CN116661426B

CN116661426B - Abnormal AI diagnosis method and system of sensor operation control system

Info

Publication number: CN116661426B
Application number: CN202310865515.7A
Authority: CN
Inventors: 高春亚; 赵正军; 张菊; 杨建国
Original assignee: Chuangyu Intelligent Changshu Netlink Technology Co ltd
Current assignee: Chuangyu Intelligent Changshu Netlink Technology Co ltd
Priority date: 2023-07-14
Filing date: 2023-07-14
Publication date: 2023-09-22
Anticipated expiration: 2043-07-14
Also published as: CN116661426A

Abstract

The embodiment of the application provides an abnormal AI diagnosis method and an abnormal AI diagnosis system for a sensor operation control system, wherein the training convergence indexes of a feedforward neural network and a naive Bayesian network are respectively adjusted, so that an abnormal root cause thermal value is combined in a first training convergence index of the feedforward neural network, and an operation access observation effect value is combined in a second training convergence index of the naive Bayesian network.

Description

Abnormal AI diagnosis method and system of sensor operation control system

Technical Field

The embodiment of the application relates to the technical field of sensors, in particular to an abnormal AI diagnosis method and system of a sensor operation control system.

Background

The sensor detection system is a product of the development of a sensing technology to a certain stage, and is an organic combination of a sensor, a measuring instrument, a conversion device, a server and the like. In engineering practice, a sensor and a plurality of measuring instruments are required to be organically combined to form a whole to finish the detection of related data, so that a sensor detection system is formed. With the continuous development of computer technology and information processing technology, the content related to a sensor detection system is also filled continuously, and in the modern production process, the detection of the process parameters of the sensor detection system is automatically controlled by a sensor operation control system. Therefore, when an unknown abnormal fault exists in the sensor operation control system, diagnosis and investigation are required to be performed in time, in the related technology, abnormal optimization and adjustment of operation access nodes can be conveniently performed by determining the abnormal root cause of the system, such as operation access nodes related to the abnormal root cause of the system, and how to improve the accuracy of analysis of the abnormal root cause is a technical problem to be solved urgently in the technical field.

Disclosure of Invention

In order to at least overcome the above-mentioned shortcomings in the prior art, an object of an embodiment of the present application is to provide an abnormal AI diagnosis method and system for a sensor operation control system.

According to an aspect of the embodiment of the present application, there is provided an abnormal AI diagnosis method of a sensor operation control system, including:

acquiring an example sensor operation control sample with a system operation abnormal root cause;

determining an operation access observation node corresponding to the sample sensor operation control sample by using the operation state description of the sample sensor operation control sample and a feedforward neural network of a sensor operation abnormality diagnosis model, wherein the operation access observation node is an operation access node for an abnormal root predicted in the sample sensor operation control sample, the operation access node is used for describing a sensor control operation of the sample sensor indicated in the sample sensor operation control sample when the abnormal state exists, and the sensor control operation is triggered after an operation access instruction is received by a sensor control system;

calculating an operation access observation effect value of the operation access observation node by using an operation access sample node corresponding to an example sensor operation control sample, and calculating a first training convergence index by using the operation access observation effect value, wherein the operation access sample node is an operation access node where an abnormal root cause marked in the example sensor operation control sample is located;

Determining an abnormal root cause thermal value of the operation access observation node by using the operation state description of the example sensor operation control sample and a naive Bayesian network of the sensor operation abnormality diagnosis model, and calculating a second training convergence index by using the abnormal root cause thermal value;

adjusting the second training convergence index by using the operation access observation effect value, and adjusting the first training convergence index by using the abnormal root cause thermal value;

adjusting the model function definition function weight of the sensor operation abnormality diagnosis model by combining the adjusted second training convergence index and the adjusted first training convergence index until the sensor operation abnormality diagnosis model is not changed any more, and generating a sensor operation abnormality diagnosis model which can be deployed and used;

the sensor operation abnormality diagnosis model capable of being deployed and used is used for diagnosing sensor operation abnormality of input target sensor operation control data.

In an alternative embodiment, before the adjusting the model function definition function weight of the sensor abnormality diagnostic model in combination with the adjusted second training convergence criterion and the adjusted first training convergence criterion, the method further comprises:

Determining an operation access prediction effect value of the operation access observation node by using an observation effect prediction network of the sensor operation abnormality diagnosis model;

determining a third training convergence index by using the operation access prediction effect value of the operation access observation node and the operation access observation effect value of the operation access observation node;

the adjusting the model function definition function weight of the sensor operation abnormality diagnosis model by combining the adjusted second training convergence index and the adjusted first training convergence index comprises:

and adjusting the model function definition function weight of the sensor abnormal operation diagnosis model by combining the third training convergence index, the adjusted second training convergence index and the adjusted first training convergence index.

In an alternative embodiment, the exception root is due to an exception fault root;

the obtaining an example sensor operation control sample having a root cause of a system operation anomaly includes:

acquiring an example sensor operation control sample with a system operation abnormal root cause and a collaborative learning sensor operation control sample corresponding to the example sensor operation control sample; the collaborative learning sensor operation control sample and the example sensor operation control sample include a matching anomaly root cause;

The determining, by using the feedforward neural network of the operational state description of the example sensor operational control sample and the sensor operational anomaly diagnostic model, an operational access observation node corresponding to the example sensor operational control sample includes:

determining a first collaborative presentation state description between the example sensor operation control sample and the collaborative learning sensor operation control sample using the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample;

importing the first collaborative expression state description into the feedforward neural network, and determining an operation access observation node corresponding to the operation control sample of the example sensor;

the determining, by the observation effect prediction network of the sensor operation anomaly diagnosis model, an operation access prediction effect value of the operation access observation node includes:

importing the first collaborative expression state description into the observation effect prediction network, and determining an operation access prediction effect value of the operation access observation node;

the determining, using the naive bayesian network of the operational state description of the example sensor operational control sample and the sensor operational anomaly diagnostic model, an anomaly root cause thermal value for the operational access observation node comprises:

Determining a second collaborative presentation state description between the example sensor operation control sample and the collaborative learning sensor operation control sample using the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample;

importing the second collaborative expression state description into the naive Bayesian network, and determining an abnormal root cause thermal value of the operation access observation node;

the first collaborative presentation state description and the second collaborative presentation state description reflect an operational logical correlation between the example sensor operational control sample and the collaborative learning sensor operational control sample.

In an alternative embodiment, prior to the determining the first collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample using the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample, the method further comprises:

acquiring an operation state description of the example sensor operation control sample and an operation state description of the collaborative learning sensor operation control sample by using the sensor operation anomaly diagnostic model;

The determining a first collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample using the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample includes:

performing feature selection of a global optimal search strategy on the operation state description of the sample sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample by using the sensor operation abnormality diagnosis model respectively, and generating a first sample global optimal search feature of the sample sensor operation control sample and a first collaborative global optimal search feature of the collaborative learning sensor operation control sample;

and carrying out blending processing on the first example global optimal search feature and the first collaborative global optimal search feature by using the sensor operation abnormality diagnosis model to generate a first collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample.

In an alternative embodiment, the sensor abnormal operation diagnosis model comprises two characteristic state description units with the same model framework configuration data and sharing model definition function weight information;

The obtaining an operational state description of the example sensor operational control sample and an operational state description of the collaborative learning sensor operational control sample using the sensor operational anomaly diagnostic model includes:

respectively importing the sample sensor operation control sample and the collaborative learning sensor operation control sample into the two feature state description units;

and synchronously generating the running state description of the running control sample of the example sensor and the running state description of the running control sample of the collaborative learning sensor by combining the two characteristic state description units.

In an alternative embodiment, the calculating, by using the operation access sample node corresponding to the operation control sample of the example sensor, the operation access observation effect value of the operation access observation node includes:

acquiring a cross access node vector between the operation access sample node and the operation access observation node;

acquiring an aggregation access node vector between the operation access sample node and the operation access observation node;

and determining the operation access observation effect value by combining the characteristic duty ratio of the cross access node vector corresponding to the operation access observation node and the convergence access node vector.

In an alternative embodiment, said calculating a second training convergence criterion using the outlier root cause thermal value includes:

acquiring the characteristic duty ratio of the cross access node vector and the convergence access node vector corresponding to the operation access observation node;

when the characteristic duty ratio is larger than a first target value, determining a second training convergence index by combining active abnormal learning basis data corresponding to the sample operation control sample of the example sensor and the abnormal root cause thermal value;

and when the characteristic duty ratio is smaller than a second target value, determining a second training convergence index by combining the negative abnormal learning basis data corresponding to the sample operation control sample of the example sensor and the abnormal root cause thermal value.

In an alternative embodiment, the exception root is due to an exception fault root; the method further comprises the steps of:

acquiring a sensor operation scheduling flow;

determining a first sensor operation control sample comprising the abnormal fault root cause in the sensor operation scheduling flow, and generating a target sensor operation control sample;

Sequentially acquiring a sensor operation control sample from a sensor operation control sample of a next node of the first sensor operation control sample, wherein the sensor operation control sample is used as a route operation control sample;

respectively importing the route operation control sample and the target sensor operation control sample into the sensor operation abnormality diagnosis model;

acquiring a first collaborative expression feature and a second collaborative expression feature between the route operation control sample and the target sensor operation control sample by using the sensor operation abnormality diagnosis model;

importing the first collaborative expression feature into a feedforward neural network of the sensor operation anomaly diagnostic model, and determining a plurality of operation access observation nodes of the routing operation control sample;

importing the second collaborative expression feature into a naive Bayesian network of the sensor operation anomaly diagnosis model, and determining an anomaly root cause thermal value of each operation access observation node of the route operation control sample;

and determining the operation access node corresponding to the abnormal fault root cause from the route operation control sample by combining the abnormal root cause thermal value of each operation access observation node of the route operation control sample.

In an alternative embodiment, the method further comprises:

acquiring input operation control data of a target sensor;

acquiring a sensor operation abnormality diagnosis model which can be deployed and used; the sensor operation abnormality diagnosis model comprises a naive Bayesian network and a feedforward neural network;

determining a plurality of operational access observation nodes of the input target sensor operational control data using the operational state description of the input target sensor operational control data and the feed forward neural network;

determining abnormal root cause thermal values of each operation access observation node of the input target sensor operation control data by using the operation state description of the input target sensor operation control data and the naive Bayesian network;

carrying out sensor operation abnormality diagnosis on the input target sensor operation control data by combining the abnormal root cause thermal value of each operation access observation node of the input target sensor operation control data, and generating an operation access node where the abnormal root cause is located;

the abnormal root is an abnormal fault root cause; the acquiring input target sensor operation control data includes:

acquiring input target sensor operation control data and a cooperative sensor operation control sample corresponding to the input target sensor operation control data; the collaborative sensor operation control sample is sample data with a system operation abnormal root cause in an operation control scheduling task where the input target sensor operation control data is located and before the input target sensor operation control data;

The operating state description using the input target sensor operating control data and the feedforward neural network determining a plurality of operating access observer nodes of the input target sensor operating control data includes:

determining a first collaborative presentation feature between the input target sensor operational control data and the collaborative sensor operational control sample using the operational state description of the input target sensor operational control data and the operational state description of the collaborative sensor operational control sample;

importing the first collaborative expression feature into the feedforward neural network, and determining a plurality of operation access observation nodes of the input target sensor operation control data;

the operating state description of the input target sensor operating control data and the naive bayes network determining abnormal root cause thermal values of each operating access observation node of the input target sensor operating control data include:

determining a second collaborative presentation characteristic between the input target sensor operational control data and the collaborative sensor operational control sample using the operational state description of the input target sensor operational control data and the operational state description of the collaborative sensor operational control sample;

Importing the second collaborative expression feature into the naive Bayesian network, and determining abnormal root cause thermal values of each operation access observation node of the input target sensor operation control data;

the sensor operation abnormality diagnosis model further comprises an observation effect prediction network;

before the abnormal root cause thermal value of each operation access observation node in combination with the input target sensor operation control data locates an abnormal root cause from the input target sensor operation control data, the method further comprises:

importing the first collaborative expression feature into the observation effect prediction network, and determining an operation access prediction effect value of each operation access observation node of the input target sensor operation control data; carrying out sensor operation abnormality diagnosis on the input target sensor operation control data by combining the abnormal root cause thermal value of each operation access observation node of the input target sensor operation control data, wherein the generation of the abnormal root cause located operation access node comprises the following steps:

weighting the operation access prediction effect value and the abnormal root cause thermodynamic value of each operation access observation node of the input target sensor operation control data to generate a target abnormal hit coefficient of each operation access observation node of the input target sensor operation control data;

And carrying out sensor operation abnormality diagnosis on the input target sensor operation control data by combining the target abnormality hit coefficients of all operation access observation nodes of the input target sensor operation control data, and generating an abnormality root cause of the operation access nodes.

According to one aspect of the embodiments of the present application, there is provided an abnormality AI diagnosis system of a sensor operation control system including a processor and a machine-readable storage medium having stored therein machine-executable instructions loaded and executed by the processor to implement the abnormality AI diagnosis method of the sensor operation control system in any one of the foregoing possible implementations.

According to an aspect of embodiments of the present application, there is provided a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The computer instructions are read from a computer-readable storage medium by a processor of a computer device, which executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations of the three aspects described above.

In the technical scheme provided by some embodiments of the present application, firstly, an example sensor operation control sample with a system operation anomaly root cause is obtained, an operation state description of the example sensor operation control sample and a feed-forward neural network of a sensor operation anomaly diagnosis model are utilized to determine an operation access observation node corresponding to the example sensor operation control sample, an operation access observation effect value of the operation access observation node is calculated by utilizing the operation access sample node corresponding to the example sensor operation control sample, a first training convergence index is calculated by utilizing the operation access observation effect value, an operation state description of the example sensor operation control sample and a naive Bayesian network of the sensor operation anomaly diagnosis model are utilized to determine an anomaly root cause heating value of the operation access observation node, a second training convergence index is calculated by utilizing the anomaly root cause heating value, then the second training convergence index is regulated by utilizing the operation access observation effect value, the first training convergence index is regulated by utilizing the anomaly root cause heating value, finally, a first training convergence index is regulated by combining the regulated second training index and the regulated first sensor convergence index, a first training convergence index, a differential function is regulated by utilizing the operation state description of the example sensor operation control sample and the naive Bayesian network of the sensor operation anomaly diagnosis model is utilized, the anomaly root cause heating value is not regulated by utilizing the first training convergence index, and the differential function is deployed in the anomaly detection model, and the anomaly detection model is used to generate an anomaly detection model, and the anomaly detection model is used to make the anomaly detection model is used to change in the anomaly detection model, after the sensor operation abnormality diagnosis model is trained through the training convergence index after adjustment, the generated deployable sensor operation abnormality diagnosis model weakens the difference between the naive Bayesian network and the feedforward neural network, and improves the accuracy of the abnormal root cause thermal value prediction of the operation access observation node, thereby improving the accuracy of the abnormal root cause analysis.

Drawings

For a clearer description of the technical solutions of the embodiments of the present application, reference will be made to the accompanying drawings, which are needed to be activated in the embodiments, and it should be understood that the following drawings only illustrate some embodiments of the present application, and therefore should not be considered as limiting the scope, and other related drawings can be extracted by those skilled in the art without the inventive effort.

FIG. 1 is a schematic flow chart of an abnormal AI diagnostic method of a sensor operation control system provided by an embodiment of the application;

fig. 2 is a schematic block diagram of an abnormality AI diagnosis system of a sensor operation control system for implementing the abnormality AI diagnosis method of a sensor operation control system according to an embodiment of the present application.

Detailed Description

The following description is presented to enable one of ordinary skill in the art to make and use the application and is provided in the context of a particular application and its requirements. It will be apparent to those having ordinary skill in the art that various changes can be made to the disclosed embodiments and that the general principles defined herein may be applied to other embodiments and applications without departing from the principles and scope of the application. Therefore, the present application is not limited to the described embodiments, but is to be accorded the widest scope consistent with the claims.

Fig. 1 is a flowchart of an abnormality AI diagnosis method of a sensor operation control system according to an embodiment of the present application, and the abnormality AI diagnosis method of the sensor operation control system will be described in detail.

Step 102, obtaining an example sensor operation control sample having a root cause of a system operation anomaly.

The system operation abnormal root can be used for representing the source of an abnormal fault point existing in the sensor operation control sample, for example, a certain sensor scheduling time, a certain control flow or a sub-flow of a certain control flow, and specifically, the abnormal root can be marked in advance. The example sensor operation control sample is used for training a sensor operation anomaly diagnostic model, and operation access sample nodes and priori anomaly root cause data exist in the example sensor operation control sample. Wherein the prior exception root factor refers to an associated operational access node of the exception root factor.

That is, the sensor operation abnormality diagnosis described in the following embodiments may refer to determining the operation access node where the abnormality root cause is located from the sensor operation control sample.

Step 104, determining an operation access observation node corresponding to the sample sensor operation control sample by using the operation state description of the sample sensor operation control sample and the feedforward neural network of the sensor operation abnormality diagnosis model.

The sensor operation abnormality diagnosis model is a machine learning model which can determine an abnormal root cause of an operation access node where an arbitrary input sensor operation control sample is located. The sensor abnormal operation diagnosis model comprises a feedforward neural network and a naive Bayesian network. The feedforward neural network is used for estimating operation access nodes and determining operation access observation nodes corresponding to the operation control samples of the example sensors. The operation access observation node may be one or more operation access nodes, where the operation access node is an operation access node where an abnormal root predicted in the operation control sample of the example sensor is located, the operation access node is used to describe a sensor control operation of the example sensor indicated in the operation control sample of the example sensor when an abnormal state exists, the sensor control operation is an operation triggered after an operation access instruction is received by a sensor control system, and may include an operation of the example sensor and an operation of the sensor control system, and the operation access instruction may be understood as an instruction generated by a computer device according to a user operation to perform a corresponding sensor control operation, and may include instruction content information of the sensor control operation. In some exemplary embodiments, the operation of the example sensor itself may include, but is not limited to, an environment sensing operation of the sensor for each measurement object, a cooperative control operation between other sensors, a signal conversion operation for measurement signals, a signal amplification operation, a signal filtering operation, a signal calibration operation, etc., and the operation of the sensor control system may include, but is not limited to, a sensor instruction issuing operation, a sensor status switching operation, a sensor data uploading control operation, etc., any of which may be used as an operation access node. The naive bayes network is used for classifying the abnormal root causes of the sample data in the operation access observation node, and the abnormal root cause classification result can be that the abnormal root causes exist or no abnormal root causes exist.

The operating state description of the example sensor operating control sample may include a sensor measurement control state description, a sensor upload control state description, a sensor co-control state description, and the like.

For some exemplary design ideas, the running state description of the example sensor running control sample may be extracted, and the running access observation node output may be performed by using the running state description of the example sensor running control sample and the feedforward neural network of the sensor running anomaly diagnostic model, so as to determine the running access observation node corresponding to the example sensor running control sample.

For some exemplary design considerations, the sensor operation anomaly diagnostic model further includes a feature state description unit that inputs the example sensor operation control sample into the feature state description unit of the sensor operation anomaly diagnostic model, generating an operation state description of the example sensor operation control sample.

For some exemplary design ideas, after the operation state description of the example sensor operation control sample is extracted, the operation state description may be imported into a feedforward neural network of the sensor operation anomaly diagnosis model to determine an operation access observation node corresponding to the example sensor operation control sample.

For some exemplary design ideas, when an abnormal root causes an abnormal fault, and when an example sensor operation control sample with a system operation abnormal root cause is obtained, a collaborative learning sensor operation control sample with the same abnormal root cause as the example sensor operation control sample is also obtained at the same time, an operation state description of the example sensor operation control sample and an operation state description of the collaborative learning sensor operation control sample are utilized to determine a first collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample, the first collaborative expression state description is imported into a feedforward neural network, and an operation access observation node corresponding to the example sensor operation control sample is determined.

Wherein the collaborative expression state description is used to represent the same operational state description between the operational state description of the example sensor operational control sample and the operational state description of the collaborative learning sensor operational control sample. That is, the collaborative presentation state description may be understood as an operation state description that exists in both the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample. The operation state description may be understood as describing data variables during operation control of the sensor, for example describing sensor state variables during control of the sensor for signal conversion operations, signal amplification operations, signal filtering operations, signal calibration operations, etc.

And 106, calculating a running access observation effect value of the running access observation node by using a running access sample node corresponding to the running control sample of the sample sensor, and calculating a first training convergence index by using the running access observation effect value.

The operation access sample node is the operation access node where the abnormal root cause marked in the example sensor operation control sample is located. The operation access observation effect value reflects the validity of the operation access observation node, for example, the reciprocal determination of the characteristic distance between the first sensor control operation corresponding to the operation access observation node and the second sensor control operation corresponding to the operation access sample node can be calculated. For example, feature vectors corresponding to different sensor control operations may be predefined, for example, the sensor instruction issuing operation A1 is predefined to be (1, 3), and the sensor state switch operation A2 is predefined to be (1, 5), then the feature distance may be a euclidean distance between A1 and A2, that is, the euclidean distance calculation result is 2, then the corresponding operation access observation effect value is 0.5, and the first training convergence index may be the operation access observation effect value, or a weighted value of the operation access observation effect value and a preset weight coefficient, which is not specifically limited. The first training convergence index reflects a loss function value between the operation access sample node and the operation access observation node, and the larger the loss function value between the operation access sample node and the operation access observation node is, the larger the first training convergence index is.

For some exemplary design considerations, the cross-scale between the operational access observation node and the operational access sample node may be utilized to determine an operational access observation effect value for the operational access observation node, and further utilize the operational access observation effect value to calculate the first training convergence index.

When the example sensor operation control sample corresponds to a plurality of operation access observation nodes, the shared feature ratio between each operation access observation node and the operation access sample node may be used to determine an operation access observation effect value for each operation access observation node. Accordingly, the operation access observation effect value of each operation access observation node can be used for respectively calculating to obtain the first training convergence index of each operation access observation node.

For some exemplary design ideas, cross access node vectors between the operation access sample nodes and the operation access observation nodes can be obtained, meanwhile, convergence access node vectors between the operation access sample nodes and the operation access observation nodes are obtained, and the operation access observation effect value is determined by combining the cross access node vectors corresponding to the operation access observation nodes and the convergence access node vectors.

For some exemplary design considerations, after calculating the operation access observation effect value, a first training convergence index may be calculated by (1-operation access observation effect value).

Step 108, determining an abnormal root cause thermal value of the operation access observation node by using the operation state description of the example sensor operation control sample and the naive Bayesian network of the sensor operation abnormality diagnosis model, and calculating a second training convergence index by using the abnormal root cause thermal value.

The abnormal root cause thermal value reflects the probability of the abnormal root cause existing in a certain operation access observation node, the second training convergence index reflects the prediction precision of the abnormal root cause thermal value, and the second training convergence index is inversely related to the prediction precision of the abnormal root cause thermal value, namely the higher the prediction precision of the abnormal root cause thermal value is, the smaller the second training convergence index is.

The calculation formula of the abnormal root cause thermal value can be expressed as follows:

wherein x is ₁ ，x ₂ ，...，x _d An operation state description for representing an example sensor operation control sample, j is any positive integer between 1 and d, d is the number of operation state description features, y _i Root cause description variable for representing candidate abnormal root causes, i is the number of candidate abnormal root causes, P (y) _i |x ₁ ，x ₂ ，...，x _d ) The probability that the operating state description representing the example sensor operating control sample matches the candidate anomaly root cause, i.e., the anomaly root cause thermal value, P (x _j ) Description of the operation states x _j Matching the prior probability of the candidate abnormal root cause, P (x) _j |y _i ) Description of the operation states x _j And matching the posterior probability of the candidate abnormal root cause.

For some exemplary design ideas, when the operation state description of the example sensor operation control sample is extracted, the operation state description of the example sensor operation control sample and the naive bayes network of the sensor operation abnormality diagnosis model can be utilized to determine an abnormal root cause heating value of each operation access observation node, for each operation access observation node, whether the operation access observation node is a positive operation access observation node or a negative operation access observation node is further judged, when the operation access observation node is a positive knowledge point, the abnormal root cause heating value and positive abnormality learning basis data are utilized to calculate a second training convergence index, and when the operation access observation node is a negative knowledge point, the abnormal root cause heating value and the negative abnormality learning basis data are utilized to calculate a second training convergence index.

For some exemplary design considerations, the running state description of the example sensor running control sample may be imported into a naive bayes network of the sensor running anomaly diagnostic model to determine an anomaly root cause thermal value for each of the running access observation nodes corresponding to the example sensor running control sample.

For some exemplary design ideas, when the example sensor operation control sample with the system operation abnormal root factor is obtained, the collaborative learning sensor operation control sample with the example sensor operation control sample including the same abnormal root factor is also obtained at the same time, and then the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample can be utilized to determine a second collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample, the second collaborative expression state description is imported into a naive bayes network, and the abnormal root factor thermal value of each operation access observation node corresponding to the example sensor operation control sample is determined. Wherein the second collaborative expression state description herein is generally a state description different from the first collaborative expression state description of the above-described embodiment.

Step 110, adjusting the second training convergence index by using the operation access observation effect value, and adjusting the first training convergence index by using the abnormal root cause thermal value.

For some exemplary design ideas, the operation access observation effect value of each operation access observation node can be weighted with the respective second training convergence index of each operation access observation node to adjust the second training convergence index of each operation access observation node, so that the auxiliary model convergence step is set, the determined accuracy of the abnormal root cause thermal value of the operation access observation node is higher, the abnormal root cause thermal value of each operation access observation node can be weighted with the respective first training convergence index of each operation access observation node to adjust the first training convergence index of each operation access observation node, the operation access observation node with higher accuracy of the abnormal root cause thermal value is more accurate, and the variability between the naive Bayesian network and the feedforward neural network is also weakened.

And step 112, adjusting the sensor operation abnormality diagnosis model by combining the adjusted second training convergence index and the adjusted first training convergence index until the sensor operation abnormality diagnosis model is not changed any more, and generating a deployable sensor operation abnormality diagnosis model.

For some exemplary design ideas, the adjusted second training convergence index and the adjusted first training convergence index may be superimposed to generate a global training convergence index, and then the sensor operation anomaly diagnostic model is trained in combination with the global training convergence index until the sensor operation anomaly diagnostic model is no longer changed, to generate a deployable sensor operation anomaly diagnostic model.

The sensor operation abnormality diagnosis model can be deployed for sensor operation abnormality diagnosis of the input target sensor operation control data. For some exemplary design ideas, after the input target sensor operation control data is acquired, determining a plurality of operation access observation nodes of the input target sensor operation control data by using a feedforward neural network of an operation state description of the input target sensor operation control data and a sensor operation abnormality diagnosis model which can be deployed, determining abnormal root cause thermal values of all operation access observation nodes by using a naive Bayesian network of the operation state description of the input target sensor operation control data and the sensor operation abnormality diagnosis model which can be deployed, and selecting the operation access observation node with the highest abnormal root cause thermal value as a corresponding target operation access observation node of the input target sensor operation control data.

By adopting the steps, firstly, an example sensor operation control sample with a system operation abnormal root cause is obtained, an operation state description of the example sensor operation control sample and a feedforward neural network of a sensor operation abnormal diagnosis model are utilized to determine an operation access observation node corresponding to the example sensor operation control sample, an operation access observation effect value of the operation access observation node is calculated by utilizing the operation access observation effect value, a first training convergence index is calculated by utilizing the operation access observation effect value, an operation state description of the example sensor operation control sample and the naive Bayesian network of the sensor operation abnormal diagnosis model are utilized to determine an abnormal root cause thermal value of the operation access observation node, a second training convergence index is calculated by utilizing the abnormal root cause thermal value, further, the second training convergence index is regulated by utilizing the operation access observation effect value, the first training convergence index is regulated by utilizing the abnormal root cause thermal value, finally, the sensor operation abnormal diagnosis model is regulated by combining the regulated second training convergence index and the regulated first training convergence index, a deployable sensor operation abnormal diagnosis index is generated when no change occurs to the sensor operation abnormal diagnosis model, the first training convergence index is regulated by utilizing the first training convergence index, the neural network is deployed in the neural network, and the first training convergence index is regulated by combining the abnormal operation abnormal root cause thermal value with the neural network after the first training convergence index is regulated by the first training convergence index, the difference between the naive Bayesian network and the feedforward neural network is weakened, and the accuracy of the abnormal root cause thermal value prediction of the operation access observation node is improved, so that the accuracy of the abnormal root cause analysis is improved.

For some exemplary design ideas, there is further provided a training method of a sensor operation abnormality diagnosis model, including the following steps 1.1 to 1.8:

1.1, acquiring an example sensor operation control sample with a system operation abnormal root cause.

1.2, determining an operation access observation node corresponding to the sample sensor operation control sample by utilizing the operation state description of the sample sensor operation control sample and a feedforward neural network of the sensor operation abnormality diagnosis model.

And 1.3, calculating an operation access observation effect value of the operation access observation node by using an operation access sample node corresponding to the operation control sample of the example sensor, and calculating a first training convergence index by using the operation access observation effect value.

And 1.4, determining an abnormal root cause thermal value of the operation access observation node by using the operation state description of the example sensor operation control sample and the naive Bayesian network of the sensor operation abnormality diagnosis model, and calculating a second training convergence index by using the abnormal root cause thermal value.

And 1.5, adjusting the second training convergence index by using the operation access observation effect value, and adjusting the first training convergence index by using the abnormal root cause thermal value.

And 1.6, determining an operation access prediction effect value of the operation access observation node by using an observation effect prediction network of the sensor operation abnormality diagnosis model.

For some exemplary design considerations, the sensor anomaly diagnostic model further includes an observation prediction network for predicting operational access observation values for each operational access observation node.

For some exemplary design considerations, the operational state descriptions of the example sensor operational control samples may be imported into an observation effect prediction network to generate operational access prediction effect values for each operational access observation node.

For some exemplary design ideas, when the abnormal root is due to the abnormal fault root, as known from the above embodiments, the running state description of the example sensor running control sample and the running state description of the collaborative learning sensor running control sample may be used to determine a first collaborative expression state description between the example sensor running control sample and the collaborative learning sensor running control sample, and then the first collaborative expression state description may be imported into the observation effect prediction network to generate the running access prediction effect value of each running access observation node.

And 1.7, determining a third training convergence index by using the operation access prediction effect value of the operation access observation node and the operation access observation effect value of the operation access observation node.

The third training convergence index reflects a loss function value between the operation access prediction effect value and the actual operation access observation effect value of the operation access observation node, and the larger the loss function value is, the larger the third training convergence index is, otherwise, the smaller the loss function value is, the smaller the third training convergence index is.

And 1.8, adjusting the sensor operation abnormality diagnosis model by combining the third training convergence index, the adjusted second training convergence index and the adjusted first training convergence index.

For some exemplary design ideas, a third training convergence index, an adjusted second training convergence index, and an adjusted first training convergence index may be superimposed to obtain a global training convergence index, and the sensor operational anomaly diagnostic model is trained using the global training convergence index.

The sensor operation abnormity diagnosis model is adjusted through the third training convergence index, the adjusted second training convergence index and the adjusted first training convergence index, the training convergence indexes of the feedforward neural network and the naive Bayesian network are respectively adjusted, so that the abnormal root cause heating power value is combined in the first training convergence index of the feedforward neural network, the operation visit observation effect value is combined in the second training convergence index of the naive Bayesian network, the difference between the naive Bayesian network and the feedforward neural network is weakened, in addition, the third training convergence index of the observation effect prediction network is added in the training convergence index, and the generated sensor operation abnormity diagnosis model capable of being deployed and used is further combined with the operation visit observation effect value, so that the sensor operation abnormity diagnosis accuracy is further improved.

For some exemplary design considerations, a training embodiment of a sensor abnormality diagnostic model is provided, specifically comprising the following steps 2.1 through 2.11:

2.1, acquiring an example sensor operation control sample with a system operation abnormal root cause and a collaborative learning sensor operation control sample corresponding to the example sensor operation control sample.

Wherein the collaborative learning sensor operation control sample and the example sensor operation control sample include matching anomaly root causes.

2.2, acquiring the running state description of the sample sensor running control sample by using a characteristic state description unit of the sensor running abnormality diagnosis model and cooperatively learning the running state description of the sample sensor running control sample.

For some exemplary design ideas, since features need to be extracted for both the example sensor operation control sample and the collaborative learning sensor operation control sample, then the feature state description unit may employ a twin network, where the feature state description unit in the sensor operation anomaly diagnosis model configures two feature state description units with the same data for the model frame and sharing the model definition function weight information. The step 2.2 specifically comprises the following steps: respectively inputting an example sensor operation control sample and a collaborative learning sensor operation control sample into two characteristic state description units; the operation state description of the sample sensor operation control sample and the operation state description of the cooperative learning sensor operation control sample are synchronously generated through the two characteristic state description units.

For some exemplary design ideas, two feature state description units are provided, which are identical in terms of model framework configuration data, model definition function weight information. The example sensor operation control sample and the collaborative learning sensor operation control sample are respectively input into the two feature state description units, so that the two feature state description units respectively extract an operation state description of the example sensor operation control sample and a second operation state description of the collaborative learning sensor operation control sample.

2.3 determining a first collaborative presentation state description between the example sensor operation control sample and the collaborative learning sensor operation control sample using the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample.

Wherein the first collaborative expression state description may be the first collaborative expression feature described above, specifically named first collaborative expression state description. The above second cooperatively expressed feature is specifically named as a second cooperatively expressed state description in the present embodiment. The collaborative expression feature fuses the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample, and can reflect the operation logic correlation degree between the example sensor operation control sample and the collaborative learning sensor operation control sample.

For some exemplary design considerations, this step 2.3 specifically includes: performing feature selection of a first global optimal search strategy on the operation state description of the sample sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample by using the sensor operation abnormality diagnosis model respectively to generate a first sample global optimal search feature of the sample sensor operation control sample and a first collaborative global optimal search feature of the collaborative learning sensor operation control sample; and carrying out blending processing on the first example global optimal search feature and the first collaborative global optimal search feature to generate a first collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample.

For some exemplary design ideas, after obtaining a first example global optimal search feature of the example sensor operation control sample and a first collaborative global optimal search feature of the collaborative learning sensor operation control sample, extracting a collaborative expression state description of the first example global optimal search feature and the first collaborative global optimal search feature, and generating a first collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample.

And 2.4, importing the first collaborative expression state description into a feedforward neural network, and determining an operation access observation node corresponding to the operation control sample of the example sensor.

And 2.5, calculating an operation access observation effect value of the operation access observation node by using an operation access sample node corresponding to the operation control sample of the example sensor, and calculating a first training convergence index by using the operation access observation effect value.

For some exemplary design ideas, acquiring a cross access node vector between an operation access sample node and an operation access observation node; acquiring an aggregation access node vector between an operation access sample node and an operation access observation node; and determining the operation access observation effect value by combining the characteristic duty ratio of the cross access node vector and the convergence access node vector corresponding to the operation access observation node.

And 2.6, importing the first collaborative expression state description into an observation effect prediction network, and determining an operation access prediction effect value of the operation access observation node.

And 2.7, determining a third training convergence index by using the operation access prediction effect value of the operation access observation node and the operation access observation effect value of the operation access observation node.

After the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample are obtained, the operation state description of the example sensor operation control sample is imported into a first global optimal search network, the operation state description of the collaborative learning sensor operation control sample is imported into a second global optimal search network, the first global optimal search network performs feature selection of a global optimal search strategy on the operation state description of the example sensor operation control sample to obtain a first example global optimal search feature of the example sensor operation control sample, the second global optimal search network performs feature selection of the global optimal search strategy on the operation state description of the collaborative learning sensor operation control sample to obtain a first collaborative global optimal search feature of the collaborative learning sensor operation control sample, then the first example global optimal search feature and the first collaborative global optimal search feature are subjected to blending processing, a first expression state description is generated, and the first collaborative expression state description is respectively input into a feedforward neural network and an effect prediction network.

2.8, determining a second collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample using the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample.

For some exemplary design ideas, feature selection of a second global optimal search strategy can be performed on the operation state description of the sample sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample by using the sensor operation anomaly diagnosis model, so as to generate a second sample global optimal search feature of the sample sensor operation control sample and a second collaborative global optimal search feature of the collaborative learning sensor operation control sample; and carrying out blending processing on the second example global optimal search feature and the second collaborative global optimal search feature to generate a second collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample.

And after obtaining the second example global optimal search feature of the example sensor operation control sample and the second collaborative global optimal search feature of the collaborative learning sensor operation control sample, extracting collaborative expression state description of the second example global optimal search feature and the second collaborative global optimal search feature, and generating a second collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample.

Importing the running state description of the example sensor running control sample into a third global optimal search network, importing the running state description of the collaborative learning sensor running control sample into a fourth global optimal search network, performing feature selection of a global optimal search strategy on the running state description of the example sensor running control sample by the third global optimal search network to obtain a second example global optimal search feature of the example sensor running control sample, and performing feature selection of the global optimal search strategy on the running state description of the collaborative learning sensor running control sample by the fourth global optimal search network to obtain a second collaborative global optimal search feature of the collaborative learning sensor running control sample. And then carrying out blending processing on the second example global optimal search feature and the second collaborative global optimal search feature to generate a second collaborative expression state description, and importing the second collaborative expression state description into a naive Bayesian network.

And 2.9, importing the second collaborative expression state description into a naive Bayesian network, determining an abnormal root cause thermal value of the operation access observation node, and calculating a second training convergence index by using the abnormal root cause thermal value.

And 2.10, adjusting the second training convergence index by using the operation access observation effect value, and adjusting the first training convergence index by using the abnormal root cause thermal value.

And 2.11, adjusting the sensor operation abnormality diagnosis model by combining the third training convergence index, the adjusted second training convergence index and the adjusted first training convergence index until the sensor operation abnormality diagnosis model is not changed any more, and generating the sensor operation abnormality diagnosis model which can be deployed and used.

For some exemplary design ideas, further steps of the embodiments of the present application specifically include:

1. and acquiring a sensor operation scheduling flow, determining a first sensor operation control sample comprising an abnormal fault root cause in the sensor operation scheduling flow, generating a target sensor operation control sample, and sequentially acquiring the sensor operation control samples from the sensor operation control sample of the next node of the first sensor operation control sample to serve as a route operation control sample.

2. And respectively inputting the route operation control sample and the target sensor operation control sample into a sensor operation abnormality diagnosis model, and acquiring a first collaborative expression characteristic and a second collaborative expression characteristic between the route operation control sample and the target sensor operation control sample by using the sensor operation abnormality diagnosis model.

3. The first collaborative expression feature is imported into a feed-forward neural network of a sensor operational anomaly diagnostic model, and a plurality of operational access observation nodes routing operational control samples are determined.

4. And importing the second collaborative expression characteristic into a naive Bayesian network of the sensor operation anomaly diagnosis model, and determining an anomaly root cause thermal value of each operation access observation node of the route operation control sample.

5. And determining the operation access node corresponding to the abnormal fault root cause from the route operation control sample by combining the abnormal root cause thermal value of each operation access observation node of the route operation control sample.

For some exemplary design ideas, the operation access observation node with the largest thermal value of the abnormal root causes can be selected to be determined as the operation access node where the abnormal root causes are located in the route operation control sample, so that the operation access node corresponding to the abnormal root causes is determined from the route operation control sample.

For some exemplary design considerations, the steps of the present application may specifically further include:

4. And leading the first collaborative expression features into an observation effect prediction network of the sensor operation abnormality diagnosis model, and generating operation access prediction effect values corresponding to all operation access observation nodes of the route operation control sample.

5. And importing the second collaborative expression characteristic into a naive Bayesian network of the sensor operation anomaly diagnosis model, and determining an anomaly root cause thermal value of each operation access observation node of the route operation control sample.

6. And weighting the operation access prediction effect value of each operation access observation node of the route operation control sample and the abnormal root cause thermodynamic value to generate a target abnormal hit coefficient corresponding to each operation access observation node of the route operation control sample.

For example, assuming that the route operation control samples correspond to two operation access observation nodes, namely an operation access observation node a and an operation access observation node B, the operation access prediction effect value of the operation access observation node a is A1, the abnormal root cause thermal value is A2, the operation access prediction effect value of the operation access observation node B is B1, the abnormal root cause thermal value is B2, the abnormal hit coefficient of the operation access observation node a is a1×a2, and the abnormal hit coefficient of the operation access observation node B is B1×b2.

7. And determining the operation access node corresponding to the abnormal fault root cause from the route operation control sample by combining the target abnormal hit coefficient corresponding to each operation access observation node of the route operation control sample.

step 202, acquiring input operation control data of the target sensor.

And 204, acquiring a sensor operation abnormality diagnosis model which can be deployed and used.

Acquiring a sensor operation abnormality diagnosis model which can be deployed and used; the sensor operation abnormality diagnosis model comprises a naive Bayesian network and a feedforward neural network; the sensor operation abnormity diagnosis model is obtained by adjusting a target first training convergence index of a feedforward neural network and a target second training convergence index of a naive Bayesian network; the target first training convergence index is obtained by adjusting the initial first training convergence index through operating and accessing the abnormal root cause thermal value of the observation node; the target second training convergence index is obtained by adjusting the initial second training convergence index through the operation access observation effect value of the operation access observation node; the operation access observation node is determined by utilizing the operation state description of the sample sensor operation control sample and the feedforward neural network of the sensor operation abnormality diagnosis model; the operation access observation effect value of the operation access observation node is calculated by using an operation access sample node corresponding to an operation control sample of the example sensor; the abnormal root cause thermal value of the operation access observation node is determined by using the operation state description of the example sensor operation control sample and the naive Bayesian network; the initial second training convergence index is calculated by using the abnormal root cause thermal value; the initial first training convergence index is calculated using the operational access observation effect value.

Step 206, determining a plurality of operation access observation nodes of the input target sensor operation control data by using the operation state description of the input target sensor operation control data and the feedforward neural network.

Step 208, determining abnormal root cause thermal values of each operation access observation node of the input target sensor operation control data by using the operation state description of the input target sensor operation control data and the naive Bayesian network.

Step 210, combining the abnormal root cause thermal values of all the operation access observation nodes of the input target sensor operation control data to locate the abnormal root cause from the input target sensor operation control data, and generating the operation access node where the abnormal root cause is located.

The sensor operation anomaly diagnosis model is obtained by adjusting a target first training convergence index of the feedforward neural network and a target second training convergence index of the naive Bayesian network, and in a training process, the training convergence indexes of the feedforward neural network and the naive Bayesian network are respectively adjusted, so that the first training convergence index of the feedforward neural network is combined with an abnormal root cause heating value, and the second training convergence index of the naive Bayesian network is combined with an operation access observation effect value, and after the sensor operation anomaly diagnosis model is trained through the adjusted training convergence index, the difference between the naive Bayesian network and the feedforward neural network is weakened, the accuracy of the abnormal root cause heating value prediction of an operation access observation node is improved, and the accuracy of the abnormal root cause analysis is improved.

For some exemplary design considerations, obtaining input target sensor operation control data includes: acquiring input target sensor operation control data and a cooperative sensor operation control sample corresponding to the input target sensor operation control data; the collaborative sensor operation control sample is sample data with a system operation abnormal root cause in an operation control scheduling task where input target sensor operation control data is located before the input target sensor operation control data; determining a plurality of operational access observation nodes of the input target sensor operational control data using the operational state description of the input target sensor operational control data and the feedforward neural network includes: determining a first collaborative expression feature between the input target sensor operation control data and the collaborative sensor operation control sample using the operation state description of the input target sensor operation control data and the operation state description of the collaborative sensor operation control sample; importing the first collaborative expression characteristic into a feedforward neural network, and determining a plurality of operation access observation nodes of input operation control data of a target sensor; determining an abnormal root cause thermal value for each operational access observation node of the input target sensor operational control data using the operational state description of the input target sensor operational control data and the naive bayes network comprises: determining a second collaborative expression feature between the input target sensor operational control data and the collaborative sensor operational control sample using the operational state description of the input target sensor operational control data and the operational state description of the collaborative sensor operational control sample; and importing the second collaborative expression characteristic into a naive Bayesian network, and determining abnormal root cause thermal values of all operation access observation nodes of the input target sensor operation control data.

For some exemplary design considerations, the sensor abnormality diagnostic model includes a feed-forward neural network, a naive bayes network, and an observation effect prediction network. The sensor abnormal operation diagnosis model is obtained by adjusting a third training convergence index of the observation effect prediction network, a target first training convergence index of the feedforward neural network and a target second training convergence index of the naive Bayesian network. The method for obtaining the target first training convergence index of the feedforward neural network and the target second training convergence index of the naive Bayesian network is the same as that in the previous embodiment, and the third training convergence index is determined by using the operation access prediction effect value of the operation access observation node and the operation access observation effect value of the operation access observation node; the operational access prediction effect value is determined using an operational state description and observation effect prediction network of the example sensor operational control sample.

step 302, acquiring input target sensor operation control data and a cooperative sensor operation control sample corresponding to the input target sensor operation control data; the collaborative sensor operation control sample is sample data with a system operation abnormal root cause in an operation control scheduling task where input target sensor operation control data is located before the input target sensor operation control data.

Step 304, determining a first collaborative presentation feature between the input target sensor operational control data and the collaborative sensor operational control sample using the operational state description of the input target sensor operational control data and the operational state description of the collaborative sensor operational control sample.

Step 306, importing the first collaborative expression feature into a feedforward neural network, and determining a plurality of operation access observation nodes of input target sensor operation control data.

Step 308, importing the first collaborative expression feature into an observation effect prediction network, and determining an operation access prediction effect value of each operation access observation node of the input operation control data of the target sensor.

Step 310, determining a second collaborative presentation feature between the input target sensor operational control data and the collaborative sensor operational control sample using the operational state description of the input target sensor operational control data and the operational state description of the collaborative sensor operational control sample.

Step 312, importing the second collaborative expression feature into a naive bayes network, and determining an abnormal root cause thermal value of each operation access observation node of the input target sensor operation control data.

And step 314, weighting the operation access prediction effect value and the abnormal root cause thermal value of each operation access observation node of the input target sensor operation control data to generate a target abnormal hit coefficient of each operation access observation node of the input target sensor operation control data.

Step 316, carrying out sensor operation abnormality diagnosis on the input target sensor operation control data by combining the target abnormality hit coefficients of all operation access observation nodes of the input target sensor operation control data, and generating an abnormality root cause of the operation access nodes.

Fig. 2 illustrates a hardware structural intent of an abnormality AI diagnosis system 100 of a sensor operation control system for implementing the abnormality AI diagnosis method of a sensor operation control system as described above according to an embodiment of the application, and as shown in fig. 2, the abnormality AI diagnosis system 100 of a sensor operation control system may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In some alternative embodiments, the anomaly AI diagnostic system 100 of the sensor operation control system may be a single server or a group of servers. The server farm may be centralized or distributed (e.g., the anomaly AI diagnostic system 100 of the sensor operation control system may be a distributed system). In some alternative embodiments, the sensor operation control system abnormality AI diagnostic system 100 may be local or remote. For example, the abnormal AI diagnostic system 100 of the sensor operation control system may access information and/or data stored in the machine-readable storage medium 120 via a network. As another example, the abnormal AI diagnostic system 100 of the sensor operation control system may be directly connected to the machine-readable storage medium 120 to access stored information and/or data. In some alternative embodiments, the anomaly AI diagnostic system 100 of the sensor operation control system may be implemented on a cloud platform. For example only, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, or the like, or any combination thereof.

The machine-readable storage medium 120 may store data and/or instructions. In some alternative implementations, the machine-readable storage medium 120 may store data acquired from an external terminal. In some alternative embodiments, the machine-readable storage medium 120 may store data and/or instructions that are used by the sensor operation control system's anomaly AI diagnostic system 100 to perform or use to complete the exemplary methods described herein. In some alternative implementations, the machine-readable storage medium 120 may include mass storage, removable storage, volatile read-write memory, read-only memory (ROM), and the like, or any combination thereof. Exemplary mass storage devices may include magnetic disks, optical disks, solid state disks, and the like. Exemplary removable memory may include flash drives, floppy disks, optical disks, memory cards, compact disks, tape, and the like. Exemplary volatile read-write memory can include Random Access Memory (RAM). Exemplary RAM may include active random access memory (DRAM), double data rate synchronous active random access memory (DDR SDRAM), passive random access memory (SRAM), thyristor random access memory (T-RAM), zero capacitance random access memory (Z-RAM), and the like. Exemplary read-only memory may include mask read-only memory (MROM), programmable read-only memory (PROM), erasable programmable read-only memory (PEROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), digital versatile disk read-only memory, and the like. In some alternative implementations, the machine-readable storage medium 120 may be implemented on a cloud platform. For example only, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any combination thereof.

In a specific implementation, the plurality of processors 110 execute computer executable instructions stored by the machine-readable storage medium 120, so that the processors 110 may execute the abnormal AI diagnosis method of the sensor operation control system according to the above method embodiment, the processors 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processors 110 may be used to control the transceiving actions of the communication unit 140.

The specific implementation process of the processor 110 may refer to the above-mentioned embodiments of the method executed by the abnormal AI diagnosis system 100 of the sensor operation control system, and the implementation principle and technical effects are similar, which are not repeated herein.

In addition, the embodiment of the application also provides a readable storage medium, wherein computer executable instructions are preset in the readable storage medium, and when a processor executes the computer executable instructions, the abnormal AI diagnosis method of the sensor operation control system is realized.

Similarly, it should be noted that in order to simplify the description of the present disclosure and thereby aid in understanding one or more embodiments of the application, various features are sometimes grouped together in a single embodiment, figure, or description thereof. Similarly, it should be noted that in order to simplify the description of the present disclosure and thereby aid in understanding one or more embodiments of the application, various features are sometimes grouped together in a single embodiment, figure, or description thereof.

Claims

1. An abnormality AI diagnosis method of a sensor operation control system, characterized by being implemented by an abnormality AI diagnosis system of the sensor operation control system, the method comprising:

the sensor operation abnormality diagnosis model which can be deployed and used is used for diagnosing sensor operation abnormality of input target sensor operation control data;

calculating the operation access observation effect value of the operation access observation node by using the operation access sample node corresponding to the operation control sample of the example sensor comprises:

determining the operation access observation effect value by combining the characteristic duty ratio of the cross access node vector corresponding to the operation access observation node and the convergence access node vector;

the abnormal root cause thermal value reflects the probability of the existence of the abnormal root cause in the operation access observation node;

before the model function definition function weight of the sensor operation abnormality diagnosis model is adjusted by combining the adjusted second training convergence index and the adjusted first training convergence index, the method further comprises:

Adjusting the model function definition function weight of the sensor operation abnormality diagnosis model by combining the third training convergence index, the adjusted second training convergence index and the adjusted first training convergence index;

2. The abnormal AI diagnosis method of a sensor operation control system according to claim 1, wherein, before the determining of the first collaborative expression state description between the example sensor operation control sample and the collaborative learning sensor operation control sample using the operation state description of the example sensor operation control sample and the operation state description of the collaborative learning sensor operation control sample, the method further comprises:

3. The abnormality AI diagnosis method of a sensor operation control system according to claim 2, characterized in that the sensor operation abnormality diagnosis model includes two feature state description units that have the same model frame configuration data and that share model definition function weight information;

4. The abnormal AI diagnosis method of the sensor operation control system according to claim 1, wherein the calculating a second training convergence index using the abnormal root cause thermal value includes:

5. The abnormal AI diagnosis method of the sensor operation control system according to claim 1, characterized by further comprising:

acquiring a sensor operation scheduling flow;

determining a first sensor operation control sample comprising an abnormal fault root cause in the sensor operation scheduling flow, and generating a target sensor operation control sample;

6. The abnormal AI diagnosis method of the sensor operation control system according to claim 1, characterized by further comprising:

acquiring input operation control data of a target sensor;

the acquiring input target sensor operation control data includes:

7. An abnormal AI diagnosis system of a sensor operation control system, characterized in that the abnormal AI diagnosis system of the sensor operation control system comprises a processor and a machine-readable storage medium, wherein machine-executable instructions are stored in the machine-readable storage medium, and the machine-executable instructions are loaded and executed by the processor to realize the abnormal AI diagnosis method of the sensor operation control system of any one of claims 1-6.