CN113537274A

CN113537274A - Equipment anomaly identification method based on machine learning technology

Info

Publication number: CN113537274A
Application number: CN202110398353.1A
Authority: CN
Inventors: 陈超; 宋彪; 韩泽文; 冯祥; 王哲
Original assignee: Inner Mongolia Weishu Data Technology Co ltd
Current assignee: Inner Mongolia Weishu Data Technology Co ltd
Priority date: 2021-04-14
Filing date: 2021-04-14
Publication date: 2021-10-22
Also published as: CN114707608A; CN114707608B

Abstract

The invention discloses a machine learning technology-based equipment anomaly identification method, and belongs to the field of medical inspection. The overall architecture is as follows: the invention relates to a data acquisition module, a data processing module, a model construction module and an abnormality identification module, which have the beneficial effects that: compared with the AON quality control method provided by the prior art, the method is superior to the AON quality control method in terms of accuracy and false positive rate, can be designed and applied to real-time abnormality monitoring of equipment, can perform quick and accurate feedback when problems occur, and can effectively shorten a large amount of time consumed by inspection personnel for finding reasons when the abnormalities occur. The invention can be used for clinical popularization and application, ensures the safety of patients and benefits clinical medical care and patients.

Description

Equipment anomaly identification method based on machine learning technology

Technical Field

The invention relates to the field of inspection medicine, in particular to an equipment abnormity identification method based on a machine learning technology.

Background

Errors in the clinical laboratory mostly occur in the early and middle stages of the examination. Pre-test errors are mostly inappropriate sample collection, transport or handling related errors. Errors in the test are mostly associated with instrumentation. At present, laboratories at the stage of inspection typically use internal quality control programs to monitor, safeguard and manage the quality of the inspection of the overall process of inspection. However, many clinical studies have indicated that current conventional indoor quality control procedures have drawbacks in detecting errors in the analysis instrumentation. In 1965, HOFFMAN and WAID proposed an "Average of normal" quality control method. AON quality control, i.e. selecting the average of consecutive patient data as a control limit, typically uses a 95% confidence interval to determine a stable patient average, and if the control limit is exceeded, the system will signal an error. However, the method still has obvious defects at present, such as poor recognition effect on critical values of glucose and total protein in biochemical clinical projects, low overall recognition accuracy, and easy influence of outliers, thereby influencing the judgment effect. The invention provides an equipment abnormity identification method based on a machine learning technology, which can effectively make up the defect that the identification accuracy is not high and is easily influenced by an outlier in an AON quality control method.

Disclosure of Invention

In order to solve the problems, the invention provides a method for identifying equipment abnormity by using a technical means of machine learning, and aims to provide a method for identifying equipment abnormity based on a machine learning technology.

The general architecture created by the invention is as follows: the device comprises a data acquisition module, a data processing module, a model construction module and an abnormality identification module. The detailed steps of the sub-modules are as follows: a data acquisition module: and acquiring quality control data in the medical inspection project, wherein the type of the data sample is serum or plasma, such as glucose, total protein, glutamic-pyruvic transaminase and the like in biochemistry.

A data processing module: and further processing the acquired project data, removing special values, unifying unit dimensions, filtering outliers, standardizing data and strengthening characteristics.

A model construction module: the module utilizes the data after the data processing module and the random forest model to construct the model classifier.

An anomaly identification module: the module identifies the equipment abnormity by using the model constructed in the step and the processed data.

The invention has the following beneficial effects: compared with the AON quality control method provided by the prior art, the equipment abnormity identification method is superior to the AON quality control method in terms of accuracy and false positive rate, can be designed and applied to real-time abnormity monitoring of equipment, can perform quick and accurate feedback when a problem occurs, and can effectively shorten a large amount of time consumed by inspection personnel for searching for reasons when abnormity occurs. The invention can be used for clinical popularization and application, ensures the safety of patients and benefits clinical medical care and patients.

Drawings

The invention is further described with the aid of the accompanying drawings, in which, however, the embodiments do not constitute any limitation to the invention, and for a person skilled in the art, without inventive effort, further drawings may be derived from the following figures.

FIG. 1 is a schematic diagram of the present invention.

Detailed Description

The overall architecture flow of the device anomaly identification method based on the machine learning technology in the embodiment of the present invention is shown in fig. 1, and the detailed description of the technical implementation method in the embodiment of the present invention will be made with fig. 1 as a main line.

Referring to fig. 1, the method of the present embodiment includes a data acquisition module, a data processing module, a model building module, and an anomaly identification module.

The data acquisition module of the embodiment is mainly test result data of the lis system, the data is data when quality control is performed, the sample type of the data is serum or plasma, and the instrument manufacturer selects rogowski equipment in a unified manner. The quality of the data obtained by the above conditions is effectively guaranteed.

Preferably, the data processing module of this embodiment includes the following steps: the special value is removed, data containing results such as character strings exist in the inspection result data, the data are not needed, and the data need to be removed in order to improve the quality of the data.

The unit dimension of the data is uniform. When data analysis is carried out, data units of each research project need to be ensured to be consistent, and the inconsistency of the units can cause great deviation of data analysis results.

And filtering outliers. The traditional AON quality control method is a great influence factor, and the final judgment effect is influenced by the fact that a certain outlier drives the overall evaluation index in the algorithm to rise. Therefore, when the machine learning technology is applied, the influence of the data on the result is solved, the scheme preferably adopts an isolated forest model to carry out outlier filtering, and the proportion of the actual filtering abnormity is properly adjusted according to the distribution condition of the data. The influence factor is reduced to a model acceptable level.

The data is being too standardized. In the embodiment, the classification model is constructed by using a machine learning technical method, and the more the data conforms to the positive distribution in the model training learning process, the better the effect of the actual model is.

And (5) strengthening the characteristics. In an actual scene, the abnormal quantity of equipment is possibly small, and the characteristic of the single observation and analysis data is not obvious enough and cannot be accurately identified by a machine learning technology, so that the invention preferably adopts N samples as a new sample to organize the data, thereby increasing the dimensionality of one sample and further strengthening the characteristic. How to obtain the best value of the N can be adjusted and optimized according to the practical analysis items and the training condition of the error comprehensive model.

Preferably, the present embodiment is a model building module. And adopting a nonlinear model to construct a model by random forests.

The depth of the trees, the number of the trees and the number of the organization samples are mainly adjusted in the process of training and tuning the random forest model.

Preferably, the present embodiment is an abnormality recognition module. The anomaly identification module needs to prepare anomaly check data that needs to be predicted, and the preparation of the data is mainly based on the situation that an anomaly may occur in an actual scene. The abnormal data is data-simulated based on the data simulation formula mentioned in the paper "Association of probability-based real-time control algorithm of analytical errors", which is:

. Data simulation post-load numberAnd carrying out anomaly identification according to the outlier filtering model, the standardization model and the classifier model.

Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the protection scope of the present invention, although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. The invention relates to an equipment abnormity identification method based on a machine learning technology, which is characterized by comprising a data acquisition module, a data processing module, a model construction module and an abnormity identification module, wherein the data acquisition module acquires quality control data in a medical inspection project, the type of a data sample is serum or plasma, the data processing module further cleans, converts and loads the data by using the acquired data, the model construction module constructs a random forest classification model by using the processed data, and the abnormity identification module identifies abnormal data by using the constructed model.

2. The method for identifying device abnormalities based on machine learning technique as claimed in claim 1, wherein said data acquisition module is quality control data in medical testing project and data sample type is serum or plasma.

3. The method for identifying the equipment abnormality based on the machine learning technology as claimed in claim 1, wherein the outlier processing method of the data in the data processing module adopts an isolated forest model to perform outlier filtering, and the proportion of the actual filtering abnormality should be adjusted properly according to the distribution of the data.

4. The method for identifying the equipment abnormality based on the machine learning technology as claimed in claim 1, wherein the model building module adopts a nonlinear model random forest building model, and the depth of trees, the number of trees and the number of organization samples are mainly adjusted in the process of training and tuning the random forest model.

5. The method for identifying the equipment abnormality based on the machine learning technology as claimed in claim 1, wherein the abnormality identification module needs to prepare a simulation of an abnormal data scene, perform outlier processing according to a specified data processing logic, perform data standardization, and finally load a model to perform the identification of the abnormality.