Processing event data streams to recognize event patterns, with conditional query instance shifting for load balancing

Abstract

A computer ( 100 ) recognizes an event pattern (ABC) for objects ( 14, 24, 34, 44 ) that belong to an event domain ( 150 ). The computer activates a pattern query ( 110 ) that corresponds to the event pattern (ABC) and that has a least one state (S 4 ) with an distinctive transition probability to a final state. The probability is derived from object observations in the event domain ( 150 ). The computer continuously receives event representations (*A 14, *A 24, *A 44, *D 44, . . . ) that are related to the objects ( 14, 24, 34, 44 ) and allocates the event representations to a first processing resource ( 101 ) to initiate instances ( 1.14, 1.24, 1.34, 1.44 ) of the query ( 110 ). It monitors the instances and, upon receiving event representations that cause the instances to reach the state (S 4 ) with the distinct transition probability, it shifts the instances to a second processing resource ( 102 ).

1 . Computer-implemented method ( 300 ) for recognizing an event pattern (ABC) for objects ( 14 ,  24 ,  34 ,  44 ) that belong to an event domain ( 150 ), the method ( 300 ) comprising:
 activating ( 310 ) a pattern query ( 110 ) that corresponds to the event pattern (ABC) and that has a least one state (S 4 ) with a distinctive transition probability (P(S 4 /S 2 ), P(S 2 /S 3 )) to a final query state (S 3 ), the probability being derived from object observations in the event domain ( 150 );   continuously receiving ( 320 ) event representations (*A 14 , *A 24 , *A 44 , *D 44 , . . . ), that are related to the objects ( 14 ,  24 ,  34 ,  44 );   allocating ( 330 ) the event representations (*A 14 , *A 24 , *A 44 , *D 44 , . . . ) to a first processing resource ( 101 ) to initiate ( 331 ) instances ( 1 . 14 ,  1 . 24 ,  1 . 34 ,  1 . 44 ) of the query ( 110 );   monitoring ( 340 ) the instances and, upon receiving event representations hat is cause the instances to reach the state (S 4 ) with the distinctive transition probability (P(S 4 /S 2 ), P(S 2 /S 3 )), shifting ( 342 ) the instances to a second processing resource ( 102 ).   
     
     
         2 . The method ( 300 ) of  claim 1 , wherein the query ( 110 ) is activated ( 310 ) such that the state (S 4 ) with the distinctive transition probability (P(S 4 /S 2 )) is not associated with the event pattern (ABC). 
     
     
         3 . Method ( 300 ) of  claim 1 , wherein
 in activating ( 310 ), the query ( 110 ) has states (S 1 , S 2 , S 3 , S 4 ) and event-driven state transitions (S 1 /S 2 , S 2 /S 3 , S 1 /S 4 , S 4 /S 2 ), wherein transition probabilities to the state (S 3 ) classify the states into a first group ( 201 ) of states (S 1 , S 2 , S 3 ) and a second group ( 202 ) of states (S 4 ), the transition probabilities being derived from object observations in the event domain ( 150 ), and the second group ( 202 ) having the at least one state with the distinctive state transition probability (P(S 4 /S 2 ));   in receiving ( 320 ), event representations (*A 14 , *A 24 , *A 44 , *D 44 , . . . ), the event representations are being received with object identifiers (* 14 , * 24 , * 34 , * 44 );   in allocating ( 330 ), the event representations (*A 14 , *A 24 , *A 44 , *D 44 , . . . ) to the first processing resource ( 101 ), the event representations are being allocated separately for different object identifiers (* 14 , * 24 , * 34 , * 44 );   in monitoring ( 340 ), the instances ( 1 . 14 ,  1 . 24 ,  1 . 34 ,  1 . 44 ) are being monitored according to the classification of the states, so that instances ( 1 . 14 ,  1 . 24 ) that reach states of the first group ( 201 , S 1 , S 2 , S 3 ), continue to be executed ( 332 ) by the first processing resource ( 101 ).   
     
     
         4 . The method ( 300 ) according to any one of the preceding claims, applied to objects ( 14 ,  24 ,  34 ,  44 ) that interact with sensors (A, B, C, D) so that the event representations (*A* 14 , *B* 14  . . . *D* 44 ) are received from the sensors (A, B, C, D). 
     
     
         5 . The method ( 300 ) according to any one of the preceding claims, wherein the is sensors (A, B, C, D) submit representations (*tA, *tB, *tC) of the event time (tA, tB, tC). 
     
     
         6 . The method ( 300 ) according to any one of the preceding claims, wherein executing the instances stops upon reaching a maximum time (Tmax). 
     
     
         7 . The method ( 300 ) according to any one of  claims 3  to  5 , wherein monitoring ( 340 ) comprises to monitor the shifted instances ( 2 . 34 ,  2 . 44 ) according to the classification of the states, for shifted instances that reach states of the first group ( 201 , S 1 , S 2 , S 3 ), re-shifting ( 343 ) the previously shifted instances ( 2 . 34 ,  2 . 44 ) to the first processing resource ( 101 ). 
     
     
         8 . The method ( 300 ) according to any of  claims 3  to  7 , wherein the query ( 110 ) is activated ( 310 ) in combination with an event relation model ( 120 ) in that most of the transition probabilities (P(S 1 /S 2 ), P(S 2 /S 3 ), P(S 1 /S 4 )) for a state that belongs to the first group (S 1 , S 2 , S 3 ) to a state that belongs to the second group (S 4 ) are different than the transition probabilities (P(S 4 /S 2 ) for the state that belongs to the second group (S 4 ) to the states that belongs to the first group (S 1 , S 2 , S 3 ). 
     
     
         9 . The method ( 300 ) according to any one of the preceding claims, wherein receiving ( 320 ) event representations (*A* 14 , *B* 14  . . . *D* 44 ) comprises to re-calculate ( 321 ) transition probabilities and re-classifying ( 322 ) the states into groups ( 201 ,  202 ). 
     
     
         10 . The method ( 300 ) according to  claim 9 , wherein monitoring ( 340 ) comprises to use the re-classified states. 
     
     
         11 . A computer program product that—when loaded into a memory of a computing device and being executed by at least one processor of the computing device—performs the steps of the computer-implemented method according to any of  claims 1  to  10 . 
     
     
         12 . Computer ( 500 ) for recognizing an event pattern (ABC) for objects ( 14 ,  24 ,  34 ,  44 ) that belong to an event domain ( 150 ), the computer ( 500 ) comprising:
 a pattern query activator ( 510 ) that corresponds to the event pattern (ABC) and that activates a pattern query ( 110 ) that has at least one state (S 4 ) with an distinctive transition probability to a final state (S 3 ) of the query ( 110 ), the probability being derived from object observations in the event domain ( 150 );   a receiver ( 520 ) that continuously receives ( 320 ) event representations (*A 14 , *A 24 , *A 44 , *D 44 , . . . ), that are related to the objects ( 14 ,  24 ,  34 ,  44 );   an allocator ( 530 ) that allocates the event representations (*A 14 , *A 24 , *A 44 , *D 44 , . . . ) to a first processing resource ( 501 ) of the computer ( 500 ) to initiate ( 331 ) instances ( 1 . 14 ,  1 . 24 ,  1 . 34 ,  1 . 44 ) of the query ( 110 ); and   a monitor ( 540 ) that monitors ( 340 ) the instances and, upon receiving event representations that causes the instances to reach the state (S 4 ) with the distinctive transition probability (P(S 4 /S 2 )), shifts ( 342 ) the instances to a second processing resource ( 502 ) of the computer ( 500 ).   
     
     
         13 . The computer ( 500 ) according to  claim 12 , wherein the pattern query activator ( 510 ) activates a query in that the state (S 4 ) with the distinctive transition probability (P(S 4 /S 2 )) is not associated with the event pattern (ABC). 
     
     
         14 . The computer ( 500 ) according to  claim 12  or  13 , wherein
 the pattern query activator ( 510 ) uses queries with states (S 1 , S 2 , S 3 , S 4 ) and event-driven state transitions (S 1 /S 2 , S 2 /S 3 , S 1 /S 4 , S 4 /S 2 ), wherein transition probabilities (P(S 1 /S 2 ), P(S 2 /S 3 ), P(S 1 /S 4 ), P(S 4 /S 2 )) classify the states into a first group ( 201 ) of states (S 1 , S 2 , S 3 ) and a second group ( 202 ) of states (S 4 ), the transition probabilities being derived from object observations in the event domain ( 150 ), and the second group ( 202 ) having the at least one state with the is distinctive state transition probability (P(S 4 /S 2 )); 
 the receiver ( 520 ) receives the event representations (*A 14 , *A 24 , *A 44 , *D 44 , . . . ) with object identifiers (* 14 , * 24 , * 34 , * 44 ); 
 the allocator ( 530 ) allocates the event representations (*A 14 , *A 24 , *A 44 , *D 44 , . . . ) separately for different object identifiers (* 14 , * 24 , * 34 , * 44 ); and 
 the monitor ( 540 ) monitors the instances according to the classification of the states, so that instances ( 1 . 14 ,  1 . 24 ) that reach states of the first group ( 201 , S 1 , S 2 , S 3 ), continue to be executed ( 332 ) by the first processing resource ( 501 ). 
 
     
     
         15 . The computer ( 500 ) according to any of  claims 12  to  14 , wherein the receiver ( 520 ) re-calculates ( 321 ) transition probabilities and re-classifies ( 322 ) the states into groups ( 201 ,  202 ).