Method for evaluating goodness of real-time comprehensive working condition in cement calcining process

Abstract

The invention discloses a real-time comprehensive working condition superiority evaluation method in a cement calcination process. The method comprises the following steps: step 1, carrying out working condition category grading on a historical operation data set in a cement production process; step 2, screening process variables with high relevancy with the key indexes to train a CVAE network to obtain a pre-trained CVAE network; 3, calculating four groups of reconstruction error control limits by adopting a kernel density estimation (KDE) method; 4, constructing an SFA network, obtaining a chronic feature mapping matrix, and calculating a statistic control limit S2; 5, fusing dual-channel monitoring results to obtain a working condition comprehensive scoring index CSI and a working condition dynamic index DI; step 6, monitoring current operation data, inputting the optimality evaluation model, calculating corresponding statistics and indexes, and outputting a current monitoring state and a real-time score; according to the invention, multi-index comprehensive optimization evaluation and visual monitoring of the quality, the yield, the energy consumption and the working condition stability of the cement calcining operation working condition can be realized.

Description

Technical Field

[0001] The invention relates to an industrial process operating condition optimization evaluation method, in particular to a cement calcination process operating condition optimization evaluation method, which is used for real-time monitoring and evaluation of the current operating state. Background Art

[0002] During the cement calcination process, due to the fluctuation of raw materials, variable operating conditions and unknown disturbances, it often exhibits non-stationary characteristics. Under non-stationary conditions, the operating conditions change frequently, energy consumption is large, and it is more likely to cause abnormal product quality and even production safety accidents. However, we found that we cannot only focus on the occurrence of unsafe operating conditions (faults) during cement calcination. The key to avoiding this problem is to reduce abnormal operating conditions and non-optimal operating conditions. Therefore, it is very important and meaningful to carry out real-time comprehensive operating condition evaluation of the cement calcination process.

[0003] In order to reduce the occurrence of failures during cement calcination, more attention should be paid to non-optimal working conditions. Therefore, if we want to achieve "safe, stable, high-quality, and long-term" operation of the cement calcination process, it is extremely important to conduct real-time working condition optimization evaluation of the production process. However, the evaluation indicators of the existing working condition evaluation methods are relatively single, and it is difficult to fully reflect the working condition of the working condition. In addition, the dynamic non-stationary characteristics of actual cement calcination have brought great difficulties to the quantitative evaluation of the current working condition. Therefore, there is an urgent need for an effective solution to effectively mine the knowledge in the historical production process data and combine the experience of advanced experts to carry out multi-indicator all-round working condition monitoring and evaluation of the cement calcination process with quality, output, energy consumption and working condition stability.

[0004] At present, there are few studies on the evaluation of the working condition in the cement calcination process. The evaluation indicators of the working condition evaluation in related industrial cases are relatively single, lacking comprehensive consideration of multiple indicators of the working condition, and most methods are to classify the working condition, that is, qualitative evaluation, while there are few studies on real-time quantitative analysis of the working condition. In addition, the cement calcination process needs to pay attention to multiple monitoring indicators such as quality, output, energy consumption, etc., and the non-stable working condition brings greater difficulties to the real-time operation evaluation.

[0005] To this end, the present invention aims to address the problem that the evaluation indicators of existing working condition evaluation-related research are relatively single and fail to fully reflect the current working condition status. During the cement calcination process, the energy consumption, output, quality and working condition stability factors of the production process are comprehensively considered to construct a dual-channel multi-indicator working condition comprehensive excellence evaluation scheme based on CVAE-SFA. Summary of the invention

[0006] In order to solve the above technical problems, the present invention proposes a real-time comprehensive working condition evaluation method for cement calcination process. Aiming at the problem that the traditional process monitoring method has a single control limit and lacks abnormal degree classification, a superiority level evaluation channel for reconstructing the optimal working condition features of the conditional variational autoencoder (CVAE) is constructed, and the "excellent, medium and poor" four control limits are designed based on the kernel density estimation (KDE) to realize the online level evaluation of the working condition. Aiming at the problem that the stable assumption monitoring method lacks dynamic characteristic monitoring of non-stationary working conditions, a slow feature analysis (SFA) is constructed. 2 The statistical dynamic characteristic abnormality monitoring channel realizes dynamic abnormality monitoring of the working condition by mining the chronic characteristics that characterize the slow changes inside the system. Finally, the dual-channel monitoring results are fused on a time scale, and the comprehensive scoring index (CSI) of the working condition is designed to realize the comprehensive evaluation and visual monitoring of the quality, output, energy consumption and working condition stability of the current operating condition.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A method for evaluating the performance of a cement calcining process in a comprehensive real-time working condition comprises the following steps:

[0009] Step 1: Classify the operating conditions of the historical operating data set in the cement production process into different categories:

[0010] Analyze the cement production process, obtain the historical operation data of the calcination system, and divide the historical operation data set into four working condition categories of "excellent, good, medium, and poor" based on the requirements of the cement calcination process quality indicators and energy consumption indicators in the national standards according to the working condition classification rules; among them, the working condition classification rules are based on comprehensive evaluation of four key indicators: quality, output, coal consumption, and electricity consumption. The quality indicator is "f-CaO content", the key variable of the output indicator is "feeding amount", the coal consumption indicator is "standard coal consumption per ton of clinker", and the electricity consumption indicator is "unit electricity consumption of the kiln system";

[0011] Step 2: Select process variables with high correlation with key indicators to train the CVAE network and obtain a pre-trained CVAE network:

[0012] The MIC method was used to calculate the correlation coefficients between 46 process variables and key indicators, and the correlation coefficients were summed and arranged in descending order. The first 20 process variables were used to train the CVAE network. The process variable operating condition data set that was screened under the "excellent" state in step 1 was selected, and the CVAE network was trained based on the minimum reconstruction error index to construct a condition division channel based on the reconstruction of CVAE key indicators. The training goal of the CVAE network was to adjust the current network structure parameters so that the mean square error (MSE) between the input monitoring key indicators in the cement calcination process and the reconstruction of the CVAE network key indicator variables was minimized, and a pre-trained CVAE network was obtained.

[0013] Among them, the key indicators monitored during cement calcination include "feeding amount", "standard coal consumption per ton of clinker", "unit power consumption of kiln system" and "f-CaO content";

[0014] Step 3: Use the kernel density estimation (KDE) method to calculate four sets of reconstruction error control limits:

[0015] The four operating conditions of "excellent, good, medium, and poor" obtained in step 1 and the data set of the screened process variables are input into the pre-trained CVAE network in step 2 to obtain the reconstruction errors under four different operating conditions. The kernel density estimation KDE method is used to calculate the reconstruction error control limits Lim under the four operating conditions with a confidence level of α. Optimal ,Lim good ,Lim general and Lim Poor ;

[0016] Step 4: construct the SFA network, obtain the chronic feature mapping matrix, and calculate S 2 Statistical control limits:

[0017] The Johansen test is used to obtain historical stable data to train the SFA network. Through two-step continuous singular value decomposition SVD, the slow feature mapping matrix is ​​solved to obtain the slow features that characterize the inherent characteristics of the system. S is calculated according to the confidence α. 2 Statistical control limits;

[0018] Step 5: Combine the dual-channel monitoring results to obtain the working condition comprehensive scoring index CSI and working condition dynamic index DI:

[0019] In order to simplify the multi-statistical index working condition analysis process and facilitate engineering application, a quantitative working condition comprehensive scoring index CSI is proposed by integrating the dual-channel monitoring results. It mainly includes three parts of information: the first is the reconstruction error of the CVAE working condition superiority level evaluation channel as the CSI benchmark BS; the second is the addition of the reconstruction error first-order difference index CMI to indicate the working condition trend; the third is the relative change RV of the SFA dynamic monitoring statistic, and the working condition dynamic index DI is defined by CMI and RV to indicate the degree of working condition abnormality;

[0020] Step 6: Monitor the current operating data, input the superiority evaluation model, calculate the corresponding statistics and indicators, and output the current monitoring status and real-time score.

[0021] 4. A further improvement of the technical solution of the present invention is that the working condition classification standard in step 1 is:

[0022] When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and the standard coal consumption per ton of clinker is ≤94kgce / t and the unit power consumption of the kiln system is ≤48kW·h / t, it is rated as "excellent";

[0023] When the key quality index of cement clinker f-CaO≤1.5 and the feed amount is greater than or equal to the average value, and the comprehensive coal consumption per unit product of clinker and the comprehensive electricity consumption per unit product of clinker are different and are greater than the third level, it is rated as "good". If the key quality index of cement clinker f-CaO≤1.5 and the feed amount is less than the average value, the comprehensive coal consumption per unit product of clinker is ≤94kgce / t and the comprehensive electricity consumption per unit product of clinker is ≤48kW·h / t, it is also rated as "good".

[0024] When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and the comprehensive coal consumption per unit product of clinker is 100kgce / t<109kgce / t and the comprehensive electricity consumption per unit product of clinker is <57kW·h / t≤61kW·h / t, it is rated as "medium". When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and the comprehensive coal consumption per unit product of clinker is ≤100kgce / t and the comprehensive electricity consumption per unit product of clinker is >6 1kW·h / t is rated as “medium”; when the key quality indicator of cement clinker f-CaO≤1.5 and the feed amount is greater than or equal to the average value and the comprehensive coal consumption per unit product of clinker>109kgce / t and the comprehensive electricity consumption per unit product of clinker≤57kW·h / t, it is also rated as “medium”; when the key quality indicator of cement clinker f-CaO≤1.5 and the feed amount is less than the average value and the comprehensive coal consumption per unit product of clinker and the comprehensive electricity consumption per unit product of clinker are different and are greater than the third level, it is rated as “medium”;

[0025] Other cases were classified as “poor”.

[0026] 5. A further improvement of the technical solution of the present invention is that the process of selecting process variables with high correlation with key indicators and training the CVAE network in step 2 is:

[0027] The MIC method was used to calculate the maximum correlation coefficients of all process variables with f-CaO content, standard coal consumption per ton of clinker, feed rate and unit power consumption of the kiln system;

[0028] Among them, MIC aims at the relationship between two variables. It uses a scatter plot to divide the current two-dimensional space into a certain number of intervals in the x and y directions in the two-dimensional space, and calculates the joint probability by checking where the current scattered points fall in each square. Let I(x, y) be the mutual information of the two variables, then MIC is expressed as:

[0029]

[0030] In the formula, a and b are the number of grids divided in the x and y directions, and B is the variable size. The sum of the four correlation coefficients is arranged in descending order, and the first 20 process variables with high correlation with key production indicators are selected to train the CVAE network;

[0031] When using CVAE for data reconstruction, from the distribution The training objective of CVAE is to maximize the conditional log-likelihood:

[0032]

[0033] Since the posterior distribution is the true distribution p θ (z|x i ,c i ) is a variational approximation; when the posterior distribution is very close to the true distribution, the KL divergence approaches 0; therefore, the objective function of CVAE is defined as:

[0034] L(θ,φ;x i ,c i ) = lgp θ (x i |c i )-KL[q φ (z|x i ,c i )||p θ (z|x i ,c i )]

[0035] Assume that the input x after variable screening is reconstructed by the CVAE network as r, and the error size of the reconstruction is calculated by MSE:

[0036]

[0037] The mean square error (MSE) of the reconstruction error of the key indicator variable of the training CVAE network is minimized.

[0038] 5. A further improvement of the technical solution of the present invention is that: the control limit in step 3 is calculated using the KDE method, wherein the input data {x1, x2…x n} is taken from a continuous distribution, then the KDE function of the overall probability density function f(x) at any x is The formula is as follows:

[0039]

[0040] Where x is the sample point of the data set, h is the bandwidth, and K is the kernel function. The Gaussian kernel function is used for KDE calculation, and the formula is updated as follows:

[0041]

[0042] The control limits are calculated using the reconstruction errors of the excellent, good, medium, and poor data sets. Given a confidence level α, the control limits can be obtained from the probability density function of the reconstruction error under the corresponding mode, that is:

[0043]

[0044] Among them, Lim RE,mode Indicates the reconstruction error under excellent, good, medium and poor working conditions, and the corresponding control limit is calculated as Lim Optimal ,Lim good ,Lim general and Lim Poor .

[0045] A further improvement of the technical solution of the present invention is that: according to the method for evaluating the real-time comprehensive working condition of cement calcining process according to claim 1, it is characterized in that: the specific operation of obtaining the steady process training data by the Johansen test in step 4 is:

[0046] Assume a matrix containing multiple time series:

[0047]

[0048] Where m is the sample size, n is the total process monitoring variables, and the input data Y is substituted into the vector autoregression model (VAR):

[0049] ΔY t =α+ΠY t-1 +ε t

[0050] Among them, ΔY t Yest The first-order difference of , α is a constant term, Π is a coefficient matrix, ε t is the error term;

[0051] Calculate Y t-1 , forming a lag matrix;

[0052] Extract the Π matrix from the model and calculate its eigenvalue λ i , by solving the following characteristic equation:

[0053] det(Π-λI)=0

[0054] Calculate statistics:

[0055]

[0056] And the largest eigenvalue statistic:

[0057]

[0058] Finally, by comparing the calculated trace statistics and the maximum eigenvalue statistics with the critical value, if LR c and LR max They are all greater than the critical value, indicating that the linear combination of time series is stationary, so as to obtain the training data of the stationary process of historical data.

[0059] A further improvement of the technical solution of the present invention is that the specific operation of training the SFA network according to the historical stable data in step 4 is:

[0060] Assume the input signal x(t)∈R m , m slow features are uniformly expressed as s = Wx, where W = [w1,…,w m ] is the model parameter matrix to be determined;

[0061] If each dimension of the input variable already has zero mean, then the optimization problem of the above linear SFA model is equivalent to the generalized eigenvalue decomposition problem:

[0062] AW=BWΩ

[0063] in, is the first-order difference covariance matrix of the input x(t), B = <xx T > is the covariance matrix of the input x(t); W = [w1,…,w m ] contains m eigenvectors, and Ω=diag{λ1,…,λ m The generalized eigenvalues ​​in} are arranged from small to large, corresponding one to one with the vectors in the W matrix, that is, the slowest latent variable is The slowness is And satisfy W ΤAW=Ω,W Τ BW = I. ;

[0064] The specific process of solving the slow feature map matrix through two consecutive singular value decompositions is:

[0065] The first step of SVD decomposition is sphering of the input data x(t) to eliminate the correlation between variables. Assume that the SVD of matrix B is as follows:

[0066] B=UΛU T

[0067] The whitened data is expressed as:

[0068] z=Λ -1 / 2 U T x

[0069] Where Q = Λ -1 / 2 U T It is called the whitening matrix; solving the matrix W on this basis is equivalent to solving the matrix P = WQ -1 This is because

[0070] s=Wx=WQ -1 z=Pz

[0071] Use the covariance of s for unified representation: <ss T > t =I

[0072] get:

[0073] <ss T > t =P <zz T > t P T =PP T =I

[0074] It implies that P is an orthogonal matrix; thus, the optimization problem of linear SFA is transformed into finding an orthogonal matrix P such that Minimum, through The covariance of is decomposed by SVD to obtain:

[0075]

[0076] Matrix Ω=diag{ω1,…,ω m(d+1) The diagonal elements in} are the optimal target values ​​of the optimization target, that is, the Δ(·) values ​​of each slow feature:

[0077]

[0078] Then construct S 2Statistics, where the choice of the number of slow features is important for the calculation of control limits, discarding input {x j The number of features that change faster is given by:

[0079] M e =card{si|Δ(s i )>max j {Δ(x j )}}

[0080] Where card{·} represents the number of elements in the set, and the number of slow features retained is calculated as:

[0081] M=m(d+1)-M e

[0082] The M slow features of the main changes in the system are expressed as The abandoned M e The slow feature is represented by

[0083] For the dynamic characteristics monitoring of the slow-changing process To design statistics, define S 2 Statistics:

[0084]

[0085] in express The covariance matrix of ; According to the statistical properties of slow features, it is assumed that If it obeys Gaussian distribution, then S 2 The statistic follows the F distribution after proportion transformation:

[0086] S 2 ~gF M,N-M-1

[0087] The parameter g is defined as:

[0088]

[0089] When the significance level α is given, S 2 The control limits of the monitoring statistics can be determined.

[0090] A further improvement of the technical solution of the present invention is that the specific operation of step 5 is:

[0091] First, the CVAE reconstruction error is combined with the control limit in step 4 for piecewise linear mapping; to avoid the overall working condition evaluation score being low when the reconstruction error is mapped from 0-Lim difference to 1-0, the reconstruction error is mapped from 0-Lim difference to 1-α as the working condition basic score BS;

[0092] Next, the first-order difference of the reconstruction error is used to characterize the operating condition trend, which is defined as the operating condition trend indicator CMI; the first-order difference of the reconstruction error E at time t is ΔE t =E t -E t-1 Expressed; CMI is expressed as follows:

[0093]

[0094] In the above formula, ΔE t If it is greater than 0, it means that the reconstruction errors of n+1 consecutive times continue to increase, that is, the working condition has a trend of deterioration, which is indicated by CMI=-1. Similarly, the working condition has a trend of improvement, which is indicated by CMI=1. CMI=0 means that the current working condition has no clear change trend, and further judgment needs to be made in combination with the dynamic information of the working condition.

[0095] Finally, based on S 2 The statistical index calculates the relative change RV of the overall working condition relative to the stable process:

[0096]

[0097] RV belongs to the real number set R. According to the calculation rule RV≥-1, the upper limit T of the maximum change of RV is set. T is selected in combination with the fluctuation of specific working conditions.

[0098] CMI×RV is expressed as the working condition dynamic index DI:

[0099]

[0100] The value range of DI is all real numbers from -T to T. If |DI|=T, it means that the operating conditions may have changed significantly beyond the acceptable range of the controller, and an alarm will be issued.

[0101] The comprehensive evaluation index CSI of the working condition is given by the following formula:

[0102] CSI=BS+γ×DI

[0103] γ represents the weight coefficient of the operating condition variation index in the comprehensive evaluation index of the operating condition, γ≤2.

[0104] Due to the adoption of the above technical solution, the technical progress achieved by the present invention is:

[0105] In view of the problem that the evaluation indicators of existing working condition evaluation related research are relatively single and fail to fully reflect the current working condition status, the energy consumption, output, quality and working condition stability factors in the cement calcination process are comprehensively considered, and a dual-channel multi-indicator working condition comprehensive superiority evaluation method based on CVAE reconstruction and SFA dynamic characteristics abnormality monitoring is constructed.

[0106] In view of the fact that most of the relevant working condition evaluation research is based on qualitative classification but lacks quantitative evaluation, the present invention calculates the reconstruction errors of four types of superiority data based on the CVAE method, designs four control limits in combination with the KDE method, and proposes corresponding control limit division evaluation indicators, which provides a new idea for the qualitative and quantitative synchronous monitoring of the current working conditions.

[0107] The present invention aims to solve the problem of insufficient information mining of dynamic process of non-stationary working condition by constructing a slow feature analysis (SFA) 2 The statistical dynamic characteristic abnormality monitoring channel realizes dynamic abnormality monitoring of the operating condition by mining the chronic features that characterize the slow changes within the system; and fuses the dual-channel monitoring results on a time scale, proposes comprehensive evaluation indicators and dynamic indicators that indicate abnormal conditions of the operating condition, and realizes a multi-angle comprehensive excellence evaluation of the current operating condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 It is the process and monitoring variable diagram of cement calcination process of the present invention;

[0109] Figure 2 The overall technical roadmap for the implementation of the present invention;

[0110] Figure 3 It is a reference standard for evaluating the working condition excellence of the cement calcination process implemented in the present invention;

[0111] Figure 4 A heat map of the correlation between process variables screening based on MIC implemented in the present invention;

[0112] Figure 5 A diagram of the CVAE network structure implemented in the present invention;

[0113] Figure 6 A flow chart of comprehensive performance evaluation of real-time working conditions of cement calcining process implemented by the present invention;

[0114] Figure 7 This is a visualization example of the experimental results of the cement calcination process implemented in the present invention. DETAILED DESCRIPTION

[0115] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0116] The embodiment of the present invention provides a method for evaluating the real-time comprehensive working condition performance of cement calcining process based on key indicator reconstruction and dynamic characteristic monitoring, and the technical roadmap of the method is as follows: Figure 2 As shown, the following steps are included:

[0117] Step 1: Through the analysis of cement production process, such as Figure 1 This is the process and monitoring variable diagram of cement calcination process of the present invention. It can be seen from the figure that it includes 46 process variables. Based on the requirements of national standards for quality indicators and energy consumption indicators of cement calcination process, combined with expert experience and knowledge, the historical operation data is divided into working condition superiority grades. The reference criteria for working condition division are as follows Figure 3 shown.

[0118] In step 1, the production process of cement clinker is first analyzed. According to the quality requirements in GB / T 21372-2024 of the Chinese national standard "Portland Cement Clinker" and the energy consumption requirements in GB16780-2021 of "Energy Consumption Limits per Unit of Cement Product", as well as the experience and knowledge of relevant experts, the historical operating condition evaluation rules of the cement calcination process are summarized. The historical data set is divided into four categories of "excellent, good, medium, and poor" as a reference for the data set selection for model training. The operating condition classification rules are comprehensively evaluated based on four indicators: quality, output, coal consumption, and electricity consumption. Figure 3 In the schematic diagram on the right, the four small boxes are from right to left: the first small box represents the requirements for the key quality indicator f-CaO of cement clinker in GB / T 21372-2024, where f-CaO≤1.5 is qualified and is indicated by a check mark, and f-CaO>1.5 is unqualified and is indicated by ×; the second small box represents the feeding amount, that is, the output situation, the middle solid line and the upward arrow indicate that the feeding amount is greater than the average, and the middle dotted line and the downward arrow indicate that the feeding amount is less than the average; the third small box represents the limit level classification of coal consumption (comprehensive coal consumption per unit product of clinker, unit: kgce / t) in GB16780-2021, where 1-4 indicates that coal consumption gradually increases; the fourth small box represents the limit level classification of electricity consumption (comprehensive electricity consumption per unit product of clinker, unit: kW·h / t, note: the electricity consumption of the raw material system is not included) in GB16780-2021, where 1-4 indicates that the electricity consumption gradually increases. A horizontal bar in the small box indicates that no matter what the current indicator value is, it will not affect the classification of working conditions. In this way, all possible working conditions are divided into four categories: "excellent, good, medium, and poor", and their rules correspond to the green, yellow, orange, and red areas in the figure.

[0119] When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and the standard coal consumption per ton of clinker is ≤94kgce / t and the unit power consumption of the kiln system is ≤48kW·h / t, it is rated as "excellent";

[0120] When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value, and the comprehensive coal consumption per unit product of clinker and the comprehensive electricity consumption per unit product of clinker are greater than the third level (the comprehensive coal consumption per unit product of clinker and the comprehensive electricity consumption per unit product of clinker are greater than 109kgce / t and 61kW·h / t respectively), it is rated as "good". If the key quality index of cement clinker f-CaO≤1.5 and the feed rate is less than the average value, and the comprehensive coal consumption per unit product of clinker is ≤94kgce / t and the comprehensive electricity consumption per unit product of clinker is ≤48kW·h / t, it is also rated as "good".

[0121] When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and 100kgce / t< comprehensive coal consumption per unit of clinker product≤109kgce / t and <57kW·h / t comprehensive electricity consumption per unit of clinker product≤61kW·h / t, it is rated as "medium". When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and the comprehensive coal consumption per unit of clinker product≤100kgce / t and the comprehensive electricity consumption per unit of clinker product>61kW·h / t, it is rated as "medium". When the index f-CaO≤1.5 and the feed rate is greater than or equal to the average value, the comprehensive coal consumption per unit product of clinker>109kgce / t and the comprehensive electricity consumption per unit product of clinker≤57kW·h / t, it is also rated as "medium". When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is less than the average value, and the comprehensive coal consumption per unit product of clinker and the comprehensive electricity consumption per unit product of clinker are greater than the third level at the same time (the comprehensive coal consumption per unit product of clinker and the comprehensive electricity consumption per unit product of clinker are greater than 109kgce / t and 61kW·h / t at the same time), it is rated as "medium";

[0122] Other cases were classified as “poor”.

[0123] Step 2: Use the MIC method to calculate the correlation coefficients between 46 process variables and key indicators, sum the correlation coefficients, arrange them in descending order, and take the first 20 process variables for training the CVAE network; select the process variable operating condition data set that has been screened under the "excellent" state in step 1, train the CVAE network based on the minimum reconstruction error index, and construct a condition division channel based on the reconstruction of CVAE key indicators; the training goal of the CVAE network is to adjust the current network structure parameters so that the mean square error (MSE) between the input monitoring key indicators in the cement calcination process and the reconstruction of the CVAE network key indicator variables is minimized, and a pre-trained CVAE network is obtained. The CVAE network structure is as follows: Figure 5 shown.

[0124] In step 2, the MIC method is used to calculate the maximum correlation coefficients of all process variables with f-CaO content, unit coal consumption, feed rate and unit electricity consumption. Figure 4This is a heat map of the correlation between process variables screening based on MIC implemented in the present invention.

[0125] Among them, MIC is aimed at the relationship between two variables. It uses a scatter plot to divide the current two-dimensional space into a certain number of intervals in the x and y directions in a discretized two-dimensional space, and calculates the joint probability by checking where the current scattered points fall in each square. Let I(x,y) be the mutual information of the two variables, then MIC can be expressed as:

[0126]

[0127] Where a and b are the number of grids divided in the x and y directions, and B is the variable size, which is generally set to the 0.6th power of the data volume. The sum of the four correlation coefficients is arranged in descending order, and the first 20 process variables with high correlation with key production indicators are selected to train the CVAE network.

[0128] When using CVAE for data reconstruction, from the distribution The latent variable z is sampled in . The training objective of CVAE maximizes the conditional log-likelihood:

[0129]

[0130] Since the posterior distribution is the true distribution p θ (z|x i ,c i ). Therefore, when the posterior distribution is very close to the true distribution, the KL divergence approaches 0. Therefore, the objective function of CVAE can be defined as:

[0131] L(θ,φ;x i ,c i ) = lgp θ (x i |c i )-KL[q φ (z|x i ,c i )||p θ (z|x i ,c i )]

[0132] Assume that the input x after variable screening is reconstructed by the CVAE network as r, and the error size of the reconstruction is calculated by MSE:

[0133]

[0134] The mean square error (MSE) of the reconstruction error of the key indicator variable of the training CVAE network is minimized.

[0135] Step 3: Use the kernel density estimation (KDE) method to calculate four sets of reconstruction error control limits.

[0136] In step 3, the four types of data selected in step 1, namely “excellent, good, medium, and poor” and the screened process variable data set are input into the pre-trained CVAE network in step 2 to obtain the reconstruction error under different working conditions. The kernel density estimation (KDE) method is used to calculate the reconstruction error control limits Lim under the four working conditions with a confidence level of α (α is generally 95%). Optimal ,Lim good ,Lim general and Lim Poor .

[0137] Assume that the input data {x1,x2…x n} is taken from a continuous distribution, then the KDE function of the overall probability density function f(x) at any x is The formula is as follows:

[0138]

[0139] Where x is the sample point of the data set, h is the bandwidth, and K is the kernel function. The Gaussian kernel function is used for KDE calculation, and the formula is updated as follows:

[0140]

[0141] The control limits are calculated using the reconstruction errors of the excellent, good, medium, and poor data sets. Given a confidence level α, the control limits can be obtained from the probability density function of the reconstruction error under the corresponding mode, that is:

[0142]

[0143] Lim RE,mode Indicates the reconstruction error under excellent, good, medium and poor working conditions, and the corresponding control limit is calculated as Lim Optimal ,Lim good ,Lim general and Lim Poor .

[0144] Step 4: Construct the SFA network, obtain the chronic feature mapping matrix, and calculate S 2 Statistical control limit: Johansen test is used to obtain historical stable data to train SFA network. Through two-step continuous singular value decomposition SVD, the slow feature mapping matrix is ​​solved to obtain the slow features that characterize the inherent characteristics of the system. S is calculated according to the confidence α. 2 Statistical control limits.

[0145] In step 4, the specific operation of Johansen test to obtain the stationary process training data is:

[0146] Assume a matrix containing multiple time series:

[0147]

[0148] Where m is the sample size. (Here n is the total number of process monitoring variables) Substitute the input data Y into the vector autoregression model (VAR): ΔY t =α+ΠY t-1 +ε t

[0149] Where ΔYt is the first-order difference of Yt, α is a constant term, Π is the coefficient matrix, and ε t is the error term.

[0150] Calculate Yt-1 to form the lag matrix. For example, if the lag period is 1, ΔYt can be expressed as Yt-1.

[0151] Extract the Π matrix from the model and calculate its eigenvalues ​​λi by solving the following characteristic equation:

[0152] det(Π-λI)=0

[0153] Calculate statistics:

[0154]

[0155] And the largest eigenvalue statistic:

[0156]

[0157] Finally, the calculated trace statistics and the maximum eigenvalue statistics are compared with the critical value (usually based on a 5% significance level). c and LR max All of them are greater than the critical value, indicating that the linear combination of time series is stable. In this way, the training data of the stable process of historical data is obtained.

[0158] Assume the input signal x(t)∈R m , m slow features are uniformly expressed as s = Wx, where W = [w1,…,w m ] is the model parameter matrix to be determined;

[0159] If each dimension of the input variable already has zero mean, then the optimization problem of the above linear SFA model is equivalent to the generalized eigenvalue decomposition problem:

[0160] AW=BWΩ

[0161] in, is the first-order difference covariance matrix of the input x(t), B = <xx T > is the covariance matrix of the input x(t); W = [w1,…,w m ] contains m eigenvectors, and Ω=diag{λ1,…,λ m The generalized eigenvalues ​​in} are arranged from small to large, corresponding one to one with the vectors in the W matrix, that is, the slowest latent variable is The slowness is And satisfy W Τ AW=Ω,W Τ BW=I.

[0162] The specific process of solving the slow feature map matrix through two consecutive singular value decompositions is:

[0163] The first step of SVD decomposition is sphering of the input data x(t) to eliminate the correlation between variables. Assume that the SVD of matrix B is as follows:

[0164] B=UΛU T

[0165] The whitened data is expressed as:

[0166] z=Λ -1 / 2 U T x

[0167] Where Q = Λ -1 / 2 U T It is called the whitening matrix; solving the matrix W on this basis is equivalent to solving the matrix P = WQ -1 This is because

[0168] s=Wx=WQ -1 z=Pz

[0169] Use the covariance of s for unified representation: <ss T > t =I

[0170] get:

[0171] <ss T > t =P <zz T > t P T =PP T =I

[0172] It implies that P is an orthogonal matrix; thus, the optimization problem of linear SFA is transformed into finding an orthogonal matrix P such that Minimum, through The covariance of is decomposed by SVD to obtain:

[0173]

[0174] Matrix Ω=diag{ω1,…,ω m(d+1) The diagonal elements in} are the optimal target values ​​of the optimization target, that is, the Δ(·) values ​​of each slow feature:

[0175]

[0176] Then construct S 2 Statistics, where the choice of the number of slow features is important for the calculation of control limits, discarding input {x j The number of features that change faster is given by:

[0177] M e =card{si|Δ(s i )>max j {Δ(x j )}}

[0178] Where card{·} represents the number of elements in the set, and the number of slow features retained is calculated as:

[0179] M=m(d+1)-M e

[0180] The M slow features of the main changes in the system are expressed as The abandoned M e The slow feature is represented by

[0181] For the dynamic characteristics monitoring of the slow-changing process To design statistics, define S 2 Statistics:

[0182]

[0183] in express The covariance matrix of ; According to the statistical properties of slow features, it is assumed that If it obeys Gaussian distribution, then S 2 The statistic follows the F distribution after proportion transformation:

[0184] S 2 ~gF M,N-M-1

[0185] The parameter g is defined as:

[0186]

[0187] When the significance level α is given, S2 The control limits of the monitoring statistics can be determined.

[0188] Step 5: Fusing the dual-channel monitoring results, calculating the comprehensive scoring index (CSI) and the dynamic index (DI) of the working condition proposed in the present invention.

[0189] In step 5, in order to simplify the multi-statistical index working condition analysis process and facilitate engineering application, the present invention integrates the dual-channel monitoring results and proposes a quantitative working condition comprehensive scoring index (Comprehensive scoring index, CSI), which mainly includes three parts of information. First, the reconstruction error of the CVAE working condition superiority grade evaluation channel is used as the CSI benchmark (Basic score, BS), second, the first-order difference index of the reconstruction error is added to indicate the working condition trend (Condition movement indication, CMI), and third, the relative variation (relative variation, RV) of the SFA dynamic monitoring statistic is integrated. The working condition dynamic index (Dynamic index, DI) is defined by CMI and RV to indicate the degree of working condition abnormality.

[0190] First, the CVAE reconstruction error is combined with the control limit in step 4 for piecewise linear mapping. In order to avoid the problem that the reconstruction error is mapped from 0-Lim difference to 1-0, which will result in a low overall operating condition evaluation score and a pessimistic attitude towards the operating status in the actual production process, we map the reconstruction error from 0-Lim difference to 1-α (α≥0). (α=0.6 is recommended) as the basic score (BS):

[0191] Next, the first-order difference of the reconstruction error is used to characterize the condition movement, which is defined as the condition movement indication (CMI). The first-order difference of the reconstruction error E at time t can be represented by ΔEt=Et-Et-1. The present invention represents CMI as follows:

[0192]

[0193] In the above formula, ΔEt greater than 0 means that the continuous n+1 reconstruction errors continue to increase, that is, the working condition has a trend of deterioration, which we use CMI=-1 to represent. Similarly, the working condition has a trend of improvement, which is represented by CMI=1. CMI=0 means that the current working condition has no clear change trend, and further judgment is required in combination with the dynamic information of the working condition. In this paper, n=2 is taken, which means that the state of the working condition change trend is perceived by paying attention to the state of 3 consecutive reconstruction error values.

[0194] Finally, based on S 2 The statistical indicators calculate the relative variation (RV) of the overall working condition relative to the stable process:

[0195]

[0196] RV belongs to the real number set R. According to the calculation rule RV≥-1, theoretically, since S2 can take infinite values, the maximum value of RV can approach positive infinity, but too large a control amount is not helpful for abnormal working condition judgment. Therefore, the upper limit T of the maximum change of RV is set. T should be selected based on the working condition fluctuation of the specific application case. This paper selects T=1 according to the empirical value, that is, when CMI×RV is expressed as the dynamic index (DI):

[0197]

[0198] The value range of DI is all real numbers from -T to T. If |DI| = TDI's absolute value = T, it means that the operating condition may have a large change beyond the controller's tolerance, and an alarm should be issued. We give the comprehensive evaluation index CSI of the operating condition as follows:

[0199] CSI=BS+γ×DI

[0200] γ represents the weight coefficient of the operating condition variation index in the comprehensive evaluation index of the operating condition. It is generally recommended that γ≤2.

[0201] The CVAE-SFA dual-channel fusion online evaluation process is as follows Figure 6 As shown. First, calculate the reconstruction error RE of the current sample to be evaluated, and dynamically monitor the statistic S2 and evaluation indicators such as DI and CMI. Then execute the judgment process to evaluate the current state.

[0202] Step 6: Monitor the current operating data, input the performance evaluation model, calculate the corresponding statistics and indicators, and output the current monitoring status and real-time score. Figure 7 shown.

[0203] The implementation cases described above are merely descriptions of the methods proposed in the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the scope of protection determined by the claims of the present invention.

Claims

1. A method for evaluating the performance of a cement calcination process in real-time comprehensive working conditions, characterized in that: The steps include: Step 1: Classify the operating conditions of the historical operating data set in the cement production process into different categories: Analyze the cement production process, obtain the historical operation data of the calcination system, and divide the historical operation data set into four working condition categories of "excellent, good, medium, and poor" based on the requirements of the cement calcination process quality indicators and energy consumption indicators in the national standards according to the working condition classification rules; among them, the working condition classification rules are based on comprehensive evaluation of four key indicators: quality, output, coal consumption, and electricity consumption. The quality indicator is "f-CaO content", the key variable of the output indicator is "feeding amount", the coal consumption indicator is "standard coal consumption per ton of clinker", and the electricity consumption indicator is "unit electricity consumption of the kiln system"; Step 2: Select process variables with high correlation with key indicators to train the CVAE network and obtain a pre-trained CVAE network: The MIC method was used to calculate the correlation coefficients between 46 process variables and key indicators. The correlation coefficients were summed and arranged in descending order. The first 20 process variables were used to train the CVAE network. The process variable operating condition data set that was screened under the "excellent" state in step 1 was selected to train the CVAE network based on the minimum reconstruction error index, and a condition division channel based on the reconstruction of CVAE key indicators was constructed. The training goal of the CVAE network was to adjust the current network structure parameters so that the mean square error (MSE) between the input monitoring key indicators in the cement calcination process and the reconstruction of the CVAE network key indicator variables was minimized, and a pre-trained CVAE network was obtained. Among them, the key monitoring indicators during cement calcination include "feeding amount", "standard coal consumption per ton of clinker", "unit power consumption of kiln system" and "f-CaO content"; Step 3: Use the kernel density estimation (KDE) method to calculate four sets of reconstruction error control limits: The four operating conditions of "excellent, good, medium, and poor" obtained in step 1 and the data set of the screened process variables are input into the pre-trained CVAE network in step 2 to obtain the reconstruction errors under four different operating conditions. The kernel density estimation KDE method is used to calculate the reconstruction error control limits Lim under the four operating conditions with a confidence level of α. Optimal ,Lim good ,Lim general and Lim Poor ; Step 4: construct the SFA network, obtain the chronic feature mapping matrix, and calculate S 2 Statistical control limits: The Johansen test is used to obtain historical stable data to train the SFA network. Through two-step continuous singular value decomposition SVD, the slow feature mapping matrix is ​​solved to obtain the slow features that characterize the inherent characteristics of the system. S is calculated according to the confidence α. 2 Statistical control limits; Step 5: Combine the dual-channel monitoring results to obtain the working condition comprehensive score index CSI and working condition dynamic index DI: In order to simplify the multi-statistical index working condition analysis process and facilitate engineering application, a quantitative working condition comprehensive scoring index CSI is proposed by integrating the dual-channel monitoring results. It mainly includes three parts of information: the first is the reconstruction error of the CVAE working condition superiority level evaluation channel as the CSI benchmark BS; the second is the addition of the reconstruction error first-order difference index CMI to indicate the working condition trend; the third is the relative change RV of the SFA dynamic monitoring statistic, and the working condition dynamic index DI is defined by CMI and RV to indicate the degree of working condition abnormality; Step 6: Monitor the current operating data, input the superiority evaluation model, calculate the corresponding statistics and indicators, and output the current monitoring status and real-time score.

2. A method for evaluating the real-time comprehensive working condition performance of cement calcination process according to claim 1, characterized in that: The working condition classification standard in step 1 is: When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and the standard coal consumption per ton of clinker is ≤94kgce / t and the unit power consumption of the kiln system is ≤48kW·h / t, it is rated as "excellent"; When the key quality index of cement clinker f-CaO≤1.5 and the feed amount is greater than or equal to the average value, and the comprehensive coal consumption per unit product of clinker and the comprehensive electricity consumption per unit product of clinker are different and are greater than the third level, it is rated as "good". If the key quality index of cement clinker f-CaO≤1.5 and the feed amount is less than the average value, the comprehensive coal consumption per unit product of clinker is ≤94kgce / t, and the comprehensive electricity consumption per unit product of clinker is ≤48kW·h / t, it is also rated as "good"; When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and the comprehensive coal consumption per unit product of clinker is 100kgce / t<109kgce / t and the comprehensive electricity consumption per unit product of clinker is <57kW·h / t≤61kW·h / t, it is rated as "medium". When the key quality index of cement clinker f-CaO≤1.5 and the feed rate is greater than or equal to the average value and the comprehensive coal consumption per unit product of clinker is ≤100kgce / t and the comprehensive electricity consumption per unit product of clinker is >6 1kW·h / t is rated as "medium", when the key quality indicator of cement clinker f-CaO≤1.5 and the feed amount is greater than or equal to the average value and the comprehensive coal consumption per unit product of clinker>109kgce / t and the comprehensive electricity consumption per unit product of clinker≤57kW·h / t, it is also rated as "medium", when the key quality indicator of cement clinker f-CaO≤1.5 and the feed amount is less than the average value and the comprehensive coal consumption per unit product of clinker and the comprehensive electricity consumption per unit product of clinker are different and greater than the third level, it is rated as "medium"; Other cases are classified as "poor".

3. A method for evaluating the real-time comprehensive working condition performance of cement calcining process according to claim 1, characterized in that: The process of selecting process variables with high correlation with key indicators and training the CVAE network in step 2 is as follows: The MIC method was used to calculate the maximum correlation coefficients of all process variables with f-CaO content, standard coal consumption per ton of clinker, feed rate and unit power consumption of the kiln system; Among them, MIC aims at the relationship between two variables. It uses a scatter plot to divide the current two-dimensional space into a certain number of intervals in the x and y directions in the two-dimensional space, and calculates the joint probability by checking where the current scattered points fall in each square. Let I(x, y) be the mutual information of the two variables, then MIC is expressed as: In the formula, a and b are the number of grids divided in the x and y directions, and B is the variable size. The sum of the four correlation coefficients is arranged in descending order, and the first 20 process variables with high correlation with key production indicators are selected to train the CVAE network; When using CVAE for data reconstruction, from the distribution The training objective of CVAE is to maximize the conditional log-likelihood: Since the posterior distribution is the true distribution p θ (z|x i ,c i ) is a variational approximation; when the posterior distribution is very close to the true distribution, the KL divergence approaches 0; therefore, the objective function of CVAE is defined as: L(θ,φ;x i ,c i )=lgp θ (x i |c i )-KL[q φ (z|x i ,c i )||p θ (z|x i ,c i )] Assume that the input x after variable screening is reconstructed by the CVAE network as r, and the reconstruction error size is calculated by MSE: The mean square error (MSE) of the reconstruction error of the key indicator variable of the training CVAE network is minimized.

4. A method for evaluating the real-time comprehensive working condition performance of cement calcination process according to claim 1, characterized in that: The control limits in step 3 are calculated using the KDE method, where the input data {x1, x2…x n } is taken from a continuous distribution, then the KDE function of the overall probability density function f(x) at any x is The formula is as follows: Where x is the sample point of the data set, h is the bandwidth, and K is the kernel function. The Gaussian kernel function is used for KDE calculation, and the formula is updated as follows: The control limits are calculated using the reconstruction errors of the excellent, good, medium, and poor data sets. Given a confidence level α, the control limits can be obtained from the probability density function of the reconstruction error under the corresponding mode, that is: Among them, Lim RE,mode Indicates the reconstruction error under excellent, good, medium and poor working conditions, and the corresponding control limit is calculated as Lim Optimal ,Lim good ,Lim general and Lim Poor .

5. The method for evaluating the real-time comprehensive working condition performance of cement calcining process according to claim 1, characterized in that: The specific operation of Johansen test in step 4 to obtain the stationary process training data is: Assume a matrix containing multiple time series: Where m is the sample size, n is the total process monitoring variables, and the input data Y is substituted into the vector autoregression model (VAR): Y t =a+PY t-1 +e t Among them, ΔY t Yes t The first-order difference of , α is a constant term, Π is a coefficient matrix, ε t is the error term; Calculate Y t-1 , forming a lag matrix; Extract the Π matrix from the model and calculate its eigenvalue λ i , by solving the following characteristic equation: det(Π-λI)=0 Calculate statistics: And the largest eigenvalue statistic: Finally, by comparing the calculated trace statistics and the maximum eigenvalue statistics with the critical value, if LR c and LR max They are all greater than the critical value, indicating that the linear combination of time series is stationary, so as to obtain the training data of the stationary process of historical data.

6. A method for evaluating the real-time comprehensive working condition performance of cement calcining process according to claim 5, characterized in that: The specific operation of training the SFA network according to the historical stable data in step 4 is: Assume the input signal x(t)∈R m , m slow features are uniformly expressed as s = Wx, where W = [w1,…,w m ] is the model parameter matrix to be determined; If each dimension of the input variable already has zero mean, then the optimization problem of the above linear SFA model is equivalent to the generalized eigenvalue decomposition problem: AW=BWΩ in, is the first-order difference covariance matrix of the input x(t), B = <xx T > is the covariance matrix of the input x(t); W = [w1,…,w m ] contains m eigenvectors, and Ω=diag{λ1,…,λ m The generalized eigenvalues ​​in} are arranged from small to large, corresponding one to one with the vectors in the W matrix, that is, the slowest latent variable is The slowness is And satisfy W Τ AW=Ω,W Τ BW = I. ; The specific process of solving the slow feature map matrix through two consecutive singular value decompositions is: The first step of SVD decomposition is sphering of the input data x(t) to eliminate the correlation between variables. Assume that the SVD of matrix B is as follows: B=UΛU T The whitened data is expressed as: z=Λ -1 / 2 U T x Where Q = Λ -1 / 2 U T It is called the whitening matrix; solving the matrix W on this basis is equivalent to solving the matrix P = WQ -1 This is because s=Wx=WQ -1 z=Pz Use the covariance of s for unified representation: <ss T > t =I get: <ss T > t =P <zz T > t P T =PP T =I It implies that P is an orthogonal matrix; thus, the optimization problem of linear SFA is transformed into finding an orthogonal matrix P such that Minimum, through The covariance of is decomposed by SVD to obtain: Matrix Ω=diag{ω1,…,ω m(d+1) The diagonal elements in} are the optimal target values ​​of the optimization target, that is, the Δ(·) values ​​of each slow feature: Then construct S 2 Statistics, where the choice of the number of slow features is important for the calculation of control limits, discarding input {x j The number of features that change faster is given by: M e =card{si|Δ(s i )>max j {Δ(x j )}} Where card{·} represents the number of elements in the set, and the number of slow features retained is calculated as: M=m(d+1)-M e The M slow features of the main changes in the system are expressed as The abandoned M e The slow feature is represented by For the dynamic characteristics monitoring of the slow-changing process To design statistics, define S 2 Statistics: in express The covariance matrix of ; According to the statistical properties of slow features, it is assumed that If it obeys Gaussian distribution, then S 2 The statistic follows the F distribution after proportion transformation: S 2 ~gF M,N-M-1 The parameter g is defined as: When the significance level α is given, S 2 The control limits of the monitoring statistics can be determined.

7. A method for evaluating the real-time comprehensive working condition performance of cement calcining process according to claim 1, characterized in that: The specific operations of step 5 are: First, the CVAE reconstruction error is combined with the control limit in step 4 for piecewise linear mapping; to avoid the overall working condition evaluation score being low when the reconstruction error is mapped from 0-Lim difference to 1-0, the reconstruction error is mapped from 0-Lim difference to 1-α as the working condition basic score BS; Next, the first-order difference of the reconstruction error is used to characterize the operating condition trend, which is defined as the operating condition trend indicator CMI; the first-order difference of the reconstruction error E at time t is ΔE t =E t -E t-1 Expressed; CMI is expressed as follows: In the above formula, ΔE t If it is greater than 0, it means that the reconstruction errors of n+1 consecutive times continue to increase, that is, the working condition has a trend of deterioration, which is indicated by CMI=-1. Similarly, the working condition has a trend of improvement, which is indicated by CMI=1. CMI=0 means that the current working condition has no clear change trend, and further judgment needs to be made in combination with the dynamic information of the working condition. Finally, based on S 2 The statistical index calculates the relative change RV of the overall working condition relative to the stable process: RV belongs to the real number set R. According to the calculation rule RV≥-1, the upper limit T of the maximum change of RV is set. T is selected in combination with the fluctuation of specific working conditions. CMI×RV is expressed as the working condition dynamic index DI: The value range of DI is all real numbers from -T to T. If |DI|=T, it means that the operating conditions may have changed significantly beyond the acceptable range of the controller, and an alarm will be issued. The comprehensive evaluation index CSI of the working condition is given by the following formula: CSI=BS+γ×DI γ represents the weight coefficient of the operating condition variation index in the comprehensive evaluation index of the operating condition, γ≤2.

Images