub-curve segments mni and sub-curve segments anj in their fluctuation curves are, that is, the more similar the corresponding positions of the two sub-curve segments in the fluctuation curve, the higher the matching degree of the corresponding sub-curve segments mni and sub-curve segments anj. mni -σ anj The smaller the value of | is, the more similar the fluctuation characteristic values of the sub-curve segments mni and anj are, and the higher the matching degree of the corresponding sub-curve segments mni and anj is.
[0065] According to the above method, the matching degree between the sub-curve segment mni and all the sub-curve segments on the fluctuation curve of the n-th dimension data on the a-th day can be calculated. The sub-curve segment with the highest matching degree is recorded as the matching sub-curve segment of the sub-curve segment mni on the fluctuation curve of the n-th dimension data on the a-th day. The sub-curve segments matching the sub-curve segment mni on the fluctuation curve of the n-th dimension data on the remaining days can be calculated using the same method.
[0066] Then use the LOF algorithm to calculate the anomaly score in the sub-curve segment mni where the i-th data is located. The LOF algorithm is a well-known technology and will not be described in detail. Each sub-curve segment can be matched to M-1 matching sub-curve segments. Then add the i-th data to the M-1 matching sub-curve segments respectively, and calculate the anomaly score of the i-th data in the other M-1 matching sub-curve segments respectively. Taking the p-th day as an example, the anomaly score of the i-th data on the p-th matching sub-curve segment is recorded as α mni,p .
[0067]
[0068]
[0069] In the formula, ρ mnirepresents the abnormality of the ith data on the fluctuation curve of the nth dimension data on the mth day, M represents the number of days of collection, α mni,p Represents the abnormal score of the p-th matching sub-curve segment of the i-th data on the fluctuation curve of the n-th dimension data on the m-th day. When p = 0, α mni,0 Represents the abnormal score of the sub-curve segment mni where the ith data is located on the fluctuation curve of the nth dimension data on the mth day; ρ mi represents the abnormality of the ith chemical production data on the mth day, and N represents the number of dimensions. The ith data on the fluctuation curve of the nth dimension data on the mth day corresponds to the ith chemical production data on the mth day.
[0070] According to the above method, the abnormality of each chemical production data in the daily chemical production data sequence can be calculated to characterize the abnormality of each chemical production data on the matching sub-curve segments in different dimensions. When the abnormality score of the i-th data on the p-th matching sub-curve segment is large, that is, the i-th data is more abnormal relative to the p-th matching sub-curve segment, the abnormality of the data is greater, that is, the abnormality of the chemical production data corresponding to the data at the time of collection is also greater.
[0071] After calculating the abnormality of each chemical production data according to the above method, the abnormality distribution of all data in each leaf node is then considered. If the abnormality distribution of the data in each leaf node is more similar and the abnormality value is smaller, the data stability in the corresponding leaf node is higher, and the adaptive coefficient of the leaf node should be increased in the subsequent update process to maintain the diversity of the data; otherwise, the adaptive coefficient of the corresponding leaf node needs to be reduced to ensure that the data in the leaf node is maintained within a certain abnormal range.
[0072] Since the newly added data in the initial clustering feature tree of BIRCH is inserted from the leaf node during the update process, if the leaf node cannot be inserted, it will be inserted into the non-leaf node. Therefore, when setting the adaptive coefficient of the leaf node, the higher the level of the leaf node, the stronger the correlation of the corresponding supply chain production data of the chemical product, and the closer the production association. Therefore, when the node is subsequently adapted, the higher the level of the leaf node, the smaller the corresponding adaptive coefficient should be, so as to avoid the strong correlation of the supply chain production data in the node being destroyed, that is, the corresponding adaptive coefficient should be relatively smaller, and the data should be inserted into the leaf node as much as possible to avoid the amount of data inserted into the non-leaf node being too large, which will increase the depth of the initial clustering feature tree. Let the number of chemical production data contained in the bth leaf node be D, and the abnormality of the dth chemical production data in the bth leaf node be Yρ bd According to the above steps, the data stability degree ω of the b-th leaf node can be calculated b.
[0073]
[0074] In the formula, ω b represents the data stability of the b-th leaf node, D represents the number of chemical production data in the b-th leaf node, norm() represents the normalization function, and Yρ bd Indicates the abnormality of the d-th chemical production data in the b-th leaf node, Represents the average abnormality of all chemical production data in the bth leaf node.
[0075] It should be noted that The smaller the value, the more stable the d-th chemical production data in the b-th leaf node is. By summing up and analyzing the stability of all chemical production data in the b-th leaf node, the data stability degree ω of the b-th leaf node is obtained. b ,ω b The larger the value, the more uniform the chemical production data within the leaf node.
[0076]
[0077] In the formula, ε b Represents the adaptive coefficient of the bth leaf node, μ b , μ c represents the average abnormality of chemical production data in the bth and cth leaf nodes, respectively. C is the number of all leaf nodes at the level of the initial clustering feature tree where leaf node b is located. H b Indicates the level of the initial clustering feature tree where the leaf node b is located, +0.1 is to avoid the denominator being 0, ω b Indicates the data stability of the b-th leaf node.
[0078] It should be noted that The larger the value of , the greater the average abnormality of the chemical production data in the b-th leaf node among all the leaf nodes at the same level, and the smaller the corresponding adaptive coefficient is. b The larger the value is, the farther the level of the b-th leaf node is from the root node, and the more similar the chemical production data is, the smaller the judgment range of the corresponding chemical production data is, so that the data in the leaf node is stabilized to a certain extent; at the same time, ω b The larger the value is, the higher the data stability in the corresponding b-th leaf node is, and the larger the corresponding adaptive coefficient is, the more data diversity characteristics should be accommodated.
[0079] δ b =m L ×(α+ε b )
[0080] In the formula, δ b represents the adaptive branching factor L of the bth leaf node 1 , m L represents the reference value of the branching factor L, which is 10 in this embodiment, α is the adjustment factor, which is 0.5 in this embodiment, and ε b Represents the adaptive coefficient of the bth leaf node. Among them, the adaptive branching factor L of the leaf node 1 The indicator construction flow chart is as follows Figure 2 shown.
[0081] It should be noted that when α+ε b When it is greater than 1, the base value of the branching factor L is enlarged, and the adaptive branching factor L of the b-th leaf node is obtained. 1 The larger the b The larger it is, the more diverse the data contained in the leaf node.
[0082] According to the above method, the adaptive branching factor L of each leaf node in the initial clustering feature tree can be calculated: 1 .
[0083] Step S003, updating the initial clustering feature tree with real-time chemical production data, judging abnormal data, and ensuring the safety of the chemical product production process.
[0084] According to the above method, the adaptive branching factor L of each leaf node in the initial clustering feature tree is calculated. 1 Then, the BIRCH method is used to update the initial clustering feature tree for the newly added new chemical production data.
[0085] First, the chemical production data with a value less than 5 in all CFs in the tree subnodes of the initial clustering feature tree are recorded as abnormal CFs. If the newly added chemical production data is classified into abnormal CFs or a new CF is added during the clustering feature tree update process, the newly added chemical production data is abnormal data; otherwise, the newly added chemical production data is judged to be normal data.
[0086] By analyzing the abnormal chemical production data, the data of each dimension in the chemical production process is adjusted accordingly. If the newly added chemical production data is abnormal data, the size between each dimension data of the chemical production data and the average threshold of the corresponding dimension data is compared. If it is greater than the average threshold, the corresponding dimension data is adjusted down, and if it is less than the average threshold, the corresponding dimension data is adjusted up to ensure the safety of the chemical product production process. The average threshold is the numerical mean of the corresponding dimension data of all chemical production data collected historically.
[0087] In summary, the embodiment of the present invention analyzes the characteristics of abnormal time series fluctuations of chemical production data in the supply chain management data of chemical products, and realizes the division of sub-curve segments based on the distribution of extreme points in the fluctuation curve, so that the division result of the sub-curve segments is more convenient for analyzing the abnormal fluctuations between adjacent sub-curve segments; by analyzing the matching between any two sub-curve segments, it is possible to more accurately identify the similarity of fluctuations under different fluctuation amplitudes, construct sub-curve segments matching them, and mine the abnormalities of data points under similar fluctuation curve conditions; based on the abnormal scores of data points in different matching sub-curve segments, it is easier to highlight abnormal values and distinguish abnormal data points; based on the distribution of each chemical production data in the time series data of different days, the abnormality of each chemical production data is calculated, and the abnormality of the chemical production data is mined at two levels of the time series characteristics and fluctuation characteristics of the time series data; according to the abnormality of the chemical production data in each leaf node in the initial clustering feature tree, the data stability in the leaf node is evaluated, and the adaptive branching factor L of each leaf node is adaptively determined. 1 , that is, the adaptive coefficient of the leaf node is adjusted accordingly, so that the data added to the leaf node has a stronger correlation while being able to accommodate diverse data. This can effectively avoid deviations in the update results of the clustering feature tree when new data is added, and at the same time avoid the situation where the amount of data inserted into the data point node is too large or too small, and constrain the growth of the clustering feature tree under the node characteristics; improve the accuracy of the update results of the clustering feature tree, and then improve the accuracy of subsequent anomaly detection results, adjust the data in the chemical product production process in real time, and ensure the safety of the chemical product production process.
[0088] 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 above is a description of a specific embodiment of this specification. In addition, 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.
[0089] 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.
[0090] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Modifications to the technical solutions recorded in the aforementioned embodiments, or equivalent replacement of some of the technical features therein, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.
Claims
1. A data intelligent analysis method for an intelligent chemical product supply chain management system, characterized in that: The method comprises the following steps: Collect historical chemical production data; cluster all chemical production data using the BIRCH algorithm to obtain an initial clustering feature tree; each node of the initial clustering feature tree is composed of various CFs containing chemical production data; The data of daily chemical production data in each dimension are used to form a fluctuation curve of the data in each dimension of each day; each sub-curve segment of the fluctuation curve is obtained; for each sub-curve segment where each data on the fluctuation curve is located, the fluctuation characteristic value of each sub-curve segment is determined according to the difference between each sub-curve segment and the adjacent sub-curve segment; the matching degree between any two sub-curve segments in the same dimension is determined according to the difference in the fluctuation characteristic value between any two sub-curve segments in the same dimension; Based on the matching degree, the matching sub-curve segments of each sub-curve segment are obtained; according to the matching sub-curve segments of each sub-curve segment where each data point is located on the fluctuation curve of each dimension data every day, the abnormality of each chemical production data of each day is determined; for each leaf node in the initial clustering feature tree, the data stability of each leaf node is determined according to the abnormality of all chemical production data in each leaf node; based on the data stability of each leaf node and the hierarchical distribution in the initial clustering feature tree, the adaptive coefficient of each leaf node is determined; according to the adaptive coefficient of each leaf node, the adaptive branching factor L of each leaf node is determined 1 ; Combined with adaptive branching factor L 1 , the newly added chemical production data will be used to update the initial clustering feature tree, and abnormal data will be adjusted in real time to ensure the safety of the chemical product production process; The step of determining the fluctuation characteristic value of each sub-curve segment according to the difference between each sub-curve segment and an adjacent sub-curve segment includes: The mean value of the slope of all data on the sub-curve segment is taken as the slope value of the corresponding sub-curve segment; The difference in slope value between each sub-curve segment and its adjacent sub-curve segments on the left and right sides is calculated respectively; the absolute value of the sum of the two differences is used as the exponent of an exponential function with a natural constant as the base, and the calculation result of the exponential function is used as the fluctuation characteristic value of each sub-curve segment; The abnormality of each chemical production data of each day is determined according to the matching sub-curve segments of each sub-curve segment where each data point on the fluctuation curve of each dimension data of each day, including: For each matching sub-curve segment of each data point on the fluctuation curve of each dimension of data every day, the LOF algorithm is used to obtain the abnormal score of each data point added to each matching sub-curve segment; Among them, the expression of the abnormality of the i-th chemical production data on the m-th day is: In the formula, ρ mni represents the abnormality of the ith data on the fluctuation curve of the nth dimension data on the mth day, M represents the number of days of collection, α mni,p Represents the abnormal score of the p-th matching sub-curve segment of the i-th data on the fluctuation curve of the n-th dimension data on the m-th day. When p = 0, α mni,0 Represents the abnormal score of the sub-curve segment mni where the ith data is located on the fluctuation curve of the nth dimension data on the mth day; ρ mi represents the abnormality of the i-th chemical production data on the m-th day, and N represents the number of dimensions; the i-th data on the fluctuation curve of the n-th dimension data on the m-th day corresponds to the i-th chemical production data on the m-th day; The adaptive coefficient of each leaf node is determined based on the data stability of each leaf node and the hierarchical distribution in the initial clustering feature tree. The expression is: In the formula, ε b Represents the adaptive coefficient of the bth leaf node, μ b , μ c represents the average abnormality of chemical production data in the bth and cth leaf nodes, respectively. C is the number of all leaf nodes at the level of the initial clustering feature tree where leaf node b is located. H b Indicates the level of the initial clustering feature tree where the leaf node b is located, ω b Indicates the data stability of the b-th leaf node; The adaptive branching factor L of each leaf node is determined according to the adaptive coefficient of each leaf node. 1 The method comprises: calculating the sum of the adaptive coefficient of each leaf node and the preset adjustment factor; taking the product of the sum and the preset reference value of the branching factor L as the adaptive branching factor L of each leaf node. 1 .
2. The data intelligent analysis method of the intelligent chemical product supply chain management system according to claim 1, characterized in that: The obtaining of each sub-curve segment of the fluctuation curve comprises: obtaining all extreme value points in the fluctuation curve by using Newton's method; and forming each sub-curve segment of the fluctuation curve by combining two adjacent extreme value points and data therebetween.
3. The data intelligent analysis method of the intelligent chemical product supply chain management system according to claim 1, characterized in that: The determining the matching degree between any two sub-curve segments in the same dimension according to the difference in the fluctuation characteristic values between any two sub-curve segments in the same dimension includes: Obtain the absolute value of the difference between the fluctuation characteristic values of any two sub-curve segments in the same dimension; take the reciprocal of the absolute value of the difference as the exponent and the exponential function obtained by taking the natural constant as the base as the first exponential function; Obtain a normalized value of the absolute value of the difference between the sub-curve segment numbers on the corresponding fluctuation curves of any two sub-curve segments in the same dimension; calculate the sum of the normalized value and the number 1; The ratio of the calculation result of the first exponential function to the sum value is used as the matching degree between any two sub-curve segments in the same dimension.
4. The data intelligent analysis method of the intelligent chemical product supply chain management system according to claim 1, characterized in that: The obtaining of matching sub-curve segments of each sub-curve segment based on the matching degree includes: obtaining sub-curve segments having the maximum matching degree with each sub-curve segment in different days under the same dimension, and using them as matching sub-curve segments of each sub-curve segment.
5. The data intelligent analysis method of the intelligent chemical product supply chain management system according to claim 1, characterized in that: Determining the data stability of each leaf node according to the abnormality of all chemical production data in each leaf node includes: Get the average value of the abnormality of all chemical production data in each leaf node; For each chemical production data in each leaf node, the difference between the mean and the abnormality of each chemical production data is calculated, the normalized value of the difference is calculated, and the mean of the normalized values of all chemical production data in each leaf node is used as the data stability of each leaf node.
6. The data intelligent analysis method of the intelligent chemical product supply chain management system according to claim 1, characterized in that: The combined adaptive branching factor L 1 , update the initial clustering feature tree with the newly added chemical production data, and adjust the abnormal data in real time, including: For the newly added chemical production data, the adaptive branching factor L of each leaf node is 1 Instead of the original branching factor L, the initial clustering feature tree is updated; The CF of chemical production data in the leaf nodes of the initial clustering feature tree that is less than 5 is recorded as abnormal CF; when the newly added chemical production data is added to the abnormal CF, or the newly added chemical production data causes the number of CFs of the initial clustering feature tree to increase during the addition process, the newly added chemical production data is judged to be abnormal data; otherwise, it is normal data; For each dimension data in the abnormal data, when the dimension data is greater than the average threshold of the corresponding dimension data, the corresponding dimension data will be reduced; otherwise, the corresponding dimension data will be increased; the average threshold is the numerical mean of all chemical production data collected historically in the corresponding dimension data.
