Difference between revisions of "PMQ"
| Line 5: | Line 5: | ||
For sepecific device or sensor data(HDD Health, System CPU Usage and so on), RMM offers prediction function(Trend Forecasting Monitor). It can help users know abnormal data status as soon as possible to prevent some damages.Predictive Tyoe of Trend Forecasting Monitor are hourly averages of data, and they contain averages of current hour and even next 3 hours. | For sepecific device or sensor data(HDD Health, System CPU Usage and so on), RMM offers prediction function(Trend Forecasting Monitor). It can help users know abnormal data status as soon as possible to prevent some damages.Predictive Tyoe of Trend Forecasting Monitor are hourly averages of data, and they contain averages of current hour and even next 3 hours. | ||
| − | <span style="font-size:medium;">'''Predictive Process'''</span> | + | <span style="font-size:medium;"></span><span style="font-size:medium;">'''Predictive Process'''</span> |
The first condiction is predictive object has at least the newest 48 history hourly averages in Database, and these samples are calculated by time-series prediction algorithm. Time-series prediction algorithm RMM adopts is Exponential Smoothing, and is based on APIs in Weka machinne learning java libary. By the above Process, it can produce values of prediction(averages of current hour and next 3 hours), and Trend Forecasting Monitor takes these values and the newest 3 history hourly averages to compare with hourly averages in yesterday. If there are significant differences between values of prediction and hourly averages in yesterday, the status of analytic result is "Potential Risk". If there are significant differences between the newest 3 history hourly averages and hourly averages in yesterday, the status is "Doubtful Risk". The standard of significant differences is based on Tukey's Range Test, and the formula is: <span style="font-size:small;">[Q<sub>1</sub>-3(Q<sub>3</sub>-Q<sub>1</sub>), Q3+3(Q<sub>3</sub>-Q<sub>1</sub>)].</span>Q<sub>1</sub> and Q<sub><span style="font-size: 12px;">3</span></sub> are the lower and upper quartile of hourly averages in yesterday. If objects are in the above range, there are not significant differences between them., then we can call status of Object is "Normal". | The first condiction is predictive object has at least the newest 48 history hourly averages in Database, and these samples are calculated by time-series prediction algorithm. Time-series prediction algorithm RMM adopts is Exponential Smoothing, and is based on APIs in Weka machinne learning java libary. By the above Process, it can produce values of prediction(averages of current hour and next 3 hours), and Trend Forecasting Monitor takes these values and the newest 3 history hourly averages to compare with hourly averages in yesterday. If there are significant differences between values of prediction and hourly averages in yesterday, the status of analytic result is "Potential Risk". If there are significant differences between the newest 3 history hourly averages and hourly averages in yesterday, the status is "Doubtful Risk". The standard of significant differences is based on Tukey's Range Test, and the formula is: <span style="font-size:small;">[Q<sub>1</sub>-3(Q<sub>3</sub>-Q<sub>1</sub>), Q3+3(Q<sub>3</sub>-Q<sub>1</sub>)].</span>Q<sub>1</sub> and Q<sub><span style="font-size: 12px;">3</span></sub> are the lower and upper quartile of hourly averages in yesterday. If objects are in the above range, there are not significant differences between them., then we can call status of Object is "Normal". | ||
Revision as of 15:37, 5 March 2017
Trend Forecasting Monitor
Function
For sepecific device or sensor data(HDD Health, System CPU Usage and so on), RMM offers prediction function(Trend Forecasting Monitor). It can help users know abnormal data status as soon as possible to prevent some damages.Predictive Tyoe of Trend Forecasting Monitor are hourly averages of data, and they contain averages of current hour and even next 3 hours.
Predictive Process
The first condiction is predictive object has at least the newest 48 history hourly averages in Database, and these samples are calculated by time-series prediction algorithm. Time-series prediction algorithm RMM adopts is Exponential Smoothing, and is based on APIs in Weka machinne learning java libary. By the above Process, it can produce values of prediction(averages of current hour and next 3 hours), and Trend Forecasting Monitor takes these values and the newest 3 history hourly averages to compare with hourly averages in yesterday. If there are significant differences between values of prediction and hourly averages in yesterday, the status of analytic result is "Potential Risk". If there are significant differences between the newest 3 history hourly averages and hourly averages in yesterday, the status is "Doubtful Risk". The standard of significant differences is based on Tukey's Range Test, and the formula is: [Q1-3(Q3-Q1), Q3+3(Q3-Q1)].Q1 and Q3 are the lower and upper quartile of hourly averages in yesterday. If objects are in the above range, there are not significant differences between them., then we can call status of Object is "Normal".
Predictive Result
Result for each object contains 3 parts: result status, status information and event record list. First, for result status. except "Potential Risk", "Doubtful Risk" and "Normal" in the above statement, there is a "Data Error" situation, and it means disable predicting because of some data problem. For instance, miss some data in the newest 48 hours or query data from database error. Second, according to result status, web UI will show corresponding information, and major statement is explain reason for status and give some suggestion to users.Final, event record list will show total records for significant change, then it can helps users to trace initial time of abnormal data.