Monitoring multiple system indicators
Multiple system indicators to be monitored in a computing system are determined. Training data based on historical data of the multiple system indicators is obtained. A long short-term memory (LSTM) model is trained using the training data. The multiple system indicators are monitored in the computing system based on the LSTM model.
1. A computer-implemented method arranged to monitor operation of an insurance system, comprising:
storing, by the insurance system, multiple system indicators in an indicator library, wherein at least one system indicator in the multiple system indicators is used to indicate a state or level of the insurance system, wherein the multiple system indicators include a number of sold insurance policies, a billing amount of sold insurance policies, a number of filed claims, a number of denied claims, and a total settlement amount, and wherein the insurance system performs multi-indicator joint monitoring using a long short-term memory (LSTM) model;
storing, by the insurance system, historical data of the multiple system indicators in a database of the insurance system;
determining, by the insurance system, a plurality of system indicators from the multiple system indicators to be monitored by the insurance system, wherein a first system indicator x of the plurality of system indicators is correlated with one or more second system indicators of the plurality of system indicators, and wherein the plurality of system indicators are selected by a user of the insurance system and from the indicator library using a graphical user interface (GUI) of the insurance system;
after determining the plurality of system indicators to be monitored by the insurance system, obtaining, by the insurance system, training data based on historical data of the plurality of system indicators, wherein a historical time period is selected by the user of the insurance system using the GUI of the insurance system, and wherein a portion of the historical data of the plurality of system indicators that is within the historical time period is obtained as the training data;
after obtaining the training data based on the historical data of the plurality of system indicators, arranging, by the insurance system, the training data in a chronological order, wherein arranging the training data in the chronological order is performed for one system indicator at a time;
after arranging the training data in the chronological order, storing, by the insurance system, the training data in the database of the insurance system;
training, by the insurance system, the LSTM model as a trained LSTM model using the training data;
simultaneously monitoring, by the insurance system, the plurality of system indicators based on the trained LSTM model, wherein simultaneously monitoring the plurality of system indicators based on the trained LSTM model comprises:
automatically obtaining real-time data of the plurality of system indicators;
automatically obtaining a second portion of the historical data of the plurality of system indicators within a second historical time period;
automatically inputting the second portion of the historical data of the plurality of system indicators within the second historical time period into the trained LSTM model;
automatically obtaining predicted data of the plurality of system indicators from the trained LSTM model; and
automatically comparing the real-time data of the plurality of system indicators with the predicted data of the plurality of system indicators; and
raising, by the insurance system, an alarm in response to detecting one or more co-movement anomalies between the first system indicator x and the one or more second system indicators, wherein a co-movement anomaly is detected when an anomaly has occurred in one of the one or more second system indicators and has not occurred in the first system indicator x.
2. The computer-implemented method of claim 1 , wherein obtaining the training data based on the historical data of the plurality of system indicators comprises:
determining the historical time period; and
for each particular system indicator of the plurality of system indicators:
collecting historical data of the particular system indicator within the historical time period; and
arranging, in the chronological order, the historical data of the particular system indicator within the historical time period.
3. The computer-implemented method of claim 2 , wherein the historical time period includes n predetermined time intervals, the plurality of system indicators include the first system indicator x and a second system indicator y, and the method further comprises:
for each particular predetermined time interval in the historical time period, constructing a matrix i, wherein i is the particular predetermined time interval in the historical time period, t is a data cycle, x i is historical data of the first system indicator x at the particular predetermined time interval i, and y i is historical data of the second system indicatory at the particular predetermined time interval i; and
constructing the training data using the constructed matrices.
4. The computer-implemented method of claim 3 , wherein training the LSTM model using the training data comprises input of each matrix i into the LSTM model with
[
x
i
y
i
]
input as a label dataset and
[
x
i
-
t
…
x
i
-
1
x
i
y
i
-
t
…
y
i
-
1
y
i
]
,
input as a feature dataset.
5. The computer-implemented method of claim 1 , wherein comparing the real-time data of the plurality of system indicators with the predicted data of the plurality of system indicators comprises:
for each particular system indicator of the plurality of system indicators, calculating an absolute value of a difference between the real-time data of the particular system indicator and the predicted data of the particular system indicator;
summing up, as summed absolute values, the calculated absolute values for the plurality of system indicators; and
determining the one or more co-movement anomalies when the summed absolute values exceeds a predetermined threshold.
6. The computer-implemented method of claim 1 , further comprising:
determining that one or more system indicators are to be added to or removed from the plurality of system indicators to generate an updated plurality of system indicators;
obtaining updated training data based on historical data of the updated plurality of system indicators; and
in response to a user instruction, retraining the LSTM model using the updated training data.
7. A non-transitory, computer-readable medium storing one or more instructions executable by an insurance system, the one or more instructions instructing the insurance system to perform operations comprising:
storing multiple system indicators in an indicator library, wherein at least one system indicator in the multiple system indicators is used to indicate a state or level of the insurance system, wherein the multiple system indicators include a number of sold insurance policies, a billing amount of sold insurance policies, a number of filed claims, a number of denied claims, and a total settlement amount, and wherein the insurance system performs multi-indicator joint monitoring using a long short-term memory (LSTM) model;
storing historical data of the multiple system indicators in a database of the insurance system;
determining a plurality of system indicators from the multiple system indicators to be monitored by the insurance system, wherein a first system indicator x of the plurality of system indicators is correlated with one or more second system indicators of the plurality of system indicators, and wherein the plurality of system indicators are selected by a user of the insurance system and from the indicator library using a graphical user interface (GUI) of the insurance system;
after determining the plurality of system indicators to be monitored by the insurance system, obtaining training data based on historical data of the plurality of system indicators, wherein a historical time period is selected by the user of the insurance system using the GUI of the insurance system, and wherein a portion of the historical data of the plurality of system indicators that is within the historical time period is obtained as the training data;
after obtaining the training data based on the historical data of the plurality of system indicators, arranging the training data in a chronological order, wherein arranging the training data in the chronological order is performed for one system indicator at a time;
after arranging the training data in the chronological order, storing the training data in the database of the insurance system;
training the LSTM model as a trained LSTM model using the training data;
simultaneously monitoring the plurality of system indicators based on the trained LSTM model, wherein simultaneously monitoring the plurality of system indicators based on the trained LSTM model comprises:
automatically obtaining real-time data of the plurality of system indicators;
automatically obtaining a second portion of the historical data of the plurality of system indicators within a second historical time period;
automatically inputting the second portion of the historical data of the plurality of system indicators within the second historical time period into the trained LSTM model;
automatically obtaining predicted data of the plurality of system indicators from the trained LSTM model; and
automatically comparing the real-time data of the plurality of system indicators with the predicted data of the plurality of system indicators; and
raising an alarm in response to detecting one or more co-movement anomalies between the first system indicator x and the one or more second system indicators, wherein a co-movement anomaly is detected when an anomaly has occurred in one of the one or more second system indicators and has not occurred in the first system indicator x.
8. The non-transitory, computer-readable medium of claim 7 , wherein obtaining the training data based on the historical data of the plurality of system indicators comprises:
determining the historical time period; and
for each particular system indicator of the plurality of system indicators:
collecting historical data of the particular system indicator within the historical time period; and
arranging, in the chronological order, the historical data of the particular system indicator within the historical time period.
9. The non-transitory, computer-readable medium of claim 8 , wherein the historical time period includes n predetermined time intervals, the plurality of system indicators include the first system indicator x and a second system indicator y, and the operations further comprise:
for each particular predetermined time interval in the historical time period, constructing a matrix i, wherein i is the particular predetermined time interval in the historical time period, t is a data cycle, x i is historical data of the first system indicator x at the particular predetermined time interval i, and y i is historical data of the second system indicatory at the particular predetermined time interval i; and
constructing the training data using the constructed matrices.
10. The non-transitory, computer-readable medium of claim 9 , wherein training the LSTM model using the training data comprises input of each matrix i into the LSTM model with
[
x
i
y
i
]
input as a label dataset and
[
x
i
-
t
…
x
i
-
1
y
i
-
t
…
y
i
-
1
]
input as a feature dataset.
11. The non-transitory, computer-readable medium of claim 7 , wherein comparing the real-time data of the plurality of system indicators with the predicted data of the plurality of system indicators comprises:
for each particular system indicator of the plurality of system indicators, calculating an absolute value of a difference between the real-time data of the particular system indicator and the predicted data of the particular system indicator;
summing up, as summed absolute values, the calculated absolute values for the plurality of system indicators; and
determining the one or more co-movement anomalies when the summed absolute values exceeds a predetermined threshold.
12. The non-transitory, computer-readable medium of claim 7 , the operations further comprising:
determining that one or more system indicators are to be added to or removed from the plurality of system indicators to generate an updated plurality of system indicators;
obtaining updated training data based on historical data of the updated plurality of system indicators; and
in response to a user instruction, retraining the LSTM model using the updated training data.
13. A computer-implemented system, comprising:
one or more computers; and
one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform one or more operations comprising:
storing multiple system indicators in an indicator library, wherein at least one system indicator in the multiple system indicators is used to indicate a state or level of an insurance system, wherein the multiple system indicators include a number of sold insurance policies, a billing amount of sold insurance policies, a number of filed claims, a number of denied claims, and a total settlement amount, and wherein the insurance system performs multi-indicator joint monitoring using a long short-term memory (LSTM) model;
storing historical data of the multiple system indicators in a database of the insurance system;
determining a plurality of system indicators from the multiple system indicators to be monitored by the insurance system, wherein a first system indicator x of the plurality of system indicators is correlated with one or more second system indicators of the plurality of system indicators, and wherein the plurality of system indicators are selected by a user of the insurance system and from the indicator library using a graphical user interface (GUI) of the insurance system;
after determining the plurality of system indicators to be monitored by the insurance system, obtaining training data based on historical data of the plurality of system indicators, wherein a historical time period is selected by the user of the insurance system using the GUI of the insurance system, and wherein a portion of the historical data of the plurality of system indicators that is within the historical time period is obtained as the training data;
after obtaining the training data based on the historical data of the plurality of system indicators, arranging the training data in a chronological order, wherein arranging the training data in the chronological order is performed for one system indicator at a time;
after arranging the training data in the chronological order, storing the training data in the database of the insurance system;
training the LSTM model as a trained LSTM model using the training data;
simultaneously monitoring the plurality of system indicators based on the trained LSTM model, wherein simultaneously monitoring the plurality of system indicators based on the trained LSTM model comprises:
automatically obtaining real-time data of the plurality of system indicators;
automatically obtaining a second portion of the historical data of the plurality of system indicators within a second historical time period;
automatically inputting the second portion of the historical data of the plurality of system indicators within the second historical time period into the trained LSTM model;
automatically obtaining predicted data of the plurality of system indicators from the trained LSTM model; and
automatically comparing the real-time data of the plurality of system indicators with the predicted data of the plurality of system indicators; and
raising an alarm in response to detecting one or more co-movement anomalies between the first system indicator x and the one or more second system indicators, wherein a co-movement anomaly is detected when an anomaly has occurred in one of the one or more second system indicators and has not occurred in the first system indicator x.
14. The computer-implemented system of claim 13 , wherein obtaining the training data based on the historical data of the plurality of system indicators comprises:
determining the historical time period; and
for each particular system indicator of the plurality of system indicators:
collecting historical data of the particular system indicator within the historical time period; and
arranging, in the chronological order, the historical data of the particular system indicator within the historical time period.
15. The computer-implemented system of claim 14 , wherein the historical time period includes n predetermined time intervals, the plurality of system indicators include the first system indicator x and a second system indicatory, and the operations further comprise:
for each particular predetermined time interval in the historical time period, constructing a matrix i, wherein i is the particular predetermined time interval in the historical time period, t is a data cycle, x i is historical data of the first system indicator x at the particular predetermined time interval i, and y 1 is historical data of the second system indicatory at the particular predetermined time interval i; and
constructing the training data using the constructed matrices.