MSDN > 論壇首頁 > Data Mining > New to Datamining using SQL 2005. Trying to do a forecast for oil production but still missing the boat
發問發問
 

已答覆New to Datamining using SQL 2005. Trying to do a forecast for oil production but still missing the boat

  • 2009年7月2日 上午 05:08Tomas Santandreu 使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     
    Hi and thank you in advance

    I have the following data
    WellID
    WellName
    OilVolume
    GasVolume
    ProductionDate
    Downtime   how long the well was down
    DownTimeReason
    LocationHierarchy(Location,Platform,Asset,Well)

    1) I am trying to do predictive analysis on this data, for future well production
    2) Do predictive analysis of downtime cause and when it might occur (downtime reason can be, technical cause, weather cause, etc)

    I am trying to see how it can be done, what algorithm to use and how to decipher the end result.
    Thank you


    Tomas

解答

所有回覆

  • 2009年7月2日 上午 08:15Thomas IvarssonMVP, 版主使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     
    Hi,

    I would start with the data mining tutorials that are part of Books On Line for SQL Server 2005. I guess that Decision Trees can be a good start. They work with all types of data. If downtime is in hours you can discretize hours into buckets like 1-5 hours, 5-10 hours and so on. In Decision Trees you set downtime as predict and the other attributes as input.

    HTH
    Thomas Ivarsson
  • 2009年7月2日 下午 03:27Tomas Santandreu 使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     

    Hi
    thank you. But I was trying to see if in SQL 2005 there was a forecasting model like there is in SQL 2008. I just can not find it anywhere.
    Thank you

    Tomas


    Tomas
  • 2009年7月2日 下午 07:21Thomas IvarssonMVP, 版主使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     

    With decision trees you train a model first and when new data arrives you use that model for prediction. As I understood you problem you would like to see what attributes that can explain downtime and then put in new data in to this model to make predictions. Forecast or Time series as the algorithm is called would probably not work in that scenario.

    BR
    Thomas Ivarsson

  • 2009年7月3日 下午 08:00Allan MitchellMVP使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     
    Hi and thank you in advance

    I have the following data
    WellID
    WellName
    OilVolume
    GasVolume
    ProductionDate
    Downtime   how long the well was down
    DownTimeReason
    LocationHierarchy(Location,Platform,Asset,Well)

    1) I am trying to do predictive analysis on this data, for future well production
    2) Do predictive analysis of downtime cause and when it might occur (downtime reason can be, technical cause, weather cause, etc)

    I am trying to see how it can be done, what algorithm to use and how to decipher the end result.
    Thank you


    Tomas
    The Time Series model type will give you an idea of future predicted volumes. I am not sure you have the right variables for predicting downtime. Even though you have the "reason" I am not sure this will be specific enough. It looks like this is an LPO system. Do you have any of the operating stats to help with the prediction of downtime? type of well, type of lift, associated plant etc etc. Allan
  • 2009年7月4日 上午 01:09Tomas Santandreu 使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     

    Hi Allan

    Yes I do have those attributes such as . Type of Well (Oil or Gas) , type of Lift (GL,FL)..but my issue is trying to predict downtime by well , how much of downtime, and then predict or Forecast Oil Production utilizing downtime information. Please note that at this moment I am using SQL 2005 still, 2008 is not until later. The same goes with Excel 2003 and Data Mining add-in.
    Thank you for any help

    Tomas


    Tomas
  • 2009年7月4日 上午 08:12Allan MitchellMVP使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     

    Hi Allan

    Yes I do have those attributes such as . Type of Well (Oil or Gas) , type of Lift (GL,FL)..but my issue is trying to predict downtime by well , how much of downtime, and then predict or Forecast Oil Production utilizing downtime information. Please note that at this moment I am using SQL 2005 still, 2008 is not until later. The same goes with Excel 2003 and Data Mining add-in.
    Thank you for any help

    Tomas


    Tomas
    Tomas 2K5 has the time series Algorithm (using ARTxP). To predict if a well is going to go "pop" Take the system attributes we have spoken about. also take the date attribute and break it up into constituent parts. Hour of day etc, day of week etc. My model would look something like The attributes would be the values some period before the actual well goes pop, maybe a day. It is pointless doing mining, learning the well may go pop and it will do so in 20 seconds as you will have no way of reacting. Make sense? As for downtime. You could start with averaging values by day. It may be that you can see a trend building over time with your system measurements that allows us to model using time how long you will have as downtime in the coming periods. You can also look at other algorithms like DTs and NN. HTH allan
  • 2009年7月4日 上午 08:28Allan MitchellMVP使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     已答覆
    it would seem from my previous post that things got stripped

    My model might look something like

    AttributeA
    AttributeB
    etc
    WellGoesPop PREDICT
    WellDowntime PREDICT


  • 2009年10月17日 上午 03:37Tomas Santandreu 使用者勳章使用者勳章使用者勳章使用者勳章使用者勳章
     
    Thank you for all the help.

    Tomas
    Tomas