KR20200075059A - Method and apparatus for diagnosing abnormal state of anaerobic digestion equipment using machine learning - Google Patents

Abstract

The present invention relates to a method and an apparatus for diagnosing the abnormal state of anaerobic digestion equipment using machine learning, in which the method includes: an input processing step of processing input data; a process analysis step of predicting operation data, analyzing process conditions, and predicting process variables by machine learning; a performance prediction step of predicting a performance variable by inputting a process variable predicted by machine learning in the process analysis step; and a result output step of outputting the analysis result of the process analysis step and the prediction result of the performance prediction step. Accordingly, the present invention develops a model capable of predicting and analyzing information on the performance and abnormal conditions of the entire process by processing operation data secured at the site of an anaerobic digestion plant through a statistical technique, and updates the model to respond to changes in the field situation through continuous learning of data, thereby providing information that allows the driver to easily determine the overall situation of the process in real time.

Description

METHOD AND APPARATUS FOR DIAGNOSING ABNORMAL STATE OF ANAEROBIC DIGESTION EQUIPMENT USING MACHINE LEARNING}

The present invention relates to an abnormality diagnosis method and an abnormality diagnosis device for an anaerobic digestion system using machine learning, and more specifically, an abnormality of an anaerobic digestion system using machine learning to diagnose an abnormal operation state of an anaerobic digestion system using machine learning. It relates to a diagnostic method and an abnormality diagnosis device.

In general, there are physical methods, chemical methods, and biological methods for the treatment of wastewater, and in the case of domestic wastewater, most biological methods are used.

Biological means decomposes organic substances in wastewater by using microorganisms as nutrients and energy sources, and biological treatment also uses aerobic treatment methods and oxygen to treat low-concentration organic wastewater through microorganisms that use oxygen. It is divided into anaerobic treatment methods that treat organic wastewater at high concentration with microorganisms that are not used.

The anaerobic wastewater treatment plant for such treatment is a plant capable of treating high-concentration organic wastewater such as waste sludge, manure, and food waste through fermentation of anaerobic microorganisms and producing biomethane gas, which is a power generation energy source, as a final by-product.

The anaerobic digestion steps include hydrolysis, in which a polymer organic substance is decomposed into a soluble organic substance by an enzyme secreted by microorganisms, an acid production step in which organic acid bacteria decompose organic substances to produce organic acids and alcohols, methane-producing bacteria are acetic acid, formic acid, and hydrogen , Proceeds with the production of methane (CH ₄ ) and carbon dioxide (CO ₂ ), the final by-products using methanol.

This treatment process may be carried out through a single digester, or may be operated separately according to the process type, because the microbial groups that dominate each stage and the appropriate operating conditions are different.

In particular, the main operating conditions for anaerobic decomposition can be largely divided into temperature, pH, and nutrients. Temperature is an important indicator because the reaction of enzyme-mediated microorganisms during anaerobic decomposition is dominated by temperature.

In general, in the case of mesophilic anaerobic microorganisms present in a large amount, the temperature is about 25°C to 45°C, and the optimum growth temperature is known to be 35°C or higher, so the decomposition rate tends to decrease rapidly when operated below 25°C. When the temperature exceeds 45°C, the decomposition rate rapidly decreases as the mesophilic anaerobic microorganism reaches the growth threshold.

In addition to anaerobic digestion using mesophilic microorganisms, reactions using high-temperature microorganisms (55℃) can be used to obtain high efficiency for a short residence time, but it is not actively utilized in general plant operation because it has the disadvantage of being more sensitive to temperature changes. .

In the case of pH operating conditions, it is one of the very important factors in the metabolism of microorganisms. In general, it is appropriate to operate at a neutral pH of about pH 6.5 to pH7.5, and when it is lower than pH 5.0 or higher than pH 8.5, most microorganisms are excluded. The growth of microorganisms is inhibited.

Looking at the correlation between pH and temperature, it can be seen that the proper pH increases as the temperature increases, which means that it is necessary to adjust the pH according to the situation in order to properly grow microorganisms.

Among the main operating conditions, it is important to properly supply nutrients (C, N, O, H, P, S, etc.) necessary for microorganisms to grow. Among the nutrients, carbon and nitrogen are the most important elements, and the C/N ratio (BOD/NH-N) is usually known to be 20:1 or 30:1.

However, due to the characteristics of the sewage treatment plant, the composition of the influent wastewater varies from time to time, and there is a problem that it is very difficult to predict and manage the correct C/N ratio because there are changes such as dilution of the concentration with the inflow of rainwater. .

On the other hand, looking at the main inhibitory factors of anaerobic decomposition, most organic and inorganic substances act as inhibitors of anaerobic digestion. When there is a low concentration of the inhibitor, the activity of the microorganism is not degraded, but when it is increased above a certain concentration, the biological activity rate decreases, and the activity becomes "0" above a certain high concentration.

It is difficult to determine the activity rate in real time in the anaerobic digestion plant because various microbial groups exist in the anaerobic digestion tank, and each microbe has different adaptability to the inhibitory concentration, and the inhibitory effect is different depending on the situation.

Organic fatty acids, called volatile fatty acids (VFAs), are representative inhibitors of anaerobic digesters, which affect environmental degradation, such as shock loading, depletion of nutrients, and the introduction of harmful substrates. The most important factor in VFAs is known as the effect of propionate, and when overloading VFAs, inhibition by VFAs itself and unresolved alcohol appears.

Another factor that inhibits the performance of the anaerobic digester is the effect of pH, and excessively inflow of sulfur oxides or ammonia decreases/increases the pH and adversely affects the growth conditions of microorganisms.

In addition, it is known that the toxicity of alkali and alkaline earth metal salts and the toxicity of heavy metals, and other organic compounds such as chloroform and phenol, as well as inorganic compounds such as cyanide, impair anaerobic digestion performance.

As such, since the anaerobic digestion wastewater treatment plant has a complex correlation between various operating factors and inhibitors, in order to operate the process stably, it is possible to understand and properly manage the behavior and products of complex microorganisms at each stage of anaerobic digestion. Know-how is required, but there is a problem that a general driver needs a lot of time and experience to have such know-how.

In addition, during the operation of an anaerobic digestion plant, if the exact cause is not clearly identified, performance often deteriorates rapidly or microorganisms die frequently, and if the driver detects a problem in advance and fails to respond, the entire process is stopped. Leads to accidents.

In particular, when the process is stopped, the inflow of wastewater must use an external consignment treatment facility, which incurs not only a high cost, but also, during normal process operation, the supply of biogas that can be produced is interrupted, thereby preventing profit generation through power generation. There was a problem that it would cause serious economic losses to the operation of the plant.

Therefore, an operation control system such as a programmable logic controller (PLC) is used for the operation of the anaerobic digestion process, and simple sensor information such as the driving status and pH, temperature, and flow rate of each device is used through the operation data received by the system. The information of each unit variable is monitored.

However, it is difficult to comprehensively judge the status of the entire process through the operation data collected through individual sensors or devices, and even if the process status is estimated through some key data, the process is due to the characteristics of the anaerobic digestion process with low processing speed and high residence time. There was a problem in that it was difficult to predict whether a change in shear would occur due to a problem situation at the end of the process.

In addition, there was a problem in that it was impossible to discriminate which factor was the main cause when a problem occurred due to this uncertainty.

In addition, there is a problem that the existing method of managing each variable within a fixed range has a limitation in that it cannot be diagnosed for a gradually changing process situation.

In particular, information such as microbial activity that cannot be obtained with a sensor in the field is periodically sampled for process water to be identified through experiments and analysis, but this also delays until the process water is sampled intermittently and samples are tested and analyzed. Because of the existence of time, there was a problem that it was impossible to immediately detect and respond to an abnormal situation.

In this situation, in order to minimize the problems that occur during the operation of the anaerobic digestion plant, an operator with excellent capability must have a high understanding of the process and always keep an eye on the operating conditions and maintain conservative operating conditions.

However, advanced manpower capable of predicting the abnormality of the entire process with only fragmentary information on the process control system is extremely rare, and because of the large variation in the competency level between the operators, the plant is operated depending on the capabilities of the operators operating alternately. There was a problem that stability was limited in the way it was operated.

Republic of Korea Patent Publication No. 10-2012-0096404 (August 30, 2012) Republic of Korea Registered Patent No. 10-1690241 (December 27, 2016) Republic of Korea Patent Publication No. 10-2006-0064271 (June 13, 2006)

The present invention has been devised to solve the above-described conventional problems, and it is possible to predict and analyze information on the performance and abnormal conditions of the entire process by processing operational data obtainable at the site of an anaerobic digestion system through statistical techniques. Anaerobics using machine learning that develops a model and updates the model to respond to changes in the field through continuous learning of data, thereby providing information that enables the operator to easily determine the overall situation of the process in real time. It is an object of the present invention to provide an abnormality diagnosis method and an abnormality diagnosis device for a fire extinguishing facility.

In addition, the present invention is based on machine learning to predict the symptoms of process abnormalities in process operation big data, and correlate each variable to determine the main cause through machine learning. And another object of providing an abnormality diagnosis device.

In addition, the present invention allows the operator to operate the process much more reliably in the field, and prevents accidents in which the factory is stopped by using machine learning, which can reduce many costs and losses that can be incurred due to equipment shutdown. Another object is to provide an abnormality diagnosis method and an abnormality diagnosis device for a fire extinguishing facility.

The present invention for achieving the above object, using a machine learning method for diagnosing an abnormal operation of an anaerobic digestion system using an anaerobic digestion system, an input processing step of processing input data; A process analysis step of predicting operation data by machine learning, analyzing process conditions, and predicting process variables; A performance prediction step of predicting performance variables by inputting process variables predicted by machine learning in the process analysis step; And a result output step of outputting the analysis result of the process analysis step and the prediction result of the performance prediction step.

The input processing step of the present invention, the operation data of the anaerobic digestion equipment and converting the experimental data and weather information data to the input data; And pre-processing the converted input data as data for a machine learning model.

The process analysis step of the present invention comprises: driving a time series analysis model to learn time series changes of individual operation data and predicting operation data by detecting abnormal behaviors and errors of facilities and sensors using autocorrelation; Learning the normal operation state of the entire process and unit process using the principal component analysis model to derive abnormality detection and cause variables to analyze the process state; And learning a correlation between a process variable and a general operation variable using the partial least squares model and predicting the process variable in real time.

The performance prediction step of the present invention is characterized in that some of the variables predicted in the process analysis step are input as parameters of the anaerobic digestion model to provide an indicator for the operator to judge the performance of the process.

The result output step of the present invention is characterized by providing an alarm of an abnormal state of the anaerobic digestion system by monitoring the entire process of the anaerobic digestion system, monitoring the status of each digester, and analyzing the state change of process variables. .

In addition, the present invention is an abnormality diagnosis apparatus of an anaerobic digestion system for diagnosing an abnormal operation state of an anaerobic digestion system using machine learning, an input processing unit 100 for processing input data; A storage management unit 200 for storing and managing data processed by the input processing unit 100; A process analysis unit 300 for predicting operation data by machine learning from the data stored in the storage management unit 200, analyzing process conditions, and predicting process variables; A performance prediction unit 400 for predicting performance variables by inputting process variables predicted by machine learning in the process analysis unit 300; And a result output unit 500 that outputs an analysis result of the process analysis unit 300 and a prediction result of the performance prediction unit 400.

The input processing unit 100 of the present invention, conversion means for converting the operation data, experimental data and weather information data of the anaerobic digestion equipment into input data; And pre-processing means for pre-processing the converted input data as data for a machine learning model.

The storage management unit 200 of the present invention accumulates data processed by the input processing unit 100 to provide a path that can be used in individual machine learning models, and the process analysis unit 300 and the performance prediction unit ( 400) is characterized by accumulating and managing analysis data and prediction data.

The process analysis unit 300 of the present invention is driven through a time series analysis model to learn the time series change of individual operation data and to predict operation data by detecting abnormal behaviors and errors of facilities and sensors using autocorrelation. Driving data prediction means; A process analysis means for learning the normal operation state of the entire process and the unit process using the principal component analysis model to derive abnormality detection and cause variables to analyze the process state; And a process variable prediction means for learning a correlation between a process variable and a general operation variable using the partial least squares model and predicting the process variable in real time.

The result output unit 500 of the present invention monitors the entire process of the anaerobic digestion system, monitors the status of each digester, analyzes the state change of the process variables, and provides an alarm of an abnormal state of the anaerobic digestion system. It is characterized by.

As described above, the present invention develops a model capable of predicting and analyzing information on the performance and anomalies of the entire process by processing operational data obtainable at the anaerobic digestion plant site through statistical techniques, and continuously By updating the model to respond to changes in the field situation through learning, it provides an effect that allows the operator to easily provide information to judge the overall situation of the process in real time.

In addition, based on machine learning, it provides the effect of predicting the signs of process anomalies in the process operation big data, and deriving the main causes through correlation between each variable.

In addition, the operator can operate the process much more reliably in the field, and provide an effect of reducing many costs and losses that may occur from the shutdown of equipment by preventing accidents in which the plant is stopped.

1 is a flow chart showing a method for diagnosing an abnormality in an anaerobic digestion system using machine learning according to an embodiment of the present invention.
Figure 2 is a block diagram showing an abnormality diagnosis apparatus of an anaerobic digestion system using machine learning according to an embodiment of the present invention.
Figure 3 is a block diagram showing the operation data predicting means of the anaerobic digestion system abnormal diagnosis apparatus using machine learning according to an embodiment of the present invention.
Figure 4 is a block diagram showing the process analysis means of the anaerobic digestion system abnormal diagnosis apparatus using machine learning according to an embodiment of the present invention.
Figure 5 is a block diagram showing the process variable prediction means of the anaerobic digestion system abnormality diagnosis apparatus using machine learning according to an embodiment of the present invention.

Hereinafter, a preferred embodiment of the present invention will be described in more detail with reference to the accompanying drawings.

1 is a flow chart showing a method for diagnosing an abnormality of an anaerobic digestion system using machine learning according to an embodiment of the present invention, and FIG. 2 is an apparatus for diagnosing an abnormality of an anaerobic digestion system using machine learning according to an embodiment of the present invention. 3 is a block diagram showing operation data prediction means of an abnormality diagnosis apparatus for an anaerobic digestion system using machine learning according to an embodiment of the present invention, and FIG. 4 is a machine according to an embodiment of the present invention It is a block diagram showing the process analysis means of the abnormality diagnosis apparatus of the anaerobic digestion system using running, and FIG. 5 is a block showing the process variable prediction means of the abnormality diagnosis apparatus of the anaerobic digestion system using machine learning according to an embodiment of the present invention It is.

1 and 2, an abnormality diagnosis method of an anaerobic digestion system using machine learning according to the present embodiment includes an input processing step (S10), a process analysis step (S20), and a performance prediction step (S30) and results. Including the output step (S40), it is a method of diagnosing an abnormality in the anaerobic digestion system by using machine learning to diagnose an abnormal operation state of the anaerobic digestion facility.

The input processing step (S10) is a step of converting driving data, experimental data, and weather information data into input data and pre-processing them. It consists of a data conversion step (S11) and a data pre-processing step (S12).

The data conversion step (S11) is a step of converting the form of data from the operation data of the anaerobic digestion system and the experimental data and the weather information data into input data, and inputs data and weather information from the plant operation and experiments to statistics. It is transformed into a form that can be used by academic processing and machine learning.

The data pre-processing step (S12) is a step of pre-processing the input data converted in the data conversion step (S11) as data for a machine learning model, to homogenize and normalize data values and remove noise while learning algorithms. Functions such as data segmentation are processed.

The process analysis step (S20) is a step of predicting the operation data by machine learning, analyzing the process state of the anaerobic digestion facility, and predicting the process parameters of the anaerobic digestion facility, the operation data prediction step (S21), and the process status analysis step (S22) and process variable prediction step (S23).

The operation data prediction step (S21) is a step of learning and predicting time series changes of individual operation data by driving through a time series analysis model, and detecting operation and errors of facilities and sensors by using autocorrelation. To predict.

The process state analysis step (S22) is a step of analyzing the process state by learning the normal operation state of the entire process and the unit process using the principal component analysis model, and analyzing the process state by deriving an abnormality detection and a cause variable.

The process variable prediction step (S23) is a step of learning the correlation between the process variable and the general operation variable and predicting the process variable using a partial least squares model, and it is preferable to predict the process variable in real time.

The performance prediction step (S30) is a step of predicting the performance variable by inputting the process variable predicted by machine learning in the process analysis step (S20), and some of the predicted variables in the process analysis step (S20) are anaerobic digestion theory By inputting as a parameter of the model, the operator provides an index to judge the performance of the process.

In this performance prediction step (S30), it is also possible to predict the main performance variables by inputting the values predicted by machine learning as parameters into the physical model for grasping the process performance using a scientifically known calculation formula.

The result output step (S40) is a step of outputting the analysis result of the process analysis step (S20) and the prediction result of the performance prediction step (S30), monitoring the entire process of the anaerobic digestion system, monitoring the status of each digester, and processing By analyzing the degree of change of the state of the variable, an alarm of an abnormal state of the anaerobic digestion system is provided.

In the result output step (S40), the operator can check the predicted result and the analyzed result through a monitoring chart, etc., so that the operator can check the abnormal status of the process and the main performance in real time using an algorithm in which the process proceeds. You will be able to respond to the abnormal situation.

Hereinafter, with reference to the accompanying drawings, an abnormality diagnosis apparatus of the anaerobic digestion system using the machine learning of this embodiment will be described in detail.

As shown in Fig. 2, the apparatus for diagnosing an abnormality in the anaerobic digestion system using machine learning according to the present embodiment includes an input processing unit 100, a storage management unit 200, a process analysis unit 300, a performance prediction unit 400, and results. It comprises an output unit 500, and it is a device for diagnosing an abnormality in the anaerobic digestion system using machine learning to diagnose an abnormal operation state of the anaerobic digestion facility.

The input processing unit 100 is a means for converting and processing operation data, experimental data, and weather information data of an anaerobic digestion system into input data, and is composed of a conversion means 110 and a pre-processing means 120.

The conversion means 110 is a means for converting the form of data from the operation data of the anaerobic digestion system and the experimental data and weather information data into input data. It will be converted into a form that processing and machine learning can utilize.

The pre-processing means 120 is a means for pre-processing the input data converted by the converting means 110 as data for a machine learning model, homogenizing and normalizing data values, and removing noise to remove data for algorithm learning. Functions such as division are handled.

The storage management unit 200 is a means for storing and managing data processed by the input processing unit 100, accumulating the data processed by the input processing unit 100, providing a path that can be used in an individual machine learning model, and processing The analysis data and prediction data of the analysis unit 300 and the performance prediction unit 400 are accumulated and managed.

The storage management unit 200 accumulates originals of input data processed from operation data, experimental data, and weather information data of various anaerobic digestion facilities and provides a path that can be used in individual machine learning models, so that each model learns and predicts data. It plays a role to accumulate and manage data.

The process analysis unit 300 is a means for predicting operation data, analyzing process conditions and predicting process variables by machine learning from data stored in the storage management unit 200, operation data prediction means 310, process analysis means It consists of 320 and process variable prediction means 330.

The driving data prediction means 310 is a means for learning time series changes of individual driving data and predicting driving data by driving through a time series analysis model, and detecting abnormal behaviors and errors of facilities and sensors using autocorrelation. The driving data is predicted.

The driving data prediction means 310 is a device and sensor data learning and prediction model, as shown in FIG. 3, and when training data and experimental data are input, variables are used to perform machine learning based on a time series analysis statistical model. Through the model building process 311, suitable models are built through the steps of model identification, parameter estimation, and model diagnosis for each operation variable.

In addition, when real-time operation data is input using the model building process 311 for each variable, the data prediction process 312 of the device and the sensor is performed to predict the future data of individual devices and sensors in real time.

The data predicted in the data prediction process 312 is performed by performing the predictive data verification process 313, and continuously comparing with actual data at the prediction time to verify the model error and in case of abnormality exceeding a certain error rate range You will be given instructions to update the model of the variable.

In the driving data prediction means 310, each model updates the model once a day regardless of the improvement order for the error rate, and compares and operates the improved model and the existing model for 1 day to continuously manage factors of the model with excellent predictive performance. Of course, it is possible.

The process analysis means 320 is a means for analyzing the process state by learning the normal operation state of the entire process and the unit process using the principal component analysis model, and analyzes the process state by deriving an abnormality detection and a cause variable.

The process analysis means 320 is a process state diagnosis and cause analysis model as shown in FIG. 4, driven by machine learning based on the principal component analysis statistical model, and co-distributes the number of main components of the data in the process abnormality diagnosis model construction process 321. The main component analysis model is developed by determining through eigenvalues and eigenvalues and eigenvectors. Based on this, the process control line is established by calculating the Hoteling T ² control line and the SPE control line.

The developed model processes the operation data input in real time and performs a process error diagnosis process (322) to determine whether the current state of the anaerobic digestion plant is in the normal range, and if it exceeds the error range, the alarm is sent to the monitoring system to cause the cause variable. The verification process 323 is performed.

If the result of the process abnormality diagnosis process 322 is normal, the process proceeds to the prediction data verification process 324, and when these values exceed a range of a certain error rate, a command of model update is issued.

Apart from this, the process status diagnosis and cause analysis model undergoes an update process once a day, and it is possible to compare and operate the predictive performance with the existing model for one day.

The process variable prediction means 330 is a means for learning a correlation between a process variable and a general operation variable and predicting the process variable using a partial least squares model, and predicts the process variable in real time.

The process variable predicting means 330 is a main process variable learning and prediction model, as shown in FIG. 5, and performs machine learning based on a partial least squares statistical model. When learning data flows, variable setting and data preprocessing After the process, the data score (Score) of the independent variable (X) and the dependent variable (Y) is calculated, respectively, and a partial least-squares model construction process (331) capable of regression analysis is performed.

In addition, the main variable is predicted in real time through the main variable prediction process 332, and the verification process 333 and the model update process of the predicted data are performed as in the previous models.

In particular, as shown in FIGS. 4 and 5, the process anomaly diagnosis model of the process analysis means 320 and the main process variable learning of the process variable prediction means 330 and the use of variables within the data used in the prediction model are asynchronous. Will be made.

The reason is that the anaerobic digestion process has a slow processing speed and a high residence time, so each variable has a different time to affect the process performance and main variables, and the delay time cannot be said to be constant.

Therefore, it is desirable to construct data in a form that can improve the predictive performance of the most model by searching for the optimal condition of the time shift combination of each variable in the data.

It is desirable that the data input in real time is also transformed and introduced according to the conditions of the time shift, and when the model is updated, the combination of the time shift is also reviewed.

In addition, the process status diagnosis and cause analysis model of the process analysis means 320 is not only operated for the entire process, but is applied to each digester and main process steps separately, and is reviewed in multiples. Of course, it is also possible to quickly and accurately determine whether a problem has occurred and analyze the cause.

The performance prediction unit 400 is a means for deriving and predicting performance variables by inputting process variables predicted by machine learning in the process analysis unit 300, and deriving driving performance based on a physical model. In 300), some of the predicted variables are input as parameters of the anaerobic digestion theory model to provide an indicator for the operator to judge the performance of the process.

The result output unit 500 is a means for outputting the analysis result of the process analysis unit 300 and the predicted result of the performance prediction unit 400, monitoring the entire process of the anaerobic digestion system and monitoring the status of each digester. By analyzing the degree of change in the state of process variables, an alarm of an abnormal condition of the anaerobic digestion system is provided.

In the result output unit 500, results derived from various models can be checked through monitoring charts and process analysis results, and the entire process monitoring and the status of each digester are monitored, and the main causes and variables of the abnormality alarm Of course, it is also possible to derive.

As described above, according to the present invention, the operation data obtainable at the anaerobic digestion plant site are processed through statistical techniques to develop a model that can predict and analyze the performance and anomalies of the entire process, and continuously By updating the model to respond to changes in the field situation through learning, it provides an effect that allows the operator to easily provide information to judge the overall situation of the process in real time.

In addition, based on machine learning, it provides the effect of predicting the signs of process anomalies in the process operation big data, and deriving the main causes through correlation between each variable.

In addition, the operator can operate the process much more reliably in the field, and provide an effect of reducing many costs and losses that may occur from the shutdown of equipment by preventing accidents in which the plant is stopped.

The present invention described above may be embodied in various other forms without departing from its technical spirit or main characteristics. Therefore, the above embodiment is merely a mere illustration in all respects and should not be interpreted limitedly.

100: input processing unit
200: storage management department
300: process analysis department
400: performance prediction unit
500: result output section

Claims (10)

As a method of diagnosing an abnormality of an anaerobic digestion system using machine learning,
An input processing step of processing input data;
A process analysis step of predicting operation data by machine learning, analyzing process conditions, and predicting process variables;
A performance prediction step of predicting performance variables by inputting process variables predicted by machine learning in the process analysis step; And
And a result output step of outputting the analysis result of the process analysis step and the predicted result of the performance prediction step; an abnormality diagnosis method of an anaerobic digestion system using machine learning. According to claim 1,
The input processing step,
Converting operation data, experimental data, and weather information data of the anaerobic digestion system into input data; And
Pre-processing the converted input data into data for a machine learning model; Method for diagnosing an abnormality in an anaerobic digestion system using machine learning. According to claim 1,
The process analysis step,
Driving through a time series analysis model to learn the time series change of individual operation data and predicting operation data by detecting abnormal behaviors and errors of facilities and sensors using autocorrelation;
Learning the normal operation state of the entire process and unit process using the principal component analysis model to derive abnormality detection and cause variables to analyze the process state; And
Using a partial least-squares model, learning the correlation between the process variable and the general operating variable and predicting the process variable in real time; Machine diagnostic method of anaerobic digestion using machine learning comprising the. According to claim 1,
The performance prediction step, anaerobic using machine learning, characterized in that by providing some of the variables predicted in the process analysis step as parameters of the anaerobic digestion model, the operator provides an index to determine the performance of the process. Method for diagnosing abnormalities in fire extinguishing equipment. According to claim 1,
The result output step monitors the entire process of the anaerobic digestion system, monitors the state of each digester, analyzes the state change of the process variables, and provides machine learning characterized by providing an alarm of an abnormal state of the anaerobic digestion system. Method for diagnosing abnormalities in the anaerobic digestion system used. As an abnormality diagnosis device of an anaerobic digestion system that diagnoses an abnormal operation state of an anaerobic digestion system using machine learning,
An input processing unit 100 for processing input data;
A storage management unit 200 for storing and managing data processed by the input processing unit 100;
A process analysis unit 300 for predicting operation data by machine learning from the data stored in the storage management unit 200, analyzing process conditions, and predicting process variables;
A performance prediction unit 400 for predicting performance variables by inputting process variables predicted by machine learning in the process analysis unit 300; And
And a result output section (500) for outputting the analysis result of the process analysis section (300) and the prediction result of the performance prediction section (400). The method of claim 6,
The input processing unit 100,
Conversion means for converting the operation data, experimental data and weather information data of the anaerobic digestion equipment into input data; And
And pre-processing means for pre-processing the converted input data as data for a machine learning model; an apparatus for diagnosing an abnormality in an anaerobic digestion system using machine learning. The method of claim 6,
The storage management unit 200 accumulates data processed by the input processing unit 100 to provide a path that can be used in individual machine learning models, and the process analysis unit 300 and the performance prediction unit 400 Analytical data and predictive data are accumulated and managed by the machine learning anaerobic digestion system abnormality diagnosis device. The method of claim 6,
The process analysis unit 300,
Driving data prediction means for predicting driving data by learning through time series analysis model to learn the time series changes of individual driving data and detecting abnormal behavior and errors of facilities and sensors using autocorrelation;
A process analysis means for learning the normal operation state of the entire process and the unit process using the principal component analysis model to derive abnormality detection and cause variables to analyze the process state; And
And a process variable prediction means for learning a correlation between a process variable and a general operation variable using a partial least-squares model and predicting the process variable in real time; an abnormality diagnosis device for an anaerobic digestion system using machine learning. The method of claim 6,
The result output unit 500, by monitoring the entire process of the anaerobic digestion facility, monitors the status of each digester, and analyzes the state change of the process variable, to provide an alarm of the abnormal state of the anaerobic digestion facility An abnormal diagnosis device for anaerobic digestion equipment using machine learning.

Images