CN104933492A

CN104933492A - On-line reliability prediction and improvement method of shopping system

Info

Publication number: CN104933492A
Application number: CN201510407544.4A
Authority: CN
Inventors: 丁佐华; 杨晓燕; 徐婷; 周远
Original assignee: Zhejiang Sci Tech University ZSTU
Current assignee: Zhejiang Sci Tech University ZSTU
Priority date: 2015-07-10
Filing date: 2015-07-10
Publication date: 2015-09-23

Abstract

The invention discloses an on-line reliability prediction and improvement method of a shopping system. In the method, system reliability is predicted by a time sequence analysis model ARIMA and improvement of the reliability is realized through positioning and reconfiguring an error component. The method especially comprises the following steps: (1) configuring a log file and collecting real-time system operation data; (2) according to the collected data, determining the ARIMA model ARIMA (p, d, q); (3) according to an obtained prediction model ARMA, predicting actual effect data in a future period of time; (4) calculating the reliability of the system in the future period of time; (5) searching a component which may have a fault through a spectrum positioning method when the predicted reliability is lower than a predicted value; (6) considering system reliability improvement degrees brought by adding a component which has a same function with a fault component and replacing the fault component respectively, and selecting a method which is used to improve the reliability maximumly to reconfigure the system so as to improve the reliability.

Description

Online reliability prediction and improvement method for shopping system

Technical Field

The invention relates to an online reliability prediction and improvement method for a shopping system, which carries out advanced reliability prediction on the real-time operation shopping system and provides a method for improving the reliability, thereby ensuring that the system can improve high-quality and high-reliability service.

Background

Software reliability is an important index for measuring software quality. There are many studies on the prediction of software reliability, but most of them are studies on static reliability, using data from static data in the test phase. The predicted reliability cannot reflect the real reliability of the system because the running environment and the dynamic behavior of the software are not considered; secondly, the static reliability cannot reflect the reliability of the system during operation, so that the system cannot be reconfigured to improve the reliability when the system reliability is not high. It has been proposed to combine a static reliability model with dynamic analysis to evaluate the reliability trend of the system during the operational phase; it has also been proposed to predict system real-time reliability by incorporating many sources of information, including the system's execution environment, while improving system reliability by reallocating executing components and changing the number of copies of components.

However, these methods have some disadvantages as follows: (1) the reliability of a certain time period next to the current state cannot be predicted according to the running data; (2) when the reliability is reduced, the components that affect the reduction in reliability cannot be positioned.

Disclosure of Invention

The invention aims to provide a method for predicting and improving online reliability of a shopping system aiming at the defects of the prior art.

The purpose of the invention is realized by the following technical scheme: a method for predicting and improving the online reliability of a shopping system comprises the following steps:

(1) collecting real-time operation data of the system: collecting real-time running data of the system by configuring a log file of the system, and obtaining a running log of the system by configuring log4j for a shopping system of a java application program;

(2) determining a time series analysis model for predicting failure data, namely an autoregressive integrated moving average model (ARIMA), according to the collected real-time operation data, wherein the model is in the form of ARIMA (p, d, q), p and q respectively represent an autoregressive term and a moving average term, and d represents the difference times when the time series becomes stable; determining parameters in the model according to the sample data;

(3) prediction of failure data: after the prediction model is determined, failure data of the system in a future period of time are predicted according to current data and past data; the method specifically comprises the following steps:

(3.1) smoothing processing of failure data: when failure data sequence { Y }₁，Y₂，…，Y_tWhen the sequence is not stable, the sequence is differentiated successively until a new stable sequence is obtained (X)₁，X₂，…，X_t-dD, wherein the difference times are the value of d; then the stationary sequence { X is repeated₁，X₂，…，X_t-dZero equalization processing

(3.2) model determination: estimating autoregressive coefficients and moving average coefficients of ARMA (p, q) by using a least square estimation method and a maximum likelihood estimation method; performing model order determination by using an AIC criterion, wherein the model with the minimum AIC value is the optimal model; finally, judging whether the residual sequence of the model is white noise or not, if so, checking to obtain a software reliability prediction model, and otherwise, recalculating;

(3.3) failure data prediction: predicting { X from the resulting model_t' } and then reduced to failure data Y_tThe predicted result of (1);

(4) and (3) predicting the reliability of the system: calculating the reliability of the system for a future period of time based on the predicted failure data and the component-based reliability calculation model; the component-based reliability model is specifically as follows:

port: port p is a tuple (M, t, c), M being a finite set of methods in port p, t representing the port type offered or required, c representing the communication type, synchronous or asynchronous;

assembly of: the component Com is a tuple (P)_p，P_r，G，W)，P_pIs to provide a limited set of ports, P_rIs a finite set of demand ports, G is a finite set of subcomponents,a port relationship representing a non-reflexive relationship, and TP ═ P_p∪P_r∪∪_C∈GC.P_r，C.P_pAnd C.P_rPort sets representing sub-component C supply and demand, respectively;

describing the dynamic behavior of a component using port activity, the basic activity of which is considered to be the exchange of information between two ports;

port reliability prediction: when a method of a port is called, the port is considered to be accessed once；n_iIndicating that port p is in a time interval 0, T]The number of times of internal access; one failure data represents one error in the prediction process, f_iRepresenting the number of errors in the prediction process; suppose port p has operation p₁→p₂Then the reliability at time tport p is defined as:

<math> <mrow> <mi>r</mi> <mrow> <mo>(</mo> <mi>p</mi> <mo>[</mo> <msub> <mi>p</mi> <mn>1</mn> </msub> <mo>&RightArrow;</mo> <msub> <mi>p</mi> <mn>2</mn> </msub> <mo>]</mo> <mo>)</mo> </mrow> <mo>=</mo> <mn>1</mn> <mo>-</mo> <mfrac> <msub> <mi>f</mi> <mi>i</mi> </msub> <msub> <mi>n</mi> <mi>i</mi> </msub> </mfrac> <mo>;</mo> </mrow> </math>

reliability prediction of the system: the reliability of a system after a period of execution is defined as:wherein r (tr)_i) Is the locus tr_iReliability of (f) (tr)_i) Is the locus tr_iThe frequency of occurrence of (c);

(5) positioning of a fault component: if the predicted reliability is lower than an expected value, searching a component causing the reduction of the system reliability, and positioning a fault component by using a frequency spectrum positioning and maximum likelihood estimation method; assuming that the system consists of M components, denoted Cj (j ∈ {1, … …, M }), there may be E erroneous components; diagnostic report D ═<…，d_k，…>For an ordered set of possible multiple erroneous component candidates, d_kRanked by likelihood of error;

establishing a spectrum matrix: the spectrum matrix represents the labels of the components contained in the dynamic behavior of the system; when the system is executed once, the passed components are marked as 1, otherwise, the passed components are 0; assuming that a total of N times are performed, the spectrum matrix is represented as a matrix of N x MA; taking into account the number of passes through the component, the matrix element a_ijIndicating whether the component Cj passes through the component and the number of times of passing when the component Cj is executed at the ith time; the execution result is stored in a vector e, which indicates that after each execution, if the operation is passed, the operation is marked as 0, and the operation failure is marked as 1;

and (3) candidate set generation: using a minimum hit set algorithm to compute a diagnostic candidate set, a spectral-based error localization technique (SFL) can well predict failure ranks in terms of component failure probability; SFL input spectrum matrices (a, e) yielding ordered ranks of component error probabilities; the components calculate the ranking with the similarity coefficient, i.e. the highest ranked component is often wrong; the specific similarity coefficient is defined as:

<math> <mrow> <mi>s</mi> <mrow> <mo>(</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = '{' close = ''> <mtable> <mtr> <mtd> <mfrac> <mrow> <msub> <mi>n</mi> <mn>11</mn> </msub> <mrow> <mo>(</mo> <mo>{</mo> <mi>j</mi> <mo>}</mo> <mo>)</mo> </mrow> </mrow> <mrow> <mi>d</mi> <mi>e</mi> <mi>n</mi> <mrow> <mo>(</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <msqrt> <mrow> <mo>(</mo> <msub> <mi>n</mi> <mn>11</mn> </msub> <mo>(</mo> <mo>{</mo> <mi>j</mi> <mo>}</mo> <mo>)</mo> <mo>+</mo> <msub> <mi>n</mi> <mn>10</mn> </msub> <mo>(</mo> <mo>{</mo> <mi>j</mi> <mo>}</mo> <mo>)</mo> <mo>)</mo> <mo>*</mo> <mo>(</mo> <msub> <mi>n</mi> <mn>11</mn> </msub> <mo>(</mo> <mo>{</mo> <mi>j</mi> <mo>}</mo> <mo>)</mo> <mo>+</mo> <msub> <mi>n</mi> <mn>01</mn> </msub> <mo>(</mo> <mo>{</mo> <mi>j</mi> <mo>}</mo> <mo>)</mo> <mo>)</mo> </mrow> </msqrt> </mrow> </mfrac> <mo>,</mo> <mi>d</mi> <mi>e</mi> <mi>n</mi> <mo>(</mo> <mi>j</mi> <mo>)</mo> <mo>&NotEqual;</mo> <mn>0</mn> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> <mo>,</mo> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mtd> </mtr> </mtable> </mfenced> </mrow> </math>

wherein,

n₁₁(j)＝|{i_∈{1,2,…,N}|a_ij>0∧e_i＝1}|

n₁₀(j)＝|{i∈{1,2,…,N}|a_ij>0∧e_i＝0}|

n₀₁(j)＝|{i∈{1,2,…,N}|a_ij＝0∧e_i＝1}|

candidate set ordering: calculating probabilities of the candidate sets using bayesian rules; according to the candidate set d_kAll observations of (d), each candidate set d_kThe probability of (c) describes the error condition of the actual system; the candidate set d under the observed observation is derived by Bayes' rule_kThe formula for calculating the posterior probability of (2) is:

let P_r(j) P denotes a component C_jA priori probability of error, given that component errors are independent, candidate set d_kThe prior probability of (a) is:

P_r(obs) is a normalization factor, which requires no computation; since each execution is independent, it is possible to perform the operations independently

Wherein Pr (obs)_i|d_k) Is defined as follows:

<math> <mrow> <msub> <mi>P</mi> <mi>r</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>obs</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <mn>1</mn> <mo>,</mo> </mtd> <mtd> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>&RightArrow;</mo> <msub> <mi>obs</mi> <mi>i</mi> </msub> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> <mo>,</mo> </mtd> <mtd> <msub> <mi>obs</mi> <mi>i</mi> </msub> <mo>^</mo> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>|</mo> <mo>=</mo> <mo>&perp;</mo> </mtd> </mtr> <mtr> <mtd> <msub> <mi>ϵ</mi> <mi>ik</mi> </msub> <mo>,</mo> </mtd> <mtd> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>&RightArrow;</mo> <mo>{</mo> <msub> <mi>obs</mi> <mn>1</mn> </msub> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msub> <mi>obs</mi> <mi>N</mi> </msub> <mo>}</mo> </mtd> </mtr> </mtable> </mfenced> </mrow> </math>

wherein:

<math> <mrow> <msub> <mi>ϵ</mi> <mi>ik</mi> </msub> <mo>=</mo> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <munder> <mi>Π</mi> <mrow> <mi>j</mi> <mo>:</mo> <mrow> <mo>(</mo> <mi>j</mi> <mo>&Element;</mo> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mo>^</mo> <mrow> <mo>(</mo> <msub> <mi>a</mi> <mi>ij</mi> </msub> <mo>></mo> <mn>0</mn> <mo>)</mo> </mrow> </mrow> </munder> <msup> <msub> <mi>h</mi> <mi>j</mi> </msub> <msub> <mi>a</mi> <mi>ij</mi> </msub> </msup> <mo>,</mo> </mtd> <mtd> <msub> <mi>e</mi> <mi>i</mi> </msub> <mo>=</mo> <mn>0</mn> </mtd> </mtr> <mtr> <mtd> <mn>1</mn> <mo>-</mo> <munder> <mi>Π</mi> <mrow> <mi>j</mi> <mo>:</mo> <mrow> <mo>(</mo> <mi>j</mi> <mo>&Element;</mo> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mo>^</mo> <mrow> <mo>(</mo> <msub> <mi>a</mi> <mi>ij</mi> </msub> <mo>></mo> <mn>0</mn> <mo>)</mo> </mrow> </mrow> </munder> <msup> <msub> <mi>h</mi> <mi>j</mi> </msub> <msub> <mi>a</mi> <mi>ij</mi> </msub> </msup> <mo>,</mo> </mtd> <mtd> <msub> <mi>e</mi> <mi>i</mi> </msub> <mo>=</mo> <mn>1</mn> </mtd> </mtr> </mtable> </mfenced> </mrow> </math>

wherein h is_j∈[0,1]Represents the probability that the component j is working properly;

thus, Pr (obs | d)_k) Is about h_jH can be obtained by maximum likelihood estimation_jThe estimated value of (c) is actually the solution of the extremum problem as follows:wherein G ═ h_j∈[0,1]:j＝1,2,…,M}；

(6) System reconfiguration: for a fault component, the component can be replaced or a normal component with the same function as the fault component can be added in the system, and the method with the highest reliability improvement is selected by calculating the reliability obtained by the two methods.

Further, in the step (2), establishing the ARIMA (p, d, q) model is completed by using Eviews statistical analysis software, which specifically includes the following processes: smoothing processing of time series data and estimation of model parameters.

Further, in the step (5), the locating of the fault component includes: and constructing a frequency spectrum matrix according to the times of each component passing each test, generating a candidate set of the fault components according to the frequency spectrum matrix, and determining the posterior probability of the candidate set by utilizing maximum likelihood estimation, wherein the largest value is the selected fault component set.

The invention has the beneficial effects that: the method of the invention carries out real-time prediction on failure data of the system in a certain time period in the future through a time series analysis model ARIMA, and predicts the reliability of the system, thereby overcoming the defects 1 in the prior art; when the predicted reliability is lower than the expected value, the components causing the system reliability to decrease are first located, and then there are two ways to reconfigure the system: the defect point 2 in the prior art is overcome by replacing the failed component and adding the component with the same function, and selecting the method with the maximum reliability improvement to reconfigure the system by judging the generated effect.

Drawings

FIG. 1 is a technical flow diagram of a service oriented system reliability prediction and enhancement method;

FIG. 2 is a partial online shopping system SCA with new components added.

Detailed Description

The present invention will be further described with reference to the accompanying drawings.

The online reliability prediction and improvement method of the shopping system provided by the invention can predict the reliability of the real-time running system at a certain time in the future and can carry out targeted reconfiguration on the system through the positioning of the fault component, thereby improving the reliability of the system. The process is described in detail below:

(1) collecting real-time dynamic data: firstly, determining reasonable variables of real-time values to be recorded by analyzing the characteristics of the system such as functional performance and the like, wherein the system with time requirement can select response time, the system with quantity requirement can select throughput and the like; second, the log file of the system is configured according to the selected variables. The present invention selects variable response times for systems with response time requirements, configuring log4j to obtain a running log of the system. Because a port-based reliability prediction model is used, failure data for all ports of the system is collected from the log. Such as the order port, has three input variables (userId, isplayed, totalpick) and two output variables (detail order, response time). By monitoring the response time value in the javaServlet and the run time value sent to the log, variable response time data is collected as shown in the following table:

time (min)	Response time (ms)	Time (min)	Response time (ms)
				1	3560	11	1980
2	3860	12	2250
				3	2214	13	2750
4	1928	14	3126
				5	1356	15	3584
6	780	16	3847
				7	972	17	4256
8	1280	18	4895
				9	1492	19	5216
10	1685	20	5872

(2) And (3) failure data prediction: and establishing an ARIMA model ARIMA (p, d, q) according to the obtained data sequence, wherein p is an autoregressive term, q is the number of moving average terms, and d is the difference times when the time sequence becomes stable. Firstly, the stationarity of the data sequence is judged according to the obtained data sequence, and a stable sequence { X ] is obtained by carrying out finite difference on an unstable sequence_t}. Continuing with the example of port order, it can be seen from the data in Table 1The data sequence is in a growing trend, so the data sequence is an unstable sequence. Smoothing the data by first logarithmizing the sequence to obtain Y_t＝lnX_t。Y_tThe results are shown in the following table.

Time (min)	Y_t	Time (min)	Y_t
				1	8.108021	11	7.700747
2	8.014996	12	7.807916
				3	7.869401	13	7.956126
4	7.704361	14	8.048468
				5	7.578656	15	8.160803
6	7.393263	16	8.243808
				7	7.1546153	17	8.347116
8	7.314552	18	8.487146
				9	7.455876	19	8.558527
10	7.590852	20	8.673341

Next, the statistical software Eviews is used to analyze Y_tStability of (2). Sequence Y obtained using Eviws_tThe unit root and the ADF value. After the ADF is subjected to the first difference processing, the ADF and the result after the first difference are both greater than 10%, so Y is_tIt is still unstable. Then, for Y again_tThe second difference was performed and the unit root was verified, and the ADF value obtained after the second difference was-4.866626, which was less than the 1% threshold. Thus Y is_tAfter a second difference being stableThe sequence, i.e. d ═ 2. Let Z_t＝D(Y_t2) represents Y_tThe values after the second difference remain in a stable sequence.

Secondly, parameter determination is carried out, and for the model ARIMA (p, q), the general mathematical formula is as follows: x_t＝φ₁X_t-1+…+φ_pX_t-p+_t+θ₁ _t-1+…+θ_q _t-qTherein-_tIs a white noise sequence, so it needs to be based on a stationary sequence { X }_tDetermining p, q, phi₁-φ_p,θ₁-θ_q. The determination of p and q requires determining the values of several groups (p, q) according to the autocorrelation coefficient and the partial autocorrelation coefficient of the stationary sequence and experience, and then determining the coefficient phi by performing maximum likelihood estimation on each group₁-φ_p,θ₁-θ_qFinally, an optimal set (p, q) and corresponding coefficients phi are selected from the set₁-φ_p,θ₁-θ_q(the decision can be made based on the Akaike Information Criterion (AIC) value, the smaller the AIC value, the better the corresponding (p, q)). After the model is determined, the variable values of the future time can be predicted in sequence, and whether the variable values are failure data or not can be judged according to the expected values. E.g.order port, when the model ARIMA (p,2, q) is analyzed by the sequence Zt from the previous step. And performing model optimization by using the AIC according to the autocorrelation and partial correlation results obtained by Zt, calculating different values of p and q by using different values of AIC, and finally obtaining the minimum value-5.354639 by using the AIC when (p, q) is equal to (6, 8). Thus, a model ARIMA (6,8) was obtained. To verify that the resulting model ARIMA (6,8) is the desired end result, the model ARIMA (6,8) residual sequence is analyzed to see if a white noise sequence is satisfied. Since Eviews is used for the analysis, it is only necessary to check whether the autocorrelation and partial correlation coefficients are within the confidence interval. The confidence interval is a measure of the reliability estimate, as known from the previously obtained values of the autocorrelation and partial correlation coefficients, which are both within the confidence interval. Therefore, the model was finally determined to be ARIMA (6,2, 8).

(3) And (3) predicting the reliability of the system: the reliability calculation model used by the invention is the reliability calculation of the service combination based on the portAnd (4) modeling. Suppose port p has operation p₁→p₂Then the reliability at time tport p is defined as:wherein n is_iIndicating that port p is in a time interval 0, T]Number of accesses within, one failure data representing one error in the prediction process, f_iRepresenting the number of errors in the prediction process; the reliability of the system is defined asWherein f (tr)_i) Represents a track tr_iThe frequency of occurrence of r (tr)_i) Represents a track tr_iDefined as the product of the port reliability on the trace: two ports p in sequential relationship₁,p₂Reliability definition of (r) (p)₁；p₂)＝r(p₁)r(p₂) The reliability of the ports of two selection relations is defined asb is a Boolean condition, if b is true, then port p is selected₁Executing, x (b) is 1, otherwise, selecting p₂Executing, wherein x (b) is 0; for a cyclic port p, reliability is defined as(b is a Boolean condition, p_r(i) Indicating the probability of exiting the loop at the ith time,or r (n × p) ═ r (p)ⁿ(n is a positive integer indicating that port p cycles n times). Still taking port order as an example, sequence Z is first predicted_tAnd then returns the result to the sequence Y_tFinally orderPredict X_t. The response time of the next 3 minutes is predicted by the model, and the prediction results are respectively as follows: 7072.03,7816.51, 8518.68. It is clear that the port order has a response time greater than 8s only at minute 23; second, assume that the order port has passed 468 times in total. Therefore, the reliability of the port order is 1-1/468 ═ 0.9978.

The same can predict the reliability at 23 minutes for each of the remaining ports, see table below.

Meanwhile, the probabilities pb4(1) ═ 0.965, pb4(2) ═ 0.024 (stolen credit card, credit insufficient, etc.), pb4(3) ═ 0.011 (wrongly written characters, etc.), frequency of signature: f (tr1) is 56, f (tr2) is 53, f (tr3) is 78, f (tr4) is 58, f (tr5) is 85, f (tr6) is 62, f (tr7) is 49, f (tr8) is 65, f (tr9) is 57, f (tr10) is 94, f (tr11) is 56, f (tr12) is 76, and f (tr13) is 51. Thus, the predicted overall system reliability is:

the reliability of the system is therefore lower than expected and for this reason it is necessary to find the faulty component that causes the problem.

(4) Positioning a fault assembly: when the predicted reliability is lower than the expected value of the system, it is then necessary to find the one (group) of components that is most likely to cause a reduction in the reliability of the system. First, from a set of test results, a spectrum matrix (a, e) is constructed, where a ═ a_ij)，a_ijIndicates the probability that the ith test passes component j, e ═ e_i)，e_iIs 0 or 1(0 means the ith test fails, 1 means the ith test succeeds); secondly, a diagnosis candidate set D is generated according to the spectrum matrix, and a candidate set D is obtained by using a minimum hit set algorithm STACCATO. The algorithm is divided into three steps: (1) the initialization stage ranks the components using correlation coefficients, (2) adds all components included in the failed set to a candidate set D, and (3) when | D | is zero<At L, the following is done for the components in the row that first exceed λ: remove component j, retain all e in (A, e)_i＝1∧a_ij>A of 0_i*The statco algorithm is run on the new (a, e), adding the returned component to D and verifying if it is a minimum hit set.

The similarity coefficient calculation formula of the component j is as follows:

n₁₁(j)＝|{i|a_ij＞0^e_i＝1}|，n₀₁(j)＝|{i|a_ij＝0∧e_i＝1}|，n₁₀(j)＝|{i|a_ij》0∧e_i0} |. Third, the posterior probability Pr (d) of each candidate set is calculated_k| obs), where obs ═ a, e.

\Pr (d_{k} | obs) = \frac{\Pr (obs | d_{k})}{\Pr (obs)},

Since each test is independent, the test is carried out

While

<math> <mrow> <mi>Pr</mi> <mrow> <mo>(</mo> <msub> <mi>obs</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <mn>1</mn> <mo>,</mo> </mtd> <mtd> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>&RightArrow;</mo> <msub> <mi>obs</mi> <mi>i</mi> </msub> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> <mo>,</mo> </mtd> <mtd> <msub> <mi>obs</mi> <mi>i</mi> </msub> <mo>^</mo> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>|</mo> <mo>=</mo> <mo>&perp;</mo> </mtd> </mtr> <mtr> <mtd> <msub> <mi>ϵ</mi> <mi>ik</mi> </msub> <mo>,</mo> </mtd> <mtd> <msub> <mi>d</mi> <mi>k</mi> </msub> <mo>&RightArrow;</mo> <mi>obs</mi> </mtd> </mtr> </mtable> </mfenced> <mo>,</mo> </mrow> </math>

Here, the_ikIs defined as

<math> <mrow> <msub> <mi>ϵ</mi> <mrow> <mi>i</mi> <mi>k</mi> </mrow> </msub> <mo>=</mo> <mfenced open = '{' close = ''> <mtable> <mtr> <mtd> <mrow> <munder> <mo>Π</mo> <mrow> <mi>j</mi> <mo>&Element;</mo> <msub> <mi>d</mi> <mi>k</mi> </msub> </mrow> </munder> <msup> <msub> <mi>h</mi> <mi>j</mi> </msub> <msub> <mi>a</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> </msup> <mo>,</mo> <msub> <mi>e</mi> <mi>i</mi> </msub> <mo>=</mo> <mn>0</mn> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mn>1</mn> <mo>-</mo> <munder> <mo>Π</mo> <mrow> <mi>j</mi> <mo>&Element;</mo> <msub> <mi>d</mi> <mi>k</mi> </msub> </mrow> </munder> <msup> <msub> <mi>h</mi> <mi>j</mi> </msub> <msub> <mi>a</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> </msup> <mo>,</mo> <msub> <mi>e</mi> <mi>i</mi> </msub> <mo>=</mo> <mn>0</mn> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>,</mo> </mrow> </math>

Wherein h is_jIndicating the probability that component j is in error. Obtaining h by maximum likelihood estimation_jAn estimate of (d). The candidate set is based on the posterior probabilityThe top candidate set is the set of components that are most likely to be in error, i.e., the set of components that need to be reconfigured, sorted from high to low. The specific calculation is as follows:

obtaining the component information and the operation condition of the system in operation from the file information of the database, and updating the frequency spectrum matrix by taking 60 seconds as the window size, as follows:

the correlation coefficient for each component is calculated as follows:

	C₁₁	C₁₂	C₂₁	C₂₂	C_cs	C₃	C_ds
								n₁₁(j)	3	2	2	5	1	1	1
n₁₀(j)	7	3	7	4	4	4	4
								n₀₁(j)	0	1	1	1	2	2	2
S(j)	0.55	0.52	0.38	0.38	0.26	0.26	0.26

thus, ordered by correlation coefficient size<C₁₁，C₁₂，C₂₁，C₂₂，C_cs，C₃，C_ds>. The candidate set returned by the minimum hit set algorithm STACCATO is { { C₁₁}，{C₁₂，C₂₁}，{C₁₂，C₂₂}}。

The probability values for each candidate set are calculated to determine the problem component. For candidate set { C₁₁And can know that:

C₁₁	1	1	1	1	1	1	1	1	1	1
											e	0	1	0	0	1	0	0	0	0	1
P_r(e_i\|d_k)	h₁	1-h₁	h₁	h₁	1-h₁	h₁	h₁	h₁	h₁	1-h₁

therefore P is_r(obs|{C₁₁})＝h₁ ⁷(1-h₁)³To h is aligned with₁Calculating its maximum likelihood estimation value h₁When Pr (e | { C) |, (0.7)₁₁})＝2.224*10^- ³. Similarly, for candidate set { C₁₂，C₂₁Knowing:

P_r(obs|{C₁₂，C₂₁})＝h₁ ³h₂ ⁷(1-h₁)(1-h₂)(1-h₁h₂) To h is aligned with₁，h₂The maximum likelihood estimation value is calculated as h₁＝0.6417，h₂0.8528, so P_r(obs|{C₁₂，C₂₁})＝2.07*10^-3。

For candidate set { C₁₂，C₂₂}：

P_r(obs|{C₁₂，C₂₂})＝h₁ ³h₂ ²⁸(1-h₁)(1-h₂ ⁴)(1-h₁h₂ ⁴) To h is aligned with₁，h₂The maximum likelihood estimation value is calculated as h₁＝0.6417，h₂0.9610, so P_r(obs|{C₁₂，C₂₂})＝2.07*10^-3。

Knowing P by formula_r(d₁|obs)＝1.86*10^-9，P_r(d₂|obs)＝1.92*10^-10，P_r(d₃|obs)＝1.92*10^-10. So the candidate platoon behavior { { C₁₁}，{C₁₂，C₂₁}，{C₁₂，C₂₂}, i.e. component C₁₁INIT error is the most likely.

After the faulty component is located, the system will be reconfigured.

(5) System reconfiguration: there are two ways to configure a system, namely adding a component that functions the same as a failed component and replacing the failed component. And selecting a proper configuration method by judging the reliability improvement degree brought by the two methods. When the components with the same functions as the fault components are added, the tracks of the system are increased, and the tracks need to be recalculated, and the probability and reliability of the tracks are improved; for replacing a failed component, the trajectory of the system and the probability of the trajectory will not change, but the reliability of each trajectory needs to be recalculated.

First, the reliability of the system after adding a new component is calculated. Additive for foodAdd and component C₁₁New component C with the same INIT function₁₁'INIT', recalculate the system reliability. As shown in FIG. 2, a new component C is added₁₁'INIT' rear system part. Adding a new component C₁₁After'' INIT, the system may select component C₁₁INIT and C₁₁' INIT begins execution in two paths, and the two start paths are mutually exclusive. Therefore, the trajectory of the system becomes twice as large. The port reliability is as follows:

the reliability of the whole system after adding the new component is as follows:

second, the system reliability after replacement of the failed component is calculated. By means of functionally identical new components C₁₁Replacement of failed component C by "═ INIT₁₁System reliability is recalculated as INIT. The port reliability is shown in the following table:

the reliability of the whole system after the replacement of the failed component is as follows:

the results of an experiment comparing the two schemes of adding a new component and replacing a failed component can be obtained, and in the experiment, when the system reliability does not reach the expected value, the effect of improving the system reliability by replacing the component is better. Therefore, in this example, replacement of the faulty group is adoptedPart C₁₁The strategy of INIT automatically improves system reliability.

Through the 5 steps, the online reliability prediction and the online reliability improvement of the shopping system can be completed. It should be noted that: when the configuration of the system is considered, only the improvement degree of the system reliability by different methods is concerned, and the time cost and the space cost of replacing components and increasing the components are not considered; when considering both time and space costs, there may be different concerns and therefore the chosen configuration scheme will also transmit changes, and therefore several improvements may be made to the system reconfiguration without departing from the principles of the present invention, which should also be considered as the scope of protection of the present invention.

Claims

1. A method for predicting and improving the online reliability of a shopping system is characterized by comprising the following steps:

assembly of: the component Com is a tuple (P)_p，P_r，G，W)，P_pIs to provide a portOf a finite set of P_rIs a finite set of demand ports, G is a finite set of subcomponents,a port relationship representing a non-reflexive relationship, and TP ═ P_p∪P_r∪∪_C∈GC.P_r，C.P_pAnd C.P_rPort sets representing sub-component C supply and demand, respectively;

port reliability prediction: when a method of a port is called, the port is considered to be accessed once; n is_iIndicating that port p is in a time interval 0, T]The number of times of internal access; one failure data represents one error in the prediction process, f_iRepresenting the number of errors in the prediction process; suppose port p has operation p₁→p₂Then the reliability at time tport p is defined as:

(5) positioning of a fault component: if the predicted reliabilityWhen the value is lower than the expected value, searching a component causing the reduction of the system reliability, wherein a frequency spectrum positioning and maximum likelihood estimation method is used for positioning a fault component; assuming that the system consists of M components, denoted Cj (j ∈ {1, … …, M }), there may be E erroneous components; diagnostic report D ═<…，d_k，…>For an ordered set of possible multiple erroneous component candidates, d_kRanked by likelihood of error;

establishing a spectrum matrix: the spectrum matrix represents the labels of the components contained in the dynamic behavior of the system; when the system is executed once, the passed components are marked as 1, otherwise, the passed components are 0; assuming that a total of N times are performed, the spectrum matrix is represented as a matrix a of N × M; taking into account the number of passes through the component, the matrix element a_ijIndicating whether the component Cj passes through the component and the number of times of passing when the component Cj is executed at the ith time; the execution result is stored in a vector e, which indicates that after each execution, if the operation is passed, the operation is marked as 0, and the operation failure is marked as 1;

<math> <mrow> <mi>s</mi> <mrow> <mo>(</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = '{' close = ''> <mtable> <mtr> <mtd> <mfrac> <mrow> <msub> <mi>n</mi> <mn>11</mn> </msub> <mrow> <mo>(</mo> <mo>{</mo> <mi>j</mi> <mo>}</mo> <mo>)</mo> </mrow> </mrow> <mrow> <mi>d</mi> <mi>e</mi> <mi>n</mi> <mrow> <mo>(</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <msqrt> <mrow> <mo>(</mo> <msub> <mi>n</mi> <mn>11</mn> </msub> <mo>(</mo> <mrow> <mo>{</mo> <mi>j</mi> <mo>}</mo> </mrow> <mo>)</mo> <mo>+</mo> <msub> <mi>n</mi> <mn>10</mn> </msub> <mo>(</mo> <mrow> <mo>{</mo> <mi>j</mi> <mo>}</mo> </mrow> <mo>)</mo> <mo>)</mo> <mo>*</mo> <mo>(</mo> <msub> <mi>n</mi> <mn>11</mn> </msub> <mo>(</mo> <mrow> <mo>{</mo> <mi>j</mi> <mo>}</mo> </mrow> <mo>)</mo> <mo>+</mo> <msub> <mi>n</mi> <mn>01</mn> </msub> <mo>(</mo> <mrow> <mo>{</mo> <mi>j</mi> <mo>}</mo> </mrow> <mo>)</mo> <mo>)</mo> </mrow> </msqrt> </mrow> </mfrac> <mo>,</mo> <mi>d</mi> <mi>e</mi> <mi>n</mi> <mo>(</mo> <mi>j</mi> <mo>)</mo> <mo>&NotEqual;</mo> <mn>0</mn> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> <mo>,</mo> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mtd> </mtr> </mtable> </mfenced> </mrow> </math>

wherein,

n₁₁(j)＝|{i∈{1,2,…,N}|a_ij>0∧e_i＝1}|

n₁₀(j)＝|{i∈{1,2,…,N}|a_ij>0∧e_i＝0}|

n₀₁(j)＝|{i∈{1,2,…,N}|a_ij＝0∧e_i＝1}|

Wherein Pr (obs)_i|d_k) Is defined as follows:

wherein:

thus, Pr (obs | d)_k) Is about h_jThe expression (h) can obtain the estimated value of hj by the maximum likelihood estimation method, and actually, the following extremum problem is solved:wherein G ═ h_j∈[0,1]:j＝1,2,…,M}；

2. The online reliability prediction and improvement method of shopping system as claimed in claim 1, wherein in the step (2), establishing ARIMA (p, d, q) model is accomplished by using Eviews statistical analysis software, which includes the following steps: smoothing processing of time series data and estimation of model parameters.

3. The online reliability prediction and improvement method of shopping system according to claim 1, wherein in said step (5), the location of the faulty component comprises: and constructing a frequency spectrum matrix according to the times of each component passing each test, generating a candidate set of the fault components according to the frequency spectrum matrix, and determining the posterior probability of the candidate set by utilizing maximum likelihood estimation, wherein the largest value is the selected fault component set.