Automatic air monitoring instrument diagnosis method and system based on identification analysis

Abstract

The invention relates to the field of equipment diagnosis, in particular to an automatic air monitoring instrument diagnosis method and system based on identification analysis. An automatic air monitoring instrument diagnosis system based on identification analysis comprises a state monitoring data set acquisition module, a data coding module, a mode matching module, an anomaly detection module and an automatic air monitoring instrument diagnosis report output module. According to the method, the state monitoring data set obtained in real time is subjected to encoding operation, then mode matching is performed based on the obtained monitoring data code and the standard data code, so that abnormality diagnosis of the automatic air monitoring instrument is realized, and data processed in the encoding operation and mode matching process is less, so that the accuracy of diagnosis is improved. According to the method, the abnormity of the automatic air monitoring instrument can be quickly diagnosed, the coded monitoring data code is in a discrete form, and mode matching can be more conveniently carried out.

Description

Technical Field

[0001] The present invention relates to the field of equipment diagnosis, and in particular to a method and system for diagnosing an automatic air monitoring instrument based on identification resolution. Background Art

[0002] Automatic air quality monitoring instruments play a vital role in modern environmental monitoring. By collecting real-time data on a variety of air pollutants, they provide a scientific basis for environmental protection and public health. However, with the widespread deployment and long-term operation of monitoring equipment, the timely detection and diagnosis of equipment failures have become increasingly critical.

[0003] Current fault diagnosis methods often rely on traditional machine learning models. Although existing machine learning methods excel in processing complex data and identifying potential fault modes, they also suffer from several significant limitations. First, they require large-scale data processing, which is computationally expensive and time-consuming, and can become a bottleneck in practical application scenarios that require rapid response. Second, these methods often lack flexibility during model training and updating, and are not very adaptable to new or unseen fault types. Furthermore, traditional models often lack real-time performance; even with real-time data stream input, they cannot guarantee immediate fault diagnosis, which limits their practicality in emergency situations. Summary of the Invention

[0004] The present invention performs encoding operations on a state monitoring data set acquired in real time, and then performs pattern matching based on the obtained monitoring data encoding and standard data encoding, thereby realizing abnormal diagnosis of the automatic air monitoring instrument. Moreover, since less data is processed during the encoding operation and pattern matching process, the abnormality of the automatic air monitoring instrument can be diagnosed quickly. The encoded monitoring data encoding is in discrete form, and pattern matching can also be performed more conveniently. When the result of the pattern matching is the occurrence of an abnormal type, the data screening operation can also be selected to be performed through the data screening strategies corresponding to the different abnormal types, and a more accurate abnormal diagnosis can be achieved through the detection model of the automatic air monitoring instrument, so that the generated automatic air monitoring instrument diagnosis report is more reference-oriented.

[0005] A diagnostic method for an automatic air monitoring instrument based on identification resolution, comprising:

[0006] Step S1: obtaining a state monitoring data set of an automatic air monitoring instrument through identification resolution technology, performing a data encoding operation on the state monitoring data set based on a data encoding rule base, and outputting a monitoring data code corresponding to the state monitoring data set. The data encoding rule base includes state monitoring data attributes and their corresponding data encoding rules. The data encoding rule base is established based on a monitoring data encoding model, and the monitoring data encoding model is established based on a decision tree model.

[0007] Step S2: Calculate the similarity between the monitoring data code and the standard data code in the standard data code library one by one, and output the data similarity corresponding to each standard data code. The standard data code library includes one-to-one corresponding standard data codes and abnormal types, and the standard data codes in the standard data code library are obtained based on the monitoring data coding model and negative samples. For each standard data code, the data similarity corresponding to the standard data code is judged against the similarity threshold. If the data similarity corresponding to the standard data code is greater than the similarity threshold, the acquired state monitoring data set is annotated with the abnormal type corresponding to the standard data code, and step S3 is entered. If the data similarity corresponding to the standard data code is not greater than the similarity threshold, the next state monitoring data set is acquired.

[0008] Step S3: Obtain the anomaly type annotated by the condition monitoring data set, traverse all anomaly types annotated by the condition monitoring data set, and perform the following operations for each anomaly type: obtain a data screening strategy according to the anomaly type, screen the condition monitoring data set based on the data screening strategy, construct the state data to be detected, and then send the state data to be detected to the air automatic monitoring instrument detection model corresponding to the anomaly type for processing, and output the detection result. The detection result includes the anomaly type and no anomaly. If the detection result is an anomaly type, the anomaly type output by the air automatic monitoring instrument detection model is stored in the air automatic monitoring instrument diagnosis report. The air automatic monitoring instrument diagnosis report is initially empty; otherwise, no operation is performed;

[0009] Step S4: until all abnormal types marked in the condition monitoring data set are traversed, the air automatic monitoring instrument diagnosis report is output.

[0010] As a preferred embodiment of the present invention, the data encoding operation is performed on the condition monitoring data set based on the data encoding rule base, specifically including the following operations:

[0011] Traverse the data encoding rules in the data encoding rule library in sequence, select the state monitoring data attribute corresponding to each data encoding rule, select the state monitoring data corresponding to the state monitoring data attribute from the state monitoring data set for matching, and output the corresponding encoding value of the successful match;

[0012] Arrange all code values ​​in traversal order to form monitoring data code.

[0013] As a preferred embodiment of the present invention, the establishment of the data encoding rule base specifically includes the following steps:

[0014] Obtain several training samples of the condition monitoring data set with labeled abnormal types, where the training samples of the condition monitoring data set with labeled abnormal types are negative samples; form a condition monitoring data set training set with several training samples of the condition monitoring data set with labeled abnormal types, and send the condition monitoring data set training set to a monitoring data encoding model with parameter initialization for training, and the splitting nodes of the monitoring data encoding model are determined a priori by a swarm optimization algorithm, with the abnormal type as the target, to determine whether the first training condition is met, and if the first training condition is met, output the trained monitoring data encoding model; otherwise, continue to train the monitoring data encoding model through the condition monitoring data set training set;

[0015] Obtain a trained monitoring data encoding model, traverse each layer of split nodes in the trained monitoring data encoding model, and perform the following operations for each layer of split nodes in the trained monitoring data encoding model: determine the state monitoring data attributes detected in the current layer based on all split nodes, and determine different intervals corresponding to the state monitoring data attributes based on all split nodes, and form a data encoding rule with the different intervals corresponding to the state monitoring data attributes and their corresponding encoding values;

[0016] All condition monitoring data attributes and their corresponding data encoding rules are organized into a data encoding rule library from top to bottom according to the number of layers.

[0017] As a preferred embodiment of the present invention, the splitting nodes of the monitoring data encoding model are confirmed by a swarm optimization algorithm, which specifically includes the following contents:

[0018] Obtain a condition monitoring data set training set, and segment the condition monitoring data set training set according to abnormality type to generate several abnormality type training sets, where the abnormality type training sets include abnormality type training samples;

[0019] Generate several simulated individuals and group all of them into a population set. The generation of each simulated individual includes the following operations: obtain all state monitoring data attributes, traverse all state monitoring data attributes, and for each state monitoring data attribute, randomly generate several split node values ​​within the value range corresponding to the state monitoring data attribute. All split node values ​​corresponding to each state monitoring data attribute constitute a state monitoring data attribute split fragment; and group all the state monitoring data attribute split fragments into simulated individuals.

[0020] For each simulated individual in the population set, perform the following operations: for each state monitoring data attribute split segment in the simulated individual, determine the different intervals corresponding to the state monitoring data attribute split segment, and form a simulation data encoding rule base by combining the state monitoring data attributes corresponding to the state monitoring data attribute split segment, the different intervals corresponding to the state monitoring data attribute split segment, and their corresponding encoding values;

[0021] Calculate the corresponding fitness for each simulated individual. The specific operation is as follows: traverse each abnormal type training set, traverse the abnormal type training samples in the abnormal type training set for each abnormal type training set, perform data encoding operation on the abnormal type training sample through the simulated data encoding rule base corresponding to the simulated individual for each abnormal type training sample, and output the simulated data code; after the abnormal type training samples in the abnormal type training set are traversed, obtain the simulated data codes corresponding to all abnormal type training samples in the abnormal type training set, and then perform cluster analysis on all simulated data codes, output the ratio between the number of simulated data codes in the cluster with the most simulated data codes and the total number of all simulated data codes, and record it as the local fitness; after all abnormal type training sets are traversed, output the average value of the local fitness corresponding to all abnormal type training sets, which is the fitness corresponding to the simulated individual;

[0022] The genetic algorithm is used to simulate the individual and perform selection and mutation operations with fitness as the goal;

[0023] The simulated individual with the highest output fitness is used to confirm the split node of the monitoring data encoding model.

[0024] As a preferred embodiment of the present invention, the establishment of the standard data encoding library includes the following steps:

[0025] Obtain the abnormal type training set and the trained monitoring data encoding model corresponding to all abnormal types, and perform the following operations for each abnormal type: traverse the abnormal type training samples in the abnormal type training set, send the abnormal type training samples to the trained monitoring data encoding model for detection, output the path with the largest value of all split nodes, record it as the target path, and perform data encoding operations on the abnormal type training samples one by one according to the split nodes on the target path. Specifically, determine the encoding value corresponding to the state monitoring data attribute according to the interval determined by the split node, and output the candidate data encoding corresponding to the abnormal type training sample; until all abnormal type training samples in the abnormal type training set are traversed, perform cluster analysis on all candidate data encodings corresponding to the abnormal type training set, and use the candidate data encoding corresponding to the cluster center corresponding to the cluster cluster with the largest number of candidate data encodings as the standard data encoding corresponding to the abnormal type;

[0026] The one-to-one correspondence between standard data codes and exception types forms a standard data coding library.

[0027] As a preferred embodiment of the present invention, the establishment of a data screening strategy includes the following steps:

[0028] In the process of training the monitoring data encoding model using the condition monitoring data training set, the Shapley value corresponding to each condition monitoring data attribute is calculated by the SHAP interpreter for each abnormality type output by the monitoring data encoding model;

[0029] The Shapley value corresponding to each condition monitoring data attribute is compared with the Shapley threshold, and the condition monitoring data attributes corresponding to all Shapley values ​​greater than the Shapley threshold are combined into a data screening strategy.

[0030] As a preferred embodiment of the present invention, the training of the detection model of the automatic air monitoring instrument corresponding to the abnormality type specifically includes the following steps:

[0031] Obtain an abnormality type training set corresponding to the abnormality type, then perform data screening on the abnormality type training set based on the data screening strategy corresponding to the abnormality type, construct a training set of state data to be detected corresponding to the abnormality type, and train an automatic air monitoring instrument detection model corresponding to the abnormality type through the training set of state data to be detected corresponding to the abnormality type. Take the abnormality type as the target and judge whether the second training condition is met. If the second training condition is met, output the trained automatic air monitoring instrument detection model corresponding to the abnormality type; otherwise, continue to train the automatic air monitoring instrument detection model corresponding to the abnormality type through the training set of state data to be detected corresponding to the abnormality type.

[0032] An air automatic monitoring instrument diagnostic system based on identification resolution, comprising:

[0033] A condition monitoring data set acquisition module is used to obtain the condition monitoring data set of the air automatic monitoring instrument through the identification resolution technology;

[0034] A data encoding module is used to perform data encoding operations on the condition monitoring data set based on a data encoding rule base, and output the monitoring data code corresponding to the condition monitoring data set. The data encoding rule base includes condition monitoring data attributes and their corresponding data encoding rules. The data encoding rule base is established based on a monitoring data encoding model, and the monitoring data encoding model is established based on a decision tree model.

[0035] The pattern matching module is used to calculate the similarity between the monitoring data code and the standard data code in the standard data code library one by one, and output the data similarity corresponding to each standard data code. The standard data code library includes one-to-one corresponding standard data codes and abnormal types, and the standard data codes in the standard data code library are obtained based on the monitoring data coding model and negative samples. For each standard data code, the data similarity corresponding to the standard data code is judged against the similarity threshold. If the data similarity corresponding to the standard data code is greater than the similarity threshold, the acquired state monitoring data set is annotated with the abnormal type corresponding to the standard data code, and step S3 is entered. If the data similarity corresponding to the standard data code is not greater than the similarity threshold, the next state monitoring data set is acquired;

[0036] The anomaly detection module is used to obtain the anomaly type marked by the condition monitoring data set, traverse all anomaly types marked by the condition monitoring data set, and perform the following operations for each anomaly type: obtain the data screening strategy according to the anomaly type, the data screening strategy is used to screen out data attributes that have a higher contribution to anomaly type detection, filter the condition monitoring data set based on the data screening strategy, construct the state data to be detected, and then send the state data to be detected to the air automatic monitoring instrument detection model corresponding to the anomaly type for processing, and output the detection result. The detection result includes the anomaly type and no anomaly. If the detection result is an anomaly type, the anomaly type output by the air automatic monitoring instrument detection model is stored in the air automatic monitoring instrument diagnosis report. The air automatic monitoring instrument diagnosis report is initially empty. Otherwise, no operation is performed;

[0037] The air automatic monitoring instrument diagnosis report output module is used to output the air automatic monitoring instrument diagnosis report.

[0038] The present invention has the following advantages:

[0039] 1. The present invention performs encoding operations on the status monitoring data set acquired in real time, and then performs pattern matching based on the obtained monitoring data encoding and standard data encoding, thereby realizing abnormal diagnosis of the automatic air monitoring instrument. Since less data is processed during the encoding operation and pattern matching process, the abnormality of the automatic air monitoring instrument can be diagnosed quickly. The encoded monitoring data encoding is in discrete form, and pattern matching can also be performed more conveniently. When the result of pattern matching is the occurrence of an abnormal type, the data screening operation can also be selected to be performed through the data screening strategy corresponding to the different abnormal types, and a more accurate abnormal diagnosis can be achieved through the detection model of the automatic air monitoring instrument, so that the generated automatic air monitoring instrument diagnosis report is more referenceable.

[0040] 2. The present invention trains the decision tree by using training samples of the state monitoring data set with labeled abnormal types, constructs a monitoring data encoding model, and determines the data encoding rules based on the splitting nodes in the trained monitoring data encoding model. This can make the generated data encoding rules more consistent with the mapping relationship between the training samples of the state monitoring data set and the abnormal types, making subsequent pattern matching more accurate.

[0041] 3. The present invention uses a genetic algorithm to perform a priori settings on the split nodes in the training of the monitoring data coding model, so that the split nodes in the training of the monitoring data coding model can basically conform to the changes in the training set at the beginning, thereby improving the convergence speed of the monitoring data coding model during the training process, and further improving the training speed of the monitoring data coding model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a schematic diagram of the structure of an automatic air monitoring instrument diagnostic system based on identity resolution adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0044] Example 1

[0045] A diagnostic method for an automatic air monitoring instrument based on identification resolution, comprising:

[0046] Step S1: Obtain a state monitoring data set of an automatic air monitoring instrument through an identification resolution technology. The state monitoring data set needs to undergo preprocessing operations, such as denoising and normalization. The identification resolution technology is used to encode and parse data from different sources to obtain original data. The state monitoring data set includes power state data, sensor state data, and sampled flow data of the automatic air monitoring instrument, which can reflect the operating state of the automatic air monitoring instrument. The state monitoring data set is encoded based on a data encoding rule base, and the monitoring data code corresponding to the state monitoring data set is output. The data encoding rule base includes state monitoring data attributes and their corresponding data encoding rules. The data encoding rule base is established based on a monitoring data encoding model, and the monitoring data encoding model is established based on a decision tree model. It should be noted that the monitoring data code in this example consists of "A, T, C, and G". Different data values ​​in the state monitoring data set will be mapped to four different intervals, and the encoding values ​​of "A, T, C, and G" will be determined according to the intervals. "A, T, C, and G" correspond to different bases in the sequence. Of course, the specific encoding method can be set by yourself. The encoding is performed to convert the state monitoring data set into a discrete form to facilitate subsequent pattern matching.

[0047] Step S2: Calculate the similarity between the monitoring data code and the standard data code in the standard data code library one by one, and output the data similarity corresponding to each standard data code. The standard data code library includes one-to-one corresponding standard data codes and abnormal types, and the standard data codes in the standard data code library are obtained based on the monitoring data code model and negative samples. The negative samples are abnormal state monitoring data sets recorded in history. The negative samples and the monitoring data code model can reveal the output of abnormal standard data codes. These standard data codes can represent state monitoring data sets of various abnormal types, and the data similarity is the similarity between the currently acquired state monitoring data set and the state monitoring data sets of various abnormal types. For each standard data code, the data similarity corresponding to the standard data code is compared with the similarity threshold. The similarity threshold is set by the operator and is generally 0.75. If the data similarity corresponding to the standard data code is greater than the similarity threshold, the acquired state monitoring data set is annotated with the abnormal type corresponding to the standard data code. The abnormal types include sensor abnormality, software abnormality, and power supply abnormality, and then enter step S3. If the data similarity corresponding to the standard data code is not greater than the similarity threshold, the next state monitoring data set is acquired.

[0048] Step S3: obtain the anomaly type annotated by the status monitoring data set, traverse all anomaly types annotated by the status monitoring data set, and perform the following operations for each anomaly type: obtain a data screening strategy according to the anomaly type, the data screening strategy is used to screen out data attributes that have a higher contribution to anomaly type detection, screen the status monitoring data set based on the data screening strategy, construct the status data to be detected, and then send the status data to be detected to the air automatic monitoring instrument detection model corresponding to the anomaly type for processing, and output the detection result. The detection result includes the anomaly type and no anomaly, that is, each air automatic monitoring instrument detection model is a two-classification model, either if there is an anomaly, the anomaly type is output, or if there is no anomaly, the detection result of no anomaly is directly output. If the detection result is an anomaly type, the anomaly type output by the air automatic monitoring instrument detection model is stored in the air automatic monitoring instrument diagnosis report, and the air automatic monitoring instrument diagnosis report is initially empty; otherwise, no operation is performed;

[0049] Step S4: until all abnormal types marked in the condition monitoring data set are traversed, the air automatic monitoring instrument diagnosis report is output. The air automatic monitoring instrument diagnosis report includes different abnormal types that occur in the air automatic monitoring instrument, which can provide a reference for maintenance personnel of the air automatic monitoring instrument.

[0050] The present application realizes abnormal diagnosis of automatic air monitoring instruments by encoding the status monitoring data set acquired in real time, and then performing pattern matching based on the obtained monitoring data encoding and standard data encoding. Since less data is processed during the encoding operation and pattern matching process, the abnormality of the automatic air monitoring instrument can be diagnosed quickly. The encoded monitoring data encoding is in discrete form, and pattern matching can also be performed more conveniently. When the result of pattern matching is the occurrence of an abnormal type, data screening operations can also be selected to be performed through data screening strategies corresponding to different abnormal types, and more accurate abnormal diagnosis can be achieved through the detection model of the automatic air monitoring instrument, so that the generated automatic air monitoring instrument diagnosis report is more referenceable.

[0051] The data encoding operation of the condition monitoring data set is performed based on the data encoding rule base, including the following operations:

[0052] Traverse the data encoding rules in the data encoding rule library in sequence. For each data encoding rule, select the state monitoring data attribute corresponding to the data encoding rule, select the state monitoring data corresponding to the state monitoring data attribute from the state monitoring data set for matching, and output the corresponding encoding value for a successful match. For example, if the data encoding rule is to judge the temperature, match the temperature data in the state monitoring data with the four different intervals corresponding to the data encoding rule, and output the corresponding encoding value for a successful match.

[0053] Arrange all code values ​​in traversal order to form monitoring data code.

[0054] The establishment of a data encoding rule base includes the following steps:

[0055] Obtain several training samples of the state monitoring data set with annotated abnormal types. The training samples of the state monitoring data set with annotated abnormal types are negative samples. The state monitoring data set training samples store the state monitoring data set of the automatic air monitoring instrument with abnormalities according to the historical records of the operator. Since the amount of data in actual operation is small, in order to avoid overfitting, the samples are expanded by software simulation. The several training samples of the state monitoring data set with annotated abnormal types form a state monitoring data set training set, and the state monitoring data set training set is sent to the parameter-initialized monitoring data encoding model for training. The splitting nodes of the monitoring data encoding model are determined a priori by the swarm optimization algorithm. The training period includes operations such as tree structure construction and pruning, which is consistent with the training method of the decision tree. The abnormal type is used as the target to determine whether the first training condition is met. The first training condition can be that the accuracy reaches a certain expectation. If the first training condition is met, the trained monitoring data encoding model is output; otherwise, the monitoring data encoding model continues to be trained through the state monitoring data set training set.

[0056] Obtain a trained monitoring data encoding model, traverse each layer of split nodes in the trained monitoring data encoding model, and perform the following operations for each layer of split nodes in the trained monitoring data encoding model: determine the state monitoring data attributes detected in the current layer based on all split nodes, and determine different intervals corresponding to the state monitoring data attributes based on all split nodes. For example, for temperature data detection, the split nodes include judgments on "0-10°C", "10°C-20°C", "20°C-30°C", and "30°C-40°C", then the corresponding intervals are (0-10], (10°C-20], (20°C-30], and (30°C-40]". The different intervals corresponding to the state monitoring data attributes and their corresponding coding values ​​form a data encoding rule;

[0057] All condition monitoring data attributes and their corresponding data encoding rules are organized into a data encoding rule library from top to bottom according to the number of layers.

[0058] This application trains a decision tree by using training samples of a condition monitoring data set with labeled abnormality types, constructs a monitoring data encoding model, and determines data encoding rules based on the splitting nodes in the trained monitoring data encoding model. This can make the generated data encoding rules more consistent with the mapping relationship between the training samples of the condition monitoring data set and the abnormality types, making subsequent pattern matching more accurate.

[0059] The split nodes of the monitoring data encoding model are confirmed through the swarm optimization algorithm, which includes the following:

[0060] Obtain a condition monitoring data set training set, and segment the condition monitoring data set training set according to abnormality type to generate several abnormality type training sets, where the abnormality type training sets include abnormality type training samples;

[0061] Generate several simulated individuals, and form all simulated individuals into a population set. The generation of each simulated individual includes the following operations: obtain all state monitoring data attributes, traverse all state monitoring data attributes, and for each state monitoring data attribute, randomly generate several splitting node values ​​within the value range corresponding to the state monitoring data attribute. In this embodiment, since the number of intervals to be obtained is 4, 3 splitting node values ​​will be generated for the state monitoring data attribute. All splitting node values ​​corresponding to each state monitoring data attribute constitute a state monitoring data attribute split segment; all state monitoring data attribute split segments are formed into a simulated individual;

[0062] For each simulated individual in the population set, perform the following operations: for each state monitoring data attribute split segment in the simulated individual, determine the different intervals corresponding to the state monitoring data attribute split segment, and form a simulation data encoding rule base by combining the state monitoring data attributes corresponding to the state monitoring data attribute split segment, the different intervals corresponding to the state monitoring data attribute split segment, and their corresponding encoding values;

[0063] The corresponding fitness is calculated for each simulated individual. The specific operation is as follows: traverse each abnormal type training set, for each abnormal type training set, traverse the abnormal type training samples in the abnormal type training set, for each abnormal type training sample, perform data encoding operation on the abnormal type training sample through the simulated data encoding rule base corresponding to the simulated individual, and output the simulated data code; after the abnormal type training samples in the abnormal type training set are traversed, obtain the simulated data codes corresponding to all abnormal type training samples in the abnormal type training set, and then perform cluster analysis on all simulated data codes, output the ratio between the number of simulated data codes in the cluster with the most simulated data codes and the total number of all simulated data codes, which is recorded as the local fitness; after all abnormal type training sets are traversed, output the average value of the local fitness corresponding to all abnormal type training sets, which is the fitness corresponding to the simulated individual, and simulate and calculate the data encoding formats corresponding to different simulated individuals, so that the data encoding format corresponding to the optimal simulated individual can meet the encoding form of most abnormal type training samples, and also make the prior splitting node in the monitoring data encoding model training more consistent with the changes in the state monitoring data set training set;

[0064] The genetic algorithm is used to simulate the simulation calculation of the simulated individuals, and the selection and mutation operations are performed with fitness as the goal. The selection operation is to select the simulated individuals based on the fitness of all simulated individuals, and the roulette algorithm is used to select the simulated individuals, and all the selected simulated individuals are composed of the parent set. The mutation operation is to randomly generate a mutation value for each simulated individual after the recombination operation, and compare it with the pre-set mutation threshold. If the mutation value is higher than the mutation threshold, any state monitoring data attribute split segment in the simulated individual is randomly set to the split node value of the simulated individual in this state monitoring data attribute split segment.

[0065] The simulated individual with the highest output fitness is used to confirm the split node of the monitoring data encoding model.

[0066] This application uses a genetic algorithm to perform a priori settings on the split nodes in the training of the monitoring data coding model, so that the split nodes in the training of the monitoring data coding model can basically conform to the changes in the training set at the beginning, thereby improving the convergence speed of the monitoring data coding model during the training process, and further improving the training speed of the monitoring data coding model.

[0067] The establishment of a standard data coding library includes the following steps:

[0068] Obtain the abnormal type training set and the trained monitoring data encoding model corresponding to all abnormal types, and perform the following operations for each abnormal type: traverse the abnormal type training samples in the abnormal type training set, send the abnormal type training samples to the trained monitoring data encoding model for detection, output the path with the largest value of all split nodes, record it as the target path, and perform data encoding operations on the abnormal type training samples one by one according to the split nodes on the target path. Specifically, determine the encoding value corresponding to the state monitoring data attribute according to the interval determined by the split node, and output the candidate data encoding corresponding to the abnormal type training sample; until all abnormal type training samples in the abnormal type training set are traversed, perform cluster analysis on all candidate data encodings corresponding to the abnormal type training set, and use the candidate data encoding corresponding to the cluster center corresponding to the cluster cluster with the largest number of candidate data encodings as the standard data encoding corresponding to the abnormal type;

[0069] The one-to-one corresponding standard data codes and exception types are combined into a standard data coding library; the standard data coding library reflects the data coding form corresponding to the condition monitoring data set under the same exception type, thereby realizing data coding pattern matching.

[0070] The establishment of a data screening strategy includes the following steps:

[0071] In the process of training the monitoring data encoding model using the condition monitoring data training set, for each abnormality type output by the monitoring data encoding model, the Shapley value corresponding to each condition monitoring data attribute is calculated by the SHAP interpreter. It should be noted that the core idea of ​​the Shapley value is to evaluate the impact of each feature on the model prediction result when it is added to a feature subset without the feature, that is, the Shapley value is calculated by considering all possible feature combinations to ensure that the contribution of each feature is fairly evaluated. The specific operations include the following: for each condition monitoring data attribute in the monitoring data encoding model, all possible condition monitoring data attribute subsets are considered, excluding the condition monitoring data attribute itself; then, for each condition monitoring data attribute, the change in the output abnormality type when the condition monitoring data attribute is added to any possible condition monitoring data attribute subset is calculated, such as the Kullback-Leibler divergence between the probability vectors corresponding to different abnormality types, and used as the marginal contribution of the condition monitoring data attribute on the condition monitoring data attribute subset. Then, the marginal contributions of the condition monitoring data attribute on all possible condition monitoring data attribute subsets are weighted averaged to obtain the Shapley value corresponding to the condition monitoring data attribute;

[0072] The Shapley value corresponding to each condition monitoring data attribute is compared with the Shapley threshold. The Shapley threshold is set in advance by the operator, and the condition monitoring data attributes corresponding to all Shapley values ​​greater than the Shapley threshold are combined into a data screening strategy. The data screening strategy is used to screen the condition monitoring data set, specifically retaining the monitoring data in the condition monitoring data set that meets the condition monitoring data attributes in the data screening strategy.

[0073] The training of the detection model of the automatic air monitoring instrument corresponding to the abnormal type includes the following steps:

[0074] Obtain an abnormality type training set corresponding to the abnormality type, then perform data screening on the abnormality type training set based on the data screening strategy corresponding to the abnormality type, construct a training set of state data to be detected corresponding to the abnormality type, and train the air automatic monitoring instrument detection model corresponding to the abnormality type through the training set of state data to be detected corresponding to the abnormality type. Take the abnormality type as the target and judge whether the second training condition is met. The second training condition can also be that the accuracy reaches a certain expectation or reaches a certain number of training times. If the second training condition is met, output the trained air automatic monitoring instrument detection model corresponding to the abnormality type; otherwise, continue to train the air automatic monitoring instrument detection model corresponding to the abnormality type through the training set of state data to be detected corresponding to the abnormality type.

[0075] Example 2

[0076] An air automatic monitoring instrument diagnosis system based on identification resolution, such as Figure 1 Shown, including:

[0077] The condition monitoring data set acquisition module is used to obtain the condition monitoring data set of the automatic air monitoring instrument through identification resolution technology. The condition monitoring data set needs to undergo preprocessing operations such as denoising and normalization. The identification resolution technology is used to encode and parse data from different sources to obtain the original data. The condition monitoring data set includes the power status data, sensor status data, and sampled flow data of the automatic air monitoring instrument;

[0078] The data encoding module is used to perform data encoding operations on the condition monitoring data set based on the data encoding rule base, and output the monitoring data code corresponding to the condition monitoring data set. The data encoding rule base includes the condition monitoring data attributes and their corresponding data encoding rules. The data encoding rule base is established based on the monitoring data encoding model, and the monitoring data encoding model is established based on the decision tree model. It should be noted that the monitoring data code in this example consists of "A, T, C and G". Different data values ​​in the condition monitoring data set will be mapped to four different intervals, and the encoding values ​​of "A, T, C and G" will be determined according to the intervals. "A, T, C and G" correspond to different bases in the sequence. Of course, the specific encoding method can be set by yourself. The purpose of encoding is to convert the condition monitoring data set into a discrete form to facilitate subsequent pattern matching;

[0079] The pattern matching module is used to: calculate the similarity between the monitoring data code and the standard data code in the standard data code library one by one, and output the data similarity corresponding to each standard data code. The standard data code library includes a one-to-one correspondence between standard data codes and anomaly types, and the standard data codes in the standard data code library are obtained based on the monitoring data code model and negative samples. The negative samples are historical records of abnormal state monitoring data sets. The negative samples and the monitoring data code model can reveal the output of standard data codes with anomalies. These standard data codes can represent state monitoring data sets of various abnormal types, and the data similarity is the similarity between the currently acquired state monitoring data set and the state monitoring data sets of various abnormal types. For each standard data code, the data similarity corresponding to the standard data code is judged against a similarity threshold. The similarity threshold is set by the operator and is generally 0.75. If the data similarity corresponding to the standard data code is greater than the similarity threshold, the acquired state monitoring data set is annotated with an abnormality type according to the abnormality type corresponding to the standard data code. The abnormality types include sensor abnormality, software abnormality, and power supply abnormality, and the process proceeds to step S3. If the data similarity corresponding to the standard data code is not greater than the similarity threshold, the next state monitoring data set is acquired.

[0080] The anomaly detection module is used to obtain the anomaly type marked by the status monitoring data set, traverse all anomaly types marked by the status monitoring data set, and perform the following operations for each anomaly type. The data screening strategy is obtained according to the anomaly type. The data screening strategy is used to screen out data attributes that have a higher contribution to anomaly type detection. The status monitoring data set is screened based on the data screening strategy, and the status data to be detected is constructed. The status data to be detected is then sent to the air automatic monitoring instrument detection model corresponding to the anomaly type for processing, and the detection results are output. The detection results include anomaly type and no anomaly, that is, each air automatic monitoring instrument detection model is a two-classification model. If there is an anomaly, the anomaly type is output, or if there is no anomaly, the detection result of no anomaly is directly output. If the detection result is an anomaly type, the anomaly type output by the air automatic monitoring instrument detection model is stored in the air automatic monitoring instrument diagnosis report. The air automatic monitoring instrument diagnosis report is initially empty; otherwise, no operation is performed;

[0081] The air automatic monitoring instrument diagnostic report output module is used to output the air automatic monitoring instrument diagnostic report. The air automatic monitoring instrument diagnostic report includes different types of abnormalities that occur in the air automatic monitoring instrument and can provide a reference for maintenance personnel of the air automatic monitoring instrument.

[0082] It should be understood that those skilled in the art may make improvements or modifications based on the above description, and all such improvements and modifications shall fall within the scope of protection of the appended claims. Any portion of this specification not described in detail is prior art known to those skilled in the art.

Claims

1. A diagnostic method for an automatic air monitoring instrument based on identification resolution, characterized in that: include: Step S1: obtaining a state monitoring data set of an automatic air monitoring instrument through identification resolution technology, performing a data encoding operation on the state monitoring data set based on a data encoding rule base, and outputting a monitoring data code corresponding to the state monitoring data set. The data encoding rule base includes state monitoring data attributes and their corresponding data encoding rules. The data encoding rule base is established based on a monitoring data encoding model, and the monitoring data encoding model is established based on a decision tree model. Step S2: Calculate the similarity between the monitoring data code and the standard data code in the standard data code library one by one, and output the data similarity corresponding to each standard data code. The standard data code library includes one-to-one corresponding standard data codes and abnormal types, and the standard data codes in the standard data code library are obtained based on the monitoring data coding model and negative samples. For each standard data code, the data similarity corresponding to the standard data code is judged against the similarity threshold. If the data similarity corresponding to the standard data code is greater than the similarity threshold, the acquired state monitoring data set is annotated with the abnormal type corresponding to the standard data code, and step S3 is entered. If the data similarity corresponding to the standard data code is not greater than the similarity threshold, the next state monitoring data set is acquired. Step S3: Obtain the anomaly type annotated by the condition monitoring data set, traverse all anomaly types annotated by the condition monitoring data set, and perform the following operations for each anomaly type: obtain a data screening strategy according to the anomaly type, screen the condition monitoring data set based on the data screening strategy, construct the state data to be detected, and then send the state data to be detected to the air automatic monitoring instrument detection model corresponding to the anomaly type for processing, and output the detection result. The detection result includes the anomaly type and no anomaly. If the detection result is an anomaly type, the anomaly type output by the air automatic monitoring instrument detection model is stored in the air automatic monitoring instrument diagnosis report. The air automatic monitoring instrument diagnosis report is initially empty; otherwise, no operation is performed; Step S4: until all abnormal types marked in the condition monitoring data set are traversed, the air automatic monitoring instrument diagnosis report is output.

2. The method for diagnosing an automatic air monitoring instrument based on identification resolution according to claim 1, characterized in that: The data encoding operation of the condition monitoring data set is performed based on the data encoding rule base, including the following operations: Traverse the data encoding rules in the data encoding rule library in sequence, select the state monitoring data attribute corresponding to each data encoding rule, select the state monitoring data corresponding to the state monitoring data attribute from the state monitoring data set for matching, and output the corresponding encoding value of the successful match; Arrange all code values ​​in traversal order to form monitoring data code.

3. The method for diagnosing an automatic air monitoring instrument based on identification resolution according to claim 2 is characterized in that: The establishment of a data encoding rule base includes the following steps: Obtain several training samples of the state monitoring data set with annotated abnormal types, and the training samples of the state monitoring data set with annotated abnormal types are negative samples; form a state monitoring data set training set with several training samples of the state monitoring data set with annotated abnormal types, and send the state monitoring data set training set to the monitoring data coding model with initialized parameters for training, and the splitting nodes of the monitoring data coding model are determined a priori by the swarm optimization algorithm, and the abnormal type is used as the target to determine whether the first training condition is met. If the first training condition is met, the trained monitoring data coding model is output; otherwise, the monitoring data coding model is continued to be trained through the state monitoring data set training set; obtain the trained monitoring data coding model, traverse each layer of splitting nodes in the trained monitoring data coding model, and perform the following operations for each layer of splitting nodes in the trained monitoring data coding model: determine the state monitoring data attributes detected in the current layer according to all splitting nodes, and determine different intervals corresponding to the state monitoring data attributes according to all splitting nodes, and form a data coding rule with the different intervals corresponding to the state monitoring data attributes and their corresponding coding values; All condition monitoring data attributes and their corresponding data encoding rules are organized into a data encoding rule library from top to bottom according to the number of layers.

4. The method for diagnosing an automatic air monitoring instrument based on identification resolution according to claim 3 is characterized in that: The split nodes of the monitoring data encoding model are confirmed through the swarm optimization algorithm, which includes the following: Obtain a condition monitoring data set training set, and segment the condition monitoring data set training set according to abnormality type to generate several abnormality type training sets, where the abnormality type training sets include abnormality type training samples; Generate several simulated individuals and group all of them into a population set. The generation of each simulated individual includes the following operations: obtain all state monitoring data attributes, traverse all state monitoring data attributes, and for each state monitoring data attribute, randomly generate several split node values ​​within the value range corresponding to the state monitoring data attribute. All split node values ​​corresponding to each state monitoring data attribute constitute a state monitoring data attribute split fragment; and group all the state monitoring data attribute split fragments into simulated individuals. For each simulated individual in the population set, perform the following operations: for each state monitoring data attribute split segment in the simulated individual, determine the different intervals corresponding to the state monitoring data attribute split segment, and form a simulation data encoding rule base by combining the state monitoring data attributes corresponding to the state monitoring data attribute split segment, the different intervals corresponding to the state monitoring data attribute split segment, and their corresponding encoding values; Calculate the corresponding fitness for each simulated individual. The specific operation is as follows: traverse each abnormal type training set, traverse the abnormal type training samples in the abnormal type training set for each abnormal type training set, perform data encoding operation on the abnormal type training sample through the simulated data encoding rule base corresponding to the simulated individual for each abnormal type training sample, and output the simulated data code; after the abnormal type training samples in the abnormal type training set are traversed, obtain the simulated data codes corresponding to all abnormal type training samples in the abnormal type training set, and then perform cluster analysis on all simulated data codes, output the ratio between the number of simulated data codes in the cluster with the most simulated data codes and the total number of all simulated data codes, and record it as the local fitness; after all abnormal type training sets are traversed, output the average value of the local fitness corresponding to all abnormal type training sets, which is the fitness corresponding to the simulated individual; The genetic algorithm is used to simulate the individual and perform selection and mutation operations with fitness as the goal; The simulated individual with the highest output fitness is used to confirm the split node of the monitoring data encoding model.

5. The method for diagnosing an automatic air monitoring instrument based on identification resolution according to claim 4 is characterized in that: The establishment of a standard data coding library includes the following steps: Obtain the abnormal type training set and the trained monitoring data encoding model corresponding to all abnormal types, and perform the following operations for each abnormal type: traverse the abnormal type training samples in the abnormal type training set, send the abnormal type training samples to the trained monitoring data encoding model for detection, output the path with the largest value of all split nodes, record it as the target path, and perform data encoding operations on the abnormal type training samples one by one according to the split nodes on the target path. Specifically, determine the encoding value corresponding to the state monitoring data attribute according to the interval determined by the split node, and output the candidate data encoding corresponding to the abnormal type training sample; until all abnormal type training samples in the abnormal type training set are traversed, perform cluster analysis on all candidate data encodings corresponding to the abnormal type training set, and use the candidate data encoding corresponding to the cluster center corresponding to the cluster cluster with the largest number of candidate data encodings as the standard data encoding corresponding to the abnormal type; The one-to-one correspondence between standard data codes and exception types forms a standard data coding library.

6. The method for diagnosing an automatic air monitoring instrument based on identification resolution according to claim 5, characterized in that: The establishment of a data screening strategy includes the following steps: In the process of training the monitoring data encoding model using the condition monitoring data training set, the Shapley value corresponding to each condition monitoring data attribute is calculated by the SHAP interpreter for each abnormality type output by the monitoring data encoding model; The Shapley value corresponding to each condition monitoring data attribute is compared with the Shapley threshold, and the condition monitoring data attributes corresponding to all Shapley values ​​greater than the Shapley threshold are combined into a data screening strategy.

7. The method for diagnosing an automatic air monitoring instrument based on identification resolution according to claim 6, characterized in that: The training of the detection model of the automatic air monitoring instrument corresponding to the abnormal type includes the following steps: Obtain an abnormality type training set corresponding to the abnormality type, then perform data screening on the abnormality type training set based on the data screening strategy corresponding to the abnormality type, construct a training set of state data to be detected corresponding to the abnormality type, and train an automatic air monitoring instrument detection model corresponding to the abnormality type through the training set of state data to be detected corresponding to the abnormality type. Take the abnormality type as the target and judge whether the second training condition is met. If the second training condition is met, output the trained automatic air monitoring instrument detection model corresponding to the abnormality type; otherwise, continue to train the automatic air monitoring instrument detection model corresponding to the abnormality type through the training set of state data to be detected corresponding to the abnormality type.

8. An air automatic monitoring instrument diagnostic system based on identification resolution, characterized in that: The system applies the air automatic monitoring instrument diagnosis method based on identification resolution according to any one of claims 1 to 7, including: A condition monitoring data set acquisition module is used to obtain the condition monitoring data set of the air automatic monitoring instrument through the identification resolution technology; A data encoding module is used to perform data encoding operations on the condition monitoring data set based on a data encoding rule base, and output the monitoring data code corresponding to the condition monitoring data set. The data encoding rule base includes condition monitoring data attributes and their corresponding data encoding rules. The data encoding rule base is established based on a monitoring data encoding model, and the monitoring data encoding model is established based on a decision tree model. The pattern matching module is used to calculate the similarity between the monitoring data code and the standard data code in the standard data code library one by one, and output the data similarity corresponding to each standard data code. The standard data code library includes one-to-one corresponding standard data codes and abnormal types, and the standard data codes in the standard data code library are obtained based on the monitoring data coding model and negative samples. For each standard data code, the data similarity corresponding to the standard data code is judged against the similarity threshold. If the data similarity corresponding to the standard data code is greater than the similarity threshold, the acquired state monitoring data set is annotated with the abnormal type corresponding to the standard data code, and step S3 is entered. If the data similarity corresponding to the standard data code is not greater than the similarity threshold, the next state monitoring data set is acquired; The anomaly detection module is used to obtain the anomaly type marked by the status monitoring data set, traverse all anomaly types marked by the status monitoring data set, and perform the following operations for each anomaly type: obtain the data screening strategy according to the anomaly type, the data screening strategy is used to screen out data attributes that have a higher contribution to anomaly type detection, filter the status monitoring data set based on the data screening strategy, construct the status data to be detected, and then send the status data to be detected to the air automatic monitoring instrument detection model corresponding to the anomaly type for processing, and output the detection result. The detection result includes the anomaly type and no anomaly. If the detection result is an anomaly type, the anomaly type output by the air automatic monitoring instrument detection model is stored in the air automatic monitoring instrument diagnosis report. The air automatic monitoring instrument diagnosis report is initially empty, otherwise, no operation is performed; the air automatic monitoring instrument diagnosis report output module is used to output the air automatic monitoring instrument diagnosis report.

Images