Incinerator system fault early warning method based on digital twinning

Abstract

The present application relates to the technical field of data processing, and especially relates to a method for fault early warning of an incinerator system based on digital twinning, which comprises the following steps: collecting operation data of the incinerator; obtaining an initial dependency degree of a target point to a neighbor point based on the distribution characteristics of the operation data in a data set coordinate system; obtaining an abnormality degree of the target point by fusing the deviation degree of the target point and recent historical operation data and the difference in fluctuation degree between recent and long-term historical operation data; obtaining a target dependency degree of the target point to the neighbor point according to the abnormality degree; obtaining an adaptive K value of the target point to the neighbor point through the target dependency degree, and monitoring the abnormality of the incinerator by the adaptive K value and realizing fault early warning in combination with digital twinning. The present application obtains an adaptive K value through the distribution characteristics of data, and breaks through the limitation of using a fixed K value.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a fault early warning method for incinerator systems based on digital twins. Background Technology

[0002] Incinerators, as equipment in industrial production, play a vital role in waste treatment and energy recovery. However, their high-temperature, high-pressure operating environment and complex combustion process make them prone to malfunctions. When an incinerator system experiences overheating or abnormal pressure during operation, it can typically lead to equipment damage and even safety accidents.

[0003] Currently, the integration of digital twin technology and intelligent algorithms provides a new path for industrial equipment monitoring. Among these, the use of digital twin technology combined with the FastABOD algorithm for incinerator fault diagnosis and early warning is gaining increasing attention. However, in related technologies, the FastABOD algorithm typically sets a fixed value for each data point when calculating the anomaly score. The value, that is, the number of neighboring points, is subsequently based on a fixed value. Anomaly scores for data points are calculated and analyzed.

[0004] However, in practical applications, the data distribution in different regions may vary significantly, making the use of a fixed... There are limitations to analyzing these values. In low-density areas, data points are typically relatively isolated, surrounded by fewer and farther-distance data points. If a fixed value is set... When the value is small, it is impossible to accurately capture anomalies at the target point. In high-density areas, data points are usually closely clustered together, and the data points are close to each other. They have high similarity in local areas, and only a small value is needed. A fixed value can effectively assess the anomaly of a data point. In this case, using a fixed value is appropriate. Setting such values ​​increases computational load and creates unnecessary redundancy. Therefore, this fixed value... The way the values ​​are set limits the applicability of the FastABOD algorithm to fault early warning for incinerators under all operating conditions. Summary of the Invention

[0005] To address the issue of using digital twin technology combined with the FastABOD algorithm for incinerator fault diagnosis and early warning, the FastABOD algorithm uses a fixed... The analysis of values ​​has limitations. This invention provides a fault early warning method for incinerator systems based on digital twins, the method comprising the following steps: The incinerator's operational data is collected at several time points. Using the current operational data point as the target point, and based on the distribution characteristics of the operational data at several time points in the dataset coordinate system, the initial dependency of the target point on its neighboring points is obtained. The neighboring points are other data points distributed near the target point. The current point's immediate and consecutive preceding data points are then used as the basis for further analysis. The data points at each time point are taken as the first data segment, and the data points immediately preceding the current time point are taken as the next data segment. The data points at each time point constitute the second data segment; the deviation degree between the target point and the running data in the first data segment, as well as the difference in the fluctuation degree of the running data in the first and second data segments, are fused to obtain the anomaly degree of the target point; the target dependency degree of the target point on its neighbors is obtained by multiplying the anomaly degree of the target point by the initial dependency degree of the target point on its neighbors; the fixed values ​​in the FastABOD algorithm are corrected using the target dependency degree of the target point on its neighbors. The value is used to obtain the adaptive value of the target point. Value, successively obtain the operating data at several time points and adapt it to the neighbor points. The FastABOD algorithm is used to monitor incinerator data for anomalies, and a digital twin model is used to provide early warning of incinerator malfunctions.

[0006] This invention constructs a dynamic dependency of a target point on its neighboring points based on data density characteristics and dual-time-period time series analysis, breaking through the fixed dependency of the traditional FastABOD algorithm. Overcoming the limitations of numerical values, this invention automatically selects the appropriate number of neighboring points based on neighborhood density characteristics. This increases the number of neighboring points in low-density areas, capturing more potential anomalies, while reducing the number in high-density areas, avoiding computational redundancy and improving detection efficiency. Simultaneously, by combining the physical mechanism verification function of the digital twin model, a dual-protection mechanism combining statistical anomaly identification and mechanism deviation verification is formed, effectively reducing false alarm rates and accurately locating fault types. This invention can capture early warning signs of anomalies in key incinerator operating data in real time, triggering fault warnings in advance, providing intelligent support for stable equipment operation, reducing unplanned downtime, and lowering maintenance costs.

[0007] Preferably, the initial dependency of the target point on its neighboring points satisfies the following relationship: ; in, It is the first The initial dependency of the runtime data on neighboring points at each time step. It is the preset number of neighboring points of the target point. It is the first The running data at time 1 and the data at time 2 The distance between neighboring points It is the first The average distance between the running data at each time point and all neighboring points. It is the average distance between all data points in the dataset's coordinate system. It is the standard normalization function.

[0008] This invention calculates the distance between a target point and preset neighbor points, combines the deviation between the average distance between the target point and its neighbors and the global average distance, and performs two normalizations to eliminate dimensional interference, accurately characterizing the density distribution of target points in the dataset coordinate system; it also transforms the spatial distribution into a quantized initial dependency for subsequent adaptive adjustment. This lays the foundation for the FastABOD algorithm to dynamically adapt the number of neighbor points it calculates to the data distribution, breaking through the limitations of fixed values. To overcome the limitations of current values, we need to improve the accuracy and efficiency of incinerator anomaly detection.

[0009] Preferably, the difference in the degree of fluctuation of the running data in the first data segment and the second data segment is quantified by the difference between the interquartile range of the running data in the first data segment and the interquartile range of the running data in the second data segment, wherein the interquartile range is used to characterize the discrete fluctuation level of the running data in each data segment.

[0010] This invention leverages the interquartile range's ability to characterize data dispersion fluctuations. By calculating the difference in the interquartile range between the first and second data segments, it captures the variation in dispersion of operational data across two time periods. This difference reflects the difference in the dispersion trend of the recent data segment relative to the long-term data segment. It provides temporal-series fluctuation pattern support for subsequent assessment of anomalies when merging deviations, enabling anomaly detection to combine data dispersion trend variations for more accurate identification of potential anomalies in incinerator operation.

[0011] Preferably, the acquisition of the difference in the degree of fluctuation of the running data in the first data segment and the second data segment includes: acquiring the first-order difference sequence of the first data segment and the first-order difference sequence of the second data segment; and acquiring the difference in the degree of fluctuation of the running data in the first data segment and the second data segment by calculating the absolute value of the difference between the variance of the first-order difference sequence of the first data segment and the variance of the first-order difference sequence of the second data segment.

[0012] Preferably, the deviation between the target point and the running data in the first data segment is obtained by calculating the absolute value of the difference between the maximum value of the running data in the first data segment and the target point.

[0013] Preferably, the adaptive The values ​​satisfy the following relation: ; in, It is the first Adaptation of runtime data at each moment value, It is the initial preset fixed value, It is the first The degree of dependency of runtime data on neighboring points at each time step. It is the floor symbol. It performs a rounding operation. yes The preset lower limit value.

[0014] This invention dynamically correlates target dependency with initial fixed This value enables dynamic adjustment of the number of neighboring points. First, let's consider... A basic association is established, then numerical adaptability is optimized through rounding, and finally, reasonableness is ensured by rounding down and setting a preset lower limit. The number of neighbor points is dynamically adjusted according to the target point's dependency; the higher the target dependency, the more neighbor points are added to capture potential anomalies; the lower the target dependency, the fewer neighbor points are added to avoid computational redundancy. This mechanism breaks through the traditional fixed... Overcoming the limitations of traditional values, this system precisely adapts to the differences in incinerator data distribution, significantly improving the efficiency and accuracy of anomaly detection.

[0015] Preferably, the second data segment includes the first data segment. .

[0016] Preferably, the operating data includes incinerator temperature data and pressure data.

[0017] Preferably, the adaptive data at the specified time points are obtained sequentially. The FastABOD algorithm is used to monitor incinerator data for anomalies, and a digital twin model is used to provide early warnings of incinerator malfunctions, including: adaptive data based on each time step. The system calculates the anomaly score of the incinerator operation data at each moment; when the anomaly score is greater than a preset anomaly score threshold, the operation data at the corresponding moment is marked as anomaly data; the operation data marked as anomaly data is input into the digital twin model to determine whether it belongs to a fault and trigger an alarm.

[0018] Preferably, the construction of the digital twin model includes: using the current operating data as real-time data and the operating data before the current time as historical data, establishing a digital twin model by combining the real-time data and historical data with the physical structure parameters and process mechanism of the incinerator; using historical data as a benchmark, correcting the parameter deviations in the digital twin model, mapping the real-time data into the digital twin model, updating the boundary conditions of the digital twin model, so that the digital twin model simulates the operating state of the incinerator; receiving the input operating data, calculating the deviation between the actual state of the incinerator corresponding to the operating data and the state simulated by the digital twin model, so as to match the fault type and trigger an alarm.

[0019] This invention optimizes the FastABOD algorithm. The method of obtaining values ​​enables it to more accurately identify anomalies based on data distribution characteristics and precisely capture discrete anomaly points. The digital twin model, based on physical mechanisms, simulates the actual operating logic of the incinerator to verify anomalies. The combination of the two forms a complementary relationship: the FastABOD algorithm is responsible for quickly screening potential anomalies, while the digital twin model completes the verification at the mechanism level. This retains the efficiency of the algorithm while using physical laws to eliminate false positives, thereby significantly improving the accuracy of fault detection, effectively reducing the false alarm rate, and achieving precise fault location.

[0020] The beneficial effects of this invention: Based on data density characteristics and dual-time-period time series analysis, this invention constructs the dynamic dependency of a target point on its neighboring points, breaking through the limitations of the traditional FastABOD algorithm which uses fixed... To overcome the limitations of numerical values, the system automatically selects the appropriate number of neighboring points based on neighborhood density characteristics. This increases the number of neighboring points in low-density areas, capturing more potential anomalies, while reducing the number of neighboring points in high-density areas, avoiding computational redundancy and improving detection efficiency. Simultaneously, by calculating the interquartile range difference between the first and second data segments, the system captures the dispersion changes of the data over the two time periods, providing temporal-series fluctuation patterns to support subsequent assessments of anomaly severity through fusion. Finally, by combining FastABOD and digital twin models, the accuracy of fault diagnosis and alarm precision are improved. Attached Figure Description

[0021] Figure 1 A flowchart of a fault early warning method for an incinerator system based on digital twins provided in an embodiment of the present invention. Detailed Implementation

[0022] This invention provides a fault early warning method for incinerator systems based on digital twins, such as... Figure 1 As shown, the method includes steps S100-S600: Step S100: Collect operating data of the incinerator at several times.

[0023] It should be noted that the core function of an incinerator is to achieve the harmless treatment of waste through high-temperature combustion. The temperature inside the furnace directly reflects the combustion efficiency and reaction state, and temperature data is a key indicator for judging whether the combustion state of the incinerator is stable and whether there is overheating or combustion failure. The pressure data inside the incinerator can reflect the stability of the airflow inside the furnace in real time and is an important basis for predicting potential problems such as flue blockage and fan failure. Therefore, this invention selects temperature data and pressure data as key monitoring parameters for fault early warning.

[0024] Specifically, temperature and pressure sensors are deployed at sampling points. Temperature and pressure data from the incinerator are continuously collected at various times by setting a sampling frequency; these are collectively referred to as operational data. The sampling frequency can be set to 10 times per second, but can be adjusted according to specific requirements.

[0025] At this point, operational data for the incinerator at several points in time were obtained.

[0026] Step S200: Using the current running data as the target point, based on the distribution characteristics of the running data at several times in the dataset coordinate system, obtain the initial dependency of the target point on neighboring points. Neighboring points are other data points distributed near the target point.

[0027] It should be noted that the FastABOD algorithm is a fast angle-based outlier detection algorithm. This algorithm significantly reduces computational costs through sampling, can handle large-scale datasets, and is suitable for unlabeled outlier detection scenarios. Based on these advantages, this invention uses the FastABOD algorithm to achieve fault early warning for incinerator systems. However, this algorithm typically uses a fixed... The limited density of the FastABOD algorithm results in limitations in processing high-density and low-density regions. Therefore, this invention improves the FastABOD algorithm to address these limitations.

[0028] Specifically, taking the current running data as the target point, if the target point is relatively isolated in the dataset coordinate system, it is difficult to find enough neighboring points. In this case, data based solely on a few neighboring points cannot reflect the overall structure of the area. Therefore, low-density areas require a larger... The value represents more neighboring points, reflecting the target point's location within the dataset and improving anomaly detection stability. Target points in this region have a greater dependence on their neighbors. If a target point is relatively clustered with other data points in the dataset coordinate system, it is less likely to be misclassified as an anomaly. Therefore, a smaller value is needed. The value can be used to assess whether the target point is abnormal. In high-density areas, the target point has less dependence on neighboring points.

[0029] Therefore, by analyzing the density characteristics of data points at their locations within the dataset coordinate system, the anomaly score of the target point can be calculated. The value is optimized and calculated.

[0030] Specifically, based on the above logic, the initial dependency of the target point on its neighboring points satisfies the following relationship: ; in, It is the first The initial dependency of the runtime data on neighboring points at each time step. It is the preset number of neighboring points of the target point. It is the first The running data at time 1 and the data at time 2 The distance between neighboring points It is the first The average distance between the running data at each time point and all neighboring points. It is the average distance between all data points in the dataset's coordinate system. It is the standard normalization function.

[0031] In this formula, It is the first The running data at each moment and its preset The sum of the distances to the nth neighboring points; the larger this value, the stronger the nth neighboring point. The data at a given time point is in a low-density region, requiring more neighboring points to reflect the target point's location within the dataset. Therefore, the calculated initial dependency on neighboring points is greater. (The formula...) It is the first The difference between the average distance of the running data at time t and its all neighboring points and the average distance between all data points in the entire dataset coordinate system. The larger this value, the more significant the difference between the data points at time t and the average distance between all data points in the entire dataset coordinate system. The higher the degree of relative isolation of the running data at time step 1 in the dataset coordinate system, the more it can be explained that the data at time step 2 is relatively isolated. The greater the initial dependency of each running data point on its neighboring points, the better.

[0032] In addition, the dataset can be set to contain 500 data points, or it can be set according to actual needs. The preset number of neighboring points U for the target point can be set to 50, or it can be set according to actual needs.

[0033] At this point, the initial dependency of the target point on its neighboring points has been obtained.

[0034] Step S300: Take the current time's immediately preceding consecutive... The data points at each time point are taken as the first data segment, and the data points immediately preceding the current time point are taken as the next data segment. The data points at each time point are the second data segment; the deviation between the target point and the running data in the first data segment, as well as the difference in the fluctuation of the running data in the first and second data segments, are fused to obtain the degree of anomaly of the target point.

[0035] It should be noted that the above steps, based on the density characteristics of the data points surrounding the target point, obtain its initial dependence on neighboring points. However, relying solely on its position and density in the dataset coordinate system is insufficient to comprehensively evaluate the target point's dependence on neighboring points. Value requirements.

[0036] This invention takes into account that the collected temperature and pressure data are time-series data, which may fluctuate significantly at different points in time. These fluctuations may be related to environmental conditions or potential faults. It is necessary to further distinguish normal fluctuations from pre-fault indicators through time-series characteristics. Especially when an incinerator malfunctions, many data points with unusual temperature and pressure values ​​will appear. These data points may cluster together, which can also affect the... The selection of values ​​and the accuracy of anomaly detection results are crucial. Therefore, it is also necessary to analyze the operational data of the target point, namely temperature and pressure data, and their numerical characteristics over their respective time series. The more unusual the characteristics, the greater the likelihood that the target point is potentially an anomaly, and thus it needs to be given more attention when calculating its anomaly score, requiring the setting of more neighboring points, that is, a larger set of neighboring points. value.

[0037] Specifically, if the temperature and pressure data of the target point exhibit unusual values ​​in their respective time series, the greater the deviation of the target point from historical data, the more significant the difference in the fluctuation of historical data. Therefore, the degree of anomaly of the target point can be measured by the degree of deviation and the difference in the degree of fluctuation.

[0038] It should be noted that to quantify these two indicators, preliminary steps are required, such as defining a historical data segment for reference.

[0039] Specifically, this is achieved by selecting recent historical data and long-term historical data. The current moment's immediately preceding consecutive historical data is used as the basis. The data points at each time point are taken as the first data segment, and the data points immediately preceding the current time point are taken as the next data segment. The data points at each time point constitute the second data segment. The second data segment contains the data from the first data segment. Combining two historical data segments to assess the degree of anomaly at a target point can improve the accuracy of anomaly assessment.

[0040] For example, select 50 data points immediately preceding the current time as the first data segment, and select 600 data points immediately preceding the current time as the second data segment. Based on the sampling frequency of 10 times per second set above, if the initial data points are insufficient, one minute of running data can be pre-collected to meet the construction requirements of the two data segments.

[0041] After the preliminary operations are completed, the next step is to explain how to quantify these two metrics.

[0042] The degree of fluctuation in historical data can be quantified by the interquartile range (ICM), a statistic that describes the dispersion of the middle 50% of a dataset. It is commonly used to identify outliers and measure the distribution characteristics of data.

[0043] Specifically, based on this characteristic, the present invention achieves this through the difference between the interquartile range of the running data in the first data segment and the interquartile range of the running data in the second data segment. The interquartile range is used to characterize the discrete fluctuation level of the running data within each data segment. The calculation of the interquartile range is existing technology and will not be elaborated upon here.

[0044] In addition, the variance of time series data can be used to quantify the differences in the degree of fluctuation of historical data, and the appropriate method can be selected according to the needs during actual implementation.

[0045] Specifically, firstly, the first-order difference sequence of the first data segment and the first-order difference sequence of the second data segment are calculated; then, the absolute value of the difference between the variances of the first and second data segments is calculated to obtain the difference in the degree of fluctuation between the two data segments. This absolute difference reflects the difference between the variation amplitude of recent historical data and the variation amplitude of long-term historical data, which is suitable for capturing anomalies in the frequency of equipment mutations.

[0046] Regarding the deviation of the target point from historical data, since most incinerator malfunctions are due to overheating or excessive pressure, only the maximum value needs to be considered. Specifically, this can be obtained by calculating the absolute value of the difference between the maximum value of the operating data in the first data segment and the target point.

[0047] The above text introduced two factors that affect the degree of anomaly of the target point. The following text explains how to integrate these two indicators to obtain the degree of anomaly of the target point.

[0048] It's important to note that analyzing either indicator in isolation has limitations: If only the deviation of the target point from historical data is considered, a large deviation coupled with small fluctuations indicates that the historical data is inherently predictable and extreme values ​​are rare. In this case, the abnormal signals reflected by the deviation are more alarming due to their breaking of conventional patterns. Conversely, if only the fluctuations in historical data are considered, large fluctuations coupled with small deviations suggest that the historical data is naturally discrete, with small fluctuations being the norm, thus reducing the risk of anomalies. Therefore, multiplying the two indicators simultaneously characterizes both the probability and the actual intensity of exceptional values, providing a more comprehensive reflection of the abnormal risks in incinerator operation data.

[0049] Specifically, to achieve the above logic, taking temperature data as an example, the target dependency of temperature data on neighboring points satisfies the following relationship: ; in, It is the first The degree of anomaly in the temperature data at each moment in its corresponding time series. It is the first Temperature data at each moment, It is the first The maximum value of the temperature data at each moment in the first data segment. This is a very small value, which can be set to 0.001, or can be set according to needs, to prevent... =0; It is the first The interquartile range of the temperature data at each moment in the dataset consisting of all data in the first data segment. It is the first The interquartile range of the temperature data at each moment in the dataset consisting of all data in the second data segment. It is the standard normalized function. It is an absolute value. The degree of fluctuation in historical data can be calculated using the interquartile range or the variance of the difference series.

[0050] In this formula, The larger it is, the more it indicates the number of... The more unique the temperature data at a given moment is in terms of its temporal sequence, the more unique the corresponding numerical value. The larger it is, the more likely it is to be the first The numerical distribution of the temperature data at time t is more discrete in the first reference segment compared to the numerical distribution in the second reference segment, which indicates that the temperature data at time t is... The greater the probability that a temperature data point at a given time will have an unusual value, the more unusual the corresponding value will be.

[0051] At this point, we have obtained the current operational data, which is the degree of anomaly at the target point.

[0052] Step S400: Obtain the target dependency of the target point on its neighboring points based on the product of the anomaly degree of the target point and the initial dependency of the target point on its neighboring points.

[0053] It should be noted that the more unusual the temperature and pressure data of the target point are in their respective time series, the greater the likelihood that the target point is a potential outlier. Therefore, it needs to be given more attention when calculating its outlier score, and more neighboring points should be set; that is, a larger set of neighboring points should be assigned to it. value.

[0054] Preferably, the target dependency of the target point on its neighboring points satisfies the following relationship: ; in, It is the first The degree of dependency of runtime data on neighboring points at each time step. It is the first The initial dependency of the runtime data on neighboring points at each time step. It is the first The degree of anomaly in the temperature data at each moment in its corresponding time series. It is the first The degree of anomaly in the pressure data at a given moment within its corresponding time series is calculated using the same logic as that for temperature data, and will not be elaborated upon here. This is an adjustment coefficient for the target point's dependency on its neighboring points, used to adapt to subsequent... Value adaptive, It can be set to 0.5 to adjust the target dependency of the target point on its neighboring points to a range. , The size can also be set according to requirements.

[0055] In this formula, The larger the value, the more likely it is to be the first. When analyzing the density characteristics of surrounding data points in the dataset coordinate system at a given time step, the higher the initial dependence on neighboring points, the greater the corresponding target dependence on neighboring points. In the formula... The larger the value, the more likely it is to be the first. The more unusual the numerical performance of the temperature and pressure data at each time point, the higher the probability that the target point belongs to potential outliers, which also means that the... The more isolated the runtime data is within the dataset coordinate system at a given time point, the greater its credibility. In this case, calculating the anomaly score requires a larger... The value, that is, more neighboring points are needed to reflect the location of the target point in the dataset, thereby improving the stability of anomaly detection. Therefore, the corresponding value is... The operational data at each time point will also have a stronger dependence on the target of neighboring points.

[0056] At this point, the target dependency of the target point on its neighboring points has been obtained.

[0057] Step S500: Correct the fixed parameters in the FastABOD algorithm by adjusting the target dependency of the target point on its neighboring points. The value is used to obtain the adaptive value of the target point. value.

[0058] It should be noted that the above steps obtain the target dependency of the target point on its neighboring points. The greater the target dependency, the more isolated the target point is in the dataset coordinate system, and the more unusual the numerical performance of its corresponding two dimensions in their respective time series. Therefore, when calculating the anomaly score of the incinerator's operating data, a larger value needs to be set. The value allows the algorithm to gain a more comprehensive understanding of the target point's location within the entire dataset, improving anomaly detection accuracy. Conversely, the lower the target point's dependence on its neighbors, the more clustered it is with other data points in the dataset's coordinate system, and the more normal the numerical performance of its corresponding two dimensions in their respective time series. Therefore, to improve the algorithm's computational efficiency, a smaller value can be set. value.

[0059] Specifically, based on the above logic, the adaptive target point The values ​​satisfy the following relation: ; in, It is the first Adaptation of runtime data at each moment value, It is the initial preset fixed value, It is the first The degree of dependency of runtime data on neighboring points at each time step. It rounds down. It performs rounding to ensure the result is an integer, which conforms to... The actual value requirements, yes The preset lower limit value can be set to 1 or set according to requirements.

[0060] In this formula, The larger the value, the more isolated the target point is, and the higher its dependence on neighboring points. The larger the value, the more neighboring points it has, which can comprehensively characterize its position in the dataset and improve the accuracy of anomaly identification. The smaller the value, the more clustered the data points are, and the lower the target dependence on neighboring points. The calculated value is... The smaller the value, the fewer neighbor points are obtained, which reduces the amount of computation and improves efficiency while ensuring the basic effect.

[0061] At this point, the adaptive target point has been obtained. value.

[0062] Step S600: Obtain the adaptive data of neighbor points at several time points. The FastABOD algorithm is used to monitor incinerator data for anomalies, and a digital twin model is used to provide early warning of incinerator malfunctions.

[0063] It should be noted that, based on the aforementioned steps, the adaptive calculation of anomaly scores can be performed by acquiring the runtime data at each time point. This addresses the issue of using a fixed value in the FastABOD algorithm. The value has limitations.

[0064] Specifically, based on the adaptive calculation The anomaly score for each incinerator's operating data is calculated using the FastABOD algorithm. A data point is considered an anomaly if its anomaly score exceeds an anomaly threshold. The anomaly threshold can be set to 10, or adjusted as needed.

[0065] It should be noted that digital twin is a technology that links a virtual copy of a physical entity, process, or system to the physical world in real time. It collects operational data from the physical entity through sensors and combines this data with simulation, modeling, and data analysis techniques to construct a digital model in virtual space that can dynamically map, simulate, and predict the entire lifecycle state of the physical entity. This invention combines a digital twin model with the FastABOD algorithm to achieve fault early warning for incinerator systems.

[0066] Specifically, the current operating data is used as real-time data, and the operating data prior to the current time is used as historical data. A digital twin model is established using real-time and historical data, combined with the physical structural parameters and process mechanisms of the incinerator. Based on the historical data, parameter deviations in the digital twin model are corrected, real-time data is mapped into the digital twin model, and the boundary conditions of the digital twin model are updated, enabling the digital twin model to simulate the operating state of the incinerator.

[0067] In practical use, when the FastABOD algorithm determines that a data point is abnormal, the corresponding operational data is input into the digital twin model. The model's boundary conditions are updated in real time, ensuring that the virtual model's operational status is synchronized with the physical incinerator system. By comparing the normal state simulated by the virtual model with the real-time operational data of the physical equipment, the deviation is calculated. If the deviation is within a preset threshold range, the system is considered to be operating normally. If the deviation exceeds the threshold, a fault mode sample library is called, and the real-time deviation characteristics are compared with the deviation patterns of historical abnormal data. The model with the highest matching degree is identified as the suspected fault type, thus triggering an early warning. The threshold range can be set to temperature deviation. Pressure deviation The threshold can also be dynamically adjusted according to the incinerator load and operating condition.

[0068] In this invention, the FastABOD algorithm identifies anomalies based on statistical characteristics, while the digital twin model verifies anomalies based on physical mechanisms. The combination of the two improves the accuracy of fault detection and reduces the false alarm rate.

[0069] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A fault early warning method for incinerator systems based on digital twins, characterized in that, Including the following steps: Collect operational data from the incinerator at various points in time; Using the current running data as the target point, and based on the distribution characteristics of the running data at several times in the dataset coordinate system, the initial dependency of the target point on neighboring points is obtained. The neighboring points are other data points distributed near the target point. The immediate preceding consecutive times of the current moment The data points at each time point are taken as the first data segment, and the data points immediately preceding the current time point are taken as the next data segment. The data points at each time point are the second data segment; the deviation degree between the target point and the running data in the first data segment, as well as the difference in the fluctuation degree of the running data in the first data segment and the second data segment, are fused to obtain the anomaly degree of the target point; The target dependency of the target point on its neighboring points is obtained by multiplying the anomaly level of the target point by the initial dependency of the target point on its neighboring points. The fixed value in the FastABOD algorithm is corrected by adjusting the target dependency of the target point on its neighboring points. The value is used to obtain the adaptive value of the target point. value; Adaptive processing of neighboring points by acquiring runtime data at several time points one by one. The FastABOD algorithm is used to monitor incinerator data for anomalies, and a digital twin model is used to provide early warning of incinerator malfunctions.

2. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The initial dependency of the target point on its neighboring points satisfies the following relationship: ； in, It is the first The initial dependency of the runtime data on neighboring points at each time step. It is the preset number of neighboring points of the target point. It is the first The running data at time 1 and the data at time 2 The distance between neighboring points It is the first The average distance between the running data at each time point and all neighboring points. It is the average distance between all data points in the dataset's coordinate system. It is the standard normalized function.

3. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The difference in the degree of fluctuation of the running data in the first data segment and the second data segment is quantified by the difference between the interquartile range of the running data in the first data segment and the interquartile range of the running data in the second data segment. The interquartile range is used to characterize the discrete fluctuation level of the running data in each data segment.

4. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The acquisition of the difference in the degree of fluctuation of the running data in the first data segment and the second data segment includes: Obtain the first-order difference sequence of the first data segment and the first-order difference sequence of the second data segment; The difference in the degree of fluctuation of the running data in the first and second data segments is obtained by calculating the absolute value of the difference between the variance of the first-order difference sequence of the first data segment and the variance of the first-order difference sequence of the second data segment.

5. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The degree of deviation between the target point and the running data in the first data segment is obtained by calculating the absolute value of the difference between the maximum value of the running data in the first data segment and the target point.

6. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The adaptive The values ​​satisfy the following relation: ； in, It is the first Adaptation of runtime data at each moment value, It is the initial preset fixed value, It is the first The degree of dependency of runtime data on neighboring points at each time step. It is the floor symbol. It performs a rounding operation. yes The preset lower limit value.

7. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The second data segment contains the first data segment. .

8. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The operational data includes temperature and pressure data inside the incinerator.

9. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The adaptive data at each of the aforementioned time points is obtained sequentially. The FastABOD algorithm is used to monitor incinerator data for anomalies, and a digital twin model is used to provide early warnings of incinerator malfunctions, including: Adaptive at every moment The value is used to calculate the anomaly score of the incinerator operation data at each moment. When the abnormal score is greater than the preset abnormal score threshold, the running data at the corresponding time is marked as abnormal data; The operational data marked as abnormal is input into the digital twin model to determine whether it constitutes a fault and trigger an alarm.

10. The fault early warning method for an incinerator system based on digital twins according to claim 1, characterized in that, The construction of the digital twin model includes: Using the current operating data as real-time data and the operating data before the current time as historical data, a digital twin model is established by combining the real-time data and historical data with the physical structural parameters and process mechanism of the incinerator. Based on historical data, the parameter deviations in the digital twin model are corrected, real-time data is mapped into the digital twin model, and the boundary conditions of the digital twin model are updated so that the digital twin model can simulate the operating status of the incinerator. It receives input operating data, calculates the deviation between the actual state of the incinerator corresponding to the operating data and the state simulated by the digital twin model, and matches the fault type to trigger an alarm.

Images