Process monitoring based on large-scale combination of time series data

A computer program product, comprising a non-transitory machine-readable storage medium having encoded therein executable code of one or more software programs, wherein the one or more software programs when executed perform the following steps:

• generating one or more multivariate statistical models from time series data for a given target time series by (a) obtaining a given group of time series data on a given working compute node, wherein said given group comprises one of a plurality of groups of time series data selected from a larger set of time series data;
• (b) generating a measurement matrix, using said given working compute node, for said given group of time series data based on a plurality of time series selected for said given group and a plurality of time lags of said selected time series;
• (c) processing, using said given working compute node, said measurement matrix for said given group to generate a first linear model for said given group using an Orthogonal Matching Pursuit algorithm with a predefined number of first independent selected variables, each first independent selected variable corresponding to a specific time series and lag;
• (d) assigning a score, using said given working compute node, to each of said first independent selected variables based on an order of an absolute value of a coefficient of said first independent selected variable within said linear model of said plurality of time series selected for said given group and said corresponding plurality of time lags;
• and (e) providing said first independent selected variables and said corresponding assigned scores to a master compute node, wherein said master compute node obtains first independent selected variables and corresponding assigned scores for each of said plurality of groups from a plurality of said working compute nodes;
• ranks the said first independent selected variables according to said corresponding assigned scores for all of said plurality of groups;
• selects a predefined number of second independent selected variables based on a final rank to create a final group of time series;
• and processes said final group of time series to generate a final linear model using said Orthogonal Matching Pursuit algorithm;
• determining whether a first difference between one or more measured values and one or more predicted values based on said one or more multivariate statistical models exceeds a predefined target prediction threshold indicating a target prediction error;
• in response to a detected target prediction error, performing evaluations of (i) a neighborhood coherence comprising an average of variables of said one or more multivariate statistical models weighted by corresponding model variable coefficients on a predefined neighborhood time window, and (ii) a second difference between a given value of at least one given variable in said one or more multivariate statistical models and an average value of said at least one given variable in said one or more multivariate statistical models based on a training dataset;
• providing notifications in response to first predefined criteria based on said evaluations being satisfied;
• and
• updating said one or more multivariate statistical models in response to second predefined criteria based on said evaluations being satisfied.

A system, comprising:

• a memory;
• and
• at least one hardware device, coupled to the memory, operative to implement the following steps:
• generating one or more multivariate statistical models from time series data for a given target time series by (a) obtaining a given group of time series data on a given working compute node, wherein said given group comprises one of a plurality of groups of time series data selected from a larger set of time series data;
• (b) generating a measurement matrix, using said given working compute node, for said given group of time series data based on a plurality of time series selected for said given group and a plurality of time lags of said selected time series;
• (c) processing, using said given working compute node, said measurement matrix for said given group to generate a first linear model for said given group using an Orthogonal Matching Pursuit algorithm with a predefined number of first independent selected variables, each first independent selected variable corresponding to a specific time series and lag;
• (d) assigning a score, using said given working compute node, to each of said first independent selected variables based on an order of an absolute value of a coefficient of said first independent selected variable within said linear model of said plurality of time series selected for said given group and said corresponding plurality of time lags;
• and (e) providing said first independent selected variables and said corresponding assigned scores to a master compute node, wherein said master compute node obtains first independent selected variables and corresponding assigned scores for each of said plurality of groups from a plurality of said working compute nodes;
• ranks the said first independent selected variables according to said corresponding assigned scores for all of said plurality of groups;
• selects a predefined number of second independent selected variables based on a final rank to create a final group of time series;
• and processes said final group of time series to generate a final linear model using said Orthogonal Matching Pursuit algorithm;
• determining whether a first difference between one or more measured values and one or more predicted values based on said one or more multivariate statistical models exceeds a predefined target prediction threshold indicating a target prediction error;
• in response to a detected target prediction error, performing evaluations of (i) a neighborhood coherence comprising an average of variables of said one or more multivariate statistical models weighted by corresponding model variable coefficients on a predefined neighborhood time window, and (ii) a second difference between a given value of at least one given variable in said one or more multivariate statistical models and an average value of said at least one given variable in said one or more multivariate statistical models based on a training dataset;
• providing notifications in response to first predefined criteria based on said evaluations being satisfied;
• and
• updating said one or more multivariate statistical models in response to second predefined criteria based on said evaluations being satisfied.