Method for classifying the process state of a biogas digester based on process variables of a biogas plant.

Abstract

A method for classifying the process state of a biogas digester based on at least one process variable of a biogas plant, comprising: a) measuring a set of values ​​for at least one process variable of the biogas digester; b) preparing a model for monitoring the stability of the biogas digester, wherein the model is trained on a dataset containing historical data of the digester of the biogas plant, and the historical data contains at least one process variable; c) implementing the set of measured values ​​of the process variable in step a) into the model in step b) with the assistance of a data processing unit; d) identifying whether an outlier has occurred in the set of values ​​in step a) after the at least one process variable has been implemented into the model; and e) classifying a process state indicating the stability of the biogas digester based on the presence or absence of the identified outlier in step d) with the assistance of a data processing unit.

Description

[Technical Field]

[0001] The present invention relates to a method for classifying the process state of a biogas digester based on one or more process variables of a biogas plant. [Background technology]

[0002] In recent years, numerous biogas plants have been constructed, making a significant contribution to energy supply. However, in reality, problems often arise in the operation of these plants, hindering their realization of their economic potential. One of the main problems is inadequate process control in biogas plants. This is due to a lack of knowledge among plant operators and a lack of equipment related to online measurement technology, which leads to unfavorable selection of substrate supply and substrate composition, especially in plants that perform co-substrate operations.

[0003] In anaerobic digestion, particularly stringent requirements are imposed on process control because substrate supply is often non-uniform in terms of spatial load and composition. As is commonly seen in biogas production from organic matter (supply material), when the digested substrate is a complex mixed culture, variations occur within the adapted bacterial population, which can result in variations in overall metabolic performance.

[0004] The worst-case scenario in biogas plant operation is the interruption of the digestion process in the digester. This can occur, for example, when the acid content in the digester reaches a critical level, making proper methane production impossible. Thus, the digestion process stops. The most common reason for digestion process interruption is the overloading of organic matter into the digester, for example, due to fluctuations in feed rate and substrate composition. The risk of process failure is particularly high in co-digestion.

[0005] Control systems have been established to detect potential interruptions in advance and prevent them by adjusting the conditions within the digester. In many cases, these control systems require an operator to periodically measure the content of volatile fatty acids (VFAs) in the digester. Based on the concentration and composition of VFAs in samples from the digester, the operator can determine or "classify" the process state of the digester. Currently, these VFA samples are collected manually and usually analyzed with the assistance of high-performance liquid chromatography (HPLC). It takes 1 to 3 days for the operator to obtain the results. In other words, the operator can only judge the past state of the digester going back 1 to 3 days and cannot say anything about the current state, so they do not have "up-to-date" information about the actual process state of the digester.

[0006] Therefore, there is a need for a method or system that can provide information about the current process status of a biogas plant.

[0007] German Patent Application Publication No. 10354406(A1) discloses an arrangement for classifying process states based on the processing of measurement data, including slow process variables. This arrangement integrates shifts in multiple process variables over time intervals into the classification of process states. The disclosed arrangement uses thresholds to classify process states. However, these thresholds are highly specific to the system being classified, depending on the system's construction and the potential drift of the sensors used to measure the process variables. Thus, while a "one-fits-all" approach is theoretically possible, the classification results may not be very representative.

[0008] International Publication No. 2022 / 187818(A1) discloses a method for analyzing fermentation processes occurring within a bioreactor. This method uses a trained computing device and one or more machine learning models to analyze the fermentation process and determine the current fermentation state. The disclosed method requires the time-consuming preparation of a large training dataset containing several historical interruption events labeled by process experts. Furthermore, such a model will need to be retrained at some point to avoid model degradation, requiring even more labeled training data. A sufficient amount of historical interruption events is necessary to properly train such a model. Since it is difficult to record thousands of historical interruption events in the same digester, datasets from different digesters may be combined, resulting in an approach that is not specifically tailored to a particular digester.

[0009] U.S. Patent Application Publication 2020 / 074307(A1) discloses a system and method for evaluating, optimizing, and / or controlling the performance of an anaerobic digestion plant. The system includes a user interface, a database for storing inputs, a server for controlling the operation of the system, and a simulation engine configured to generate a biochemical methane potential and estimate the biomethane, electricity, and heat production of the plant. [Overview of the Initiative] [Problems that the invention aims to solve]

[0010] Therefore, the problem that the present invention aims to solve is to overcome the aforementioned drawbacks of the prior art and to provide an improved method for classifying the process state of a digester based on process variables of a biogas plant. [Means for solving the problem]

[0011] According to the present invention, this problem is solved by the method described in claim 1, the device described in claim 12, the biogas plant described in claim 13, and the model described in claim 14. Preferred embodiments of the present invention are described in the dependent claims.

[0012] The present invention provides a method for classifying the process state of a biogas digester based on at least one process variable of a biogas plant. The method includes the steps of measuring (directly and indirectly) a set of values ​​for at least one process variable of the biogas digester (step a)) and preparing a model for monitoring the stability of the biogas digester. The model is trained on a dataset containing historical data of the biogas plant digester, the historical data containing the measured at least one process variable (step b)). In a further step c), the measured set of values ​​for at least one process variable is implemented into the model with the assistance of a data processing unit. In a subsequent step d), the model is used to identify whether outliers have occurred in the implemented set of values ​​for at least one process variable. Based on the presence or absence of identified outliers, the process state of the biogas digester, indicating digester stability, is determined with the assistance of the data processing unit (step e)). [Modes for carrying out the invention]

[0013] For the purposes of this invention, the term “set of values” is understood to mean a series of values ​​measured over a time interval, for example, values ​​measured over an hour or a day. This set of values ​​preferably includes at least two data points, each data point having a first value representing a measurement from a process variable and a second value representing the time of the acquired measurement of the process variable, thereby the two data points representing different time points within the time interval.

[0014] For the purposes of this invention, the term “process variable” is understood to refer to a variable that characterizes the digestion process. Examples of such process variables include pH, the concentration of a particular metabolite, such as volatile fatty acid or methane, or temperature.

[0015] For the purposes of this invention, the term "stability monitoring model" is understood to refer to a computer model that includes an algorithm for managing the stability of the digestion process in a digester.

[0016] For the purposes of the present invention, the term “dataset including historical data” is understood to mean a dataset including at least one data point, the data point including a first value representing a previously measured measurement (a past measurement taken before the current measurement) and a second value representing the time of the previously measured measurement. Preferably, the dataset includes at least one value representing a number of measurements taken in the past at different points in time from the past for a particular process variable. It will be apparent to those skilled in the art that the dataset may also include a number of values ​​from a number of past points in time for a number of process variables.

[0017] For the purposes of this invention, the term "outlier" is used to describe a data point within a set of values ​​for a process variable that is unusually far from other data points in the same dataset, or that is significantly different from the historical data points of a particular process variable. In the context of this invention, an anomaly can be considered a synonym for an outlier.

[0018] The method of the present invention will be described below with reference to preferred examples for better understanding. In this regard, it should be noted that the methods disclosed below are examples and are not in any way limiting to the methods referred to in the present invention. In these preferred examples, the method uses hydrogen concentration and methane concentration in the digester as measured process variables for classifying the process state.

[0019] In the first step, the hydrogen concentration [H2] and methane concentration [CH4] in the digester are measured over a 24-hour interval (t1~t). n Measured at multiple time points within n>1). Compared with historical data from the digester, the combination of rising hydrogen concentration and decreasing methane concentration suggests that the digestion reaction in the digester was not occurring under ideal conditions, and that intermediate products (in this case, hydrogen) were present at concentrations higher than desirable. Time interval t1-t n Set of measured hydrogen and methane concentration values ​​([H2]t1-t n ) and ([CH4]t1-t n This will be implemented into an existing model for monitoring digester stability. This model was pre-trained using a historical digester dataset containing hydrogen and methane concentrations at various time points prior to t1. Time interval t1~t n Set of values ​​for hydrogen and methane concentrations ([H2]t1~t n ) and ([CH4]t1~t n Once the model is implemented, it can be used to identify possible outliers within the set of values. In this example, the time interval t1~t n A set of measured hydrogen concentration values ​​was found to be higher than expected, and the measured methane concentration was found to be lower than expected based on historical data for these two process variables. These outliers were detected by the model, and the data processing unit subsequently classified the digester as "unstable" because the outliers in hydrogen and methane concentrations indicate that the digestion reaction in the digester was not proceeding under ideal conditions.

[0020] In the context described above, the unstable process is considered to be the beginning of hyperacidification of the digester, which can occur when the hydrolysis process intensifies or when the acid-degrading ability of microorganisms is reduced due to ammonia inhibition.

[0021] Surprisingly, the method according to the present invention for classifying process states based on the presence or absence of identified outliers has been found to significantly improve the classification accuracy compared to using "filtered" process variables from which outliers have been excluded as described in the prior art. Furthermore, the method according to the present invention has the advantage that it does not use absolute values (thresholds), unlike the methods described in the prior art.

[0022] The implementation of a model trained on a dataset containing the historical data of a digester has the advantage that the classification accuracy is improved because the dataset used to train the model is highly specialized for that digester. In particular, this highly specialized dataset is more accurate than a "standardized training dataset" (i.e., a single dataset used to train multiple models for the stability monitoring of various digesters) with respect to the specific settings of the digester and its sensors, thereby avoiding misclassification due to sensor drift. Drift is a natural phenomenon for sensors. This basically affects all sensors and is caused by physical changes in the sensors. The only way to know if a sensor has drifted is to perform calibration. A typical example of sensor drift is when a sensor measures a parameter that does not change, e.g., a fixed methane concentration, several times, but the resulting values are not the same and are "drifting". Another example is when a sensor is measuring a parameter that is actually changing but reporting the same value.

[0023] In a preferred embodiment of the present invention, at least one process variable is selected from the group consisting of methane concentration in biogas, hydrogen concentration in biogas, the ratio of the generated biogas to the fermentation biomass, and combinations thereof. Here, the term "biogas" refers to the gas generated in a biogas digester (also known as "raw biogas" to those skilled in the art). The concentration of metabolites in the biogas is measured in a sample from the digester. The term "the ratio of the generated biogas to the fermentation biomass" refers to dividing the average biogas flow rate on the day (t1) of process classification by the average mass of the biomass fed to the digester (preferably over the past 2 to 14 days, more preferably over the past 5 days, including day t1). The preferred unit of the ratio of the generated biogas to the fermentation biomass is Nm 3 / (h*T), that is, normal cubic meters (Nm 3 ) per hour (h) per metric ton (t).

[0024] For each of the above preferred process variables, several preferred outlier detection methods will be described below.

[0025] When at least one process variable is the methane concentration in the biogas, the detection of outliers in step d) of the method of the present invention preferably involves using a bootstrap sampling approach and subsequently identifying outliers that deviate from the predicted value by a specific number of standard deviations.

[0026] When at least one process variable is the hydrogen concentration in the biogas, the detection of outliers in step d) of the method of the present invention preferably involves using an algorithm for detecting "outliers" or "anomalies" in a Gaussian distribution dataset. A preferred algorithm is the "elliptic_envelope" in Python.

[0027] When at least one process variable is the hydrogen concentration of the biogas, another preferred approach is described below: Firstly, a low-pass Butterworth filter is used to reduce noise in the hydrogen concentration dataset signal.

[0028] Secondly, local peaks are detected in the filtered dataset. As mentioned above, the dataset can be viewed as a series of data points, each containing the value of the measured process variable (in this case, hydrogen concentration in biogas) and the value at the time of measurement. In a Cartesian coordinate system with time on the X-axis and the process variable (in this case, hydrogen concentration) on the Y-axis, we can assume a function connecting all the data points. In this function, local peaks are assumed to be the mathematical minimum and maximum values, with the local minimum always being followed by the local maximum.

[0029] In the third step, the mean absolute vertical distance (hydrogen concentration) on the Y axis between any two adjacent peaks is calculated, and significant peaks are selected. Significant peaks are preferably defined as peaks whose vertical distance to its adjacent peak is more than 10-30%, more preferably more than 20%, of the mean vertical distance calculated above. For all significant peaks, the mean vertical distance (μ p ) and standard deviation (α p ) is calculated.

[0030] In the fourth step, the standard peak factor (SPF) is calculated using the following equation (I).

[0031]

number

[0032] In the fifth step, equation (II) is used to determine the hydrogen concentration [H] of the last data point in the dataset. dp From the last localized peak hydrogen concentration [H] lp By subtracting this, the final peak difference (LPD) is calculated.

[0033]

number

[0034] To assess whether the observed trend is significant, LPD is compared to SPF. If the absolute value of LPD (regardless of its sign) is higher than the SPF value, the final trend (the trend arising from the last peak) is considered significant. If the final trend is increasing and significant, it is concluded that the hydrogen value is increasing towards an outlier. This combination of a significant increase in hydrogen and a decrease in methane value is sufficient to classify the system as unstable.

[0035] If at least one process variable is the ratio of the biogas produced to the fermented biomass, it is preferable to use an algorithm that detects changes to lower values ​​of the underlying distribution using Hoeffding's bounds by moving average tests. A preferred algorithm for detecting outliers in step d) is the Python script "skmultiflow.drift_detection.HDDM_A".

[0036] A shift to a low baseline value, occurring when throughput is stable or increasing, biogas production has not fluctuated significantly over the past few days, and biogas production has decreased significantly in the past day, indicates biological instability. Therefore, among the biogas data points produced from fermenting biomass, only those that detect a shift to a low baseline value and that also meet these conditions are identified as "outliers."

[0037] Implementing one or a combination of the preferred process variables described above yielded remarkably reliable results regarding the classification of the process state of biogas digesters. Another advantage of using these process variables is that, in contrast to the complex and time-consuming measurement of volatile fatty acids (VFAs) proposed in the prior art, this method enables a "state-of-the-art" classification of the digester's process state.

[0038] In a preferred embodiment, the set of measured values ​​in step a) includes a combination of hydrogen concentration and methane concentration in the biogas as process variables, and each process variable is processed separately by method steps a) to d). Thus, steps a) to d) are repeated twice in total: once for hydrogen concentration and once for methane concentration. Subsequently, the final step e) is performed, and the process state is classified based on the identification of outliers determined for one or both process variables in step d). Thus, the order in which the two process variables are processed does not matter.

[0039] A preferred method for classifying the process state of the digester will be described in more detail for the above preferred embodiment.

[0040] i. A process is classified as "unstable" if an outlier is detected in the set of methane concentration values ​​within a specific time interval, and an outlier is also detected in the set of hydrogen concentration values ​​within the same time interval.

[0041] ii. If outliers are detected only in the set of hydrogen concentration values ​​within a given time interval, and no outliers are detected in the set of methane concentration values ​​within the same time interval, the process is classified as "stable".

[0042] iii. If outliers are detected only in the set of methane concentration values ​​within a specific time interval, and no outliers are detected in the set of hydrogen concentration values ​​within the same time interval, the process is classified as "stable".

[0043] iv. If no outliers are detected in either the set of methane concentration values ​​or the set of hydrogen concentration values ​​within a given time interval, the process is classified as "stable".

[0044] If outliers are detected in both the set of methane concentration values ​​and the set of hydrogen concentration values, and these outliers do not occur at the same time interval, the process is classified as "stable".

[0045] In the cases i. to v. above, the time interval is preferably 1 to 30 days, more preferably 21 days, and most preferably 5 days.

[0046] If the process is classified as unstable, this often indicates that hyperacidification of the digester has occurred. Hyperacidification can be caused by oversupply to the digester or ammonia inhibition.

[0047] Preferably, if the process state shifts from stable to unstable, an alarm is activated to warn the operator.

[0048] In another preferred embodiment, the set of measured values ​​in step a) includes a combination of hydrogen concentration and methane concentration in the biogas as process variables, and step a) further includes measuring CO2 concentration in the biogas as a process measurement validation variable. This process measurement validation variable is measured according to the above-described method for measuring hydrogen or methane concentration in the biogas.

[0049] Using CO2 as a process measurement validation variable means using the CO2 concentration in the biogas to validate the methane concentration in the biogas. If the sum of the CO2 concentration and methane concentration in the biogas is less than 99% of the total concentration of all metabolites in the biogas, the measured methane concentration is considered incorrect, and therefore the state of the digester cannot be classified using the methane concentration. Accordingly, in embodiments in which the CO2 concentration is measured, if the sum of the methane and CO2 concentrations in the biogas is less than 99% of the total concentration of all compounds in the biogas, it is preferable that the classification in step e) is not performed, or that the process is classified as "unclassifiable".

[0050] In a preferred embodiment of the present invention, the historical data used to train the digester in step b) includes digester historical data for the past month, preferably the past three months, more preferably the past six months, even more preferably the past nine months, and most preferably the past twelve months. Here, the term "includes digester historical data for at least the past X months" refers to a dataset containing measured process variable values ​​from the past up to the classification date.

[0051] Preferably, in step b), the model is trained using an algorithm for unsupervised machine learning. The term “unsupervised learning” refers to a machine learning technique that does not require training the model using labeled data. Instead, the model can operate independently using unlabeled data when discovering information. Training a model with unsupervised learning has the advantage of eliminating the need for time-consuming data annotation because it does not depend on labeled data.

[0052] If an outlier is identified, step d) preferably further includes subsequently filtering the set of measured process variable values ​​from step a) to detect the root cause of the outlier. This filtering in step d) has the advantage that each set of measured process variable values ​​does not need to be filtered beforehand, and only the measured variable values ​​that generate the outlier are filtered after the outlier has been identified.

[0053] In the above preferred embodiment, steps d) and e) preferably include the following:

[0054] -If at least one outlier is identified in step d), the process variables measured in step a) are considered to contain values ​​that require review. This correction involves further processing of the measured process variables, particularly filtering, to find the root cause of the outlier. If the root cause is found, the reviewed values ​​are considered "valid," and in step e), the digester is classified as "unclassifiable." In other words, an identified outlier with a root cause refers to a digester that cannot be classified as "stable" or "unstable," and is therefore "unclassifiable."

[0055] - On the other hand, if the root cause cannot be identified, the reviewed value is considered "invalid," and in step e), the digester is classified as "unstable." In other words, an identified outlier without a root cause refers to a digester that is "unstable" because there is no explanation for why the outlier occurred.

[0056] In an alternative, preferred embodiment, each set of measured process variable values ​​in step a) is filtered before implementation into the model in step c) to detect the root cause of potential outliers. This method, with default filtering in step a), has the advantage that only the filtered set of values ​​is implemented into the model, and outlier identification in step d) can be limited to "valid" values. In this embodiment, the method preferably includes: After measuring a set of values ​​for at least one process variable in step a), each set of values ​​is labeled as a “review value”. These review values ​​are further processed in filtering step a2) to identify outliers. If no outliers are found in a set of review values, that set of values ​​is implemented into the model in step c).

[0057] If at least one outlier is identified, the root cause of this outlier is determined. If a root cause is found, the set of values ​​under review is considered "valid," and subsequent steps b), c), and d) are skipped, and in step e), the digester is classified as "unclassifiable." In other words, an outlier with a identified root cause refers to a digester that cannot be considered "stable" or "unstable" and is therefore labeled as "unclassifiable."

[0058] If at least one outlier is identified and no root cause is found, the set of values ​​under review is considered "invalid," and subsequent steps b), c), and d) are skipped, and in step e), the digester is classified as "unstable." In other words, an identified outlier without a root cause refers to a digester that is "unstable" because there is no evidence to suggest why the outlier occurred.

[0059] Preferably, the step of filtering values ​​to detect the root cause of outliers uses a variable selected from the group consisting of agitator torque, supply gradient, and supply pause time as the filtering criterion.

[0060] In the context of this invention, the term "stirrer torque" refers to the torque of a stirrer in a digester used to stir or mix the biomass in the digester.

[0061] For the purposes of this invention, the term "supply gradient" refers to the amount of biomass supplied to the digester at specific time intervals. This specific time interval is called a "supply cycle." Since biomass is generally supplied to the digester in batches, a supply cycle begins when supply starts and ends when supply ends. Therefore, the supply gradient refers to the ratio of biomass [kg] to time [minutes] in each cycle. The time between two supply cycles is called the supply interval.

[0062] For the purposes of this invention, the term "supply downtime" is used to represent the sum of all supply intervals in a day that exceed the average supply interval plus a standard deviation. Supply downtime represents the cumulative amount of unexpected downtime during a day.

[0063] Surprisingly, these three variables have proven to be remarkably accurate in determining the root cause of outliers identified in the set of values ​​for at least one process variable.

[0064] For example, an increase in the process variable, the agitator torque, may indicate a greater amount of green waste in the biomass, resulting in lower methane production than that of food waste. “Green waste” in this specification refers to organic waste derived from plants, including mowed grass, wood, and leaves, but excluding food. On the other hand, an increase in the agitator torque may also indicate a higher biomass density in the fermenter, potentially limiting the release of the biogas produced.

[0065] Furthermore, an increase in the supply gradient value of the process variables may indicate a rapid supply pace, meaning that more biomass is being supplied to the digester than it can ferment into biogas. Consequently, the quality of the biogas deteriorates, the relative concentration of methane in the biogas decreases, and the relative concentration of by-products or intermediates such as H2 in the biogas increases.

[0066] Furthermore, an increase in the value of the process variable supply pause time suggests that the pause between two supply cycles may become longer. Such extended pauses can reduce biogas yields over multiple cycles because biomass (and therefore substrate that can be converted to biogas) is not supplied to the fermenter during the pause period.

[0067] For example, a combination of reduced supply downtime and increased supply gradient increases the total amount of biomass supplied to the digester. This may result in a short-term increase in H2 concentration and a decrease in CH4 concentration. Therefore, if a significant increase in throughput is detected, and the H2 and CH4 values ​​quickly return to normal levels, the deviations in H2 and CH4 concentrations are not considered to classify the system. In a preferred embodiment of the present invention, steps a) to d) are repeated twice with different process variables. In the subsequent step e), the classification of the process state of the biogas digester is preferably based on the outliers identified in step d) for all the process variables evaluated. It has been found that the classification of the process state is more reliable when based on two or more process variables that are examined separately.

[0068] Preferably, the classified process state of step e) is disclosed to the operator. The operator can then manually modify the digestion process based on their knowledge. This combination of implementing a trained model and a skilled operator enables improved control of biogas production.

[0069] The present invention further relates to a device comprising a training data generation unit, a model building unit, and a data processing unit. The training data generation unit is configured to generate training data based on historical data of a biogas plant digester. The model building unit is configured to build a model by learning using the training data. The data processing unit is configured to input a set of values ​​for at least one process variable into the model, identify outliers in the set of values, and obtain a classification of the process state of the biogas digester.

[0070] Unlike devices disclosed in the prior art, the device according to the present invention has the advantage of not using absolute values ​​(thresholds). As described above, the implementation of the device, which includes a model trained on a dataset containing digester history data, has the advantage of improved classification accuracy because the dataset for training the model is highly specialized for the digester in question. Therefore, by implementing this device in existing biogas plants, the stability of the digester can be improved and the profitability of the biogas plant can be increased.

[0071] The present invention further relates to a biogas plant comprising a biogas digester for generating biogas, a sensor for measuring a set of values ​​of at least one process variable of the biogas digester, and the aforementioned device.

[0072] The biogas plant according to the present invention has the advantage of not using absolute values ​​(thresholds) like the biogas plants disclosed in the prior art. Improved digester stability leads to increased profitability of the biogas plant.

[0073] The present invention further relates to a model for monitoring the stability of a biogas digester and enabling a data processing unit within a biogas plant to perform a method for classifying process conditions as described above.

[0074] The model according to the present invention has the advantage of improving digester stability because even small changes can be detected at an early stage, allowing the operator to intervene as needed. This improved digester stability increases the profitability of the biogas plant.

[0075] This specification further discloses a method for creating a training dataset for classifying the process state of a biogas digester. The method includes, firstly, measuring a set of values ​​for at least one process variable of the biogas digester, and secondly, implementing the set of values ​​for at least one process variable from step a) into an existing or newly created training dataset. These two steps are repeated multiple times over multiple time points to generate a training dataset for classifying the process state of a biogas digester.

[0076] The method described above for creating training datasets has the advantage of being able to detect even slight changes early, allowing operators to intervene as needed, and thus enabling the creation of training datasets to improve digester stability. This improved digester stability leads to increased profitability of the biogas plant.

Claims

1. A method for classifying the process state of a biogas digester based on at least one process variable of a biogas plant, a) A step of measuring a set of values ​​for at least one process variable of a biogas digester, b) A step of preparing a model for monitoring the stability of the biogas digester, wherein the model is trained on a dataset including historical data of the digester of the biogas plant, and the historical data includes at least one process variable, c) A step of implementing the set of measured values ​​of the process variables in step a) into the model in step b) with the assistance of a data processing unit, d) After implementing the at least one process variable in the model, a step of identifying whether an outlier has occurred in the set of values ​​in step a), e) A step of classifying the process state indicating the stability of the biogas digester based on the presence or absence of the identified outliers in step d), with the assistance of the data processing unit. A method that includes this.

2. The method according to claim 1, wherein the at least one process variable is selected from the group consisting of the methane concentration in the biogas, the hydrogen concentration in the biogas, the ratio of the generated biogas to the fermented biomass, and combinations thereof, and preferably the at least one process variable is the hydrogen concentration in the biogas.

3. The method according to claim 2, wherein the at least one process variable is a combination of the concentration of methane in the biogas and the concentration of hydrogen in the biogas.

4. Step a) uses the CO2 in the biogas as a process validation variable. ２ The method according to claim 3, further comprising measuring the concentration of the substance.

5. The method according to any one of claims 1 to 4, wherein the dataset includes historical data of the digester for the past month, preferably the past three months, more preferably the past six months, even more preferably the past nine months, and most preferably the past twelve months.

6. The method according to any one of claims 1 to 5, wherein in step b), the model is trained using an algorithm for unsupervised machine learning.

7. The method according to any one of claims 1 to 6, further comprising, if an outlier is identified, step d) filtering the set of values ​​of the at least one process variable measured in step a) to detect the root cause of the outlier.

8. The method according to any one of claims 1 to 6, wherein the measured values ​​of at least one process variable are filtered to detect potential outlier root causes before the set of measured values ​​of the process variable is implemented in the model in step c).

9. The method according to claim 7 or 8, wherein the filtering for detecting the root cause of the outlier or the potential outlier uses a variable selected from the group consisting of agitator torque, supply gradient, and supply pause time as a filter criterion.

10. The method according to any one of claims 1 to 9, wherein steps a) to d) are repeated twice with different process variables.

11. The method according to any one of claims 1 to 10, wherein the classified process state of step e) is passed to an operator.

12. A training data generation unit that generates training data based on historical data of the digester tank of a biogas plant, A model building unit that constructs a model by learning using the aforementioned training data, A data processing unit inputs a set of values ​​for at least one process variable into the model, identifies outliers in the set of values, and obtains a classification of the process state indicating the stability of the biogas digester based on the presence or absence of the identified outliers. A device equipped with the following features.

13. A biogas digester for generating biogas, A sensor for measuring a set of values ​​of at least one process variable of the biogas digester, The device according to claim 12, A biogas plant equipped with [specific features / equipment].

14. A model for monitoring the stability of a biogas digester and operating a data processing unit within a biogas plant to perform a method for classifying the process state according to any one of claims 1 to 11.

Images