CN112116198A - Data-driven process industrial state perception network key node screening method - Google Patents
Data-driven process industrial state perception network key node screening method Download PDFInfo
- Publication number
- CN112116198A CN112116198A CN202010774423.4A CN202010774423A CN112116198A CN 112116198 A CN112116198 A CN 112116198A CN 202010774423 A CN202010774423 A CN 202010774423A CN 112116198 A CN112116198 A CN 112116198A
- Authority
- CN
- China
- Prior art keywords
- node
- state
- network
- data
- variables
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 114
- 230000008569 process Effects 0.000 title claims abstract description 41
- 238000012216 screening Methods 0.000 title claims abstract description 34
- 230000008447 perception Effects 0.000 title claims abstract description 16
- 238000011156 evaluation Methods 0.000 claims abstract description 119
- 230000001364 causal effect Effects 0.000 claims abstract description 97
- 238000012544 monitoring process Methods 0.000 claims abstract description 92
- 230000003993 interaction Effects 0.000 claims abstract description 15
- 238000010606 normalization Methods 0.000 claims description 24
- 230000009467 reduction Effects 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 14
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 claims description 13
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 claims description 13
- 241000039077 Copula Species 0.000 claims description 12
- 238000009826 distribution Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000012512 characterization method Methods 0.000 claims description 6
- 238000011425 standardization method Methods 0.000 claims description 6
- 239000002131 composite material Substances 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 230000002452 interceptive effect Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000000844 transformation Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000005457 optimization Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 239000000126 substance Substances 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000000498 cooling water Substances 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/18—Network design, e.g. design based on topological or interconnect aspects of utility systems, piping, heating ventilation air conditioning [HVAC] or cabling
Abstract
The invention discloses a data-driven process industry state perception network key node screening method which comprises the steps of firstly, calculating causal relationship values among variables by a Copula-Granger method, introducing Gaussian noise to effectively eliminate false causal relationship among the variables, and establishing a causal network model capable of accurately reflecting information interaction modes among monitoring variables of a system; calculating LR values of all nodes in the network by using a Leader Rank algorithm, ranking all nodes of the causal network, and dividing a node set by a preset proportion difference according to ranking results to form a plurality of groups of key node sets; the method comprises the steps of collecting historical monitoring data of a system, respectively evaluating the state of the system by using each group of key node sets and all nodes, establishing a multi-index weighted comprehensive index MCI (multi-index weighted comprehensive index) to evaluate the accuracy and timeliness of each key node set on the evaluation result of the state of the system by taking the evaluation result of all nodes on the state of the system as a reference, obtaining the optimal key node determination proportion, and realizing effective screening of the process industry state perception network key node set.
Description
Technical Field
The invention relates to the technical field of state evaluation and optimization of complex electromechanical systems, in particular to a data-driven method for screening key nodes of a process industrial state perception network.
Background
The process industry represented by energy chemical industry plays a very important role in national economy of China, and production equipment of modern chemical enterprises is increasingly large in scale along with the continuous improvement of market demands. In order to effectively schedule, evaluate and optimize the production state of the equipment, enterprises install tens of thousands of digital instruments and sensors in each production equipment to monitor the production state of the system in real time, and a state sensing network capable of effectively sensing the state of the system is formed. Due to the characteristics of multiple points and large data volume, the process industry state perception network has an obvious value sparse characteristic, and it is particularly necessary to screen out key nodes capable of accurately reflecting the system state from tens of thousands of state perception network nodes.
In order to solve the problem of screening key nodes of a complex system, Guocheng and the like have the characteristic of invariance aiming at a system structure, an information flow model of a plurality of subsystems reflecting the overall characteristics of the system is established according to a production process flow diagram of the system, and key information nodes of the system are obtained through the model based on a PageRank algorithm; liu Zhonghua and the like propose a TOPSIS-based multi-attribute decision method according to the physical topological structure of large hydraulic engineering to identify key nodes in a complex hydraulic engineering network, and the survivability of the hydraulic engineering is improved by protecting the screened key nodes; liu xiaoli etc. are big to the electric wire netting scale, and the characteristics that the risk node is difficult to be effectively discerned synthesize two indexes of the active global influence of node and idle global influence and confirm the fragile node in the electric wire netting. The researches provide different node importance degree evaluation methods aiming at the key node screening problems of different complex systems, and the screening target of the key nodes is realized. However, these researches are directed to developing key node identification for a process diagram or a physical topological structure of a complex system, and a complex system key node screening method based on the process diagram or the physical topological structure cannot accurately reflect an information interaction mode between monitoring variables of a system state sensing network, so that the method cannot be effectively applied to screening of key nodes of the complex electromechanical system state sensing network with large node scale and ambiguous correlation relationship between nodes.
Disclosure of Invention
In order to solve the above problems, an object of the present invention is to provide a data-driven method for screening key nodes of a process industry state awareness network, which solves the problems that the existing state evaluation technology needs to calculate the correlation of all system variables to evaluate the state of a complex electromechanical system, so that a large amount of computing resources are consumed in the evaluation process, and the existing method for screening key nodes of a complex system cannot effectively identify key nodes capable of accurately reflecting the state of the system based on state monitoring data of each node of the system.
In order to achieve the purpose, the invention adopts the following technical scheme:
the data-driven process industry state perception network key node screening method comprises the following steps:
step 1), state monitoring time sequence data of a plurality of monitoring variables of a complex electromechanical system are obtained, and noise reduction and normalization processing are respectively carried out on the state monitoring time sequence data by adopting wavelet noise reduction and min-max standardization methods;
step 2), analyzing the causal relationship among the state monitoring time sequence data subjected to noise reduction and normalization in the step 1) by using a Copula-Granger method to obtain a causal relationship matrix among variables, and setting the autocorrelation coefficients of the variables to 0;
step 3), generating a Gaussian noise with the mean value of 0 and the variance of 1, analyzing the causal relationship between the monitoring time sequence data of each state and the Gaussian noise after noise reduction and normalization in the step 1) by using a Copula-Granger method, and simplifying the causal relationship matrix obtained in the step 2) by using a 1.5-time value of the causal relationship between each monitoring variable and the Gaussian noise as a threshold value;
step 4), constructing a causal relationship network model reflecting an information interaction mode among the monitoring variables by taking the monitoring variables as nodes and the causal relationship values among the monitoring variables as side weights;
step 5), calculating the LR value of each node of the causal relationship network model obtained in the step 4) by using a Leader Rank algorithm, ranking each node, dividing a node set according to a ranking result and a preset proportion difference, and acquiring a plurality of groups of key node sets;
and 6) evaluating the state of the complex electromechanical system by using the obtained multiple groups of key node sets respectively, calculating absolute percentage errors MAPE and contour similarity CS of the evaluation results of the key node sets on the system state and the evaluation results of all nodes on the system state, and comprehensively considering the multi-index weighted comprehensive index MCI and the number of the key nodes to obtain the optimal key node proportion, so as to screen out the key nodes meeting the evaluation requirements of the system state.
Further, the step 1) of normalizing the data by the min-max normalization method is to perform normalization on each variable time series x1,x2,…xnThe following transformations are performed:
obtain a time series y1,y2,…ynI.e. the time series after normalization, where yn∈[0,1]。
Further, the step 2) of analyzing the causal relationship among the variables after the pretreatment of the monitoring data by using the Copula-Granger method comprises the following steps:
(1) obtaining state monitoring data of variable X and variable YAnd wherein m and n respectively represent the delay dimensions of the variable X and the variable Y;
(2) calculating the edge distribution of the variables X and Y by a kernel function estimation method according to the monitoring data of the variables X and YAnd
(3) based on the edge distribution calculated in the step (2), estimating an empirical Copula density function by adopting a rank statistic methodThe Copula density function estimation process for the continuous binary joint distribution F (x, y) is as follows:
wherein Representing K samples, x, subject to a continuous binary joint distribution F (x, y)(i) and y(j)Representing the order statistics in the respective unary samples, 1 ≦ i, j ≦ K;
(4) applying a Bernstein approximation method to the optimal estimation of the Copula density function, and optimizing the Copula density function estimation in the step (3), wherein the expression of the Bernstein polynomial approximation is as follows:
wherein ,uk=F(xk,yk),u1,…,ukFor a k-dimensional uniform random variable obtained by sample calculation,mjrepresents the size of Copula probability density function, 0 ≦ vj≤mj;
(5) Taking logarithm of the Copula density function obtained through the optimal estimation in the step (4), calculating expectation of a sample, and obtaining a causal relationship among monitoring variables:
further, the simplified cause and effect relationship matrix in step 3) is an n × n dimensional matrix with all 0 diagonal lines, as follows:
in the formula, each element cijRepresenting a causal relationship value between variable i and variable j.
Further, the causal network model established in step 4) is a directional weighting network, the direction of which is determined according to the causal relationship value, cijRepresenting directed connections from variable i to variable j;
further, in the step 5), the LR value of each node in the causal relationship network model is calculated by using a Leader Rank algorithm, and the specific steps are as follows:
(1) adding a background node G into the causal relationship network model obtained in the step 4);
(2) calculating a Google relation matrix A of each node according to the causal relation network model obtained in the step 4), wherein:
(3) assigning the initial importance of each node in the network to be 1;
(4) calculating LR values of all nodes in the causal relationship network, and iterating until the algorithm converges, wherein the LR value calculation formula is as follows:
wherein j is the number of variables in the causal network,weighting out degree of variable j, and t is iteration frequency;
(5) bisecting the LR value of the background node, adding the halved LR value into the LR value after convergence of each node, and obtaining the global LR value of each node:
further, according to the obtained global LR value, ranking each node according to the sequence from large to small;
further, the step of evaluating the state of the complex electromechanical system by using the key node set in the step 6) is as follows:
(1) extracting historical monitoring data of the complex electromechanical system in a non-steady operation state within a period of time, and evaluating the system state by using a sliding window method;
(2) calculating the causal relationship of the time sequence of the key variables in each window by using a Copula-Granger method to obtain a dynamic network reflecting the key variable information interactive evolution mode in real time, extracting the network structure entropy characteristics of the causal network of each window, and evaluating the system state to obtain an evolution curve of the system state, wherein the calculation formula of the network structure entropy is as follows:
in the formula ,P(ki) As intensity distribution of network points, kiIs the point strength of node i and N is the number of network nodes.
(3) And evaluating the importance of each node set by adopting a multi-index weighted composite index MCI, wherein the node set with the lowest MCI index is a key node set capable of effectively representing the system state.
Further, the MCI is obtained by weighting and summing the characterization evaluation result accuracy evaluation index Era and the characterization evaluation result timeliness evaluation index Ute, and the weights of the indexes are obtained by an entropy weight method, and the method comprises the following steps:
(1) the information entropies of the two evaluation indexes Era and Ute are calculated respectively:
wherein ,j represents the jth evaluation index, i represents the ith dimension characteristic of each evaluation index, and Yij=[era1,ute1;era2,ute2;…,…;eran,uten],EjInformation entropy for each evaluation index;
(2) information entropy E of evaluation index Era obtained in (1)1And information entropy E of evaluation index Ute2And calculating the weight of each evaluation index:
(3) for a determined indicator Era era1,era2,…eranAnd Ute ═ ute1,ute2,…utenNormalizing the dimension data of the two indexes to obtain normalized indexes of Era 'and Ute' respectively;
(4) the MCI evaluates the time consumption index Ute 'and the corresponding weight W thereof through the evaluation result accuracy index Era' and the unit window1、W2To obtain the product of:
MCI=Era′·W1+Ute′·W2。
further, the evaluation index Ute is the time required to calculate the evaluation result for each sliding window.
Further, Era is determined by the combination of the contour similarity CS and the absolute percentage error MAPE of each node set evaluation result X and the full node evaluation result Y, and includes the following steps:
(1) calculating the average absolute percentage error MAPE of each node set evaluation result X and the whole node evaluation result Y:
wherein N is the total number of samples of the evaluation result, xi,yiRespectively obtaining evaluation results under the ith sliding window of each node set evaluation result X and the full node evaluation result Y;
(2) normalizing the two curves respectively to eliminate the influence of numerical value difference on the similarity judgment result;
(3) the contour similarity CS of the normalized curve is calculated using the pearson correlation coefficient as follows:
wherein ,andrespectively averaging all samples of the partial node set evaluation result X and the full node set evaluation result Y;
(4) calculating the root mean square of the dissimilarity degree 1-CS and the absolute percentage error MAPE to obtain an accuracy evaluation index Era of the characterization evaluation result:
further, a key node set with the minimum weighted composite index MCI value is selected as a key node set of the system state perception network, and the proportion of the key node set can be used universally in the energy chemical equipment state evaluation with similar data characteristics.
Compared with the prior art, the invention has at least the following beneficial technical effects:
the method starts from time sequences of monitoring variables of the state sensing network of the complex system, analyzes causal relationships among the monitoring variables of the state sensing network of the system, establishes a directed weighting network model capable of accurately reflecting information interaction modes of the nodes, provides a method for effectively identifying key nodes in the model, realizes screening of the key nodes of the state sensing network of the process industry, and further provides a better, rapid and effective solution for state evaluation and optimization of the complex electromechanical system of the process industry.
The method comprises the steps of screening key nodes of a comprehensive complex network and screening a key node set which has a large influence on the system state by a weighting comprehensive evaluation method based on an entropy weight method, calculating causal relationship values among monitoring time sequence data of each state after noise reduction and normalization by using a Copula-Granger method, introducing Gaussian noise to effectively eliminate false causal relationship among the monitoring time sequence data of each state after noise reduction and normalization, and establishing a causal network model capable of reflecting an information interaction mode among monitoring variables of the system;
furthermore, the LR value of each node in the network is calculated by adopting a Leader Rank algorithm, each node of the causal network is ranked, and a node set is screened in a certain proportion according to the ranking result to form a plurality of groups of key node sets; collecting historical monitoring data of the complex electromechanical system, and evaluating the system state by using each group of key node set and all nodes respectively; and establishing a multi-index weighted comprehensive index MCI (multi-index-weighted index) to evaluate the accuracy and timeliness of each key node set on the system state evaluation result by taking the evaluation results of all nodes on the system state as a reference, obtaining the optimal key node determination proportion, and realizing effective screening of the process industrial state perception network key node set.
Compared with other complex system key node identification methods, the method provided by the invention starts from the monitoring data of each node of the system state sensing network, and the established model can reflect the information interaction mode among the nodes of the system more accurately, so that the key node set which has a large influence on the system state is accurately identified, the key variable determination basis can be provided for the data-driven complex system state evaluation technology, and the real-time performance of the evaluation result is improved to the maximum extent.
Drawings
FIG. 1 is a flow chart of a complex electromechanical system state perception network key node screening;
FIG. 2 is a diagram showing results before and after pretreatment of monitoring data;
FIG. 3 is a Gaussian noise sequence used to determine causal thresholds between variables;
FIG. 4 is a causal relationship network model among variables;
FIG. 5 shows the Leader Rank importance and Rank of each node of the causal network;
FIG. 6 is a diagram of the evaluation results of each group of key node sets on the system state;
FIG. 7 is a graph of the trend of the main evaluation indicators of each group of key nodes for the system state evaluation results;
FIG. 8 is a graph of all nodes and key nodes causal versus heat;
FIG. 9 is a comparison graph of the state labeling results of all nodes and key nodes.
Detailed Description
The invention is described in further detail below with reference to the attached drawing figures:
as shown in fig. 1 to 9, the method for screening key nodes of a data-driven process industry state awareness network specifically includes the following steps:
step 1), state monitoring time sequence data of a plurality of monitoring variables of a complex electromechanical system are obtained, and noise reduction and normalization processing are carried out on the state monitoring time sequence data by adopting a wavelet noise reduction and min-max standardization method in sequence;
step 2), analyzing the causal relationship among the state monitoring time sequence data subjected to noise reduction and normalization in the step 1) by using a Copula-Granger method to obtain a causal relationship matrix among variables, and setting the autocorrelation coefficients of the variables to 0;
step 3), generating a Gaussian noise with the mean value of 0 and the variance of 1, analyzing the causal relationship between the monitoring time sequence data of each state and the Gaussian noise after noise reduction and normalization in the step 1) by using a Copula-Granger method, and simplifying the causal relationship matrix obtained in the step 2) by using a 1.5-time value of the causal relationship between each monitoring variable and the Gaussian noise as a threshold value;
step 4), constructing a causal relationship network model reflecting an information interaction mode among the monitoring variables by taking the monitoring variables as nodes and the causal relationship values among the monitoring variables as side weights;
step 5), calculating the LR value of each node of the causal relationship network model obtained in the step 4) by using a Leader Rank algorithm, ranking each node, dividing a node set by a proportion difference of 10% according to a ranking result, and obtaining a plurality of groups of key node sets;
and 6) evaluating the state of the complex electromechanical system by using the obtained multiple groups of key node sets respectively, calculating absolute percentage errors MAPE and contour similarity CS of the evaluation results of the key node sets on the system state and the evaluation results of all nodes on the system state, and comprehensively considering the multi-index weighted comprehensive index MCI and the number of the key nodes to obtain the optimal key node proportion, so as to screen out the key nodes meeting the evaluation requirements of the system state.
Furthermore, the obtained monitoring data of each variable is required to be guaranteed to reflect the real running state of the system, and the monitoring data of faults of the digital instrument and the sensor are eliminated as far as possible.
Further, the step 1) of normalizing the data by the min-max normalization method is to perform normalization on each variable time series x1,x2,…xnThe following transformations are performed:
obtain a time series y1,y2,…ynI.e. the time series after normalization, where yn∈[0,1]。
Further, in the step 2), the cause-and-effect relationship among variables after the monitoring data are preprocessed is analyzed by a Copula-Granger method, delay dimensions m and n of the monitoring data of the variables need to be considered, and the delay dimensions are obtained by combining the characteristics of a process industrial DCS system and using a monitoring variable experiment with strong correlation.
Further, the causal relationship matrix obtained in step 3) is obtained by removing causal relationship values smaller than a threshold value, which are 1.5 times of the causal relationship between each monitored variable and gaussian noise, and all the autocorrelation causal relationship values between the variables are set to 0, as follows:
in the above formula, cijRepresenting a causal relationship value between variable i and variable j.
Further, the background node G introduced when ranking the importance of each node in the causal network by using the Leader Rank algorithm in step 5) does not have any physical significance, and is only to solve the problems that all the LR values of all nodes finally become 0 and the iteration speed of the algorithm becomes slow due to the existence of the suspended nodes in the network.
Further, in step 5), a plurality of groups of key node sets are obtained according to the LR value by a ratio difference of 10%, and the selection of the ratio needs to comprehensively consider the number of all monitoring nodes and the number of the key node sets with the minimum ratio for comprehensive division.
Further, the system state monitoring data collected in the step 6) should reflect fluctuation in the system operation process, rather than historical monitoring data when the system is operated stably at rated load.
Further, the step 6) of evaluating the system state by using the obtained multiple sets of key node sets and all nodes respectively comprises the following steps:
(1) denoising and normalizing the acquired variable state monitoring data of the system by using wavelet denoising and min-max standardization methods in sequence, and selecting a certain group of key node set n from the preprocessed variable monitoring data1,n2,…nkCorresponding monitoring variable calendarHistory monitoring dataWherein k is the number of nodes in the key node set, and l is the length of the historical monitoring data;
(2) carrying out window division on the historical monitoring data with the length of l by using a sliding window method, wherein the width of a window is T, and the sliding step length is SST, so as to obtain n groups of historical monitoring data samples:
n=floor((l-T)/SST)
(3) acquiring state monitoring data of all variables in a key node set under each sliding window, calculating a causal relationship value of time series data of each key variable in the window by using a Copula-Granger method to obtain a dynamic causal network capable of reflecting an information interaction evolution mode among the variables in real time, extracting network structure entropy characteristics of the causal network in each window, and evaluating the system state, wherein a calculation formula of the network structure entropy is as follows:
in the formula ,P(ki) As intensity distribution of network points, kiThe node strength of the node i is defined, and N is the number of all nodes in the causal network; the key variable time sequence is monitoring time sequence data generated by monitoring point positions in each group of key node sets.
Further, the step 6) of obtaining the optimal key node set determination ratio by using the accuracy and the effectiveness of the multi-index weighted synthesis index MCI comprehensive consideration evaluation result comprises the following steps:
(1) calculating the average absolute percentage error MAPE of each node set evaluation result X and the whole node evaluation result Y:
wherein N is the total number of samples of the evaluation result, xi,yiEvaluating results for each node set separatelyX and the ith evaluation result of the full node evaluation result Y.
(2) Normalizing the two curves respectively to eliminate the influence of numerical value difference on the similarity judgment result;
(3) the contour similarity CS of the normalized curve is calculated using the pearson correlation coefficient as follows:
wherein ,andrespectively averaging all samples of the partial node set evaluation result X and the full node set evaluation result Y;
(4) calculating the root mean square of the contour similarity CS and the absolute percentage error MAPE to obtain an evaluation index Era of the accuracy of the evaluation result:
(5) obtaining unit window evaluation time-consuming indexes Ute of each group of key nodes according to the state evaluation calculation time of each sliding window;
(7) the information entropies of the two evaluation indexes Era and Ute are calculated respectively:
wherein ,j represents the jth evaluation index, i represents the ith dimension characteristic of each evaluation index, and Yij=[era1,ute1;era2,ute2;…,…;eran,uten],EjThe information entropy for each evaluation index.
(8) Obtaining the information entropy E of each index according to the step (7)1 and E2And calculating the weight of each index:
(6) for a determined indicator Era era1,era2,…erakAnd Ute ═ ute1,ute2,…utekNormalizing the dimension data of the two indexes to obtain normalized indexes of Era 'and Ute' respectively;
(9) the MCI evaluates the time consumption index Ute 'and the corresponding weight W thereof through the evaluation result accuracy index Era' and the unit window1、W2And obtaining:
MCI=Era′·W1+Ute′·W2
furthermore, a key node set with the minimum MCI value is selected as a key node of the state perception network of the complex electromechanical system, and the determined proportion of the key node set can be used in the state evaluation and optimization research of energy chemical equipment with similar data characteristics.
The data-driven process industry state perception network key node screening method of the invention is described in detail with reference to the following specific embodiments:
example 1:
in this embodiment, monitoring data of 22 continuous monitoring variables in the TE process in a normal state and multiple fault states are selected for analysis, and the specific steps are as follows:
the method comprises the following steps: monitoring data analysis and preprocessing
And acquiring state monitoring data of the TE process equipment in a normal operation state and continuous monitoring data of various fault states to obtain state monitoring data of 22 variables shown in the table 1. As Gaussian noise exists in state monitoring data of the TE process, the dimension of each variable monitoring data is different, and in order to prevent the Gaussian noise and the dimension level difference from influencing the effectiveness of an analysis result, wavelet denoising and min-max standardization methods are respectively used for denoising and normalization before analysis.
Step two: multivariate causal relationship analysis and modeling
And analyzing the causal relationship among the variables after the monitoring data is preprocessed by using a Copula-Granger method to obtain a causal relationship matrix among the variables, and setting the autocorrelation causal relationship of each variable to be 0. And generating a Gaussian noise with the mean value of 0 and the variance of 1, and removing false causal relations among the variables by taking a value which is 1.5 times of the causal relation between the variables and the Gaussian noise as a threshold. And (3) constructing a causal relationship network model reflecting the information interaction mode among the variables by taking the monitoring variables as nodes and the causal relationship values among the variables as boundary weights.
Step three: importance evaluation of each node of causal network
And calculating the LR value of each node in the causal relationship network model by using a Leader Rank algorithm, and ranking the importance of each node according to the LR value. And acquiring a plurality of groups of key node sets in a certain proportion based on the ranking result.
Step four: state-aware network key node screening
And evaluating the multi-fault state of the system by using the obtained multiple groups of key node sets respectively, calculating indexes such as absolute percentage errors, contour similarity, unit window calculation time consumption and the like of the evaluation results of the system state of each group of key node sets and the evaluation results of the system state of all nodes, fusing the indexes by using an entropy weight method to obtain a multi-index weighted comprehensive index MCI, and obtaining the optimal key node determination proportion by comprehensively considering the accuracy and the timeliness of the evaluation results, thereby screening out the key node sets capable of meeting the system state evaluation requirements.
1 analysis and preprocessing of monitoring data
In order to verify the effectiveness of the method proposed herein, a public data set of the TE process is chosen for verification. State monitoring data of 22 variables of the TE process in a normal operation state and a plurality of fault operation states are simulated respectively, and the fault types comprise faults of step of process variables, random change of cooling water inlet temperature and feeding components, slow drift of reaction kinetics and the like. A detailed description of the 22 continuously monitored variables is shown in table 1:
TABLE 1 TE Process 22 continuous monitoring variable description
Since the TE process data contains gaussian noise and the dimensional levels of the variables are different, before analysis, denoising and normalization are performed by wavelet denoising and min-max normalization methods, respectively, and the data after the variable 1 monitoring data preprocessing is shown in fig. 2.
2 multivariate causal relationship analysis and modeling
The cause-and-effect relationship between the monitoring variables is calculated by using a Copula-Granger method, and no matter whether the cause-and-effect relationship exists between the two variables, the method can give out a cause-and-effect relationship value between the two variables, so that a false cause and effect exists in a result, and an information interaction mode between the monitoring variables of the system cannot be accurately reflected. For this purpose, the method introduces a gaussian noise with a mean value of 0 and a variance of 1, and determines the threshold value of causal relationship between each variable by calculating the causal relationship value between each monitored variable and the gaussian noise as shown in fig. 3. In order to avoid the identification of false causal relationship to the maximum extent, 1.5 times of the causal relationship between each monitoring variable and Gaussian noise is selected as a threshold value, and false causal relationship between the variables is eliminated.
The cause-and-effect relationship matrix obtained after the false cause-and-effect relationship and the cause-and-effect relationship of the variables are eliminated can effectively reflect the information interaction mode among the monitoring variables of the system. Taking each monitored variable as a node and a causal relationship value between each variable as an edge, a causal relationship network model reflecting an information interaction mode between each variable of the system is constructed, as shown in fig. 4, it can be seen intuitively in the network that nodes 7, 13, 16, 11, 18, 19 and the like are closely connected with other nodes in the network, and the connection strength between the links is higher than that of other nodes, which indicates that the nodes are in a more critical position in the network. It can be seen from table 1 that nodes 7, 13, 16, 11, 18, and 19 represent reactor pressure, separator pressure, stripper pressure, separator temperature, stripper temperature, and stripper flow, respectively, and according to the analysis of the relevant knowledge, these monitoring parameters are closely related to other parts in the system, and play an important role in ensuring the normal operation of the system, and enterprises usually use these variables as key variables for monitoring whether the system is in a normal state. Therefore, the real information interaction mode among the variables of the system can be effectively reflected by the causal relationship network model among the monitoring variables established by the method.
3 evaluation of importance of each node of causal network
The evaluation of the importance of the nodes of the complex network mainly ranks the nodes by extracting features which can reflect the interactive behaviors of the nodes in the network. Because the network constructed by the method is a directed weighting network, a Leader Rank algorithm is selected to evaluate the importance of each node, and each node in the network is ranked by calculating the LR value of each node in the causal network. The Leader Rank algorithm comprehensively considers the degree of entrance and the importance degree of each node in the network, and can accurately Rank the importance of each node in the network. In order to solve the problem of slow ranking of the traditional algorithm, a background node is added into the network by the Leader Rank algorithm, and the node is in bidirectional connection with all nodes in the network, so that the connectivity of the network is improved. And ranking each node in the causal network by using a Leader Rank algorithm, wherein the result is shown in FIG. 5.
According to the ranking results shown in fig. 5, factors such as the total number of nodes and the validity of the key node to the system state evaluation result are comprehensively considered, and the ranked nodes are divided into 8 key node sets according to the proportions of the top 20%, 30%, 40%, 50%, 60%, 70%, 80% and 90%, as shown in table 2.
4 state aware network key node screening
Acquiring state monitoring data of 22 variables in a normal running state and multiple fault states in the TE process, and performing noise reduction and normalization processing on the data by using wavelet noise reduction and min-max standardization methods respectively. And (3) carrying out window division on the historical monitoring data of each variable by using a sliding window method, wherein the width T of a window is 500, and the sliding step SST is 50, so as to obtain 278 groups of historical monitoring data samples.
The system state is evaluated by using the obtained multiple groups of key node sets and all nodes respectively, state monitoring data of all variables in each group of node sets under each sliding window are obtained, a cause-and-effect relationship value of time series data of each variable in each window is calculated by using a Copula-Granger method, a dynamic cause-and-effect network capable of reflecting an information interaction evolution mode among all variables in real time is obtained, network structure entropy characteristics of the cause-and-effect network in each window are extracted, and the network structure entropy characteristics are evaluation results of the system state, and are shown in fig. 6.
And respectively calculating the average absolute percentage error MAPE and the normalized curve contour similarity curve CS of each key node set and all the node evaluation results, and solving the root mean square of the two indexes to obtain an index Era capable of effectively representing the accuracy of the evaluation result, wherein the optimal key node set is required to meet the requirements of minimum MAPE and maximum CS value, and 1-CS is adopted for analysis in order to facilitate analysis. The unit window evaluation elapsed time indicator Ute is the time spent evaluating the unit window of each node set. And (3) calculating weights of Era and Ute by an entropy weight method, and obtaining a multi-index weighted comprehensive index MCI reflecting the validity degree of the evaluation result of each node set through weighted summation. The evaluation result accuracy evaluation index Era, the unit window evaluation time consumption index Ute and the multi-index weighted composite index MCI of each node set are shown in table 2.
Table 1 node numbers of node sets and evaluation indexes
The three indexes Era, Ute and MCI which reflect the effective degree of the evaluation result of each node set are normalized to obtain a variation trend graph of the three indexes shown in fig. 7 along with the expansion of the selection ratio, and it can be seen from the graph that along with the expansion of the selection ratio of the key node, the multi-index weighted comprehensive index MCI which can comprehensively reflect the evaluation result of the system state by each node set is firstly reduced and then increased, and reaches an extreme value when the selection ratio of the key node is 50%. As can be seen from fig. 8, the causal relationship of each node in the screened key node set is more significant compared with that of all node sets, which illustrates that most of the nodes having insignificant influence on the system state are screened by the method. The evaluation results of the system states of 50% of the key node sets and all the node sets are marked by using a system state marking method, as shown in fig. 9, 50% of the key sets can reflect all the states in the system operation process, and the evaluation time of a unit window is only 20% of the evaluation time of all the node sets, thereby further proving the effectiveness of the method provided by the invention.
In summary, compared with the traditional method for screening the key nodes of the complex system based on the process flow, the method can apply the screened key nodes to the field of state evaluation of the complex electromechanical system, and through verification, compared with all the nodes, the key nodes obtained by the method can accurately reflect the state of the system while effectively reducing the system state evaluation time, and have very important significance for developing state evaluation and optimization research of the complex electromechanical system of the data-driven process industry.
Claims (10)
1. The data-driven process industry state perception network key node screening method is characterized by comprising the following steps:
step 1), state monitoring time sequence data of a plurality of monitoring variables of a complex electromechanical system are obtained, and noise reduction and normalization processing are carried out on the state monitoring time sequence data by adopting a wavelet noise reduction and min-max standardization method in sequence;
step 2), analyzing the causal relationship among the state monitoring time sequence data subjected to noise reduction and normalization in the step 1) by using a Copula-Granger method to obtain a causal relationship matrix among variables, and setting the autocorrelation coefficients of the variables to 0;
step 3), generating a Gaussian noise with the mean value of 0 and the variance of 1, analyzing the causal relationship between the monitoring time sequence data of each state and the Gaussian noise after noise reduction and normalization in the step 1) by using a Copula-Granger method, and simplifying the causal relationship matrix obtained in the step 2) by using a 1.5-time value of the causal relationship between each monitoring variable and the Gaussian noise as a threshold value;
step 4), constructing a causal relationship network model reflecting an information interaction mode among the monitoring variables by taking the monitoring variables as nodes and the causal relationship values among the monitoring variables as side weights;
step 5), calculating the LR value of each node of the causal relationship network model obtained in the step 4) by using a Leader Rank algorithm, ranking each node, dividing a node set according to a ranking result and a preset proportion difference, and acquiring a plurality of groups of key node sets;
and 6) evaluating the state of the complex electromechanical system by using the obtained multiple groups of key node sets respectively, calculating absolute percentage errors MAPE and contour similarity CS of the evaluation results of the key node sets on the system state and the evaluation results of all nodes on the system state, and comprehensively considering the multi-index weighted comprehensive index MCI and the number of the key nodes to obtain the optimal key node proportion, so as to screen out the key nodes meeting the evaluation requirements of the system state.
2. The data-driven state-aware network key node screening method according to claim 1, wherein the step 1) of normalizing the data by the min-max normalization method is performed by performing normalization on each variable time series x1,x2,…xnThe following transformations are performed:
obtain a time series y1,y2,…ynI.e. the time series after normalization, where yn∈[0,1]。
3. The data-driven process industry state awareness network key node screening method according to claim 1, wherein the step 2) of analyzing and monitoring causal relationships among variables after data preprocessing by using a Copula-Granger method comprises the following steps:
(1) obtaining the shape of variable X and variable YState monitoring dataAnd wherein m and n respectively represent the delay dimensions of the variable X and the variable Y;
(2) calculating the edge distribution of the variables X and Y by a kernel function estimation method according to the monitoring data of the variables X and YAnd
(3) based on the edge distribution calculated in the step (2), estimating an empirical Copula density function by adopting a rank statistic methodThe Copula density function estimation process for the continuous binary joint distribution F (x, y) is as follows:
wherein Representing K samples, x, subject to a continuous binary joint distribution F (x, y)(i) and y(j)Representing the order statistics in the respective unary samples, 1 ≦ i, j ≦ K;
(4) applying a Bernstein approximation method to the optimal estimation of the Copula density function, and optimizing the Copula density function estimation in the step (3), wherein the expression of the Bernstein polynomial approximation is as follows:
wherein ,uk=F(xk,yk),u1,…,ukFor a k-dimensional uniform random variable obtained by sample calculation,mjrepresents the size of Copula probability density function, 0 ≦ vj≤mj;
(5) Taking logarithm of the Copula density function obtained through the optimal estimation in the step (4), calculating expectation of a sample, and obtaining a causal relationship among monitoring variables:
4. the data-driven screening method for key nodes of the process industry state-aware network of claim 1, wherein the simplified causal relationship matrix in step 3) is an n × n dimensional matrix with all 0 diagonal lines, as follows:
in the formula, each element cijRepresenting a causal relationship value between variable i and variable j.
5. The data-driven process industry state aware network key node screening method of claim 1, wherein the causal network model established in step 4) is a directional weighting network with a direction according to causal and causal networkDetermination of relationships, cijRepresenting a directed connection from variable i to variable j.
6. The data-driven process industry state awareness network key node screening method according to claim 1, wherein a Leader Rank algorithm is used to calculate LR values of variables in a causal relationship network model in step 5), and the specific steps are as follows:
(1) adding a background node G into the causal relationship network model obtained in the step 4);
(2) calculating a Google relation matrix A of each node according to the causal relation network model obtained in the step 4), wherein:
(3) assigning the initial importance of each node in the causal relationship network model to be 1;
(4) calculating LR values of all nodes in the causal relationship network model, and iterating until the algorithm converges, wherein the LR value calculation formula is as follows:
wherein j is the number of variables in the causal network,weighting out degree of variable j, and t is iteration frequency;
(5) bisecting the LR value of the background node, adding the halved LR value into the LR value after convergence of each node, and obtaining the global LR value of each node:
(6) and ranking the nodes according to the obtained global LR values in the descending order.
7. The data-driven process industry state awareness network key node screening method according to claim 1, wherein the step of evaluating the state of the complex electromechanical system by using the key node set in step 6) is as follows:
(1) extracting historical monitoring data of the complex electromechanical system in a non-steady operation state within a period of time, and evaluating the system state by using a sliding window method;
(2) calculating the causal relationship of the time sequence of the key variables in each window by using a Copula-Granger method to obtain a dynamic network reflecting a key variable information interactive evolution model in real time, extracting the network structure entropy characteristics of the causal network of each window, and evaluating the system state to obtain an evolution curve of the system state, wherein the calculation formula of the network structure entropy is as follows:
in the formula ,P(ki) As intensity distribution of network points, kiThe point strength of the node i and the number of the network nodes N are shown;
(3) and evaluating the importance of each node set by adopting a multi-index weighted composite index MCI, wherein the node set with the lowest MCI index is a key node set capable of effectively representing the system state.
8. The data-driven process industry state perception network key node screening method of claim 7, wherein the MCI is obtained by weighted summation of a characterization evaluation result accuracy evaluation index Era and a characterization evaluation result timeliness evaluation index Ute, and each index weight is obtained by an entropy weight method, comprising the following steps:
(1) the information entropies of the two evaluation indexes Era and Ute are calculated respectively:
wherein ,j represents the jth evaluation index, i represents the ith dimension characteristic of each evaluation index, and Yij=[era1,ute1;era2,ute2;…,…;eran,uten],EjInformation entropy for each evaluation index;
(2) information entropy E of evaluation index Era obtained in (1)1And information entropy E of evaluation index Ute2And calculating the weight of each evaluation index:
(3) for a determined indicator Era era1,era2,…eranAnd Ute ═ ute1,ute2,…utenNormalizing the dimension data of the two indexes to obtain normalized indexes of Era 'and Ute' respectively;
(4) the MCI evaluates the time consumption index Ute 'and the corresponding weight W thereof through the evaluation result accuracy index Era' and the unit window1、W2To obtain the product of:
MCI=Era′·W1+Ute′W2。
9. the data-driven process industry state aware network key node screening method of claim 8, wherein the evaluation index Ute is a time required to calculate an evaluation result for each sliding window.
10. The data-driven process industry state perception network key node screening method of claim 8, wherein Era is determined by the profile similarity CS and the absolute percentage error MAPE of each node set evaluation result X and the full node evaluation result Y, including the steps of:
(1) calculating the average absolute percentage error MAPE of each node set evaluation result X and the whole node evaluation result Y:
wherein N is the total number of samples of the evaluation result, xi,yiRespectively obtaining evaluation results under the ith sliding window of each node set evaluation result X and the full node evaluation result Y;
(2) normalizing the two curves respectively to eliminate the influence of numerical value difference on the similarity judgment result;
(3) the contour similarity CS of the normalized curve is calculated using the pearson correlation coefficient as follows:
wherein ,andrespectively averaging all samples of the partial node set evaluation result X and the full node set evaluation result Y;
(4) calculating the root mean square of the dissimilarity degree 1-CS and the absolute percentage error MAPE to obtain an accuracy evaluation index Era of the characterization evaluation result:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010774423.4A CN112116198B (en) | 2020-08-04 | 2020-08-04 | Data-driven process industrial state perception network key node screening method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010774423.4A CN112116198B (en) | 2020-08-04 | 2020-08-04 | Data-driven process industrial state perception network key node screening method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112116198A true CN112116198A (en) | 2020-12-22 |
CN112116198B CN112116198B (en) | 2023-06-20 |
Family
ID=73799205
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010774423.4A Active CN112116198B (en) | 2020-08-04 | 2020-08-04 | Data-driven process industrial state perception network key node screening method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112116198B (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112836936A (en) * | 2021-01-08 | 2021-05-25 | 国家电网有限公司 | Risk factor screening method and system for power sensing network perception layer planning |
CN112906787A (en) * | 2021-02-10 | 2021-06-04 | 宁波诺丁汉新材料研究院有限公司 | Industrial boiler fault identification method and system |
CN113344759A (en) * | 2021-06-30 | 2021-09-03 | 北京建筑大学 | Analysis method for mobile source pollution emission |
CN114826936A (en) * | 2022-04-14 | 2022-07-29 | 中国人民解放军国防科技大学 | Method and system for determining key node set of weighted network communication efficiency |
CN116719293A (en) * | 2023-08-04 | 2023-09-08 | 江苏甬金金属科技有限公司 | Production monitoring method and system for cold-rolled steel strip |
CN117078043A (en) * | 2023-08-22 | 2023-11-17 | 华能澜沧江水电股份有限公司 | State evaluation method for computer monitoring system of hydropower plant |
CN117078043B (en) * | 2023-08-22 | 2024-04-30 | 华能澜沧江水电股份有限公司 | State evaluation method for computer monitoring system of hydropower plant |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101446827A (en) * | 2008-11-06 | 2009-06-03 | 西安交通大学 | Process fault analysis device of process industry system and method therefor |
US20120078678A1 (en) * | 2010-09-23 | 2012-03-29 | Infosys Technologies Limited | Method and system for estimation and analysis of operational parameters in workflow processes |
CN105117849A (en) * | 2015-08-28 | 2015-12-02 | 湖南大学 | Electrical LeaderRank algorithm based power network node importance evaluation method |
CN108228586A (en) * | 2016-12-12 | 2018-06-29 | 上海看榜信息科技有限公司 | A kind of cross-platform network leader of opinion ranks method |
CN108763729A (en) * | 2018-05-24 | 2018-11-06 | 西安交通大学 | Process industry Mechatronic Systems couple state appraisal procedure based on network structure entropy |
CN111008363A (en) * | 2019-11-21 | 2020-04-14 | 西安交通大学 | Multivariable causal-driven complex electromechanical system service safety situation evaluation method |
-
2020
- 2020-08-04 CN CN202010774423.4A patent/CN112116198B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101446827A (en) * | 2008-11-06 | 2009-06-03 | 西安交通大学 | Process fault analysis device of process industry system and method therefor |
US20120078678A1 (en) * | 2010-09-23 | 2012-03-29 | Infosys Technologies Limited | Method and system for estimation and analysis of operational parameters in workflow processes |
CN105117849A (en) * | 2015-08-28 | 2015-12-02 | 湖南大学 | Electrical LeaderRank algorithm based power network node importance evaluation method |
CN108228586A (en) * | 2016-12-12 | 2018-06-29 | 上海看榜信息科技有限公司 | A kind of cross-platform network leader of opinion ranks method |
CN108763729A (en) * | 2018-05-24 | 2018-11-06 | 西安交通大学 | Process industry Mechatronic Systems couple state appraisal procedure based on network structure entropy |
CN111008363A (en) * | 2019-11-21 | 2020-04-14 | 西安交通大学 | Multivariable causal-driven complex electromechanical system service safety situation evaluation method |
Non-Patent Citations (1)
Title |
---|
尹璐;李斌;陈立章;王晓莹;: "基于因果关系的网络安全事件传导路径分析方法", 信息网络安全, no. 05 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112836936A (en) * | 2021-01-08 | 2021-05-25 | 国家电网有限公司 | Risk factor screening method and system for power sensing network perception layer planning |
CN112836936B (en) * | 2021-01-08 | 2023-11-24 | 国家电网有限公司 | Risk factor screening method and system for power sensing network perception layer planning |
CN112906787A (en) * | 2021-02-10 | 2021-06-04 | 宁波诺丁汉新材料研究院有限公司 | Industrial boiler fault identification method and system |
CN112906787B (en) * | 2021-02-10 | 2023-09-15 | 宁波诺丁汉新材料研究院有限公司 | Industrial boiler fault identification method and system |
CN113344759A (en) * | 2021-06-30 | 2021-09-03 | 北京建筑大学 | Analysis method for mobile source pollution emission |
CN114826936A (en) * | 2022-04-14 | 2022-07-29 | 中国人民解放军国防科技大学 | Method and system for determining key node set of weighted network communication efficiency |
CN114826936B (en) * | 2022-04-14 | 2023-06-09 | 中国人民解放军国防科技大学 | Method and system for determining key node set of weighted network communication efficiency |
CN116719293A (en) * | 2023-08-04 | 2023-09-08 | 江苏甬金金属科技有限公司 | Production monitoring method and system for cold-rolled steel strip |
CN116719293B (en) * | 2023-08-04 | 2023-11-17 | 江苏甬金金属科技有限公司 | Production monitoring method and system for cold-rolled steel strip |
CN117078043A (en) * | 2023-08-22 | 2023-11-17 | 华能澜沧江水电股份有限公司 | State evaluation method for computer monitoring system of hydropower plant |
CN117078043B (en) * | 2023-08-22 | 2024-04-30 | 华能澜沧江水电股份有限公司 | State evaluation method for computer monitoring system of hydropower plant |
Also Published As
Publication number | Publication date |
---|---|
CN112116198B (en) | 2023-06-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112116198B (en) | Data-driven process industrial state perception network key node screening method | |
CN111222290B (en) | Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment | |
CN111062508B (en) | Method for evaluating real-time running state of wind turbine generator based on big data technology | |
US11650968B2 (en) | Systems and methods for predictive early stopping in neural network training | |
CN108399434B (en) | Analysis and prediction method of high-dimensional time series data based on feature extraction | |
CN110757510B (en) | Method and system for predicting remaining life of robot | |
CN113240201B (en) | Method for predicting ship host power based on GMM-DNN hybrid model | |
CN113065702A (en) | Landslide displacement multi-linear prediction method based on ST-SEEP segmentation method and space-time ARMA model | |
CN114692758A (en) | Power communication fault analysis method and device, terminal equipment and medium | |
CN114861788A (en) | Load abnormity detection method and system based on DBSCAN clustering | |
CN112070322A (en) | High-voltage cable line running state prediction method based on long-short term memory network | |
CN115982141A (en) | Characteristic optimization method for time series data prediction | |
CN114548493A (en) | Method and system for predicting current overload of electric energy meter | |
CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
CN110244690B (en) | Multivariable industrial process fault identification method and system | |
CN112380763A (en) | System and method for analyzing reliability of in-pile component based on data mining | |
CN115718880A (en) | Prediction method for degradation stage of complex equipment | |
CN112329535B (en) | CNN-based quick identification method for low-frequency oscillation modal characteristics of power system | |
CN113821401A (en) | WT-GA-GRU model-based cloud server fault diagnosis method | |
CN113878613B (en) | Industrial robot harmonic reducer early fault detection method based on WLCTD and OMA-VMD | |
CN110750848A (en) | Method for estimating remaining life of software-hardware degradation system by considering software operation | |
CN117647725B (en) | Aging test method and system for PCBA | |
CN116780524B (en) | Industrial enterprise short-term load prediction method based on LSTM deep learning | |
CN117193184A (en) | VAE and deep neural network mixed intelligent manufacturing factory process quality monitoring method | |
CN117009752A (en) | Power consumption demand prediction method based on POA-VMD-SSA-KELM model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |