Unified governance system and method based on multi-source heterogeneous industrial data

Abstract

The invention discloses a unified management system and method based on multi-source heterogeneous industrial data, and relates to the technical field of industrial data management, and the method comprises the steps: S1, collecting furnace condition monitoring data in real time, and carrying out the preprocessing of the collected data; s2, constructing a semantic classification data set, establishing a working condition label, periodically evaluating the disturbance level of the hearth temperature, and performing disturbance attribute labeling on the semantic classification data set; s3, screening out a temperature distortion candidate period, quantifying the distortion risk of the hearth temperature, and performing differential repair on the temperature sequence; and S4, evaluating the risk intensity of the multi-source furnace condition features in the fusion stage, screening out an abnormal period, and executing field consistency check and priority coverage correction. The problems that temperature reading abnormity cannot be repaired in time and multi-source data fusion is unreliable due to the fact that an existing industrial data management technology cannot identify temperature distortion and cannot conduct consistency verification on multi-source furnace condition data are solved.

Description

Technical Field

[0001] This invention relates to the field of industrial data governance technology, specifically to a unified governance system and method based on multi-source heterogeneous industrial data. Background Technology

[0002] With the continuous and in-depth development of the Industrial Internet, intelligent manufacturing, and digital factories, the scale of multi-source heterogeneous data generated in industrial sites is constantly expanding, encompassing multi-dimensional information such as equipment operation data, process monitoring data, environmental sensing data, and control execution data. Along with the gradual popularization of edge computing, the Industrial Internet of Things (IIoT), and big data technologies, industrial enterprises are increasingly demanding the collection, governance, and sharing of massive amounts of industrial data. Consequently, the application scope of big data resource services is constantly expanding, and research and application surrounding the construction of data governance systems are gradually becoming an important part of industrial digitalization. Against this backdrop, various big data governance methods for the industrial sector have been proposed and applied to different business scenarios.

[0003] For example, the invention with publication number CN116628059A relates to a data governance method based on industrial big data, including the following steps: S1, acquiring multi-source heterogeneous industrial data; S2, cleaning and transforming the data acquired in S1; S3, auditing and improving the data quality according to published data standards; S4, making the audited and improved data publicly available through API sharing to provide data application services. Using this method, not only can the validity and format consistency of the data be guaranteed, but also its applicability. It also allows for the full utilization of the audited and improved data, thereby solving the problems of low data quality and difficulty in data utilization in the industrial field.

[0004] For example, the invention with publication number CN117332023A discloses an industrial data governance method based on data classification management, including the following steps: S1, classifying metadata; S2, collecting data from the classified metadata; S3, cleaning and transforming the collected data; S4, performing data governance on the cleaned and transformed data; S5, transforming the data assets after data governance. The industrial data governance method based on data classification management proposed in this invention, by classifying metadata and implementing classified management of data collection and data governance processing, makes the data governance processing more targeted to different data categories, which is conducive to more effective implementation of data asset transformation after data governance. It can effectively reduce the difficulties in data application caused by low data quality and solve the technical problems of large amount of data, low quality, and difficulty in data utilization in the industrial field.

[0005] However, the aforementioned existing technologies primarily focus on the abstract-level governance of industrial data, such as executing processing flows like data collection, cleaning, transformation, classification, management, and unified standardization. They are largely geared towards improving data quality and asset management at a general level. For high-frequency furnace condition monitoring data generated in industrial settings, which contains multi-dimensional coupled characteristics such as temperature, pressure, disturbances, and combustion status, and is significantly affected by changes in operating conditions, disturbance fluctuations, and measurement errors, existing methods struggle to perform fine-grained identification and governance of real-time disturbance behavior, temperature distortion characteristics, and the consistency of data fusion across multiple sources. Furthermore, existing technologies typically lack a unified, process-level governance mechanism for the dynamic semantics of furnace condition data, sources of disturbance, multi-channel differences, and fusion risks. This limits the ability to perform in-depth analysis and decision support based on furnace condition monitoring data in industrial operation scenarios.

[0006] Therefore, in order to address the above problems, there is an urgent need for a unified governance system and method based on multi-source heterogeneous industrial data. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a unified governance system and method based on multi-source heterogeneous industrial data. This solves the problems of existing industrial data governance technologies being unable to identify temperature distortion and perform consistency checks on multi-source furnace condition data, resulting in the failure to promptly repair abnormal temperature readings and unreliable multi-source data fusion.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a unified governance method based on multi-source heterogeneous industrial data, comprising: S1, real-time acquisition of furnace condition monitoring data, and performing clock synchronization, noise suppression, anomaly removal, missing data imputation, and normalization on the acquired furnace condition monitoring data to generate preprocessed furnace condition monitoring data; S2, constructing a semantic classification dataset based on the preprocessed furnace condition monitoring data, establishing operating condition labels, periodically evaluating the disturbance level of furnace temperature, performing disturbance attribute labeling on the semantic classification dataset, and constructing a structured furnace condition feature set; S3, selecting candidate periods for temperature distortion based on the structured furnace condition feature set, quantifying the distortion risk of furnace temperature within the candidate periods, performing differentiated repair on the temperature sequence according to the operating condition stage, and generating repaired furnace condition monitoring data; S4, evaluating the risk intensity of multi-source furnace condition features in the fusion stage based on the repaired furnace condition monitoring data, selecting abnormal periods, performing field consistency checks and priority coverage correction on the abnormal periods, and structuring furnace condition records to achieve unified governance of multi-source heterogeneous industrial data.

[0009] Furthermore, the specific steps for generating preprocessed furnace condition monitoring data are as follows: Real-time acquisition of furnace condition monitoring data, including furnace temperature, thermocouple voltage, flue gas velocity, flue gas negative pressure, flue gas oxygen content, burner gas flow rate, combustion air flow rate, heater power, furnace pressure, ambient temperature, furnace door opening / closing status, and exhaust fan operating frequency; For the acquired furnace condition monitoring data, cross-node time alignment is achieved based on a unified clock synchronization protocol, and a hybrid algorithm of polynomial smoothing filtering and amplitude limiting filtering is used to suppress and smooth continuous time-series data; Abnormal observation points are identified using a local anomaly factor algorithm to remove outlier data caused by flue gas disturbances and furnace disturbances; Missing data caused by monitoring interruptions is interpolated using a K-nearest neighbor interpolation algorithm; The furnace condition monitoring data is normalized using a standard deviation normalization algorithm to achieve numerical scale uniformity.

[0010] Furthermore, the specific steps for constructing a semantic classification dataset and establishing operating condition labels based on the preprocessed furnace condition monitoring data are as follows: Extract the preprocessed furnace condition monitoring data, establish semantic labels according to the physical source and measurement meaning of the data, classify furnace temperature and thermocouple voltage as temperature sensing data, classify flue gas velocity, flue negative pressure, and flue gas oxygen content as flue gas disturbance data, classify burner gas flow rate, combustion air flow rate, and heater power as combustion operating condition data, and classify furnace pressure, ambient temperature, furnace door opening and closing status, and exhaust fan operating frequency as furnace body status data, thus forming a semantic classification dataset; set a fixed-width sliding time window as an evaluation period, periodically identify the changing trend of furnace temperature based on the semantic classification dataset, divide the current operating condition into a heating stage, a constant temperature stage, and a cooling stage, and establish operating condition labels.

[0011] Furthermore, the specific steps for periodically assessing the disturbance level of furnace temperature are as follows: Extract the furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency within the current assessment period; calculate the instantaneous changes in furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency, respectively; and calculate the change in furnace temperature within the assessment period. Divide the instantaneous change in furnace temperature by the periodic change in furnace temperature and add a minimum term to obtain the temperature change ratio term. Squat the instantaneous changes in flue gas velocity, flue negative pressure, and exhaust fan operating frequency sequentially, sum them, take the square root, add one, and take the natural logarithm to obtain the disturbance coupling term. Multiply the temperature change ratio term and the disturbance coupling term, and use the product as input to take the hyperbolic tangent value to obtain the furnace temperature disturbance assessment value.

[0012] Furthermore, the specific steps for performing perturbation attribute annotation on the semantic classification dataset and constructing a structured furnace condition feature set are as follows: The furnace temperature perturbation evaluation values... and multi-level perturbation threshold and A comparison is performed, and perturbation attribute annotations are applied to the semantic classification dataset within the evaluation period: when ≤ When, the current semantic classification dataset segment is marked as a normal furnace condition segment; when < < When, the current semantic classification dataset segment is marked as the segment affected by smoke disturbance; when ≥ At that time, the current semantic classification dataset segment is marked as a suspected thermocouple distortion segment; and the operating condition label, furnace temperature disturbance evaluation value and disturbance attribute label are attached to the corresponding time index in the semantic classification dataset to construct a structured furnace condition feature set.

[0013] Furthermore, the specific steps for selecting candidate periods for temperature distortion based on the structured furnace condition feature set and quantifying the distortion risk of furnace temperature within the candidate periods are as follows: Based on the semantic classification dataset and furnace temperature perturbation evaluation values, cross-feature cross-validation is performed on the temperature sensing data within the evaluation period. By jointly comparing the instantaneous change in furnace temperature, the amplitude of thermocouple voltage change, and the periodic change in furnace temperature, when the furnace temperature and thermocouple voltage changes are inconsistent and do not conform to the operating trend, the current evaluation period is marked as a candidate period for temperature distortion. For all candidate periods for furnace temperature distortion, the corresponding furnace condition monitoring data is extracted, the mean and variance of the furnace temperature are calculated, and the second-order difference energy of the furnace temperature is further calculated. The furnace temperature and thermocouple voltage changes are then compared. The Pearson correlation coefficient of voltage is calculated. Simultaneously, the flue gas velocity, flue negative pressure, and exhaust fan operating frequency are sequentially differentially sampled at adjacent sampling points. All differential values ​​are squared and summed to obtain the overall disturbance energy. The second-order differential energy of furnace temperature is incremented by one and its natural logarithm is taken to obtain the temperature gradient term. The square of the Pearson correlation coefficient between furnace temperature and thermocouple voltage is subtracted from one, and the resulting difference is squared and added to the ratio of the overall disturbance energy to the minimum term, yielding the correlation bias term. The temperature gradient term is multiplied by the correlation bias term to obtain the distortion enhancement term. The square of the furnace temperature variance is incremented by one, its natural logarithm is taken, and then one is added again to obtain the variance constraint term. The distortion enhancement term is divided by the variance constraint term to obtain the furnace temperature distortion confidence assessment value.

[0014] Furthermore, the specific steps for performing differentiated repair on the temperature sequence according to the operating condition stage to generate the repaired furnace condition monitoring data are as follows: The furnace temperature distortion confidence assessment value and confidence threshold are compared. When the furnace temperature distortion confidence assessment value is less than or equal to the confidence threshold, the candidate period for temperature distortion is determined to be a false distortion and no processing is performed. When the furnace temperature distortion confidence assessment value is greater than the confidence threshold, the candidate period for temperature distortion is determined to have temperature distortion, and the data repair process begins: During the heating stage, local linear regression interpolation is performed on the furnace temperature sequence; during the isothermal stage, moving average interpolation is used; and during the cooling stage, exponential smoothing interpolation is used. The repaired furnace temperature sequence and the thermocouple voltage sequence are compared using first-order differential synchronization. When the signs of their differences are consistent, the original furnace temperature within the corresponding time range is replaced with the repaired furnace temperature sequence; otherwise, the repair process continues until the signs of their differences are consistent.

[0015] Furthermore, the specific steps for assessing the risk intensity of multi-source furnace condition characteristics during the fusion phase based on the repaired furnace condition monitoring data are as follows: Extract the repaired furnace condition monitoring data, resample the furnace condition monitoring data obtained from different sampling periods and different acquisition nodes onto a unified time axis to form a time-synchronized furnace condition dataset; Based on the time-synchronized furnace condition dataset, calculate the changes in furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency within the current assessment period, and calculate the absolute value of the difference between the four changes and the furnace temperature disturbance assessment value. Sum the four absolute values ​​of the difference to obtain the collaborative deviation; Square the collaborative deviation, add one, and take the natural logarithm to obtain the collaborative growth term; Square the furnace temperature distortion confidence assessment value and add one to obtain the distortion amplification term; Multiply the collaborative growth term and the distortion amplification term, use the negative of the product as the exponent, take the power function value of the natural constant e, and then subtract the obtained power function value from the constant to obtain the furnace condition fusion risk assessment value.

[0016] Further, abnormal periods are identified, and field consistency checks and priority coverage corrections are performed on these abnormal periods. The furnace condition records are then structured to achieve unified governance of multi-source heterogeneous industrial data. The specific steps are as follows: The furnace condition fusion risk assessment value is compared with the risk threshold. When the furnace condition fusion risk assessment value is less than or equal to the risk threshold, the multi-source furnace condition record for the corresponding assessment period is marked as a consistency pass record and directly written into the unified governance dataset. Otherwise, the corresponding assessment period is marked as a consistency anomaly record and enters the cross-channel data verification process: Field consistency checks are performed on the furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency of each sampling channel within the assessment period. The verification process... In the process, the variation range of each field between adjacent sampling points is compared, and the variation range is compared with the coordination deviation in the same period. Fields with inconsistent directions are filtered out and recorded as conflict fields. According to the set field priority rules, the reliable field value is selected to cover the abnormal field under the same time index. For the furnace condition monitoring data after the field conflict is processed, the temperature sensing data, flue gas disturbance data, combustion condition data and furnace status data in the same evaluation period are combined into a single record according to the time index and written into the unified governance data table in sequence. A time index, operating condition index and equipment index are established for each record. The record position and corresponding index key are registered in the index table to form a data entry that can be retrieved by index.

[0017] The second aspect of this invention provides a unified governance system based on multi-source heterogeneous industrial data, comprising: a data acquisition and preprocessing module, a semantic parsing feature construction module, a disturbance identification and distortion removal module, and a multi-source fusion unified governance module. The data acquisition and preprocessing module is used to acquire furnace condition monitoring data in real time and perform clock synchronization, noise suppression, anomaly removal, missing data imputation, and normalization processing on the acquired furnace condition monitoring data to generate preprocessed furnace condition monitoring data. The semantic parsing feature construction module is used to construct a semantic classification dataset based on the preprocessed furnace condition monitoring data, establish operating condition labels, and periodically evaluate the disturbance level of furnace temperature. The system performs perturbation attribute annotation on the semantic classification dataset to construct a structured furnace condition feature set. A perturbation identification and distortion removal module is used to filter candidate periods for temperature distortion based on the structured furnace condition feature set, quantify the distortion risk of furnace temperature within the candidate periods, perform differentiated repair on the temperature sequence according to the operating stage, and generate repaired furnace condition monitoring data. A multi-source fusion and unified governance module is used to assess the risk intensity of multi-source furnace condition features during the fusion stage based on the repaired furnace condition monitoring data, filter out abnormal periods, perform field consistency checks and priority coverage corrections on abnormal periods, and structure furnace condition records to achieve unified governance of multi-source heterogeneous industrial data.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) A unified governance system and method based on multi-source heterogeneous industrial data, which divides heterogeneous monitoring data such as furnace temperature, flue gas disturbance, combustion conditions and furnace status into semantic classification datasets according to physical attributes, and constructs condition labels based on sliding time windows, so that the data governance process is transformed from unstructured input to structured semantic expression, which greatly improves the understanding of physical scenarios in the process of industrial data governance and realizes semantic-level parsing capabilities that traditional methods do not have.

[0019] (2) A unified governance system and method based on multi-source heterogeneous industrial data. A furnace temperature disturbance assessment mechanism is constructed by combining temperature change ratio term, disturbance coupling term and hyperbolic tangent. A furnace temperature distortion confidence assessment model is established by introducing multi-dimensional features such as second-order differential energy, Pearson correlation coefficient and disturbance comprehensive energy. This enables quantitative characterization of temperature distortion behavior, making disturbance source identification and abnormal behavior identification more accurate and breaking through the limitations of the traditional simple threshold method.

[0020] (3) A unified governance system and method based on multi-source heterogeneous industrial data, which uses local linear regression interpolation, moving average interpolation and exponential smoothing interpolation respectively to achieve temperature sequence repair based on the results of heating, constant temperature and cooling stages. The reliability of the repair results is verified by synchronous verification in the first-order difference direction, so that the temperature sequence repair process is transformed from fixed algorithm repair to stage adaptive repair, which significantly improves the physical consistency and credibility of the furnace temperature after repair.

[0021] (4) A unified governance system and method based on multi-source heterogeneous industrial data. By constructing a furnace condition fusion risk assessment formula based on collaborative deviation, collaborative growth term and distortion amplification term, the deviation of multi-source furnace condition data in the fusion stage can be accurately quantified. Cross-channel data conflicts are resolved through field consistency verification and priority coverage correction. Finally, a unified governance data entry that can be retrieved by time index, operating condition index and equipment index is formed, realizing the consistency guarantee and high-quality fusion of multi-source data that is difficult to achieve in traditional industrial scenarios. Attached Figure Description

[0022] Figure 1 A flowchart of a unified governance method based on multi-source heterogeneous industrial data; Figure 2 This is a structural diagram of a unified governance system based on multi-source heterogeneous industrial data; Figure 3 A line graph showing the temperature distortion determination based on the furnace temperature distortion confidence assessment value; Figure 4 Overall flowchart for unified management of furnace condition data. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figures 1-4 This invention provides a technical solution: a unified governance method based on multi-source heterogeneous industrial data, comprising: S1, real-time acquisition of furnace condition monitoring data, and performing clock synchronization, noise suppression, anomaly removal, missing data imputation, and normalization processing on the acquired furnace condition monitoring data to generate preprocessed furnace condition monitoring data; S2, constructing a semantic classification dataset based on the preprocessed furnace condition monitoring data, establishing operating condition labels, periodically evaluating the disturbance level of furnace temperature, performing disturbance attribute labeling on the semantic classification dataset, and constructing a structured furnace condition feature set; S3, selecting candidate periods for temperature distortion based on the structured furnace condition feature set, quantifying the distortion risk of furnace temperature within the candidate periods, performing differentiated repair on the temperature sequence according to the operating condition stage, and generating repaired furnace condition monitoring data; S4, evaluating the risk intensity of multi-source furnace condition features in the fusion stage based on the repaired furnace condition monitoring data, selecting abnormal periods, performing field consistency checks and priority coverage correction on abnormal periods, and structuring furnace condition records to achieve unified governance of multi-source heterogeneous industrial data.

[0025] Specifically, the process involves real-time acquisition of furnace condition monitoring data, followed by clock synchronization, noise suppression, anomaly removal, missing data interpolation, and normalization to generate pre-processed furnace condition monitoring data. The specific steps are as follows: Real-time acquisition of furnace condition monitoring data, including furnace temperature, thermocouple voltage, flue gas velocity, flue negative pressure, flue gas oxygen content, burner gas flow rate, combustion air flow rate, heater power, furnace pressure, ambient temperature, furnace door opening / closing status, and exhaust fan operating frequency. During the acquisition process, the original time sequence of raw measurement points from different sources is recorded according to their respective sampling timestamps to ensure data integrity during the acquisition phase. The furnace temperature is determined by measurements taken from key locations within the furnace. The K-type thermocouple continuously outputs a temperature signal, and the thermocouple voltage is acquired in real time via a thermocouple signal acquisition card; the flue gas velocity is measured by a thermal velocity sensor installed in the flue, and the flue negative pressure is acquired by a pressure transmitter connected to the flue; the flue gas oxygen content is continuously monitored by an online flue gas oxygen analyzer; the burner gas flow rate is recorded by a gas flow meter, and the combustion air flow rate is acquired by an air flow sensor installed in the blower duct; the heater power is acquired in real time by an electricity meter; the furnace pressure is acquired by a micro-pressure transmitter on the furnace wall; the ambient temperature is measured by a temperature probe located outside the furnace; the furnace door opening / closing status is fed back by a magnetic switch installed at the furnace door; and the exhaust fan operating frequency is read from the fan inverter's operation monitoring interface.For the collected furnace condition monitoring data, cross-node time sequence alignment is achieved based on a unified clock synchronization protocol. During the alignment process, the collection timestamps are uniformly calibrated to ensure that the time series of furnace temperature, thermocouple voltage, flue gas velocity, flue gas negative pressure, flue gas oxygen content, burner gas flow rate, combustion air flow rate, heater power, furnace pressure, ambient temperature, furnace door opening / closing status, and exhaust fan operating frequency are correlated with the sampling points under the same reference clock. After time sequence alignment, a hybrid algorithm of polynomial smoothing filtering and amplitude limiting filtering is used to suppress and smooth the continuous time series data for noise. The amplitude constraint threshold of the amplitude limiting filter is set according to the mean of the field within the evaluation period plus or minus three standard deviations. Alternatively, the 95th percentile of the rectangular window can be used empirically to effectively suppress spike noise and avoid the cumulative amplification of high-frequency interference in the furnace temperature, flue gas velocity, and flue gas negative pressure sequences. Subsequently, anomaly observation points are identified through a local anomaly factor algorithm to eliminate those caused by flue gas disturbances and Outlier data caused by furnace disturbances are identified, and the reliability of distorted readings is improved by incorporating the local density characteristics of thermocouple voltage during the identification process. After outlier removal, the K-nearest neighbor interpolation algorithm is used to interpolate missing data caused by monitoring interruptions. The value of K ranges from three to fifteen, with five being preferred, to ensure that the interpolation neighborhood is both stable and reflects the changing trends of furnace temperature gradient, flue gas velocity fluctuation, and exhaust fan operating frequency before and after the time interval, making the interpolation results more consistent with the actual furnace condition evolution. Finally, the furnace condition monitoring data is normalized using a standard deviation normalization algorithm to unify the numerical scale of different physical quantities after processing, while maintaining the relative change characteristics of furnace temperature, thermocouple voltage, flue gas velocity, flue negative pressure, flue gas oxygen content, burner gas flow rate, combustion air flow rate, heater power, furnace pressure, ambient temperature, furnace door opening and closing status, and exhaust fan operating frequency without weakening them, providing a consistent and reliable data foundation for subsequent feature analysis based on physical sources.

[0026] In this implementation plan, the data sources for furnace temperature, thermocouple voltage, flue gas velocity, flue gas negative pressure, flue gas oxygen content, burner gas flow rate, combustion air flow rate, heater power, furnace pressure, ambient temperature, furnace door opening / closing status, and exhaust fan operating frequency are clearly defined during the data acquisition phase. Subsequent synchronization alignment, noise suppression, outlier removal, missing data interpolation, and normalization processes ensure that all data are processed under the same standard. This results in pre-processed furnace condition monitoring data with more stable temporal continuity, higher reading reliability, and stronger comparability. This refined design from acquisition to pre-processing not only strengthens the structural correlation between furnace temperature and thermocouple voltage but also improves the completeness of the expression of flue gas disturbance indicators, combustion condition indicators, and furnace status indicators in subsequent analysis. This allows subsequent semantic segmentation, disturbance identification, distortion judgment, and multi-source fusion to be based on a more reliable data foundation, thereby enhancing the stability of subsequent feature quantification results and the rigor of the analysis chain.

[0027] Specifically, the steps for constructing a semantic classification dataset and establishing operating condition labels based on preprocessed furnace condition monitoring data are as follows: Extract the preprocessed furnace condition monitoring data, maintaining time synchronization and dimensional consistency across all monitoring fields during the extraction process. Establish semantic labels based on the physical source and measurement meaning of the data. Classify furnace temperature and thermocouple voltage as temperature sensing data according to measurement attributes that directly reflect furnace thermal changes. Classify flue gas velocity, flue gas negative pressure, and flue gas oxygen content as flue gas disturbance data according to process characteristics describing flue gas disturbance behavior. Classify burner gas flow rate, combustion air flow rate, and heater power according to… Operating characteristics reflecting the strength of fuel supply and combustion conditions are categorized as combustion condition data. Furnace pressure, ambient temperature, furnace door opening and closing status, and exhaust fan operating frequency are categorized as furnace status data according to monitoring characteristics reflecting the external environmental conditions and structural boundary conditions of the furnace body, forming a semantic classification dataset. A fixed-width sliding time window is set as an evaluation cycle, and the length of this time window is required to be no less than three sampling points. During the gradual sliding of the window, it is ensured that the data segment corresponding to each evaluation cycle can simultaneously contain a complete record of temperature sensing data, flue gas disturbance data, combustion condition data, and furnace status data. Based on a semantic classification dataset, the changing trend of furnace temperature is periodically identified. By comparing the direction, magnitude, and duration of temperature changes, the current operating condition is divided into three stages: heating, isothermal, and cooling. Operating condition labels are then established. During the identification process, the furnace temperature within the evaluation period is first differentially analyzed between adjacent sampling points to determine the direction of temperature change. Then, the cumulative sum of continuous differential values ​​is calculated to determine whether the magnitude of change is continuously increasing or decreasing. Simultaneously, the shortest duration for which the differential sign remains consistent is recorded to determine the duration of change. When the differential value is continuously positive and the cumulative magnitude exceeds a set increase threshold, the current evaluation period is determined to be in the heating stage. When the differential value fluctuates within the differential threshold range and the cumulative magnitude is lower than both the increase and decrease thresholds, the current evaluation period is determined to be in the isothermal stage. When the differential value is continuously negative and the cumulative magnitude exceeds a set decrease threshold, the current evaluation period is determined to be in the cooling stage. Finally, the identified stage information is registered as an operating condition label and bound to the corresponding time index.

[0028] In this implementation scheme, by precisely segmenting furnace condition monitoring data based on physical sources during the semantic classification stage, and introducing a joint determination mechanism of differential direction, differential accumulation, and differential persistence during the condition identification process, the constructed semantic classification dataset has clearer data boundaries, more complete physical meaning expression, and a more stable time segment structure. The condition labels formed in this way can strengthen the determination criteria for stage segmentation while maintaining the correlation between temperature sensing data, flue gas disturbance data, combustion condition data, and furnace status data. This allows subsequent furnace temperature disturbance analysis, temperature distortion identification, and data fusion processing to be based on a more reliable condition benchmark, helping to reduce chain deviations caused by stage misjudgments and improving the overall coherence and stability of the furnace condition analysis chain.

[0029] Specifically, the steps for periodically assessing the disturbance level of furnace temperature are as follows: Extract the furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency within the current assessment period. During extraction, maintain consistency of the four monitoring quantities under the same time index. Calculate the instantaneous changes in furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency, and calculate the change in furnace temperature within the assessment period. In the instantaneous change calculation stage, perform absolute value calculation on the difference between adjacent sampling points to eliminate directional interference. Divide the instantaneous change in furnace temperature by the periodic change in furnace temperature and add a minimum term to obtain the temperature change ratio term. The minimum term is a very small but non-zero positive real number used to avoid numerical instability caused by division by zero during calculation. Its value range is [insert range here]. arrive Unless otherwise specified, all subsequent minima shall adopt this definition and value range; the instantaneous changes in flue gas velocity, flue negative pressure, and exhaust fan operating frequency shall be squared sequentially, summed, and the square root shall be taken, then one shall be added and the natural logarithm shall be taken to obtain the disturbance coupling term. When constructing the disturbance coupling term, the sensitivity to high-amplitude disturbances is enhanced by summing by squaring, and the amplification effect caused by abnormal peak values ​​is suppressed by taking the natural logarithm; the temperature change ratio term is multiplied by the disturbance coupling term, and the product is used as input to take the hyperbolic tangent value to obtain the furnace temperature disturbance evaluation value. In the hyperbolic tangent calculation, its output range limitation effect is used to compress extreme disturbance inputs to ensure that the evaluation results remain within a stable range.

[0030] The specific formula for calculating the furnace temperature disturbance assessment value is as follows: ; In the formula, This indicates the furnace temperature disturbance assessment value. This indicates the instantaneous change in furnace temperature. This indicates the periodic change in furnace temperature. This represents the instantaneous change in flue gas velocity. This indicates the instantaneous change in negative pressure in the flue. represents the instantaneous change in the operating frequency of the smoke exhaust fan, represents a minterm.

[0031] In this implementation, by introducing a strict extraction process for instantaneous changes, a stabilization construction strategy for change ratio terms, and an amplitude-sensitive enhancement mechanism for disturbance coupling terms during the disturbance level assessment process, the disturbance response of the furnace temperature is expressed in a more fine-grained form. The hyperbolic tangent function is used to perform a compression mapping on the combined disturbance input, improving the controllability of the disturbance results in high-fluctuation scenarios, and enabling the furnace temperature disturbance evaluation value to exhibit a smoother and more continuous change structure in consecutive evaluation cycles. As a result, the evaluation results maintain higher stability when dealing with sampling noise, short-term abnormal shocks, and cross-variable amplitude differences, thereby enhancing the reliability of furnace temperature disturbance determination and providing a more robust disturbance input basis for the subsequent temperature distortion identification process.

[0032] Specifically, the specific steps for performing disturbance attribute annotation on the semantic classification dataset and constructing a structured furnace condition feature set are as follows: Compare the furnace temperature disturbance evaluation value K with the multi-level disturbance thresholds K1 and K2, maintaining the time series continuity of all monitoring data within the current evaluation cycle during the comparison process, and ensuring that the disturbance attribute annotation is performed at the same time resolution. Perform disturbance attribute annotation on the semantic classification dataset within the evaluation cycle: When K ≤ K1, mark the current semantic classification dataset segment as a normal furnace condition segment, and record the low-amplitude fluctuation characteristics of the temperature change within this segment while marking, enabling this segment to reflect the temperature sensing data under stable operating conditions; when K1 < K < K2, mark the current semantic classification dataset segment as a segment affected by flue gas disturbance, and register the transient abnormal fluctuations of the flue gas flow rate, flue duct negative pressure, and the operating frequency of the smoke exhaust fan within this segment when marking, enabling the influence range of the disturbance source to be retained with time indexing; when K ≥ K2, mark the current semantic classification dataset segment as a segment suspected of thermocouple distortion, and synchronously record the deviation of the temperature sensing data and the thermocouple voltage within this segment during the marking process, enabling the time period when suspected distortion occurs to be locatable in the structured description; and attach the operating condition label, furnace temperature disturbance evaluation value, and disturbance attribute label to the corresponding time index in the semantic classification dataset, maintaining strict alignment between the label content and the corresponding monitoring fields during the attachment process to construct a structured furnace condition feature set. [[ID=IO]]

[0033] In this implementation, by performing perturbation attribute annotation on the semantic classification dataset and attaching operating condition labels, furnace temperature perturbation evaluation values, and perturbation attribute labels under the corresponding time index, the furnace condition monitoring data possesses clear temporal characteristic indication capabilities before entering subsequent processing. Since different perturbation levels are precisely divided during the annotation process and kept synchronized with temperature sensing data, flue gas perturbation data, combustion operating condition data, and furnace status data, this method can form furnace condition feature fragments with clear boundaries during the data structuring stage. This improves the ability to discriminate the source of perturbation behavior during subsequent distortion identification, data repair, and feature fusion processes, enabling each processing step to proceed on a more reliable semantic basis and achieving a fine-grained expression of dynamic changes in furnace conditions.

[0034] Specifically, the steps for selecting candidate periods for temperature distortion based on a structured furnace condition feature set and quantifying the distortion risk of furnace temperature within the candidate periods are as follows: Based on the semantic classification dataset and the furnace temperature perturbation evaluation value, cross-feature cross-validation is performed on the temperature sensing data within the evaluation period. By jointly comparing the instantaneous change in furnace temperature, the amplitude of thermocouple voltage change, and the periodic change in furnace temperature, when the furnace temperature and thermocouple voltage change are inconsistent and do not conform to the operating condition trend, the current evaluation period is marked as a candidate period for temperature distortion. During the marking process, the mean value of furnace temperature is calculated to provide a stable benchmark level for subsequent variance and difference analysis, which is used to support statistical stability analysis. For all candidate periods of furnace temperature distortion, the corresponding furnace condition monitoring data are extracted, and the mean and variance of the furnace temperature are calculated. The furnace temperature variance is used to characterize the stability of temperature fluctuations within the evaluation period. A larger furnace temperature variance indicates more unstable temperature changes, while a smaller furnace temperature variance indicates a more stable temperature sequence. Furthermore, the second-order difference energy of the furnace temperature is calculated to measure the severity of furnace temperature changes. A larger second-order difference energy indicates more obvious abrupt changes, while a smaller second-order difference energy indicates smoother temperature changes. The Pearson correlation coefficient between furnace temperature and thermocouple voltage is calculated to assess the temporal synchronicity of temperature and voltage changes. A Pearson correlation coefficient close to 1 indicates strong synchronicity, while a low Pearson correlation coefficient suggests that the thermocouple output may deviate from the actual temperature change. Simultaneously, the flue gas velocity, flue negative pressure, and exhaust fan operating frequency are sequentially differentially sampled from adjacent sampling points. The squared values ​​of all differentials are then summed to obtain the comprehensive disturbance energy, which describes the intensity of external disturbances. A larger comprehensive disturbance energy indicates more significant environmental fluctuations, while a smaller comprehensive disturbance energy indicates weaker external disturbances to the furnace. The temperature gradient term is obtained by adding one to the second-order difference energy of the furnace temperature and taking its natural logarithm. This temperature gradient term smooths and compresses the degree of temperature abrupt changes, mapping drastic changes to a controllable range. The correlation bias term is obtained by subtracting the square of the Pearson correlation coefficient between the furnace temperature and the thermocouple voltage from one, squared the difference, and adding it to the ratio of the total disturbance energy to the minimum term of the total disturbance energy. The internal synchronization bias component is used to reinforce the inconsistency between temperature and voltage changes, while the disturbance bias component reflects external disturbances. The contribution to the probability of distortion is as follows: Multiplying the temperature gradient term with the relevant deviation term yields a distortion enhancement term, which is used to amplify the distortion signal when internal deviations and external disturbances coexist, making key risks more prominent; Adding one to the square of the furnace temperature variance, taking the natural logarithm, and adding one more yields a variance constraint term, which is used to suppress the amplification of cross-cycle extreme values ​​caused by drastic temperature fluctuations, making the final assessment more stable and reliable; Dividing the distortion enhancement term by the variance constraint term yields the furnace temperature distortion confidence assessment value, which serves as an indicator for comprehensively quantifying the probability of temperature distortion.

[0035] The specific formula for calculating the confidence assessment value of furnace temperature distortion is as follows: ; In the formula, This indicates the confidence assessment value for furnace temperature distortion. This represents the second-order difference energy of the furnace temperature. This represents the variance of furnace temperature. The Pearson correlation coefficient represents the relationship between furnace temperature and thermocouple voltage. This represents the total energy of the disturbance. Indicates a minus term.

[0036] In this embodiment, Table 1 is a data table of furnace temperature distortion confidence assessment values, listing all parameter data used to quantify the risk of furnace temperature distortion in the five candidate periods of temperature distortion within this assessment period. Specifically: Candidate period 1 of temperature distortion: the second-order difference energy of furnace temperature is 0.80, the variance of furnace temperature is 2.0, the Pearson correlation coefficient between furnace temperature and thermocouple voltage is 0.95, the comprehensive disturbance energy is 1.20, and the confidence assessment value of furnace temperature distortion is 0.226. Candidate period 2 of temperature distortion: the second-order difference energy of furnace temperature is 1.50, the variance of furnace temperature is 3.0, the Pearson correlation coefficient between furnace temperature and thermocouple voltage is 0.85, the comprehensive disturbance energy is 2.50, and the confidence assessment value of furnace temperature distortion is 0.298. Candidate period for temperature distortion 3: Second-order difference energy of furnace temperature is 0.30, furnace temperature variance is 1.0, Pearson correlation coefficient between furnace temperature and thermocouple voltage is 0.60, total disturbance energy is 0.80, and confidence assessment value for furnace temperature distortion is 0.217. Candidate period for temperature distortion 4: Second-order difference energy of furnace temperature is 2.00, furnace temperature variance is 4.0, Pearson correlation coefficient between furnace temperature and thermocouple voltage is 0.40, total disturbance energy is 3.00, and confidence assessment value for furnace temperature distortion is 0.488. Candidate period for temperature distortion 5: Second-order difference energy of furnace temperature is 1.00, furnace temperature variance is 2.5, Pearson correlation coefficient between furnace temperature and thermocouple voltage is 0.75, total disturbance energy is 1.50, and confidence assessment value for furnace temperature distortion is 0.275.

[0037] Table 1. Confidence Assessment Values ​​for Furnace Temperature Distortion like Figure 3As shown in the figure, the furnace temperature distortion confidence assessment values ​​for five candidate temperature distortion periods are displayed, along with the corresponding temperature distortion judgment results. The line graph uses different colors to distinguish the nature of distortion: blue dots represent false distortion periods, and red dots represent temperature distortion periods. The figure uses black dashed lines to indicate the distortion confidence threshold, used to distinguish between high-risk and low-risk distortion periods. It can be seen from the figure that the furnace temperature distortion confidence assessment values ​​for candidate temperature distortion periods 1, 2, 3, and 5 are all below the confidence threshold, indicating that although there are fluctuations in the furnace temperature readings during these periods, the fluctuation amplitude remains consistent with the thermocouple voltage changes and flue gas disturbance energy, and can be judged as false distortion, not triggering the repair process. The furnace temperature distortion confidence assessment value for candidate temperature distortion period 4 exceeds the confidence threshold, and the temperature gradient term is significantly increased while the temperature-voltage correlation is significantly reduced, belonging to a typical temperature distortion period, and should enter the phased temperature sequence repair process to restore the effective furnace temperature signal. Figure 3 This paper presents a intuitive demonstration of a furnace temperature distortion confidence assessment mechanism based on second-order differential energy of furnace temperature, temperature variance, temperature-voltage correlation coefficient, and comprehensive disturbance energy. Through comprehensive quantitative analysis of the consistency, stability, and coupling deviation of actual acquired signals under disturbance conditions, the mechanism can accurately identify temperature distortion cycles. This mechanism can promptly detect distortion anomalies during furnace temperature monitoring, providing a reliable basis for subsequent temperature repair, anomaly removal, and unified management of multi-source furnace condition data.

[0038] In this implementation scheme, multi-dimensional coupled analysis of furnace temperature, thermocouple voltage, and flue gas disturbance data is performed within the candidate period for temperature distortion. A hierarchical quantification process is constructed using the furnace temperature mean, furnace temperature variance, second-order difference energy of furnace temperature, Pearson correlation coefficient between furnace temperature and thermocouple voltage, and comprehensive disturbance energy as inputs. This allows for the step-by-step separation of internal temperature variation consistency, external disturbance intensity, and temperature fluctuation stability within the same mathematical framework. Through the joint construction of temperature gradient terms, correlation bias terms, distortion enhancement terms, and variance constraint terms, a clear distinction is made between weak and significant distortions in terms of magnitude. This achieves high recognition capability for true temperature distortion during the candidate period selection stage, improves the reliability of temperature distortion judgment, reduces the probability of misjudging short-term fluctuations in furnace temperature, and provides a reliable basis for subsequent temperature sequence repair operations.

[0039] Specifically, the steps for performing differentiated repair on the temperature sequence according to the operating condition stage to generate the repaired furnace condition monitoring data are as follows: The furnace temperature distortion confidence assessment value and the confidence threshold are compared. During the comparison process, furnace temperature, thermocouple voltage, and flue gas disturbance data are first extracted based on the time index of the corresponding assessment period to ensure that the judgment of the furnace temperature distortion confidence assessment value and the confidence threshold is based on the same time series benchmark. When the furnace temperature distortion confidence assessment value is less than or equal to the confidence threshold, the candidate period for temperature distortion is determined to be a false distortion. In this case, the original furnace temperature sequence remains unchanged, so that normal temperature changes are not mistakenly repaired. When the confidence assessment value of furnace temperature distortion is greater than the confidence threshold, temperature distortion is determined to have occurred in the candidate period of temperature distortion, and the data repair process begins: During the heating stage, local linear regression interpolation is performed on the furnace temperature sequence. Before interpolation, continuous temperature segments during the heating stage are extracted, and the regression window is constrained to only cover intervals with consistent trends, so that the repaired temperature change can maintain the true shape of the heating slope; During the isothermal stage, moving average interpolation is used. Before performing moving average, the width of the moving window is fixed, and the temperature value within the window is extracted point by point from the isothermal temperature sequence, so that the repaired temperature sequence maintains low fluctuation characteristics; During the cooling stage, exponential smoothing interpolation is used. Before performing exponential smoothing, the smoothing coefficient is adaptively selected according to the cooling slope, so that the repaired temperature sequence can maintain a natural decay trend. The repaired furnace temperature sequence and the thermocouple voltage sequence are compared synchronously using a first-order differential method. During the comparison, the difference sign, difference amplitude, and difference consistency duration are calculated simultaneously. When the difference signs are consistent, the original furnace temperature within the corresponding time range is replaced with the repaired furnace temperature sequence to ensure that the repair result is consistent with the true thermal response of the thermocouple voltage in terms of direction and continuity. Otherwise, the repair process continues. By reselecting the interpolation window and adjusting the interpolation parameters, the repaired furnace temperature sequence gradually approximates the true temperature change trajectory, ensuring that the difference sign between the repaired sequence and the thermocouple voltage remains consistent.

[0040] In this implementation scheme, local linear regression interpolation, moving average interpolation, and exponential smoothing interpolation are used during the heating, isothermal, and cooling stages, respectively. After the repair is completed, a first-order differential synchronous comparison is performed between the repaired furnace temperature sequence and the thermocouple voltage sequence. This allows the temperature repair process to automatically calibrate the repair results based on the true thermal response characteristics. This method not only avoids trend distortion caused by a uniform interpolation strategy but also ensures that the direction, rate, and continuity of temperature change after repair better conform to the physical laws of the furnace thermal state. This significantly improves the availability and accuracy of furnace temperature within the distortion period, laying a highly reliable foundation for subsequent disturbance identification, anomaly screening, and multi-source feature fusion.

[0041] Specifically, the steps for assessing the risk intensity of multi-source furnace condition characteristics during the fusion phase based on the repaired furnace condition monitoring data are as follows: Extract the repaired furnace condition monitoring data, maintaining consistency in the timestamp accuracy of all monitoring fields during extraction, and ensuring that furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency all originate from the same sampling reference; reconstruct the furnace condition monitoring data obtained from different sampling periods and different acquisition nodes according to a unified sampling interval, filling in the missing sampling positions during the resampling process using linear interpolation to form a time-synchronized furnace condition dataset; based on the time-synchronized furnace condition dataset, calculate the changes in furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency for the current assessment period, respectively, using adjacent sampling data as a reference during the calculation process. Sample point differencing is used to obtain the true change magnitude; the absolute value of the difference between the four changes and the furnace temperature disturbance assessment value is calculated, and after obtaining the absolute difference sequence, the four absolute differences are summed to obtain the collaborative deviation; the collaborative deviation is squared, incremented by one, and the natural logarithm is taken to obtain the collaborative growth term, where the squaring operation is used to amplify the contribution corresponding to the large deviation; the furnace temperature distortion confidence assessment value is squared and incremented by one to obtain the distortion amplification term, where the squaring operation is used to enhance the impact of distortion accumulation in the fusion stage; the collaborative growth term and the distortion amplification term are multiplied, and the opposite of the product is taken as the exponent, and then the exponential mapping is performed in the form of a power function of the natural constant e. The fusion sensitivity response is formed through the exponential decay mechanism, and then the furnace condition fusion risk assessment value is obtained by subtracting the obtained power function value from the constant one.

[0042] The specific formula for calculating the risk assessment value of furnace condition integration is as follows: ; In the formula, This indicates the risk assessment value for furnace condition integration. This represents the amount of coordination deviation. This indicates the confidence assessment value for furnace temperature distortion.

[0043] In this implementation scheme, by performing unified timeline reconstruction, co-analysis of changes, and exponential mapping quantification on the repaired furnace condition monitoring data, a comprehensive and sensitive identification capability for multi-source furnace temperature characteristics and disturbance behavior can be formed during the fusion stage. This allows for the accurate identification of potential consistency deviations caused by differences in time series, acquisition nodes, and disturbance amplitudes. Through a joint enhancement mechanism of co-variance deviation and furnace temperature distortion confidence assessment value, the potential risks of abnormal cycles can be effectively exposed in advance before fusion, ensuring that the subsequent data fusion process is entirely based on verified data consistency. This method can significantly improve the stability of multi-source furnace condition monitoring data during the fusion stage, ensuring that temperature sensing data, flue gas disturbance data, combustion condition data, and furnace status data maintain a high degree of coordination before entering unified governance, thereby enhancing the reliability of the overall data governance chain and the credibility of the fused data.

[0044] Specifically, the following steps are taken to achieve unified governance of multi-source heterogeneous industrial data: The abnormal periods are screened out, and field consistency checks and priority overwrite corrections are performed on these periods. The furnace condition records are then structured. The specific steps are as follows: The furnace condition fusion risk assessment value is compared with the risk threshold. During the comparison, the time index consistency of all data fields is maintained. When the furnace condition fusion risk assessment value is less than or equal to the risk threshold, the multi-source furnace condition record for the corresponding assessment period is marked as a consistency-passing record. Simultaneously, the original values ​​of complete temperature sensing data, flue gas disturbance data, combustion condition data, and furnace status data are retained and directly written into the unified governance dataset. Otherwise, the corresponding assessment... Records with inconsistent data are marked as such and then enter a cross-channel data verification process. This process involves checking the reliability of data for each sampling channel, performing field consistency checks on furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency for each sampling channel within the evaluation period. During the verification, the actual variation of each field between adjacent sampling points is first calculated. This variation is then compared item by item with the corresponding periodic deviation. Fields whose variation direction deviates from the coordinated trend are identified and recorded as conflicting fields. Subsequently, based on the set field priority rules, reliable field values ​​are selected under the same time index to cover the conflicting values. The data consists of constant fields. Field priority rules are applied based on the direct correlation between each field and the furnace thermal state, its sensitivity to furnace condition trends, and the stability of historical monitoring data. Specifically, furnace temperature in temperature sensing data takes precedence over thermocouple voltage; flue gas negative pressure in flue gas disturbance data takes precedence over flue gas velocity; burner gas flow rate in combustion condition data takes precedence over combustion air flow rate; and furnace pressure in furnace status data takes precedence over ambient temperature. This ensures that field values ​​with stronger physical correlation and higher stability are prioritized for correction during the coverage process. Furthermore, the coverage values ​​are ensured to meet the collaborative bias requirements. The differential trend requirement ensures that the correction results remain consistent with the overall furnace condition changes. For furnace condition monitoring data that has completed field conflict processing, temperature sensing data, flue gas disturbance data, combustion condition data, and furnace status data within the same evaluation period are combined into a single record according to the time index. During the combination process, the field arrangement order of each data category is ensured to be consistent with the field specifications of the governance data table, and they are written into the unified governance data table in sequence. A time index, operating condition index, and equipment index are established for each record, and the record location and corresponding index key are registered in the index table. The index key enables fast and accurate retrieval and positioning as needed, so that multi-source furnace condition records have structured access capabilities after governance.

[0045] In this implementation scheme, by performing field consistency checks and priority overwrite corrections during abnormal cycles, multi-source furnace condition records can establish stable data associations within a single cycle before being written into the governance data table. This ensures that temperature sensing data, flue gas disturbance data, combustion condition data, and furnace status data remain consistent under the same time index. When field conflicts occur, this method can identify the deviation pattern based on the amount of coordinated deviation and overwrite abnormal fields with reliable fields, resulting in a continuous, stable, and complete data structure after governance. The furnace condition records constructed in this way possess highly consistent temporal logic and reliable numerical expression capabilities, forming accurate, standardized, and searchable data entries in the storage medium, laying a rigorous data foundation for the unified governance of furnace condition monitoring data.

[0046] like Figure 2 As shown, the second aspect of this invention provides a unified governance system based on multi-source heterogeneous industrial data, comprising: a data acquisition and preprocessing module, a semantic parsing feature construction module, a disturbance identification and distortion removal module, and a multi-source fusion unified governance module. The data acquisition and preprocessing module is used to acquire furnace condition monitoring data in real time and perform clock synchronization, noise suppression, anomaly removal, missing data imputation, and normalization processing on the acquired furnace condition monitoring data to generate preprocessed furnace condition monitoring data. The semantic parsing feature construction module is used to construct a semantic classification dataset based on the preprocessed furnace condition monitoring data, establish operating condition labels, and periodically evaluate disturbances in furnace temperature. The system comprises four modules: a perturbation attribute annotation module for semantic classification datasets to construct a structured furnace condition feature set; a perturbation identification and distortion removal module to filter candidate periods for temperature distortion based on the structured furnace condition feature set, quantify the distortion risk of furnace temperature within the candidate periods, perform differentiated repair on temperature sequences according to operating conditions, and generate repaired furnace condition monitoring data; and a multi-source fusion unified governance module to assess the risk intensity of multi-source furnace condition features during the fusion stage based on the repaired furnace condition monitoring data, filter out abnormal periods, perform field consistency checks and priority coverage corrections on abnormal periods, and structure furnace condition records to achieve unified governance of multi-source heterogeneous industrial data. The specific governance process is as follows: Figure 4 As shown.

[0047] This implementation plan establishes a continuous governance chain across the stages of data acquisition, analysis, identification, and fusion. This ensures a stable correlation between furnace temperature, thermocouple voltage, flue gas velocity, flue gas negative pressure, flue gas oxygen content, burner gas flow rate, combustion air flow rate, heater power, furnace pressure, ambient temperature, furnace door opening / closing status, and exhaust fan operating frequency throughout the entire process from raw records to final structured storage. This governance chain possesses complete cleaning, synchronization, and calibration capabilities before data enters semantic classification, operating condition identification, temperature distortion repair, and multi-source fusion, enabling consistent representation of multi-source furnace condition monitoring data within a single framework. The resulting furnace condition records have clear semantic boundaries, identifiable operating condition stages, verifiable temperature reliability, and quantifiable fusion risk characteristics. This allows multi-source heterogeneous industrial data to be presented in a unified format, with unified logic and a unified time base after governance, providing robust technical support for data quality improvement.

[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0049] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A unified governance method based on multi-source heterogeneous industrial data, characterized in that, Includes the following steps: S1 collects furnace condition monitoring data in real time and performs clock synchronization, noise suppression, anomaly removal, missing data imputation and normalization on the collected furnace condition monitoring data to generate preprocessed furnace condition monitoring data. S2. Based on the preprocessed furnace condition monitoring data, construct a semantic classification dataset, establish operating condition labels, periodically evaluate the disturbance level of furnace temperature, perform disturbance attribute labeling on the semantic classification dataset, and construct a structured furnace condition feature set. S3, based on the structured furnace condition feature set, selects candidate periods for temperature distortion, quantifies the distortion risk of furnace temperature within the candidate periods, performs differentiated repair on the temperature sequence according to the working condition stage, and generates repaired furnace condition monitoring data. S4 assesses the risk intensity of multi-source furnace condition characteristics during the fusion stage based on the repaired furnace condition monitoring data, filters out abnormal cycles, performs field consistency checks and priority coverage corrections on abnormal cycles, and structures furnace condition records to achieve unified governance of multi-source heterogeneous industrial data.

2. The unified governance method based on multi-source heterogeneous industrial data according to claim 1, characterized in that: The specific steps for acquiring real-time furnace condition monitoring data and performing clock synchronization, noise suppression, anomaly removal, missing data imputation, and normalization on the acquired furnace condition monitoring data to generate preprocessed furnace condition monitoring data are as follows: Real-time collection of furnace condition monitoring data, including furnace temperature, thermocouple voltage, flue gas velocity, flue negative pressure, flue gas oxygen content, burner gas flow rate, combustion air flow rate, heater power, furnace pressure, ambient temperature, furnace door opening and closing status, and exhaust fan operating frequency. For the collected furnace condition monitoring data, cross-node time alignment is achieved based on a unified clock synchronization protocol, and a hybrid algorithm of polynomial smoothing filtering and amplitude limiting filtering is used to suppress noise and smooth continuous time series data. Anomaly observation points are identified through the local anomaly factor algorithm to remove outlier data caused by flue gas disturbance and furnace disturbance. Missing data caused by monitoring interruption is interpolated through the K-nearest neighbor interpolation algorithm. The furnace condition monitoring data is normalized through the standard deviation normalization algorithm to achieve numerical scale uniformity.

3. The unified governance method based on multi-source heterogeneous industrial data according to claim 1, characterized in that: The specific steps for constructing a semantic classification dataset and establishing operating condition labels based on the preprocessed furnace condition monitoring data are as follows: Extract the preprocessed furnace condition monitoring data, establish semantic labels based on the physical source and measurement meaning of the data, classify furnace temperature and thermocouple voltage as temperature sensing data, flue gas velocity, flue negative pressure, and flue gas oxygen content as flue gas disturbance data, burner gas flow rate, combustion air flow rate, and heater power as combustion condition data, and furnace pressure, ambient temperature, furnace door opening and closing status, and exhaust fan operating frequency as furnace body condition data, forming a semantic classification dataset; A fixed-width sliding time window is set as an evaluation period. Based on the semantic classification dataset, the changing trend of furnace temperature is periodically identified. The current working condition is divided into a heating stage, a constant temperature stage, and a cooling stage, and working condition labels are established.

4. The unified governance method based on multi-source heterogeneous industrial data according to claim 1, characterized in that: The specific steps for periodically assessing the disturbance level of the furnace temperature are as follows: Extract the furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency within the current evaluation period. Calculate the instantaneous changes in furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency, and calculate the change in furnace temperature within the evaluation period. Divide the instantaneous change in furnace temperature by the periodic change in furnace temperature and add a minimum term to obtain the temperature change ratio term; square the instantaneous changes in flue gas velocity, flue negative pressure, and exhaust fan operating frequency in sequence, sum them, take the square root, add one, and take the natural logarithm to obtain the disturbance coupling term; multiply the temperature change ratio term and the disturbance coupling term, take the hyperbolic tangent value as input, and obtain the furnace temperature disturbance evaluation value.

5. The unified governance method based on multi-source heterogeneous industrial data according to claim 4, characterized in that: The specific steps for performing perturbation attribute annotation on the semantic classification dataset and constructing a structured furnace condition feature set are as follows: Furnace temperature disturbance assessment value and multi-level perturbation threshold and A comparison was performed, and perturbation attribute annotations were applied to the semantic classification dataset within the evaluation period: when ≤ At that time, the current semantic classification dataset segment is marked as a normal furnace condition segment; when < < At that time, the current semantic classification dataset segment is marked as the segment affected by smoke disturbance; when ≥ At that time, the current semantic classification dataset segment is marked as a suspected thermocouple distortion segment; The operating condition label, furnace temperature disturbance assessment value, and disturbance attribute label are then attached to the corresponding time index in the semantic classification dataset to construct a structured furnace condition feature set.

6. The unified governance method based on multi-source heterogeneous industrial data according to claim 1, characterized in that: The specific steps for selecting candidate periods for temperature distortion based on a structured furnace condition feature set and quantifying the distortion risk of furnace temperature within the candidate periods are as follows: Based on the semantic classification dataset and the furnace temperature disturbance evaluation value, cross-feature cross-validation is performed on the temperature sensing data within the evaluation period. By jointly comparing the instantaneous change in furnace temperature, the magnitude of thermocouple voltage change, and the periodic change in furnace temperature, when the furnace temperature and thermocouple voltage change are inconsistent and do not match the operating condition trend, the current evaluation period is marked as a candidate period for temperature distortion. For all candidate periods of furnace temperature distortion, the corresponding furnace condition monitoring data are extracted, the mean and variance of the furnace temperature are calculated, and the second-order difference energy of the furnace temperature is further calculated. Calculate the Pearson correlation coefficient between furnace temperature and thermocouple voltage. Simultaneously, perform differential sampling on adjacent sampling points for flue gas velocity, flue negative pressure, and exhaust fan operating frequency. Squaring all differential values ​​and summing them yields the comprehensive disturbance energy. Adding one to the second-order differential energy of furnace temperature and taking its natural logarithm yields the temperature gradient term. Subtracting the square of the Pearson correlation coefficient between furnace temperature and thermocouple voltage from one, squaring the resulting difference, and adding it to the ratio of the comprehensive disturbance energy to the minimum term yields the correlation bias term. Multiplying the temperature gradient term and the correlation bias term yields the distortion enhancement term. Adding one to the square of the furnace temperature variance, taking its natural logarithm, and adding one again yields the variance constraint term. Dividing the distortion enhancement term by the variance constraint term yields the furnace temperature distortion confidence assessment value.

7. The unified governance method based on multi-source heterogeneous industrial data according to claim 6, characterized in that: The specific steps for performing differentiated repair on the temperature sequence according to the operating condition stage to generate repaired furnace condition monitoring data are as follows: The confidence assessment value of furnace temperature distortion is compared with the confidence threshold. When the confidence assessment value is less than or equal to the confidence threshold, the candidate period for temperature distortion is determined to be false distortion and no action is taken. When the confidence assessment value is greater than the confidence threshold, the candidate period for temperature distortion is determined to have temperature distortion, and the data repair process begins. During the heating phase, local linear regression interpolation is performed on the furnace temperature sequence. During the isothermal phase, moving average interpolation is used, and during the cooling phase, exponential smoothing interpolation is used. The repaired furnace temperature sequence is compared with the thermocouple voltage sequence using first-order differential synchronization. When the signs of their differences are consistent, the original furnace temperature within the corresponding time range is replaced with the repaired furnace temperature sequence. Otherwise, the repair process continues until the signs of their differences are consistent.

8. The unified governance method based on multi-source heterogeneous industrial data according to claim 1, characterized in that: The specific steps for assessing the risk intensity of multi-source furnace condition characteristics during the fusion phase based on the repaired furnace condition monitoring data are as follows: Extract the repaired furnace condition monitoring data, and resample the furnace condition monitoring data obtained from different sampling periods and different acquisition nodes onto a unified time axis to form a time-synchronized furnace condition dataset; Based on the time-synchronized furnace condition dataset, the changes in furnace temperature, flue gas velocity, flue negative pressure, and exhaust fan operating frequency within the current assessment period are calculated. The absolute values ​​of the differences between these four changes and the furnace temperature disturbance assessment values ​​are calculated, and the sum of the four absolute values ​​is obtained to obtain the collaborative deviation. The collaborative deviation is squared, one is added, and the natural logarithm is taken to obtain the collaborative growth term. The furnace temperature distortion confidence assessment value is squared and one is added to obtain the distortion amplification term. The collaborative growth term and the distortion amplification term are multiplied, and the negative of the product is used as the exponent. The power function value of the natural constant e is taken, and the constant e is subtracted from the power function value to obtain the furnace condition fusion risk assessment value.

9. A unified governance method based on multi-source heterogeneous industrial data according to claim 8, characterized in that: The specific steps for filtering out abnormal periods, performing field consistency checks and priority overwrite corrections on abnormal periods, and structuring furnace condition records to achieve unified governance of multi-source heterogeneous industrial data are as follows: The furnace condition fusion risk assessment value is compared with the risk threshold. When the furnace condition fusion risk assessment value is less than or equal to the risk threshold, the multi-source furnace condition record of the corresponding assessment period is marked as a consistency pass record and directly written into the unified governance dataset. Otherwise, the corresponding assessment period is marked as a consistency anomaly record and enters the cross-channel data verification process: the furnace temperature, flue gas velocity, flue negative pressure and exhaust fan operating frequency of each sampling channel in the assessment period are checked for field consistency. During the verification process, the change range of each field between adjacent sampling points is compared and the change range is compared with the collaborative deviation amount in the same period. Fields with inconsistent directions are filtered out and recorded as conflict fields. According to the set field priority rules, the trusted field value is selected to cover the abnormal field under the same time index. For furnace condition monitoring data that has completed field conflict resolution, temperature sensing data, flue gas disturbance data, combustion condition data and furnace status data within the same evaluation period are combined into a single record according to the time index and written into the unified governance data table in sequence; For each record, a time index, operating condition index, and equipment index are created. The record location and corresponding index key are registered in the index table to form a data entry that can be retrieved by index.

10. A unified governance system based on multi-source heterogeneous industrial data, characterized in that, include: The module comprises a data acquisition and preprocessing module, a semantic parsing feature construction module, a disturbance identification and distortion removal module, and a multi-source fusion and unified governance module, among which: The data acquisition and preprocessing module is used to acquire furnace condition monitoring data in real time, and to perform clock synchronization, noise suppression, anomaly removal, missing data imputation and normalization on the acquired furnace condition monitoring data to generate preprocessed furnace condition monitoring data. The semantic parsing feature construction module is used to construct a semantic classification dataset based on the preprocessed furnace condition monitoring data, establish operating condition labels, periodically evaluate the disturbance level of furnace temperature, perform disturbance attribute labeling on the semantic classification dataset, and construct a structured furnace condition feature set. The disturbance identification and distortion removal module is used to screen out candidate periods of temperature distortion based on a structured furnace condition feature set, quantify the distortion risk of furnace temperature within the candidate periods, perform differentiated repair on the temperature sequence according to the working condition stage, and generate repaired furnace condition monitoring data. The multi-source fusion unified governance module is used to assess the risk intensity of multi-source furnace condition characteristics during the fusion stage based on the repaired furnace condition monitoring data, screen out abnormal cycles, perform field consistency checks and priority overwrite corrections on abnormal cycles, and structure furnace condition records to achieve unified governance of multi-source heterogeneous industrial data.

Images