Application system availability prediction method based on markov chain and regression analysis

Abstract

The application relates to a Markov chain and regression analysis-based application system availability prediction method, and belongs to the field of fault prediction. The application determines an index set influencing the availability probability of an application system, defines the state space of each index in the index set, determines the relationship between the state space of each index and the availability probability of the application system by using a regression analysis method, constructs a model for predicting the state distribution of each index by using a Markov chain model, determines the initial state of each index, defines the time range of the availability probability prediction of the application system, predicts the state distribution of each index of the application system, and predicts the availability probability of the application system after a specified time according to the state distribution of each index and the relationship between the state space of each index and the availability probability of the application system. The application effectively improves the availability of the application system, prolongs the stability time length of the application system, and improves the user satisfaction.

Description

Technical Field

[0001] This invention belongs to the field of fault prediction, specifically relating to an application system availability prediction method based on Markov chains and regression analysis. Background Technology

[0002] As enterprises continue to develop, the number of information systems gradually increases, and the scale of these systems expands. Almost all daily office work and production within the enterprise rely on these application systems. Therefore, the stability of these information systems directly impacts office efficiency and production progress. Because application system failures occur at unpredictable times and cannot be detected promptly, traditional information system fault handling often involves addressing problems only after they occur, resulting in long troubleshooting cycles, significant impacts, and ultimately, low user satisfaction with enterprise IT systems and daily operations.

[0003] Based on the above situation, enterprises urgently need a fault prediction method. By monitoring parameters that affect system availability, the time when system failures will occur can be predicted in advance, achieving the goal of early prevention. By handling potential failures in advance, the stable operation time of the system can be extended, thereby improving the efficiency of enterprise operation. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] The technical problem this invention aims to solve is how to provide an application system availability prediction method based on Markov chains and regression analysis, in order to address the problem that traditional information system fault handling often involves processing after a fault occurs, resulting in long troubleshooting cycles and significant impacts.

[0006] (II) Technical Solution

[0007] To address the aforementioned technical problems, this invention proposes a method for predicting the availability of application systems based on Markov chains and regression analysis. This method includes the following steps:

[0008] S1. Determine the set of indicators that affect the probability of application system availability;

[0009] S2. Define the state space of each indicator in the indicator set;

[0010] S3. Use regression analysis to determine the relationship between the state space of each indicator and the probability of application system availability.

[0011] S4. Construct a model that uses a Markov chain model to predict the state distribution of each indicator;

[0012] S5. Determine the initial state of each indicator;

[0013] S6. Define the time range for predicting the availability probability of the application system, and then predict the state distribution of each indicator of the application system based on the model in step S4 and the initial state in step S5.

[0014] S7. Based on the state distribution of each indicator predicted in step S6 and the relationship between the state space of each indicator determined in step S3 and the availability probability of the application system, predict the availability probability of the application system after a specified time.

[0015] Further, step S1 specifically includes: by statistically analyzing application system log data, and based on the accumulated experience of operations and maintenance personnel, determining indicators that have a significant impact on the availability of the application system, and establishing an indicator set E = {E1, E2, ..., E...} n}

[0016] Furthermore, step S2 specifically includes: the operation and maintenance personnel determining the specific status of each indicator based on experience or statistical data;

[0017] Assume E1 = (S1, N1), where S1 represents the state space of index E1, the state space represents the specific states contained in E1, and N1 represents the number of states of index E1. Then the state space of E1 is:

[0018] Similarly, we can conclude that:

[0019] The state space of E2=(S2,N2) is:

[0020] …

[0021] E n =(S n N n The state space of ) is:

[0022] Furthermore, step S3 specifically includes the following steps:

[0023] S31. Assuming that the system availability is only related to E1, the system availability probability is represented by U1. Then the system availability probability U1 will only be related to different states in the E1 state space.

[0024] S32, Use Let X1 be the vector of factors influencing index E1, satisfying: make

[0025] S33. Given the historical statistical data of the application system, only the index values ​​in the state space of E1 and the corresponding availability probability of the application system are taken as the initial dataset. Let: the calculated availability probability be... The actual availability probability is U1, the number of data rows is n, and the loss function is defined as:

[0026]

[0027] By continuously adjusting the vector values ​​in X1, data training and calculation are performed, and the vector value in X1 corresponding to the minimum value of L is taken as the influence factor vector of indicator E1.

[0028] S34. Repeat S31 to S33, using the same method, to calculate the influence factor vectors of E2, ..., En respectively. If the system availability probability is denoted as U, then the relationship between the system availability probability and various influencing indicators is expressed as follows:

[0029] U = min(S1*X1,S2*X2,…,S) n *X n ).

[0030] Furthermore, step S4 specifically includes the following steps:

[0031] S41. Calculate the state transition matrix of index E1. Taking E1 as an example, let A1 represent the state transition matrix of index E1. The elements in A1 are: Where C1(i,j) represents the number of times state j follows state i in the statistical data, and A1(i,j) represents the probability of transitioning from state i to state j; based on the statistics of historical data, the state transition matrix of index E1 is calculated.

[0032] S42. Repeat step S41 to calculate the state transition matrix for all indicators in step S1; assume indicators E1, E2, ..., E n The state transition matrices are: A1, A2, ..., A n Assuming the initial state distribution of E1 is π1(0), then according to the Markov property, the state distribution of the E1 index after iteration t is: π1(t)=π1(0)A1 t ;

[0033] Similarly, the state distribution of the E2 index after iteration t can be predicted as: π2(t)=π2(0)A2 t ,

[0034] Predicting E n The state distribution of the index after t iterations is: π n (t)=π n (0)A n t ;

[0035] S43. Define the iteration time for each metric.

[0036] Based on the characteristics of each indicator, an iteration time is defined for each indicator. Let the indicators be E1, E2, ..., E... n The iteration times are set to t1, t2, ..., t n .

[0037] Further, step S5 specifically includes: querying the initial states of all indicators based on the specific predicted demand time point; assuming at a certain moment, indicators E1, E2, ..., E n The initial state distributions are respectively

[0038] Further, step S6 specifically includes: predicting the availability probability of the application system after h hours. Then, the indicators E1, E2, ..., E... n After h hours, the state distribution of the indicators is as follows:

[0039] Further, step S7 specifically includes: obtaining the formula for the availability probability of the application system based on the formulas obtained in steps S3 and S6, as follows:

[0040]

[0041] Furthermore, after step S7, the method further includes: setting a threshold W based on the importance of the application system; when the availability probability U of the application system is less than the set threshold W, it indicates that the system availability probability will not meet the unit's usage requirements by the predicted time, and maintenance personnel need to conduct system problem investigation and pre-emptive maintenance.

[0042] (III) Beneficial Effects

[0043] This invention proposes an application system availability prediction method based on Markov chains and regression analysis. This invention uses a Markov chain prediction model and regression analysis to predict system failures. By identifying indicators that affect system stability, the Markov model is used to predict the state of these indicators. At the same time, regression analysis is used to determine the relationship between the state of the indicators and system availability, ultimately achieving the prediction objective. Detailed Implementation

[0044] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to examples.

[0045] S1. Determine the set of indicators that affect the probability of application system availability.

[0046] By analyzing application system log data and based on the experience accumulated by operations and maintenance personnel, indicators that have a significant impact on application system availability were identified, and the determined indicator set E = {E1, E2, ..., E...} n}

[0047] S2. Define the state space of each indicator in the indicator set.

[0048] Based on experience or statistical data, maintenance personnel determine the specific status included in each indicator.

[0049] Assume E1 = (S1, N1), where S1 represents the state space of index E1, the state space represents the specific states contained in E1, and n1 represents the number of states of index E1. Then the state space of E1 is:

[0050] Similarly, we can conclude that:

[0051] The state space of E2=(S2,N2) is:

[0052] …

[0053] E n =(S n N n The state space of ) is:

[0054] S3. Use regression analysis to determine the relationship between each influencing indicator and the probability of application system availability.

[0055] Based on statistical data, regression analysis is used to find the relationship between each state of each influencing indicator and the probability of system availability, thereby determining the influencing factor vector for each indicator. Different indicators have different numbers of states, therefore their state spaces differ, requiring individual calculation for each indicator.

[0056] S31. Assuming that system availability depends only on E1, and the system availability probability is represented by U1, then the system availability probability U1 will depend only on the different states in the E1 state space.

[0057] S32, Use Let X1 be the vector of factors influencing index E1, satisfying: make

[0058] S33. Given the historical statistical data of the application system, take only the index values ​​in the state space of E1 and the corresponding availability probability of the application system as the initial dataset. Let: the calculated availability probability be... The actual availability probability is U1, the number of data rows is n, and the loss function is defined as:

[0059]

[0060] By continuously adjusting the vector values ​​in X1 for data training and calculation, the vector value in X1 corresponding to the minimum L value is taken as the influence factor vector of indicator E1.

[0061] S34. Repeat S31 to S33, using the same method, to calculate E2, ..., E respectively. n Influence factor vector

[0062] If the system availability probability is denoted as U, then the relationship between the system availability probability and various influencing indicators is expressed as follows:

[0063] U = min(S1*X1,S2*X2,…,S) n *X n )

[0064] S4. Construct a model that uses a Markov chain model to predict the state distribution of each indicator.

[0065] S41. Calculate the state transition matrix of index E1.

[0066] Taking E1 as an example, let A1 represent the state transition matrix of index E1, and the elements in A1 are: Where C1(i,j) represents the number of times state j immediately follows state i in the statistical data, and A1(i,j) represents the probability of transitioning from state i to state j. Based on the statistics of historical data, the state transition matrix of index E1 can be calculated.

[0067] S42. Repeat step S41 to calculate the state transition matrix for all indicators in step S1. Assume indicators E1, E2, ..., E... n The state transition matrices are: A1, A2, ..., A n Assuming the initial state distribution of E1 is π1(0), then according to the Markov property, the state distribution of the E1 index after iteration t is: π1(t)=π1(0)A1 t .

[0068] Similarly, the state distribution of the E2 index after iteration t can be predicted as: π2(t)=π2(0)A2 t ,

[0069] Predicting E n The state distribution of the index after t iterations is: π n (t)=π n (0)A n t

[0070] S43. Define the iteration time for each metric.

[0071] Based on the characteristics of each metric, define an iteration time for each metric. For example, the iteration time for memory utilization is 2 seconds, and the iteration time for the number of online users of the application system is 10 minutes. Assume metrics E1, E2, ..., E... n The iteration times are set to t1, t2, ..., t n .

[0072] S5. Determine the initial state of each indicator.

[0073] Based on the specific forecast time point, query the initial state of all indicators. Assume that at a certain moment, indicators E1, E2, ..., E... n The initial state distributions are respectively

[0074] S6. Define the time frame for predicting the availability probability of the application system, and then, based on the model from step S4 and the initial state from step S5, predict the state distribution of each indicator of the application system.

[0075] This refers to predicting the availability probability of a certain application system after h hours. The indices E1, E2, ..., E... n After h hours, the state distribution of the indicators is as follows:

[0076] S7. Based on the state distribution of each indicator predicted in step S6 and the relationship between the state space of each indicator determined in step S3 and the availability probability of the application system, predict the availability probability of the application system after a specified time.

[0077] Based on the formulas obtained in steps S3 and S6, the formula for the availability probability of the application system is as follows:

[0078]

[0079] Depending on the importance of the application system, a threshold W can be set. When the availability probability U of the application system is less than the set threshold W, it proves that the system availability probability will not meet the unit's usage needs by the predicted time. Therefore, the operation and maintenance personnel can conduct system problem investigation in advance and carry out pre-operation and maintenance. Through pre-operation and maintenance, the usability of the application system can be effectively improved, the stability time of the application system can be extended, and user satisfaction can be improved.

[0080] This invention predicts system failures using a Markov chain prediction model and regression analysis. It identifies indicators affecting system stability, predicts the state of these indicators using a Markov model, and determines the relationship between these indicators and system availability through regression analysis, ultimately achieving the predictive goal. This invention effectively improves the usability of application systems, extends their stability duration, and enhances user satisfaction.

[0081] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the availability of application systems based on Markov chains and regression analysis, characterized in that, This method is used for application system fault prediction and includes the following steps: S1. Determine the set of indicators that affect the probability of application system availability. ={ , … The metrics include: memory utilization and number of online users. S2. Define the state space of each indicator in the indicator set; S3. Use regression analysis to determine the relationship between the state space of each indicator and the probability of application system availability. S4. Construct a model that uses a Markov chain model to predict the state distribution of each indicator; S5. Determine the initial state of each indicator; S6. Define the time range for predicting the availability probability of the application system, and then predict the state distribution of each indicator of the application system based on the model in step S4 and the initial state in step S5. S7. Based on the state distribution of each indicator predicted in step S6 and the relationship between the state space of each indicator determined in step S3 and the availability probability of the application system, predict the availability probability of the application system after a specified time. in, The state space is: ={ , , ..., }; Similarly, we can conclude that: The state space is: ={ , , ..., }; … The state space is: ={ , , ..., }; Step S3 specifically includes the following steps: S31. Assume that system availability is only related to Regarding this, the system availability probability at this point is expressed as: Then the system availability probability Will only with It relates to different states in the state space; S32, Use ={ , , ..., } represents the indicator Influence factor vector satisfy: …+ = 1. Order = ; S33. Given the historical statistical data of the application system, only take... Using the index values ​​in the state space and the availability probability of the application system as the initial dataset, let: the calculated availability probability be... The actual availability probability value is Given n rows of data, the loss function is defined as: Through continuous adjustments The vector values ​​in the data are used for training and calculation, and the value corresponding to the minimum L value is selected. Vector values ​​in the index Influence factor vector; S34. Repeat S31~S33, using the same method to calculate the results respectively. … Influence factor vector ={ , , ..., },…, ={ , , ..., The system availability probability is represented by U, and the relationship between the system availability probability and each influencing indicator is expressed as follows: ； Step S4 specifically includes the following steps: S41, Calculation Indicators The state transition matrix, for ,set up Indicators The state transition matrix, Elements in: ;in Indicates the state in statistical data Stay in the state The number of times after, Indicates from state Transition to state The probability; based on statistical analysis of historical data, the index is calculated. The state transition matrix; S42. Repeat step S41 to calculate the state transition matrix for all indicators in step S1; assume the indicators , … The state transition matrices are as follows: , … ; Assumption The initial state distribution is According to the Markov property, The state distribution of the index after t iterations is as follows: ; Similarly, it can be predicted The state distribution of the index after t iterations is as follows: , predict The state distribution of the index after t iterations is as follows: ; S43. Define the iteration time for each metric. Based on the characteristics of each metric, an iteration time is defined for each metric. Assume the metric... , , ..., The set iteration times are respectively , , ..., ; In step S5, the specific steps include: querying the initial state of all indicators based on the specific forecast demand time point; assuming at a certain moment, the indicators... , , ..., The initial state distributions are respectively , , ..., ; In step S6, the availability probability of the application system is predicted after h hours; then the index , , ..., After h hours, the state distribution of the indicators is as follows: , , ..., ; Step S7 specifically includes: Based on the formulas obtained in steps S3 and S6, the formula for the availability probability of the application system is as follows: U=min( * , * ,… * 。 2. The application system availability prediction method based on Markov chain and regression analysis as described in claim 1, characterized in that, Step S7 is followed by setting a threshold W based on the importance of the application system. When the availability probability U of the application system is less than the set threshold W, it proves that the system availability probability does not meet the unit's usage requirements by the predicted time, and maintenance personnel need to conduct system problem investigation and pre-operation maintenance in advance.

Images