TWI747452B - System, method and storage medium for intelligent monitoring of case field anomaly detection using artificial intelligence - Google Patents

Abstract

A method for intelligent monitoring of case field anomaly detection using artificial intelligence, which includes (a) receiving a case field sensing data set and obtaining an input data set based on it; (b) applying the input data set to the first prediction The engine outputs the prediction result to predict whether an abnormal event occurs; (c) For the prediction result indicating that a predicted abnormal event occurs, use the input data set to generate a notification message for the predicted abnormal event; (d) Receive and associate the notification message (E) Update the first prediction engine based on the at least one feedback message to obtain the second prediction engine; wherein the second prediction engine is used to replace the first prediction engine based on the subsequent input data set of the case To predict whether there will be any abnormal events.

Description

System and method for intelligent monitoring of case field abnormality detection by artificial intelligence And storage media

The present invention relates to a technology for detecting anomaly in a case, and more particularly to an intelligent monitoring system, method and storage medium for detecting anomaly in a case using artificial intelligence.

Case site anomaly detection technology analyzes and detects various data collected in the case site, and then determines whether an abnormality has occurred, so as to issue warnings or notices to the case site personnel, thereby reducing or avoiding case site operations The risk of loss or danger can also promote the efficiency of operating businesses. For example, the monitoring system of a solar power plant or the monitoring system of a manufacturing product factory warns of abnormal occurrences, and dispatches technicians to the site to conduct inspections, repairs, and maintenance in response to the warnings to deal with or solve the abnormal situation.

The abnormal judgment method of the conventional monitoring system is generally based on the change rule of the data detected in the case when the abnormality occurs. For example, the temperature of a certain machine in the case is higher than 80℃ and the output voltage is lower than the rated 12V. , It means that the machine is abnormal. When a new abnormality occurs in the case, on the one hand, technicians need to find out the rule of thumb for judging the abnormality, on the other hand, the programmer needs to adjust or rewrite the program of the monitoring system according to the found rule of thumb to respond. The above process involves the work of multiple personnel, and the rule of thumb for judging abnormalities is not easy to obtain or may be difficult to quantitatively describe, and the program of the monitoring system that is adjusted or rewritten also needs to be tested or verified repeatedly before it can be officially launched. It can be seen that the above-mentioned method is inefficient, and the conventional monitoring system has systematic inflexibility and scalability in terms of reflecting the update or change of abnormal problems.

In view of the above-mentioned shortcomings of the prior art, one purpose of the present invention is to provide a technology for intelligent monitoring of case field abnormality detection using artificial intelligence. In this way, the system implemented according to this technology can have flexibility and expandability in terms of reflecting the update or change of abnormal problems.

In order to achieve at least the above objectives, the present invention provides an intelligent monitoring system for case field abnormality detection using artificial intelligence, which includes: at least one processing unit; and at least one storage device, coupled to the at least one processing unit and storing multiple Instructions, when the instructions are executed by the at least one processing unit, the at least one processing unit realizes multiple operations, the operations including: (a) receiving a set of sensing data from a case site and obtaining an input data accordingly (B) The input data set is applied to a first prediction engine and a prediction result is output to predict whether an abnormal event occurs, wherein the first prediction engine includes a plurality of prediction models, and the prediction results are determined by the predictions At least one of the models is generated; (c) for the prediction result indicating the occurrence of a predicted abnormal event, use the input data set to generate a notification message for the predicted abnormal event; (d) receive a notification message associated with the notification message At least one feedback message; (e) updating the first prediction engine according to the at least one feedback message to obtain a second prediction engine; wherein the second prediction engine is used to replace the first prediction engine to follow the subsequent steps of the case Enter the data set to predict whether an abnormal event will occur.

In order to achieve at least the above objectives, the present invention proposes a method for intelligent monitoring of case field abnormality detection using artificial intelligence. The method is executed by at least one computing device. The method includes: (a) receiving a case field sensing Data set and get an input data set accordingly; (b) Apply the input data set to a first prediction engine and output a prediction result to predict whether an abnormal event occurs, wherein the first prediction engine includes a plurality of prediction models, and the prediction results are determined by the prediction models (C) For the prediction result indicating the occurrence of a predicted abnormal event, use the input data set to generate a notification message for the predicted abnormal event; (d) Receive at least one associated with the notification message Feedback information; (e) updating the first prediction engine according to the at least one feedback information to obtain a second prediction engine; wherein the second prediction engine is used to replace the first prediction engine to follow the input data of the case Set to predict whether an abnormal event will occur.

In order to achieve at least the above objectives, the present invention further provides a storage medium that stores instructions that can be read by a computing device, wherein when the instruction is executed by at least one computing device, the at least one computing device realizes the aforementioned artificial intelligence. Intelligent monitoring method for field anomaly detection or its embodiments.

The system implemented by the intelligent monitoring technology for detecting anomalies in a case using artificial intelligence provided above in the present invention can be flexible and expandable in terms of reflecting the update or change of anomalous problems. For example, a network service application that can help realize intelligent monitoring of case anomaly detection, and also help to derive other applications such as a work order system based on the monitoring results.

1: The case

2: machine

3: data concentrator

5: Communication network

9: Terminal device

10.10A, 10B: Intelligent monitoring system for case site anomaly detection with artificial intelligence

11: Monitoring system

12: Work Order System

12A: Management module

110: Pre-processing module

111: prediction engine

112: Post-processing module

115: control module

120: Communication module

121: Control Module

122: User Interface Module

125: memory unit

200: prediction engine

210: Unsupervised learning module

211: Unsupervised learning model

220: Supervised learning module

220_1~220_N: Supervised learning model

220_N+1: Supervised learning model

300: prediction engine

310_1~310_N: Supervised learning module

311: Supervised learning model

320: Unsupervised learning module

321: Unsupervised learning model

M11, M12: Model adjustment module

M13: Model new module

S10~S50: steps

S31, S33, S35: steps

S51, S53: steps

S110~S150: steps

S310~S332: steps

S1110~S1180: steps

A10~A60: Arrow

B10~B30: square

B120, B122, B124: square

B210~B260: square

A10-1~A70-1: Arrow

A110-1~A160-1: Arrow

B10-1~B40-1: Block

DB: database

DS: input data set

DS1: The first sub-data set

DS2: The second sub-data set

SR_1~SR_N: prediction result

N1~N5, W1, W2: feedback information

UR: forecast result

NR: prediction result

RA: abnormal event

RB: normal event

GA: First group

GB: The second group

FIG. 1 is a schematic diagram of an application scenario of an intelligent monitoring system for detecting anomaly in a case using artificial intelligence according to an embodiment of the present invention.

FIG. 2 is a schematic flowchart of an embodiment of an intelligent monitoring method for detecting anomalies in a case using artificial intelligence that can be applied to the system of FIG. 1.

FIG. 3 is a schematic flowchart of an embodiment of step S30 in FIG. 2.

FIG. 4 shows a schematic diagram of an embodiment in which the input data set is a multi-dimensional data set.

FIG. 5 is a schematic flowchart of an embodiment of step S50 in FIG. 2.

FIG. 6A is a schematic block diagram of an embodiment of the system of FIG. 1. FIG.

FIG. 6B is a schematic block diagram of another embodiment of the system of FIG. 1. FIG.

Fig. 6C is a schematic block diagram of an embodiment of the monitoring system of Fig. 6A.

Fig. 6D is a schematic block diagram of an embodiment of the work order system of Fig. 6A.

FIG. 7 is a schematic sequence diagram of using a monitoring system and a work order system to implement an embodiment of the method of FIG. 2.

Fig. 8 is a schematic block diagram of an embodiment of a prediction engine of a monitoring system.

FIG. 9 is a schematic flowchart of some embodiments of implementing step S20 in the monitoring system.

Fig. 10 is a schematic flow chart of an embodiment of the work order system.

Fig. 11 is a schematic diagram of an embodiment of a feedback method in the system.

Figure 12 is a schematic diagram of an embodiment of the monitoring system.

FIG. 13 is a schematic sequence diagram of another embodiment of the method of FIG. 2 by using a monitoring system and a work order system.

FIG. 14 is a schematic block diagram of some embodiments of step S50 in FIG. 2.

FIG. 15 is a schematic block diagram of another embodiment of the prediction engine in the system of FIG. 1. FIG.

FIG. 16 is a schematic block diagram of an embodiment of the supervised learning module in FIG. 15.

FIG. 17 is a schematic block diagram of an embodiment of the unsupervised learning module in FIG. 15.

In order to fully understand the purpose, features, and effects of the present invention, the following specific embodiments are used in conjunction with the accompanying drawings to give a detailed description of the present invention. The description is as follows: please refer to Figure 1, which is based on this A schematic diagram of an application scenario of an intelligent monitoring system using artificial intelligence to detect anomalies in a case according to an embodiment of the invention. As shown in Fig. 1, the intelligent monitoring system (or system for short) 10 for detecting anomaly in the case with artificial intelligence can be implemented by at least one computing device such as a server, and can be configured to connect to a communication network 5 for computing In the environment, it is used to receive a sensing data set of case 1, and use a prediction engine to detect abnormal events accordingly. If the system 10 detects the occurrence of a predicted abnormal event, it will generate a notification message, such as sending to one or more terminal devices 9, outputting to a database or memory storage or other suitable output methods to notify Relevant personnel of the case 1, management personnel or technical personnel of the system 10. The notification message can not only carry the event data of the predicted abnormal event (such as the time of the abnormal occurrence, which may further include machine information, etc.), but also the abnormal related information of the predicted abnormal event (such as the cause of the abnormality, the solution Plan, etc.) for reference or application of relevant personnel. The person can send at least one feedback message to the system 10 through the terminal device 9 or other terminal devices, so as to feedback the notification message and the actual situation of the case 1 to the system 10 to determine how to update the prediction engine. The detection result of the abnormal event of the prediction engine of the system 10 is made more accurate, and the system can also provide more accurate abnormal-related information in subsequent notification messages. After multiple executions of the above application scenarios, the abnormal event detection capability of the system 10 and the content of abnormal information related to the abnormality can be more optimized, thereby improving the performance of the abnormal detection of the case 1.

For example, multiple identical or different machines 2 can be set in the case 1, in order to detect abnormal events for a specific machine among the multiple machines 2 in the case 1, the specific machine (for example, reverse Inverter, used to convert DC to AC machines) continuously generate N sensing data (where N>1) within a period of time, which can be regarded as a sensing data set, where the sensing data is generally Chronological The physical quantity instead of automatic speech. For example, the specific machine has one piece of sensing data every minute, and the time interval is assumed to be one day, so a total of 60*24 pieces of sensing data become a sensing data set of the specific machine, and the sensing data set contains time-series Of multiple sensing data. The time interval can be set to any length of time (such as 1, 5, 10 minutes, or 0.5, 1, 2 hours, etc.) and is not limited by the above examples. In addition, the sensing data set may have fields with one or more dimensions, and may also have data units or text annotations. However, the main data recorded in the sensing data set is the physical quantity sensed. For example, the case 1 refers to a work site, such as a company, a community, a house, a school, a public institution, a factory, a solar power plant, or other places. In one embodiment, the case site 1 may be a solar power plant, and the sensing data set may be time-series power generation data at a certain time interval, such as voltage, current, power, temperature, and sunshine value (such as sunshine intensity or solar radiation). ) At least one or any combination, so the sensing data set can be high-dimensional data. In addition, as far as data transmission is concerned, in some application scenarios, one or more data concentrators 3 can be set up in the case 1, and the data concentrator 3 transmits the sensing data collected in the case 1. In the system 10, the data concentrator 3 can be implemented by any computing device or communication device with wired or wireless communication functions. However, the implementation of the present invention is not limited by the above examples.

In some implementations, the system 10 can be implemented using one or more of various technologies such as web services, script engines, web applications, or web application programming interface (API) servers, so as to be compatible with the case. Field 1 communication to receive a sensing data set, and to provide application services for the user's (such as field personnel, technicians or engineering personnel) browsers, applications, etc. to send a notification message or receive a feedback message. From the perspective of the user, the terminal device 9 on the user side can be used to receive a notification message from the system 10 or send a feedback message to the system 10 in various suitable ways. For example, applications, browsers or other programs can be implemented on the terminal device 9 to provide a user interface. The user can view data from the system 10 on the user interface A notification message, or optionally, a feedback message is further sent to the system 10. The application, browser or other program on the terminal device 9 can send the request data to the system 10, and after receiving the reply data, display the relevant information of the sunshine value on the field of the relevant result output of the user interface . In one embodiment, the system 10 can be implemented as a server that provides network API services. The terminal device 9 is implemented on the terminal device 9 to send request data to the system 10 by calling the network API in an application or script program. This gets the reply data. However, the implementation of the present invention is not limited by the above examples. The system 10 can also be implemented to communicate with the terminal device 9 through e-mail or short message. For example, the terminal device can be any computing device with wireless or wired communication capabilities, such as a smart device, a wearable device, a notebook computer, a desktop computer, or a specific electronic device.

Please refer to FIG. 2, which is a schematic flowchart of an embodiment of a method for intelligent monitoring of case field anomaly detection using artificial intelligence (or the method for short) that can be applied to the system 10 of FIG. 1. As shown in Figure 2, the method includes steps S10 to S50.

As shown in step S10, the system 10 receives a set of sensed data from a case and obtains an input data set accordingly.

As shown in step S20, the system 10 applies the input data set to a first prediction engine and outputs a prediction result to predict whether an abnormal event occurs, wherein the first prediction engine includes a plurality of prediction models, and the prediction result Generated by at least one of these predictive models.

As shown in step S30, if the prediction result indicates that a predicted abnormal event has occurred, the system 10 uses the input data set to generate a notification message for the predicted abnormal event.

As shown in step S40, the system 10 receives at least one feedback message associated with the notification message.

As shown in step S50, the system 10 updates the first prediction engine according to the at least one feedback message to obtain a second prediction engine.

The second prediction engine obtained in step S50 can be used to replace the first prediction engine, so that the system 10 can predict whether an abnormal event occurs based on the subsequent input data set of the case; if the second prediction engine According to the subsequent input data set, a subsequent abnormal event is predicted to occur, and the system 10 uses the subsequent input data set to generate a corresponding notification message for the predicted subsequent abnormal event.

In addition, in some embodiments, the system 10 can also be implemented to replace the current prediction engine with an updated prediction engine and then repeat at least one or more of the steps S10 to S50, or further update the update again. Prediction engine. In the above embodiment, the prediction engine is updated multiple times, so that the detection capability of the abnormal event of the system 10 and the content of the abnormal information related to the abnormality are more optimized, thereby improving the performance of the abnormal detection of the case 1.

Regarding the step S10, in one embodiment, the sensing data set includes sensing data from one of the machines 2 of the case 1 as shown in FIG. 1. The system 10 performs data processing on the sensing data set to become the input data set. For example, the sensing data is subjected to data format conversion, sorting, deletion of unnecessary data, or other processing, so that the input data set can be suitable It is applied to subsequent steps, such as applied to the first prediction engine in step S20. For example, the step S10 can also be implemented as copying the sensing data set as the input data set under appropriate circumstances. However, the implementation of the present invention is not limited by the above examples.

Regarding the step S20, in one embodiment, the first prediction engine may include a supervised learning model, and the supervised learning model is a trained model used to detect an abnormal event. In another embodiment, the first prediction engine may include an unsupervised learning model. The supervised learning model performs clustering or classification operations on the input data set, and then uses it to discover or detect unknown abnormal events. Regarding the first prediction engine, an embodiment will be described later. If the prediction result indicates that the predicted abnormal event has occurred, the method further executes the step S30. Optionally, if the prediction result indicates that no abnormal event has occurred, the method may execute step S10 to receive a subsequent sensing data set and perform subsequent steps, or optionally wait for data or perform other operations. However, the implementation of the present invention is not limited by the above examples.

Regarding the step S30, in an embodiment, as shown in FIG. 3, the step S30 may include steps S31, S33, and S35 to generate the notification message in response to the prediction result indicating that the predicted abnormal event has occurred.

In step S31, the system 10 collectively obtains an event data according to the input data corresponding to the prediction result, and the event data includes an abnormal occurrence time value.

In step S33, the system 10 obtains predicted abnormality related information related to the prediction result from a database, wherein the prediction result corresponds to one of the first prediction models of the prediction models.

In step S35, the system 10 generates the notification message based on the event data and the predicted abnormality related information.

Regarding the step S31, in an embodiment, the step S31 may be implemented as: the system 10 finds, in the input data set, the data that corresponds to the abnormal prediction result earlier in time series. For example, the first and second columns of Table 1 below are examples of the content of the input data set, representing the sensor data parameters of a machine in the case 1, such as the output voltage value of the machine on August 26, 2019. The third column of Table 1 is the prediction result corresponding to the content of the input data set, where the prediction result is Y represents abnormality, and N represents no abnormality or normal. For the example in Table 1, the predicted result corresponding to the voltage value from 11:30 to 12:00 is abnormal. The event data obtained based on the step S31 includes an abnormal occurrence time value, and the abnormal occurrence time value is 2019/8/26, 11:30.

In addition, the input data set can also be a multi-dimensional data set. For example, please refer to Figure 4. Columns 1 to 13 in the table shown in Figure 4 are examples of the content of the input data set, representing the sensor data parameters of the case 1, such as the output voltage value of the machine on a certain day. The fields A, B to L represent the values of 12 kinds of parameters that change over time; in an embodiment of the foregoing step S20, the first prediction engine may include an unsupervised learning model, and the unsupervised learning model is used for For example, perform clustering or classification calculations on normal or abnormal data on the input data set. For example, the unsupervised learning model has a high correlation between field C and field A and field G, and learns to determine abnormal data; the last row in the table shown in Figure 4 is the input data set The prediction result corresponding to the content; in this example, the unsupervised learning model judges that an abnormality occurs at 14:30 and later based on the changes in field C, field A, and field G, so the prediction result is Y. The prediction result of the input data set can also be stored in the system 10 for searching and tracing. However, the implementation of the present invention is not limited by the above examples.

Regarding the step S33, in one embodiment, a database may be implemented in the system 10, and the codes of the prediction models in the first prediction engine are recorded in the database (as shown in the first column of Table 2 below). ), and record the predicted abnormal events associated with these prediction models (such as the example in the second column of Table 2 below) and the predicted abnormal information (such as the example in the third column of Table 2 below). Suppose that there are N prediction models (such as events M1, M2, M3) in the first prediction engine, which correspond to N different types of models, respectively. For common events (such as events A1, A2, A3), if the prediction result obtained in step S20 (for example, an abnormal event is predicted) is generated by the prediction model whose code is M2 among the prediction models, then the prediction model M2 is The first prediction model described in this step S33. The system 10 can obtain the predicted abnormal event data (such as the description represented by A1) and abnormal related information (such as the reason R2 and the solution SN2) associated with the prediction model M2 from the database based on the step S33. The database is realized by, for example, files, data structures, relational databases, or any other suitable method or combination thereof, and the number is not limited. However, the implementation of the present invention is not limited by the above examples.

Regarding the step S35, the notification message generated by the system 10 is, for example, any form of message such as an email, a short message, a web page, or an electronic file. In one embodiment, the notification message is sent to at least one associated terminal device 9 in the form of a work order, so that it can be used by, for example, engineers, maintenance personnel, or management personnel. For example, the notification message can be used as a work order, in addition to predicting abnormal events, event data, and predicting abnormal related information, it can also include business data, as shown in Table 3 below. For example, the notification message is not limited to the presentation content form, and can use one or more methods, such as text, image, and voice; the notification message can include other information, such as geographic location or map. In practice, the system 10 can use the aforementioned database, and can aggregate information to the target object by searching for other related databases, such as a database for business data or other databases, such as a database for personnel management, etc. . However, the implementation of the present invention is not limited by the above examples.

Regarding the step S40, in one embodiment, the at least one feedback message can be implemented as a data field including a first feedback data and a second feedback data. The first feedback data is used to characterize whether the predicted abnormal event is an abnormality in the case by the personnel (such as engineering personnel, maintenance personnel or management personnel) of the case. The second feedback data is used to represent the abnormality-related information that the personnel of the case site feedback on the abnormality of the case site. With this, when the person who receives the notification message of step S30 (such as engineering personnel, maintenance personnel or management personnel) confirms that the predicted abnormal event is the abnormality of the case 1, or further discovers the abnormality of the case 1 After the cause and solution, the person can send a feedback message through the terminal device 9 one or more times to send back to the system 10. For example, the at least one feedback message is, for example, any form of e-mail, text message, web page or electronic document, such as the example in Table 4 below, where the abnormality is located in the system category: such as terminal device, power line, network line, At least one or a combination of relay box, low-voltage system, high-voltage system, dirty module, abnormal line, leakage, inverter load reduction, etc. For example, the feedback message is not limited to the form of presenting or recording content, and can use one or more methods, such as text Words, images, voices; the feedback message can be used to return other related data, such as the progress or status of on-site processing or inspection, such as geographic location or recording inspection time. For example, the feedback information can be further aggregated and analyzed by the system 10 for use in this method, and can be further used for other applications, such as generating corresponding repair reports, or the system 10 can be used as a network service for users Quickly search by keywords or other methods. However, the implementation of the present invention is not limited by the above examples.

Regarding the step S50, in an embodiment, as shown in FIG. 5, the step S50 may include steps S51 and S53. As shown in step S51, it is determined according to the at least one feedback message to update an adjustment instruction value of the first prediction engine. For example, in response to the content of at least one feedback message, the system 10 decides to set the adjustment indicator value to indicate a fine-tuning model for one of the prediction models that is related to the prediction result, or set it to indicate the A prediction model is added to the first prediction engine. For example, the adjustment instruction value is expressed in numerical value or text. For example, {01,0002} means to fine-tune the model of the prediction module whose representative code is 0002 (01 means update action), such as {11} means to add a new one Forecast module. However, the implementation of the present invention is not limited by the above examples.

As shown in step S53, the system 10 fine-tunes one of the prediction models related to the prediction result according to the adjustment instruction value, or adds a prediction model to the first prediction engine. Regarding this step S50, another embodiment will be described later.

According to the method described in FIG. 2, the system 10 can be implemented in various ways. Please refer to FIG. 6A, which is a schematic block diagram of an embodiment of the intelligent monitoring system for detecting anomaly in a case using artificial intelligence in FIG. 1. As shown in FIG. 6A, the intelligent monitoring system 10A (hereinafter referred to as the system 10A) for detecting anomalies in the case using artificial intelligence is an embodiment of the system 10 in FIG. 1. The system 10A includes a monitoring system 11 and a work order The system 12, wherein the monitoring system 11 includes a prediction engine 111. The monitoring system 11 and the work order system 12 can be implemented by a computing device such as a server, or can be implemented by two or more computing devices. The monitoring system 11 and the work order system 12 can also be regarded as two subsystems or modules in the software system. However, the implementation of the present invention is not limited by the above examples.

For example, as shown in FIG. 6B, it is another embodiment of the system 10. As shown in FIG. 6B, the system 10B includes a prediction engine 111, a control module 115, and a work order system 12. The work order system 12 can be configured to cooperate with the control module 115, and the control module 115 can notify the work order system 12 according to the prediction result of the prediction engine 111. Optionally, the work order system 12 may be configured to directly or indirectly receive the prediction result of the prediction engine 111. In addition, the system 10B can also be used as an embodiment of the system 10A in FIG. 6A. For example, the monitoring system 11 may include the prediction engine 111 and the control module 115.

The monitoring system 11 and the work order system 12 can be respectively used to implement the steps S10 to S50 of the method based on FIG. 2.

In an embodiment, the monitoring system 11 can implement the steps S10, S20, and S50 (such as S53) based on the method in FIG. 2, and the work order system 12 can implement the steps S30, S40, and S50 based on the method in FIG. 2. (Such as S51).

In another embodiment, the monitoring system 11 can implement the steps S10, S20, and S50 based on the method in FIG. 2, and the work order system 12 can implement the steps S30 and S40 based on the method in FIG. 2.

In yet another embodiment, the monitoring system 11 can implement the steps S10, S20, S40, and S50 based on the method in FIG. 2, and the work order system 12 can implement the step S30 based on the method in FIG. 2.

FIG. 6C is a schematic block diagram of an embodiment of the monitoring system 11 of FIG. 6A. As shown in FIG. 6C, an embodiment of the monitoring system 11 may include a pre-processing module 110, a prediction engine 111, a post-processing module 112, and a control module 115. The pre-processing module 110 is used to implement the step S10, for example. The control module 115 is used to manage the prediction engine 111 to implement the steps S20 and S50. The post-processing module 112 is optional, and is used to process the output of the control module 115 or the prediction engine 111. However, the implementation of the present invention is not limited by the above examples.

FIG. 6D is a schematic block diagram of an embodiment of the work order system 12 of FIG. 6A. As shown in FIG. 6D, an embodiment of the work order system 12 may include a communication module 120, a control module 121, a user interface module 122, and a memory unit 125. The communication module 120 is used to communicate with the terminal device 9. The control module 121 is used to process the prediction result from the monitoring system 11, and accordingly control the communication module 120 to implement the steps S30, S40, and S50. The user interface module 122 is optional. For example, when the work order system 12 communicates with the terminal device 9, the user interface module 122 is used to communicate with the user (such as a technician) in the notification message or the Processing of feedback messages. The memory unit 125 includes, for example, a main memory and an auxiliary memory. However, the implementation of the present invention is not limited by the above examples.

In an embodiment, the work order system 12 may be implemented as the database in the foregoing embodiment having the step S30, for example, by using the memory unit 125. In another embodiment, the database in the previous embodiment of the step S30 can also be implemented outside the work order system 12, and the work order system 12 communicates with the database through the communication module 120. In addition, the communication module 120 can also be implemented as, The database is used to communicate with the monitoring system 11. However, the implementation of the present invention is not limited by the above examples.

Please refer to FIG. 7, which is a schematic sequence diagram of an embodiment of the method in FIG. 2 using the monitoring system and the work order system. In FIG. 7, arrows or squares are used to represent operations, functions, or steps that can be implemented in the monitoring system and work order system based on the method in FIG. 2 in this embodiment, for example, can be used in the monitoring system or work order system At least one program or software module can be implemented.

As shown by the arrow A10 in Figure 7, Case 1 sends out the sensing data set.

As indicated by the arrow A20, the monitoring system 11 receives the sensing data set and obtains an input data set accordingly to implement step S10.

As shown in block B10, the monitoring system 11 applies the input data set to a first prediction engine (such as the prediction engine 111 in FIG. 6) and outputs a prediction result to predict whether an abnormal event occurs, thereby implementing step S20. As indicated by the arrow A30, the monitoring system 11 outputs the prediction result to the work order system 12.

As indicated by arrow A35, if the prediction result indicates that a predicted abnormal event occurs, the work order system 12 uses the input data set to generate a notification message for the predicted abnormal event, and sends the notification message to the terminal device 9. As shown by arrow A40, step S30 is realized.

According to the notification message, the terminal device 9 can display, for example, the predicted abnormal event and corresponding abnormal information (such as abnormal solutions, and/or reasons, etc.) for reference or application by relevant personnel, and relevant personnel can use them accordingly Actually observe or deal with related matters at the scene or machine of the predicted abnormal event. Relevant personnel can send feedback messages through the terminal device 9 after actually observing or handling related matters, as shown by arrow A45.

According to step S40, the monitoring system 11 can receive at least one feedback message associated with the notification message, as shown by arrow A50; the work order system 12 can receive at least another feedback message associated with the notification message, such as arrow Shown in A60. However, the implementation of the present invention is not limited by the number of feedback messages and the sending or receiving order in the above examples.

As shown in block B20, the work order system 12 can adjust the related abnormal information according to the at least one feedback message.

As shown in block B30, the monitoring system 11 can update the first prediction engine according to the at least another feedback message to obtain a second prediction engine, so as to implement step S50. The second prediction engine (such as the updated prediction engine 111) is used to replace the first prediction engine to predict whether an abnormal event occurs based on the subsequent input data set of the case.

In addition, the monitoring system 11 and the work order system 12 can be implemented respectively to replace the current prediction engine with an updated prediction engine and then repeat at least one or more steps based on the steps S10 to S50, or further update again The updated forecasting engine. According to the above-mentioned embodiment, the prediction engine can be updated multiple times to enhance the detection capability of abnormal events of the system 10A, thereby improving the performance of abnormal detection of the case 1, such as the accuracy of abnormal detection and response rate, etc. .

According to FIG. 2 and FIG. 7, various implementations of the monitoring system 11 and the work order system 12 are further illustrated below.

Regarding block B10 based on step S20, in one embodiment, the prediction engine 200 shown in FIG. 8 may be implemented in the monitoring system 11 as the first prediction engine. In FIG. 8, the prediction engine 200 includes an unsupervised learning module 210 and a supervised learning module 220. The unsupervised learning module 210 includes at least one unsupervised learning model 211. The supervised learning module 220 includes at least one supervised learning model (e.g., supervised Learning model 210_1~210_N), where N≧1. The output of the unsupervised model is located upstream of the input of the unsupervised model.

Regarding block B10 based on step S20, in one embodiment, the input data set DS may be applied to the unsupervised learning module 210 of the prediction engine 200, for example, the input data set DS may be input to the unsupervised learning model 211 , To execute the clustering or classification algorithm to divide the input data set DS into at least two types or at least one of two types, for example, the first sub-data set DS1 that is temporarily identified as "normal data" and "abnormal data" The second sub-data set DS2. Since the so-called "normal data" and "abnormal data" are determined by the unsupervised learning model 211 and are not necessarily facts, they need to be further processed, for example, judged by the supervised learning module 220. Therefore, when the unsupervised learning module 210 generates the second sub-data set DS2, the second sub-data set DS2 can be further input to each supervised learning model of the supervised learning module 220 (for example, represented by 210_1, 210_2~210_N ). Each supervised learning model (such as the supervised learning model 210_P, where 1≦P≦N) is used to predict whether the second sub-data set DS2 meets the abnormal event (or abnormal type) corresponding to the supervised learning model, and output Corresponding prediction results (for example, represented by SR_1, SR_2~SR_N).

Regarding block B10 based on step S20, in one embodiment, as shown in step S110 of FIG. 9, in the monitoring system 11, the prediction results of each supervised learning model are received. As shown in step S120, it is judged whether at least one abnormal event matches, that is, it is judged whether the prediction result indicates that at least one predicted abnormal event has occurred. If yes, step S130 is executed to transmit the predicted abnormal event to the work order system 12.

If the judgment of step S120 is no, that is, the second sub-data set DS2 does not conform to any supervised learning model, other processing can be performed, such as a processing action for preparing a new supervised learning model. Optionally, in an embodiment, step S140 is performed to mark the second sub-data set DS2 as conforming to a new supervised learning model (as represented by 210_N+1 in FIG. 8), and the step may be performed S150: Mark the first sub-data set DS1 (if any) output by the unsupervised learning module 210 as not conforming to the new supervised learning model, as a preparation for adding the new supervised learning model. Thus, for example, the second sub-data set DS2 that has been marked as conforming can be used to train the new supervised learning model, and the trained new supervised learning model can be added to the supervised learning module 220. In one embodiment, the monitoring system 11 may further notify relevant personnel through the work order system 12 or the terminal device 9 and receive feedback from the relevant personnel on the actual situation of the case, such as details of abnormal events, solutions, etc. The feedback can be used as anomaly related information and stored by the work order system 12 and associated with the new supervised learning model. However, the implementation of the present invention is not limited by the foregoing example regarding the determination of step S120 as negative.

In addition, it is further possible to update the "normal data" or "normal information" or "for the first sub-data set DS1 or the second sub-data set DS2" with the help of at least one feedback message obtained from the method in FIG. 2 or the subsequent steps in FIG. Abnormal data" is identified, and the prediction engine 200 is updated accordingly. This will be illustrated later.

Continuing with step S130 in FIG. 9, the predicted abnormal event is transmitted to the work order system 12. The work order system 12 (or the system 10, 10A, or 10B) can be implemented to manage the prediction model (such as the supervised learning model 210_1, 210_2~210_N) and the corresponding abnormal information in the first prediction engine (such as the prediction engine 111) The association between the prediction model and the abnormal information can be stored or realized in the database. For example, as shown in FIG. 10, the work order system 12 can be implemented to include a management module 12A and a database DB, where the database DB is used to realize the association between the aforementioned prediction model and abnormal information. When the notification message is generated based on the step S30, the management module 12A of the work order system 12 (or the system 10, 10A, or 10B) can perform some operations, such as searching, obtaining from the database DB, and based on the step The predicted abnormality related information associated with the predicted abnormal result obtained in S20, and the operation of generating the notification message based on it, for example, block B120 represents the program of these operations.

When processing the feedback message based on the step S40, the work order system 12 (or the system 10, 10A, or 10B) can use the data in the feedback message (such as the abnormal information related to the feedback) to update (such as delete, add, The content of the abnormality-related information corresponding to the modified content or association relationship). For example, the block B122 represents the operation of updating the details of the abnormal event, and the block B124 represents the operation of updating the corresponding abnormality-related information.

In one embodiment, a feedback program can be implemented, as shown by arrow A45, to allow relevant personnel to feedback the actual situation of the abnormality in the case 1; for example, whether an abnormal event has occurred, and whether the type of the abnormal event is consistent with the prediction Whether the abnormal event matches the abnormal event, and whether the abnormal related information (such as the solution of the abnormality) is consistent. In addition, the information on the actual situation of the abnormality in the case 1 returned by the relevant personnel can be fed back to the monitoring system 11 to determine how to update the forecasting engine, or to the work order system 12 to adjust the related abnormal information.

As shown in FIG. 11, it is an embodiment of the method for realizing the feedback program. As shown in step S310, it is determined whether the user feedback data indicates whether the case corresponding to the predicted abnormal event is abnormal.

If it is determined in step S310 that the user feedback data indicates that the case corresponding to the predicted abnormal event is not abnormal, the feedback prediction does not match, and at least one feedback message is generated to the monitoring system 11, the feedback message includes, for example, feedback Information N1 and N3. For example, the feedback information N1 indicates that the adjustment indicator value is set to represent that the corresponding second sub-data set DS2 is marked as "normal" for the unsupervised learning module 210, and the unsupervised learning module 210 is executed accordingly. Fine-tuning; feedback information N3 means that the adjustment indicator value is set to represent the second subdata set DS2 corresponding to the supervised learning model (such as the supervised learning model 210_P, where 1≦P≦N) corresponding to the predicted abnormal event Mark it as "non-conforming", and fine-tune the supervised learning model 210_P accordingly.

If it is determined in step S310 that the user feedback data indicates that the case corresponding to the predicted abnormal event is abnormal, at least one feedback message is generated to the monitoring system 11 and step S320 is executed. The feedback information includes, for example, the feedback information N2. For example, the feedback information N2 indicates that the adjustment indicator value is set to represent that the corresponding second sub-data set DS2 is marked as "abnormal" for the unsupervised learning module 210, and the unsupervised learning module 210 is executed accordingly. Fine-tuning.

As shown in step S320, it is determined whether the user feedback data indicates whether the predicted abnormal event is consistent with the abnormality of the case.

If the step S320 determines that the user feedback data indicates that the predicted abnormal event does not match the abnormality of the case, at least one feedback message is generated to the monitoring system 11 and step S322 is executed. The feedback information includes, for example, the feedback information N3. For example, the feedback information N3 indicates that the adjustment indicator value is set to represent the second sub-data corresponding to the supervised learning model (such as the supervised learning model 210_P, where 1≦P≦N) corresponding to the predicted abnormal event The set DS2 is marked as "non-conforming", and the supervised learning model 210_P is fine-tuned accordingly.

As shown in step S322, it is determined whether the user feedback data indicates whether the abnormality of the case meets other abnormal types (such as abnormal events corresponding to other supervised learning models in the supervised learning module 220).

If it is determined in step S322 that the user feedback data indicates that the abnormality of the case matches other existing abnormality types, such as the abnormal event corresponding to another supervised learning model 210_Q (where 1≦Q≦N, P≠Q) , At least one feedback message is generated to the monitoring system 11. The feedback information includes, for example, the feedback information N3.

If it is determined in step S322 that the user feedback data indicates that the abnormality of the case does not conform to other existing abnormality types, at least one feedback message is generated to the monitoring system 11. The feedback For example, the feedback information N5 is included and step S324 is executed. For example, the feedback information N5 indicates that the adjustment indicator value is set to represent that the second sub-data set DS2 is marked as conforming to a new supervised learning model, and a new supervised learning model is added (as shown in Figure 8 210_N+1 means).

As shown in step S324, the user's feedback on the actual situation of the case is further received, such as the details of the abnormal event, the solution, etc., and at least one feedback message is generated to the work order system 12 accordingly. The feedback information includes, for example, the feedback information W1 and W2. The feedback information W1 and W2 can be used as abnormal information and stored by the work order system 12 and associated with the new supervised learning model. For example, the abnormality-related information of the feedback is the details and solutions of the abnormality that the user feedbacks after confirming the case.

If the step S320 determines that the user feedback data indicates that the predicted abnormal event is consistent with the abnormality of the case, at least one feedback message is generated to the monitoring system 11 and step S330 is executed. The feedback information includes, for example, the feedback information N4. For example, the feedback information N4 indicates that the adjustment indicator value is set to represent the second sub-data corresponding to the supervised learning model (such as the supervised learning model 210_P, where 1≦P≦N) corresponding to the predicted abnormal event The set DS2 is marked as "conforming", and the supervised learning model 210_P is fine-tuned accordingly.

As shown in step S330, it is determined whether the solution in the abnormality-related information corresponding to the predicted abnormal event is consistent (for example, whether it is suitable for solving the abnormality of the case).

If the step S330 determines that the solution in the abnormality-related information corresponding to the predicted abnormal event is consistent, step S331 is executed. Step S331 is, for example, ending the flow of FIG. 11 or performing other steps.

If the step S330 determines that the solution in the abnormality-related information corresponding to the predicted abnormal event is not consistent, step S332 is executed. Step S332 is, for example, to further receive feedback from the user on the actual situation of the case, such as solutions to abnormal events. After step S332, at least A feedback message is sent to the work order system 12, and the feedback message includes, for example, the feedback information W2. The feedback information W2 can be used as anomaly related information and stored by the work order system 12 and associated with the supervised learning model (such as 210_P) corresponding to the predicted anomalous event. For example, the feedback information W2 is a solution provided by the user after confirming the case.

Regarding step S50 based on FIG. 2, please refer to FIG. 12, which is a schematic diagram of an embodiment of the monitoring system. As shown in FIG. 12, the monitoring system 11 can be implemented to include model adjustment modules M11 and M12 and a model addition module M13.

The model adjustment module M11 is used to implement the operations represented by the blocks B210 and B220, and to fine-tune the unsupervised learning module 210 accordingly. As shown in block B210, the model adjustment module M11 receives or processes the feedback information N1 in the feedback message, and sets the adjustment indicator value to represent the unsupervised learning module 210 marking the corresponding second sub-data set DS2 as " Normal", and fine-tune the unsupervised learning module 210 accordingly. As shown in block B220, the model adjustment module M11 receives or processes the feedback information N2 in the feedback message, and sets the adjustment indicator value to represent the unsupervised learning module 210 marking the corresponding second sub-data set DS2 as " Abnormal", and fine-tune the unsupervised learning module 210 accordingly.

The model adjustment module M12 is used to implement the operations represented by the blocks B230 and B240, and to fine-tune the supervised learning module 220 accordingly. As shown in block B230, the model adjustment module M12 receives or processes the feedback information N3 in the feedback message, and sets the adjustment indicator value to represent the supervised learning model corresponding to the predicted abnormal event (e.g., the supervised learning model 210_P stands for) Mark the corresponding second sub-data set DS2 as "non-conforming", and fine-tune the supervised learning model 210_P accordingly. As shown in block B240, the model adjustment module M12 receives or processes the feedback information N4 in the feedback message, and sets the adjustment indicator value to represent the supervised learning model corresponding to the predicted abnormal event (for example, by supervised The supervised learning model 210_P represents) marking the corresponding second sub-data set DS2 as “conforming”, and fine-tuning the supervised learning model 210_P accordingly.

The model addition module M13 is used to implement the operations represented by the blocks B250 and B260, and according to which the supervised learning model is added to the supervised learning module 220. As shown in block B250, the model addition module M13 receives or processes the feedback information N5 in the feedback message, and sets the adjustment indicator value to represent that the second sub-data set DS2 is marked as conforming to a new supervised learning model. And add the new supervised learning model (as represented by 210_N+1 in Figure 8). As shown in block B260, the model addition module M13 marks the first sub-data set DS1 as "non-conforming", and executes the new supervised learning model (as represented by 210_N+1 in Figure 8) train.

In addition, the above feedback program can be implemented on the terminal device 9, the monitoring system 11, or the work order system 12. However, the implementation of the present invention is not limited by the feedback program in the above example.

The foregoing embodiment updates the prediction engine multiple times, so that the detection capability of the abnormal event of the monitoring system 11 and the content of the abnormal information related to the abnormality can be more optimized. When the monitoring system 11 detects an abnormal event (referring to the aforementioned predicted abnormal event), the work order system 12 can provide more optimized abnormal related information. Managers or technicians, etc. can refer to the abnormal information in the notification message provided by the work order system 12, and obtain effective assistance in the detection and troubleshooting of abnormal events. Thereby, the system 10A (or the system 10) can promote the efficiency and effectiveness of the operation and maintenance of the case.

Please refer to FIG. 13, which is a schematic sequence diagram of an embodiment of the system 10A in FIG.

As shown by the arrow A10-1 in Figure 13, a case 1 sent a sensing data set.

As indicated by the arrow A20-1, the monitoring system 11 receives the sensing data set and obtains an input data set accordingly.

As shown in block B10-1, the monitoring system 11 applies the input data set to a first prediction engine (such as the prediction engine 111 in FIG. 6C) and outputs a prediction result to predict whether an abnormal event occurs. As indicated by the arrow A30-1, the monitoring system 11 outputs the prediction result to the work order system 12.

If the prediction result indicates that a predicted abnormal event occurs, the work order system 12 uses the input data set to generate a notification message for the predicted abnormal event, and sends the notification message to the terminal device 9, such as arrow A40- 1 shown.

The work order system 12 receives at least one feedback message associated with the notification message, as indicated by an arrow A50 or A60.

As shown in block B20-1, the work order system 12 determines an adjustment instruction value for the first prediction engine according to the at least one feedback message. As indicated by the arrow A70-1, the work order system 12 outputs the adjustment instruction value to the monitoring system 11.

As shown in block B30-1, the monitoring system 11 updates the first prediction engine according to the adjustment indication value to obtain a second prediction engine. The second prediction engine (such as the updated prediction engine 111) is used to replace the first prediction engine to predict whether an abnormal event occurs based on the subsequent input data set of the case.

As indicated by the arrow A110-1, the case site 1 sends out subsequent sensing data sets.

As indicated by the arrow A120-1, the monitoring system 11 receives the sensing data set and obtains an input data set accordingly.

As shown in block B40-1, the monitoring system 11 applies the input data set to the second prediction engine (such as the updated prediction engine 111) and outputs a prediction result to predict whether an abnormal event occurs. As shown by the arrow A130-1, the monitoring system 11 outputs the prediction result to the work order system 12.

If the prediction result indicates that a predicted abnormal event occurs, the work order system 12 uses the input data set to generate a notification message for the predicted abnormal event, and sends the notification message to the terminal device 9, such as arrow A140- 1 shown.

The work order system 12 receives at least one feedback message associated with the notification message, as shown by arrow A160-1. The work order system 12 determines an adjustment instruction value for the second prediction engine according to the at least one feedback message. As indicated by the arrow A170-1, the work order system 12 outputs the adjustment instruction value to the monitoring system 11. Thereby, the monitoring system 11 can update the second prediction engine again according to the adjustment instruction value as shown in the aforementioned block B30-1 to obtain an updated second prediction engine. The updated second prediction engine (such as the updated prediction engine 111) is used to replace the second prediction engine to predict whether an abnormal event occurs based on the subsequent input data set of the case.

In addition, the monitoring system 11 and the work order system 12 can be respectively implemented to replace the current prediction engine with an updated prediction engine and then repeat at least one or more steps based on the steps S10 to S50, or further update the Updated forecasting engine. In the above-mentioned embodiment, the prediction engine is updated multiple times, which can strengthen the detection capability of the abnormal event of the system 10A, thereby improving the performance of the abnormal detection of the case 1.

In addition, the work order system 12 (or the system 10, 10A, or 10B) can be used to manage the prediction model in the prediction engine (such as the prediction engine 111) and to associate the prediction model with corresponding abnormal information. Management or exception-related information can be realized by using the database. When generating a notification message based on the step S30, the work order system 12 (or the system 10, 10A, or 10B) can obtain the predicted abnormality related information associated with the abnormal result of the step S20 from the database and generate it accordingly The notification message. When processing the feedback message based on the step S40, the work order system 12 (or the system 10, 10A, or 10B) can match the data in the feedback message (such as the abnormal information related to the feedback) with the corresponding information in the database. Information (such as predicted abnormality-related information) is compared to determine the adjustment method of the prediction engine, and the content of the corresponding abnormality-related information can be updated (such as deleting, adding, modifying content or relationship). The foregoing embodiment updates the prediction engine multiple times, so that the detection capability of the abnormal event of the monitoring system 11 and the content of the abnormal information related to the abnormality can be more optimized. When the monitoring system 11 detects an abnormal event (referring to the aforementioned predicted abnormal event), the work order system 12 can provide more optimized abnormal related information. Managers or technicians, etc. can refer to the abnormal information in the notification message provided by the work order system 12, and obtain effective assistance in the detection and troubleshooting of abnormal events. Thereby, the system 10A (or the system 10) can promote the efficiency and effectiveness of the operation and maintenance of the case.

In some embodiments, the operation of receiving feedback information of the work order system 12 in FIG. 13 and various related embodiments (as indicated by arrows A50-1 or A60-1) or determining adjustments to one of the first prediction engines The operation of the indicator value (as shown in the block B20-1) can be partially or fully implemented in the monitoring system 11. In other embodiments, the operation of the monitoring system 11 in FIG. 13 and various related embodiments can also be implemented by using the control module 115 in the system 10B of FIG. 6B.

The above-mentioned work order system 12 (or the system 10, 10A, or 10B) can be used to manage the prediction model in the prediction engine (such as the prediction engine 111) and the operation of associating the prediction model with the corresponding abnormality-related information is the same as that in Fig. 2 Step S50 is related. In the step S50, the system 10 (or 10A or 10B) determines an adjustment instruction value of the first prediction engine (or the current prediction engine) according to the at least one feedback message. The following examples illustrate the step S50 Various implementations.

Please refer to FIG. 14, which is a schematic block diagram of some embodiments based on step S50 of FIG. 2. For ease of description, it is assumed here that the at least one feedback message can be realized as a data field including a first feedback data and a second feedback data. The first feedback data is used to characterize whether the predicted abnormal event is an abnormality in the case by the personnel (such as engineering personnel, maintenance personnel or management personnel) of the case. The second feedback data is used to represent the abnormality-related information that the personnel of the case site feedback on the abnormality of the case site. However, the implementation of the present invention is not limited by the above examples.

As shown in FIG. 14, an embodiment of step S50 in FIG. 2 includes steps S1110 and S1120. As shown in step S1110, it is determined whether the first feedback data (or the at least one feedback message) indicates that the predicted abnormal event is an abnormality of the case.

If it is determined in step S1110 that the first feedback data (or the at least one feedback message) indicates that the predicted abnormal event is not an abnormality of the case, then as shown in step S1115, the prediction error is fed back, for example, the input data is integrated with A piece of data related to the prediction result is marked as a prediction error, and the adjustment indication value is set to indicate that the first prediction model is fine-tuned.

If it is determined in step S1110 that the first feedback data (or the at least one feedback message) indicates that the predicted abnormal event is an abnormality in the case, then as shown in step S1120, further determine the abnormality-related information referred to by the predicted abnormal event Whether it meets the actual situation. For example, it is determined whether the abnormality-related information of the feedback (or the at least one feedback message) matches the predicted abnormality-related information to determine the adjustment instruction value. For example, the abnormality-related information of the feedback is the reason and solution of the abnormality returned by the user after confirming the case, and the predicted abnormality-related information is, for example, the predicted reason and solution of the abnormality.

In one embodiment, if the judgment result of step S1120 is completely consistent, for example, the reason and solution of the abnormality in the feedback are consistent with the predicted reason and solution of the abnormality, then as shown in step S1130, the feedback prediction is correct. For example, marking the piece of data as accurate for prediction, and setting the adjustment indication value to indicate that the first prediction model is fine-tuned.

In this step S1120, there are many possibilities for whether the abnormal information referred to by the predicted abnormal event conforms to the actual situation, such as complete conformance, partial conformity, or none conformity. Therefore, when step S1120 is implemented, at least one of the following implementations is optionally adopted Examples to be realized.

In one embodiment, the step S50 further includes: if the judgment result of the step S1120 is partially inconsistent, for example, the reason for the abnormality of the feedback is consistent with the reason of the predicted abnormality, but the solution of the feedback is the same as the predicted solution If they are inconsistent, as shown in step S1140, it is instructed to update the abnormal information. For example, the system 10 (or 10A or 10B) marks the piece of data as accurately predicted, and sets the adjustment indicator value to indicate that the first prediction model is fine-tuned, and based on the second feedback data (or the at least one Feedback information) update the database accordingly. For example, the system 10 (or 10A or 10B) updates the predicted anomaly related information corresponding to the first prediction model in the database.

In one embodiment, the step S50 further includes: if the determination result of the step S1120 is not consistent, for example, the abnormal cause and solution of the feedback are inconsistent with the predicted abnormal cause and solution, then as in step S1160 It shows that the system 10 (or 10A or 10B) further determines whether the feedback related information about the abnormality (or the at least one feedback message) conforms to other known abnormalities.

Regarding the step S1160, for example, the system 10 (or 10A or 10B) determines whether the feedback related abnormal information (or the at least one feedback message) at least partially corresponds to other prediction models in the first prediction engine, to Decide on the adjusted indication value. For example, please refer to the example of the database in Table 2 above. The system 10 (or 10A or 10B) can search for anomaly related information predicted by the database (as in the third row in Table 2) based on the step S1160 Whether there is data that matches the reason in the abnormal information of the feedback. If the reason in the abnormality-related information of the feedback matches the predicted abnormality-related information associated with a prediction module in the database (such as the reason R1 in the third column of Table 2), the prediction model (such as prediction Model M1) is regarded as having a corresponding relationship with the abnormal related information of the feedback.

In one embodiment, the step S50 further includes: if the step S1160 determines that at least part of the abnormal information related to the feedback (or the at least one feedback message) is at least part of the second prediction model of the first prediction engine The prediction model (such as the prediction model M1 in Table 2) has a corresponding relationship, such as step S1170 As shown, the feedback prediction is correct. For example, mark the piece of data as accurate prediction, and set the adjustment indicator value to indicate that the second prediction model is fine-tuned, and the data is updated correspondingly according to the second feedback data (or the at least one feedback message) Library. For example, the system 10 (or 10A or 10B) updates the predicted anomaly related information corresponding to the first prediction model in the database.

In one embodiment, the step S50 further includes: if the determination result of the step S1160 is that there is no corresponding relationship, as shown in the step S1180, instructing to add a new model. For example, the system 10 (or 10A or 10B) marks the piece of data as abnormal data, and sets the adjustment indicator value to represent a new prediction model added to the first prediction engine, and based on the second feedback data ( Or the at least one feedback message) correspondingly update the database. For example, the system 10 (or 10A or 10B) updates the predicted anomaly related information corresponding to the first prediction model in the database; another example, adds information about the newly added prediction module, such as predicted anomaly The event, the predicted abnormal information, the representative code of the newly added prediction module, and their relationship are as illustrated in Table 2.

In addition, in another embodiment, the step S1150 can be implemented in the following manner: the system 10 (or 10A or 10B) marks the piece of data as a prediction error, and sets the adjustment indicator value to indicate the first The prediction model fine-tunes the model, and correspondingly updates the database according to the second feedback data (or the at least one feedback message). For example, the system 10 (or 10A or 10B) updates the predicted anomaly related information corresponding to the first prediction model in the database.

In some embodiments, at least one or any combination of the above-mentioned embodiments regarding the step S50 can be implemented by the system 10 (or the system 10A), so it can be implemented by the work order system 12 (such as the control module 121) Or the monitoring system 11 (such as the control module 115) can be implemented. However, the implementation of the present invention is not limited by the above examples.

In the above embodiment, the first prediction engine of the system 10 (or system 10A) includes a plurality of prediction models, and the prediction result is generated by one of the prediction models. Please refer to FIG. 15, which is a schematic block diagram of an embodiment of the prediction engine in the system 10 of FIG. 1. As shown in FIG. 15, the prediction engine 300 includes at least one supervised learning module (such as supervised learning modules 310_1 to 310_N) and an unsupervised learning module 320, where N≧1. The supervised learning module is a trained model used to distinguish whether the input data set input to the supervised learning module is abnormal or non-abnormal (that is, normal). The unsupervised learning module is used to divide the input data set into different groups, such as at least two groups, and assume good (or normal) and bad (or abnormal), and thereby discover or detect unknown abnormal problems.

As shown in FIG. 15, the output of the supervised model (such as the supervised learning modules 310_1 to 310_N) is located upstream of the input of the unsupervised model 320. In FIG. 15, the system 10 (or system 10A) first applies the input data set (represented by DS) to the at least one supervised learning module (such as supervised learning modules 310_1~310_N), and then applies it to the non- Supervised learning module 220.

If the input data set DS is applied to the prediction result of any supervised learning module 310_K (K is any one of 1, 2 to N) representing an abnormality, then the prediction result is used as one of the prediction results SR_K of the prediction engine 300 . If the at least one supervised learning module produces no abnormal prediction results, the system 10 (or system 10A) applies the input data set DS to the unsupervised learning module 320 and generates corresponding prediction results.

If the corresponding prediction result generated by the unsupervised learning module 320 is normal, the prediction result is used as the prediction result NR of the prediction engine 300. If the corresponding prediction result generated by the unsupervised learning module 320 is abnormal, the prediction result is used as the prediction result UR of the prediction engine 300. The prediction result UR represents an unknown anomaly. Therefore, the system 10 (or the system 10A) outputs a corresponding notification message according to the step S30 in FIG. 2 to notify the user that there is an unknown abnormal event. Next, the department The system 10 (or the system 10A) receives the corresponding at least one feedback message according to the step S40 of FIG. 2, so that it can be known whether the unknown abnormal event is indeed an abnormal event occurred in the case, or further feedback can be received Exception related information (such as cause, solution). If the system 10 (or system 10A) determines according to the step S50 of FIG. 2 that the feedback message indicates that the unknown abnormal event is indeed an abnormal event that occurred in the case, it can be further based on the abnormal information related to the feedback (such as (Cause, Solution) feedback to add a new model, such as according to the step S50 in FIG. 2.

Regarding the step S50 based on FIG. 2, in one embodiment, the step S50 includes: the system 10 determines whether to add a prediction model to the current prediction engine (such as the first prediction engine) according to the adjustment indicator value; If the adjustment indicator value indicates that a new prediction model is to be added, the operation of the new prediction model is performed to update the current prediction engine. For example, by adding a new training model (Training model from scratch), a piece of data related to the prediction result (which represents an abnormal event) in the input data set is marked as abnormal data, and the system 10 (or the system 10A, such as the work order system 12) or the database records the abnormal information (such as cause, solution) of the feedback. Then, mark the later data in the input data set as normal data. The reason is that when the feedback message is received, it means that the abnormal event has ended. Therefore, the sensor data starting at the time when the feedback message is received can be used as normal data. Assume normal data. Then, use the above-mentioned piece of data and the data before or after the piece of data as a training data set, and based on this, newly train a new supervised learning model and add it to the prediction engine of the system 10 (or the system 10A) . For example, the new supervised learning model can be added to the supervised learning model of FIG. 15 after the supervised learning module 310_N and before the unsupervised learning module 320.

Regarding the step S50, in one embodiment, the step S50 includes: the system 10 determines whether to perform a prediction on one of the prediction models that is related to the prediction result according to the adjustment indicator value. Fine-tune the model; if the adjustment indication value indicates that a specified prediction model is to be fine-tuned, then the specified prediction model will be fine-tuned to update the current prediction engine.

Regarding the fine-tuning model, for example, a feedback fine-tuning model can be applied. The following are examples of whether the prediction is correct or not based on user feedback.

If the feedback message indicates that the user feedbacks that the prediction is correct, a piece of data related to the prediction result (which represents the predicted abnormal event) in the input data is marked as abnormal data. Then, mark the later data in the input data set as normal data. The reason is that when the feedback message is received, it means that the abnormal event has ended. Therefore, the sensor data starting at the time when the feedback message is received can be used as normal data. Assume normal data. Then, use the above-mentioned segment of data and the data before or after the segment of data as the training data set, and apply it to the specified prediction model for fine-tuning. After the specified prediction model is fine-tuned, it can be used in the prediction engine of the system 10 (or the system 10A) to replace the original prediction model.

If the feedback message indicates that the user feedbacks a prediction error, that is, the user confirms that the related predicted abnormal event is an erroneous prediction, then the input data is collected and a segment of data related to the predicted result (which represents the predicted abnormal event) is marked It is normal information. Then, take the piece of data and the data previously marked as abnormal data during training of the specified prediction model as the training data set, and apply the specified prediction model to fine-tune the training data set. After the specified prediction model is fine-tuned, it can be used in the prediction engine of the system 10 (or the system 10A) to replace the original prediction model.

Please refer to FIG. 16, which is a schematic block diagram of an embodiment of the supervised learning module of FIG. 15. As shown in FIG. 16, the supervised learning module 310_1 includes a supervised learning model 311, and is used to output a prediction result RA representing an abnormal event or a prediction result RB representing a normal event according to the input data set DS. The supervised learning model 311 is, for example, implemented based on a nonlinear regression model, for example, based on deep learning Model-like neural networks, such as deep convolutional neural network (DNN), recurrent neural network (recurrent neural network RNN), convolutional neural network (CNN), residual neural network Road (residual neural network, ResNet), deconvolutional neural network (deconvolutional neural network), long short-term memory (long short-term memory, LSTM) or other neural networks, etc. at least one model, or both or A combination of the above. For example, the supervised learning model 311 may include an input layer, one or more hidden layers, and an output layer. For another example, the supervised learning model 311 may further include a long short-term memory (LSTM), where the long short-term memory model includes repeated long and short-term memory cells (LSTM cells), each of which can be Contains multiple interactive layers, such as those called forget gates, input gates, and output gates. The other supervised learning modules 310_2~310_N can also be implemented similarly. Regarding the above example of fine-tuning the model, the output layer of the supervised learning model can be fine-tuned, or other suitable methods can be used for fine-tuning. However, the implementation of the present invention is not limited by the above examples.

Please refer to FIG. 17, which is a schematic block diagram of an embodiment of the unsupervised learning module of FIG. 15. As shown in FIG. 17, the unsupervised learning module 320 includes an unsupervised learning model 321, and is used to distinguish data into different groups according to the input data set DS, such as the first group GA or the second group GB. For example, it can be assumed that the first group of GA with a small amount of data is used as the prediction result of an abnormal event, and the second group of GB with a small amount of data is used as the prediction result of a normal event. The unsupervised learning model 321 is implemented by, for example, cluster analysis, dimension reduction or other suitable unsupervised learning techniques, for example, k-nearest neighbors algorithm (k-NN). )). However, the implementation of the present invention is not limited by the above examples.

In some embodiments, a storage medium is provided, which stores an instruction readable by a computing device, wherein the instruction is executed by at least one computing device (such as the system 10 or 10A or 10B), and the At least one computing device implements at least one of the aforementioned multiple embodiments of the intelligent monitoring method for detecting anomalies in a case using artificial intelligence. This method can be one or any combination of all the above-mentioned embodiments including the method in FIG. 2. For example, the program code is, for example, one or more programs or program modules, such as used to implement steps S10 to S50 according to FIG. 2, and can be executed in any suitable sequence. When at least one computing device (such as the system 10 or 10A or 10B) executes this code, it can cause the computing device to execute the embodiment based on the operating method of FIG. 2. Examples of these storage media include, but are not limited to: optical information storage media, magnetic information storage media, hard drives, solid state drives, or memory, such as memory cards, firmware, or ROM or RAM. For example, the computing device includes a communication unit, at least one processing unit, and at least one storage device, wherein the processing unit is electrically coupled to the communication unit and the memory unit. The storage device includes, for example, a primary storage device and/or an auxiliary storage device, such as any of the foregoing storage media. The at least one processing unit is used to communicate with the communication network in a wireless or wired manner through the communication unit, so as to communicate with other computing devices such as terminal devices. The processing unit may include one or more processors, and the computing device may also include other devices such as graphics processors to perform operations. In some embodiments, the computing device can run an operating system, and can further utilize one or more of various technologies such as web services, script engines, web applications, or web application programming interface (API) servers. And to achieve, to provide application services for the user's browser, applications, etc. to use.

The system implemented by the intelligent monitoring technology for detecting anomalies in a case using artificial intelligence provided above in the present invention can be flexible and expandable in terms of reflecting the update or change of anomalous problems. For example, a network service application that can help realize intelligent monitoring of case anomaly detection, and also help to derive other applications such as a work order system based on the monitoring results.

The present invention has been disclosed above in a preferred embodiment, but those skilled in the art should understand that the embodiment is only used to describe the present invention and should not be construed as limiting the scope of the present invention. It should be noted Yes, all changes and substitutions equivalent to this embodiment should be included in the scope of the present invention. Therefore, the protection scope of the present invention should be defined by the scope of the patent application.

S10~S50: steps

Claims (14)

An intelligent monitoring system for case field abnormality detection using artificial intelligence, comprising: at least one processing unit; and at least one storage device, coupled to the at least one processing unit and storing a plurality of instructions, when the instructions are When at least one processing unit is executed, the at least one processing unit realizes multiple operations, and the operations include: (a) receiving a set of sensing data from a case and obtaining an input data set based on it; (b) the input data The set is applied to a first prediction engine and outputs a prediction result to predict whether an abnormal event occurs, wherein the first prediction engine includes a plurality of prediction models, and the prediction result is generated by at least one of the prediction models; c) For the prediction result indicating the occurrence of a predicted abnormal event, use the input data set to generate a notification message for the predicted abnormal event; (d) receive at least one feedback message associated with the notification message; (e) basis The at least one feedback message updates the first prediction engine to obtain a second prediction engine; wherein the second prediction engine is used to replace the first prediction engine to predict whether there is an abnormal event based on the subsequent input data set of the case occur. The system according to claim 1, wherein the prediction models of the first prediction engine include an unsupervised model and at least one supervised model, and one of the output of the unsupervised model is located at the input of the at least one supervised model Upstream. The system according to claim 2, wherein in the operation (b), the input data set is applied to the unsupervised learning module to classify the input data set into at least a first sub-data set and a At least one of the second sub-data sets, wherein the second sub-data set corresponding to the abnormal data is applied to the at least one supervised model, and each of the at least one unsupervised model outputs a corresponding prediction result. The system according to claim 1, wherein the operation (c) includes: According to the input data corresponding to the prediction result, an event data is collected, and the event data includes an abnormal occurrence time value; the predicted abnormality related information associated with the prediction result is obtained from a database, wherein the prediction result corresponds to the One of the first prediction models among other prediction models; the notification message is generated based on the event data and the abnormal information related to the prediction. The system according to claim 1, wherein the operation (e) includes: (e1) determining an adjustment indicator value for updating the first prediction engine based on the at least one feedback message; and (e2) determining an adjustment indicator value based on the adjustment indicator value Fine-tune one of the prediction models related to the prediction result, or add a new prediction model to the first prediction engine. The system according to claim 1, wherein the prediction models of the first prediction engine include at least one supervised model and an unsupervised model, and the output of the at least one supervised model is located upstream of the input of the unsupervised model . A method for intelligent monitoring of case field anomaly detection using artificial intelligence. The method is executed by at least one computing device. The method includes: (a) receiving a case field sensing data set and obtaining an input data accordingly (B) The input data set is applied to a first prediction engine and a prediction result is output to predict whether an abnormal event occurs, wherein the first prediction engine includes a plurality of prediction models, and the prediction results are determined by the predictions At least one of the models is generated; (c) for the prediction result indicating the occurrence of a predicted abnormal event, use the input data set to generate a notification message for the predicted abnormal event; (d) receive a notification message associated with the notification message At least one feedback message; (e) updating the first prediction engine according to the at least one feedback message to obtain a second prediction engine; The second prediction engine is used to replace the first prediction engine to predict whether an abnormal event occurs based on the subsequent input data set of the case. The method according to claim 7, wherein the prediction models of the first prediction engine include an unsupervised model and at least one supervised model, and one of the output of the unsupervised model is located at the input of the at least one supervised model Upstream. The method according to claim 8, wherein in the step (b), the input data set is applied to the unsupervised learning module, so as to classify the input data set into at least a first sub-data set and a At least one of the second sub-data sets, wherein the second sub-data set corresponding to the abnormal data is applied to the at least one supervised model, and each of the at least one unsupervised model outputs a corresponding prediction result. The method according to claim 7, wherein the step (c) includes: obtaining an event data collectively according to the input data corresponding to the prediction result, the event data including an abnormal occurrence time value; The predicted abnormality-related information associated with the predicted result, where the predicted result corresponds to a first prediction model of one of the prediction models; the notification message is generated based on the event data and the predicted abnormality-related information. The method according to claim 7, wherein in the step (d), the at least one feedback message includes a first feedback data and a second feedback data; the first feedback data is used to characterize the predicted abnormal event Whether it is the feedback of the abnormality of the case; the second feedback data is used to characterize the abnormality-related information of the abnormality of the case. The method according to claim 7, wherein the step (e) includes: (e1) determining an adjustment indicator value for updating the first prediction engine according to the at least one feedback message; and (e2) According to the adjustment instruction value, fine-tune one of the prediction models related to the prediction result, or add a prediction model to the first prediction engine. The method according to claim 7, wherein the prediction models of the first prediction engine include at least one supervised model and an unsupervised model, and the output of the at least one supervised model is located upstream of the input of the unsupervised model . A storage medium storing instructions that can be read by an arithmetic device, wherein when the instruction is executed by at least one arithmetic device, the at least one arithmetic device implements artificial intelligence as described in any one of claims 7 to 13 Intelligent monitoring method for field anomaly detection.

Images