Intelligent gas micro-differential pressure monitoring system based on multi-sensor fusion

Abstract

The invention discloses an intelligent gas micro-differential pressure monitoring system based on multi-sensor fusion, and relates to the technical field of gas micro-differential pressure monitoring, a sensor network arranged at a pipeline collects and verifies process state data, a component fluctuation index is established to quantify the fluctuation degree of gas components, and if the component fluctuation index is abnormal, the component fluctuation index is determined to be abnormal. Calling a dynamic physical property compensation function to output a correction coefficient and a differential pressure reference value; if the accumulative deviation between the actually measured differential pressure and the theoretical differential pressure reference value exceeds a set threshold value, locally adjusting and optimizing or globally re-calibrating and updating the model; outputting a real-time correction value and deviation measurement through a fusion algorithm, and when an error continuously exceeds the standard, dynamically adjusting a fusion weight or triggering model calibration; if the obtained value exceeds the preset threshold value, early warning or alarming is sent out and fed back to the previous step.

Description

Technical Field

[0001] The present invention relates to the technical field of gas differential pressure monitoring, and specifically to an intelligent gas differential pressure monitoring system based on multi-sensor fusion. Background Art

[0002] In the fields of modern industrial processes and high-precision environmental control, the accurate monitoring of gas differential pressure is of great significance for ensuring production safety, improving product quality, and achieving energy conservation and consumption reduction. Common application scenarios include clean room pressure control, microbalance of ventilation systems, exhaust gas monitoring during combustion processes, and flow measurement of chemical pipelines, etc. In these scenarios, the gas composition is often no longer limited to a single component, and the actual working environment will also change continuously with factors such as temperature and humidity, flow conditions, and operating loads, resulting in obvious multi-variable coupling characteristics of differential pressure fluctuations. To balance the accuracy of real-time measurement and the robustness of the system, the intelligent gas differential pressure monitoring technology based on multi-sensor fusion has gradually become a solution widely concerned in the industry.

[0003] However, in the face of multiple types of gas sensors and complex on-site working conditions, traditional single sensors or fixed calibration modes are often unable to cope with long-term drift, dynamic changes in gas composition, and on-site interference. Especially in the measurement of differential pressure levels (such as Pascal level or millimeter water column level), any zero drift of the sensor itself, sensitivity mismatch, or minor fluctuations in the process will significantly amplify the measurement error. Therefore, it is necessary to systematically integrate gas composition analysis, physical property compensation models, and adaptive fusion algorithms to ensure the accuracy and reliability of differential pressure measurement through multi-source data complementarity and real-time error correction.

[0004] In the actual scenario where the gas composition fluctuates or changes continuously (such as the proportion of mixed gases is adjusted according to the process stage, and multiple gases flow alternately in the pipeline), how to correct the differential pressure measurement deviation caused by composition differences in real time. Specifically, when the molecular weight, viscosity, or thermal physical property parameters of each component in the gas flow change, if a single compensation coefficient is still used or dynamic updates are ignored, the system will be unable to accurately reflect the true pressure difference, and even large errors within the range of the measuring range or threshold will occur; at the same time, without differentiated management of different gas characteristics, minor air leakage, condensation, and water vapor interference will further exacerbate the measurement instability. This problem not only affects the instantaneous pressure monitoring results in a small range, but also accumulates significant deviations during long-term operation, bringing potential risks to subsequent process control and equipment safety.

[0005] Therefore, the present invention provides an intelligent gas differential pressure monitoring system based on multi-sensor fusion. Summary of the Invention

[0006] (1) Technical Problems to be Solved In view of the deficiencies of the prior art, the present invention provides an intelligent gas differential pressure monitoring system based on multi-sensor fusion. The sensor network arranged at the pipeline collects process state data and verifies it, and establishes a component fluctuation index to quantify the fluctuation degree of the gas component. If the component fluctuation index is abnormal, the dynamic physical property compensation function is called to output a correction coefficient and a differential pressure reference value. If the deviation between the measured differential pressure and the theoretical differential pressure reference value accumulates and exceeds the set threshold, local optimization or global recalibration is performed and the model is updated. The real-time correction value and deviation measurement are output through the fusion algorithm. When the error continuously exceeds the standard, the fusion weight is dynamically adjusted or the model calibration is triggered. If the obtained value exceeds the preset threshold, a warning or alarm is issued and feedback is sent back to the previous step. For the industrial site with dynamic changes in multiple gas components, it is necessary to deeply couple multiple links such as front-end sensor acquisition, gas physical property modeling, and measurement fusion processing, so as to achieve both the monitoring accuracy and response speed of the differential pressure, thus solving the technical problems recorded in the background art.

[0007] (2) Technical solution To achieve the above objectives, the present invention is realized through the following technical solutions: An intelligent gas differential pressure monitoring system based on multi-sensor fusion, including that the sensor network arranged at the pipeline collects process state data and verifies it, and establishes a component fluctuation index to quantify the fluctuation degree of the gas component. If the component fluctuation index is abnormal, the corresponding time interval is marked. The effective target process data at the same moment is aligned in time series, and the dynamic physical property compensation function is called to output a correction coefficient and a differential pressure reference value. If the deviation between the measured differential pressure and the theoretical differential pressure reference value accumulates and exceeds the set threshold, local optimization or global recalibration is performed and the model is updated. The theoretical differential pressure correction coefficient, the compensated theoretical differential pressure reference value, the micro differential pressure sensor reading, and the target process data are paired, and the real-time correction value and deviation measurement are output through the fusion algorithm. When the error continuously exceeds the standard, the fusion weight is dynamically adjusted or the model calibration is triggered. Define a multi-scale function and perform multi-scale analysis on the corrected differential pressure and deviation. If the obtained value exceeds the preset threshold, a warning or alarm is issued and feedback is sent back to the previous step. If there is still an abnormality after maintenance, maintenance suggestions are output directionally or global recalibration is performed.

[0008] Furthermore, the following types of sensors are arranged at the key nodes of the sensor network pipeline or the gas source to collect the gas component ratio, temperature , pressure , humidity , gas flow and real-time density ; At the moment The collected data is initially packaged and its validity is verified to form the verified data packet 101, which is uniformly stored in the first-step data set 101.

[0009] Furthermore, a reference benchmark distribution is constructed ; For the component data obtained from real-time monitoring An ingredient fluctuation index is established Quantify it, and the specific form is as follows: Among them, Is the weight coefficient of the th ingredient, Is a power exponent greater than 1, Is an additional correction function; If , it is considered that the current gas component deviation is within the normal range; if , it is marked as abnormal or warning status.

[0010] Furthermore, obtain the target process data for the same timestamp from the first-step data set 101: gas components, environmental and process parameters, flow or density parameters, align the target data in time series based on the timestamp information, and summarize it into the second-step data set 201; model the influence of the gas component ratio and process conditions on the differential pressure measurement, and construct a dynamic physical property compensation function : In the formula, Represents the gas component vector at the current moment , Represents the environmental and process conditions, including And flow or density information; Is a constant or semi-empirical coefficient related to the th ingredient, Is a local characteristic function of the ingredient , temperature And pressure ; Is an additional compensation term.

[0011] Furthermore, introduce a comprehensive correction operator , map the output of the dynamic physical property compensation function to a parameter set that can be directly applied to the correction of differential pressure readings, and can be defined as: In the formula: Represents the real-time theoretical differential pressure correction coefficient; Represents the compensated theoretical differential pressure reference value, where Is the mapping relationship obtained through experimental calibration or numerical simulation; Execute the dynamic physical property compensation function on the valid records from the second-step dataset 201 to evaluate and obtain the corresponding moment of , and summarize the result with the original input data into the second-step dataset 202

[0012] Furthermore, obtain the physical property compensation parameters and correction coefficients in the second-step dataset 202. The sensor network obtains the measured data at the latest moment. For the same timestamp , pair and with the differential pressure sensor reading and the outputs of each auxiliary sensor, and store the qualified data into the third-step dataset 301 Integrate the physical property compensation parameters, correction coefficients and the readings of on-site multi-channel sensors, and output a relatively more accurate differential pressure value. Select a fusion algorithm and set at the initial stage of the algorithm: state vector: , where represents the differential pressure obtained after fusion; observation vector and noise term

[0013] Furthermore, introduce a non-linear cross correction term to measure the coupling degree between multiple sensor readings. The formula is as follows where: is an adaptive fusion coefficient within the range of [0,1], represents a non-linear cross correction function based on and ; after obtaining the fused differential pressure , compare it with the compensated theoretical differential pressure reference value : This error and the fusion coefficient will be stored in the third-step dataset 302 together at each moment

[0014] Furthermore, if it is detected that exceeds a certain threshold in multiple consecutive sampling periods, increase the prior weight of the model or the measured weight on site, and mark the adjustment timestamp and the reason for adjustment When large deviations continuously occur or the differential pressure readings seriously deviate from the process expectations, trigger the on-line calibration of the subsequent steps: and send an alarm instruction or a maintenance prompt; or, request to trace back the physical property compensation model generated in the second step to check whether recalibration is required , or check the gas composition sensor

[0015] Furthermore, retrieve the continuously recorded and real-time corrected differential pressure readings during the time period ​ And deviation measurement , perform associated retrieval on the process stage markers at the same moment , and merge them to form the fourth-step dataset 401; introduce a method that combines hierarchical cumulative measurement and short-cycle measurement, and define a multi-scale function , for the historical sampling window , conduct a comprehensive evaluation: In the formula: represents the instantaneous error measured at time , represents the coefficient on scale after wavelet transform of the error signal , is the index set, including several scales or frequency bands of interest , represents the integration window length for scale ; is the power exponent, is the aggregation exponent; is the weight coefficient for scale .

[0016] Furthermore, when designing, multiple-level thresholds are usually configured for different process sensitivity levels. If , it indicates that the error is within the normal or acceptable range; if exceeds but does not reach , a warning prompt is generated; if , it is regarded as a serious deviation, and feedback or maintenance measures must be taken immediately; When it is detected that has exceeded the warning threshold or the alarm threshold , a feedback trigger record is made in the fourth-step dataset 402, and notifications are automatically sent to the first step and the second step: Each feedback trigger or warning action is recorded in the fourth-step dataset 402, including: trigger type, trigger time, the value at the time of trigger, and what level of feedback request is sent to which link.

[0017] (III) Beneficial effects The present invention provides an intelligent gas differential pressure monitoring system based on multi-sensor fusion, having the following beneficial effects: 1. By calculating the component fluctuation index , abnormal changes in process gases (such as sudden increase or decrease in component ratio) can be detected in advance at the sensor level, thus providing a clear mark for the key attention moments in subsequent steps. When When it exceeds the threshold, it is convenient to preferentially process the data of this time period in subsequent physical property models and fusion algorithms.

[0018] 2. Dynamic physical property compensation function It takes into account various factors affecting differential pressure measurement, can realize the fusion of complex relationships in a single operator, and achieve precise guidance for subsequent differential pressure measurement.

[0019] 3. Through local tuning, fine-tuning can be performed on individual components or specific intervals (online) without terminating production, reducing the cumulative error caused by operating condition fluctuations; when the operating condition returns to normal, the model can automatically adapt to the new equilibrium point. Global recalibration ensures that the model will not fail as a whole after equipment overhaul, gas replacement, or long-term operation, providing more calibration means in multiple dimensions. The local tuning mechanism avoids the situation of overall shutdown and calibration only due to small-range deviations in traditional practices, improving the continuous operation efficiency of the production line. During major renovations or equipment maintenance, using standard gas for global calibration can solve the cumulative deviation at one time, reducing the possibility of frequent alarms and failures in the next cycle.

[0020] 4. Combining the original differential pressure readings with the prior information of the model can handle problems that are difficult to overcome by a single sensor, such as high-frequency noise, sudden interference, or zero drift. The non-linear cross-correction term can quickly respond to outliers, amplify or reduce their effects, and further reduce error accumulation.

[0021] 5. If the fusion output error exceeds the threshold at multiple consecutive moments, the weight of the prior model can be immediately increased to avoid large oscillations of the system under sudden interference; or vice versa, the trust in on-site measurements can be increased. This online adjustment avoids frequent manual intervention and helps maintain the normal production process. If the error persists at a high level, a request for model update or calibration in the second step is triggered, realizing a closed-loop cycle of measurement fusion - error detection - model correction. When the fusion weight is updated, the traceability of strategy optimization can be achieved.

[0022] 6. Using multi-scale functions can capture high-frequency perturbations and low-frequency drift characteristics simultaneously, distinguish possible equipment vibrations, frequent valve operations, or long-term accumulated problems, and through the combination with phase information or process stage markers, the stages with the highest error risk can be accurately located, so as to focus on optimizing the corresponding links.

[0023] 7. The expert rule library can be called to generate maintenance suggestions targeted at the symptoms, such as cleaning the pipeline, recalibrating the sensor, checking for seal leakage, etc. This avoids the waste of manpower or downtime losses caused by comprehensive troubleshooting in traditional practices and can better trace the effectiveness of maintenance operations. If the problem remains unresolved after multiple targeted maintenance operations, further prompts for global calibration or replacement of key equipment are provided to ensure that the system can restore accuracy at a higher dimension. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 FIG. is a schematic flow structure diagram of the intelligent gas differential pressure monitoring system based on multi-sensor fusion according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] Please refer to Figure 1 , the present invention provides an intelligent gas differential pressure monitoring system based on multi-sensor fusion, including Step 1: The sensor network arranged at the pipeline collects process status data for verification and establishes a composition fluctuation index to quantify the fluctuation degree of the gas composition. If the composition fluctuation index is abnormal, the corresponding time interval is marked; The content of the above Step 1 includes: Step 101: The following types of sensors are arranged at key pipeline nodes or gas sources to construct a sensor network, including a composition sensor for collecting the proportion of gas components ; an environmental sensor for monitoring temperature , pressure , humidity , etc.; an environmental / process condition flow or density sensor for obtaining gas flow and real-time density ; The data collected at time is initially packaged and recorded as the original data packet 101, which includes: original gas composition data environmental / process data flow / density data or timestamp ; Perform validity check on the original data packet 101: If a certain measurement value exceeds the expected range or the sensor drops offline, add a failure flag to the data packet. If all measurement values are within the available range, mark it as valid. After the check is completed, form the verified data packet 101 and uniformly store it in the first-step data set 101 for retention. During use, data noise and incorrect input can be reduced. By performing real-time checks on the range and communication status of each sensor, the probability of invalid data flowing into the backend calculation module due to factors such as sensor failure, range overflow, or disconnection is significantly reduced. Add a "failure flag" or "valid flag" to each record at the "original data packet 101" stage to prevent subsequent complex algorithms (such as dynamic physical property compensation, fusion algorithms) from being interfered by outliers. Improve real-time performance and system security. Once a sudden disconnection of the sensor or a measurement value far exceeds the expected range occurs, the system can issue a warning or temporarily exclude the relevant data immediately, reducing the chain reaction caused by abnormal input to the system. In extreme cases, to prevent safety accidents, emergency shutdown or process switching strategies can also be triggered in combination with process threshold limits.

[0027] Step 102: To evaluate the fluctuation degree of gas components, construct a reference baseline distribution, denoted as ; Under real-time monitoring, establish a component fluctuation index for the component data in each verified data packet 101, and the specific form is as follows: where, is the weight coefficient of the th component, indicating its sensitivity to the overall process or measurement accuracy, is a power exponent greater than 1, used to amplify the larger deviations of certain components, is an additional correction function, used to make additional adjustments to the overall fluctuation degree according to the changes in gas flow or density ; When the flow or density changes drastically, can be increased accordingly to reflect a more significant impact on the differential pressure monitoring system; Compare the calculated with the preset threshold . If , it is considered that the current gas component deviation is within the normal range; if , mark it as an abnormal or warning state, which can prompt that the data at this moment needs to be focused on in the subsequent steps; Generate the analyzed data packet 102 according to the above results, and attach fields such as the component fluctuation index and status identifier and store it in the first-step data set 102.

[0028] During use, the process gas component fluctuations are pre-identified, and by calculating the component fluctuation index , abnormal changes in the process gas (such as sudden increases or decreases in component ratios) can be detected in advance at the sensor level, thus providing a clear mark for the subsequent step of "key attention moment". When exceeds the threshold, it is marked as "abnormal / warning", which is convenient for preferentially processing the data of this time period in the subsequent physical property model and fusion algorithm.

[0029] It can ensure the applicability of the compensation model. If it is found that the gas component change exceeds a certain threshold at this stage, it also means that more frequent or more refined adjustment of the physical property compensation parameters is required in the follow-up to avoid the rapid amplification of measurement deviation caused by using a single calibration. Factory operators can flexibly arrange maintenance or switch the operation mode accordingly to improve the overall production efficiency.

[0030] Step 2: Align the effective target process data at the same moment in time series, and call the dynamic physical property compensation function to output the correction coefficient and differential pressure reference value; if the deviation between the measured differential pressure and the theoretical differential pressure reference value accumulates and exceeds the set threshold, perform local optimization or global recalibration and update the model; The said Step 2 includes the following contents:

[0031] Step 201: Obtain the following target process data for the same timestamp from the first-step dataset 101: gas components: , environmental and process parameters: ; flow or density parameters: or . If some data are marked as invalid due to sensor failure, they are excluded or not used temporarily at this stage to ensure the integrity and reliability of the input data.

[0032] Align the target data according to the timestamp information, summarize the qualified records into data packet 201, and save it in the form of the second-step dataset 201; During use, with consistent and complete data structures, the component data, environmental and process parameters, and flow / density information at the same moment are aligned and then input into the model, ensuring that the model input has complete indicators at each moment. After excluding invalid or missing records, the model calculation process is more targeted, reducing the interpolation error in subsequent fitting or interpolation.

[0033] Improve the accuracy of model calculation because all data are screened and paired in time series at this stage, reducing the probability of the model making meaningless corrections to abnormal data. Reducing the secondary interpolation or compensation for data missing makes the physical property model have a more realistic working condition mapping relationship.

[0034] Step 202: To obtain the theoretical differential pressure correction coefficient or physical property compensation parameter, first define a dynamic physical property compensation function , and model the influence of gas composition ratio and process conditions on micro differential pressure measurement. Let: In the formula, represents the gas composition vector at the current moment , represents the environment and process conditions, such as and flow rate or density information; is a constant or semi-empirical coefficient related to the th component, is the local characteristic function about the component , temperature and pressure , which is used to reflect the influence of different proportions of components on the differential pressure measurement accuracy under specific temperature and pressure conditions; is an additional compensation term, which combines humidity and flow rate or density and other factors to further dynamically optimize the overall compensation value; Introduce a comprehensive correction operator , and map the output of the dynamic physical property compensation function to a parameter set that can be directly applied to the correction of micro differential pressure readings. It can be defined as: In the formula: represents the real-time theoretical differential pressure correction coefficient, which is related to gas composition and process conditions; represents the compensated theoretical differential pressure reference value, which can be used to compare with the actual differential pressure reading. Among them, is the mapping relationship obtained through experimental calibration or numerical simulation (such as CFD simulation), which converts the output of the relatively abstract physical property function into numerical parameters that can be directly called by the next-step algorithm; Evaluate the dynamic physical property compensation function for the valid records from the second-step dataset 201, and obtain the at the corresponding moment , and summarize the results with the original input data as the second-step dataset 202. Denote the output structured data object as the second-step dataset 202, which includes the time stamp , gas composition , environment and process parameters , correction coefficient and differential pressure reference value ; When in use, the dynamic physical property compensation function Multiple factors affecting differential pressure measurement (such as component ratio, temperature, pressure, humidity, flow / density, etc.) are considered. It can fuse complex relationships in a single operator, enabling the system to obtain two key outputs, namely the "theoretical differential pressure correction coefficient" and the "theoretical differential pressure reference value", based on the current gas composition and environmental conditions, thereby achieving precise guidance for subsequent differential pressure measurement. It is convenient for online correction in the control system or edge computing node, reducing the burden of writing complex logic on-site. If new gas components are discovered in experiments or simulations and the model needs to be supplemented or corrected, only the function Ψ or its mapping operator Ω needs to be updated to automatically adapt to the new scenario.

[0035] Step 203: If the cumulative deviation between the measured differential pressure and the theoretical differential pressure reference value exceeds the set threshold, it indicates that the current physical property compensation function or its mapping operator may be inaccurate and needs to be updated or calibrated online. When it is detected that the deviation exceeds the normal range, small-scale tuning of the parameters of 、 or can be performed, or a more substantial recalibration can be carried out. The common practices typically include: Local tuning: Only within the intervals where large-scale offsets occur in certain gas components, the corresponding or ; parameters are slightly adjusted. Global recalibration: During equipment maintenance or downtime, standard gases or reference environmental conditions are introduced to recalibrate the entire function structure; The updated function or parameters are written into the second-step dataset 203, and at the same time, meta-information such as the timestamp of the update and the reason for the correction is retained. In this way, in the third step, when calling the corresponding parameters, the latest version of the physical property compensation model can be automatically switched; For local tuning and global recalibration, further details are provided below; I. Local tuning

[0036] Trigger condition: When the system detects that within a specific gas component change interval or specific process conditions, the error, recorded as shows obvious periodic high values or is particularly prominent within certain component ratio ranges, and the error magnitude exceeds the predefined local tuning threshold , the local tuning process is triggered.

[0037] Objective: Only make small adjustments to the main parameters causing the deviation Idea: Locate the most relevant parameters for partial update while ensuring that the overall influence of the remaining environment or components remains undamaged; For the gas components recorded in the second-step dataset 202 and the error Perform segmental statistics to find component indicators that are highly correlated with errors When a component is found When the error increases rapidly within a specific range, or Considered as the main tuning target.

[0038] Introduce a gradient update formula for a single component or a small number of components, such as: in: is the cumulative or weighted error index within a certain time window (such as or a comprehensive error function based on process weighting); To locally tune the learning rate, the value is usually small to avoid large fluctuations; Characterization pair The sensitivity of can be obtained through historical data or local linear approximation; if the main influence comes from Internal function, which can be Piecewise linearization, making small translations or slope adjustments at key nodes; The corrected parameters Or a new function form The local update record stored in the second step data set 203 includes: the tuning timestamp, the corresponding component or operating range, the corrected parameter value, the correction amplitude and the reason. When subsequently calculating the new compensation coefficient and differential pressure reference value, the latest local tuning result can be directly referenced. 2. Global Recalibration If deviations increase across different gas composition ranges or operating conditions and cannot be brought to an acceptable level through local tuning, or if overall failure of the parameter function is detected, global recalibration is required. Common triggering scenarios include: replacement or overhaul of key equipment resulting in significant changes in overall pipeline characteristics; Introducing new gas categories on site, Routine overall calibration is performed when new components appear in the system for more than a predetermined period of time or accumulated operating hours.

[0039] Objective: To refit the main structure of the physical property compensation model using standard gases or benchmark environmental conditions to obtain the optimal global parameter set.

[0040] Idea: When the system is shut down or under maintenance, measure multiple sets of calibration points, solve or iterate 、 and The overall form and coefficients of .

[0041] During maintenance or designated downtime, introduce a controlled flow of standard gas combinations, such as , and cooperate with adjustable environmental conditions , to form several calibrated operating points. At each operating point, record the actual differential pressure and the compensated theoretical differential pressure calculated by the system , and save them to a dedicated calibration dataset; For multiple operating points collected , and the corresponding , substitute them into the model: Use iterative numerical methods (such as quasi-Newton method, trust region method, etc.) to solve the optimal parameters: where represents the global error objective function, which can be designed in combination with weighted process requirements (for example: giving priority to ensuring the accuracy when a certain component content is high), to ensure that the fitting result has global optimality; After calculating the new parameters and function form, package them into a new version of the physical property compensation model and write it into the recalibration record area in the second-step dataset 203, mark the recalibration time, the number of operating points participating in the calibration, and the convergence accuracy during the iteration process, etc. After resuming operation, this new version of the model will be called when calculating the differential pressure correction in real time, so as to restore or improve the measurement accuracy within a wider range of component and process conditions.

[0042] During use, through "local optimization", fine-tuning can be performed on individual components or specific intervals (online) without stopping production, reducing the accumulated error caused by operating condition fluctuations; when the operating condition returns to normal, the model can automatically adapt to the new equilibrium point. "Global recalibration" ensures that the model will not fail as a whole after major repairs, gas replacement, or long-term operation, providing more calibration means. The local optimization mechanism avoids the situation of "shutting down the whole line for calibration just because of small-range deviations" in traditional practices, improving the continuous operation efficiency of the production line. During major renovations or equipment maintenance, using standard gas for global calibration can solve the cumulative deviation at one time, reducing the possibility of frequent alarms and failures in the next cycle.

[0043] Step 3: Pair the theoretical differential pressure correction coefficient, the compensated theoretical differential pressure reference value, the micro differential pressure sensor reading, and the target process data, and output the real-time correction value and deviation measurement through the fusion algorithm; when the error continuously exceeds the standard, dynamically adjust the fusion weight or trigger model calibration.

[0044] The said step 3 includes the following contents: Step 301: Obtain the physical property compensation parameters and correction coefficients in the second-step dataset 202, that is where: represents the theoretically calculated differential pressure correction coefficient obtained dynamically; Denote the theoretical differential pressure reference value after compensation; obtain the measured data at the latest moment from the sensor network constructed by the differential pressure sensor and other auxiliary sensors (temperature, pressure, humidity, flow rate, density, etc.), and record it as the original sensor data packet 301; For the same timestamp t, pair and with the micro differential pressure sensor reading and the outputs of each auxiliary sensor. If the data of some sensors is missing or abnormal at this moment, make a failure mark in the third-step data set 301 to prevent calculation errors during subsequent fusion operations; store the data that passes the verification in the third-step data set 301 to provide the basic input for the next fusion and correction algorithms; During use, when reading the second-step data sets 202 and 203, information such as their versions or update times will be loaded simultaneously to ensure that the physical property model used in the third-step fusion can be located, which helps to quickly trace back in case of anomalies. The on-site original sensor data and the model output parameters are aligned in time series, supporting real-time fusion operations without additional temporary retrieval or interpolation. If the measured values of some sensors are missing or abnormal, they are marked as failed at this point to prevent the third-step fusion algorithm from producing invalid or extreme results due to bad inputs and ensure the stability of the fusion process.

[0045] Step 302: Combine the physical property compensation parameters, correction coefficients with the readings of multiple on-site sensors, and output a relatively more accurate micro differential pressure value. Select a fusion algorithm with certain non-linear processing capabilities, such as the extended Kalman filter or the adaptive particle filter, and set at the initial stage of the algorithm: State vector: , where represents the micro differential pressure obtained after fusion; Observation vector: Combine and and other compensation information; Noise term: The appropriate process noise and measurement noise distributions can be defined according to the analysis of gas components and environmental parameters in the second step; During the fusion process, to avoid the overly simple operation of ordinary variance or standard deviation, introduce a non-linear cross correction term to measure the coupling degree between multiple sensor readings, thereby dynamically adjusting the fusion weight. The formula is as follows where: is an adaptive fusion coefficient within the range of [0,1], used to balance the original sensor readings and the theoretical differential pressure reference, represents the non-linear cross correction function based on and , for example: In the formula: and is a parameter to be calibrated or updated online, controlling the intensity of the correction term and the sensitivity to the original differential pressure reading. When it is greater than 1, it can amplify large deviations and achieve rapid correction of outliers or sudden changes. Through the above fusion operator, the system can retain the prior information of the theoretical model when the gas composition fluctuates greatly. and also take into account the dynamic characteristics of on-site real-time measurement; of the dynamic characteristics; After obtaining the fused differential pressure , to evaluate the measurement accuracy, it is necessary to compare with the compensated theoretical differential pressure reference value : This error and the fusion coefficient will be stored in the third-step dataset 302 at each moment, which is convenient for subsequent steps (such as alarm judgment or performance evaluation); When in use, combining the original differential pressure reading with the prior information of the model can cope with problems that are difficult to overcome by a single sensor, such as high-frequency noise, sudden interference, or zero drift. The non-linear cross correction term can "quickly respond" to outliers, amplify or reduce their effects, and further reduce error accumulation.

[0046] After each fusion is completed, it can be calculated and stored in the third-step dataset 302, which provides basic data for the subsequent fourth-step trend analysis. By recording the fusion coefficient, it can be judged in later diagnosis when the system relies more on the prior model and when it relies more on on-site sensors, which is convenient for optimizing the fusion strategy.

[0047] Step 303, if it is detected in multiple consecutive sampling periods that exceeds a certain threshold , it means that the existing fusion strategy is insufficiently sensitive or excessive to anomalies or mutations, and the and and other weight coefficients can be adjusted in real time, appropriately increasing the prior weight of the model or the on-site measurement weight, writing the adjusted parameters into the online fusion weight update record in the third-step dataset 303, and marking the adjustment timestamp and adjustment reason; when large deviations continuously occur or the differential pressure reading seriously does not match the process expectation, trigger the online calibration of the subsequent steps: and send an alarm instruction or a maintenance prompt; or, request to trace back the physical property compensation model generated in the second step to check whether recalibration is required , or check the gas composition sensor; When in use, if the fusion output error is detected If the threshold is exceeded multiple times in a row, the weighting of the a priori model can be immediately increased to prevent large system oscillations under sudden disturbances; conversely, to enhance confidence in field measurements. This online adjustment avoids frequent manual intervention and helps maintain normal production processes. If the error remains excessively high, a model update or calibration request is triggered in the second step, completing a closed-loop cycle of measurement fusion, error detection, and model correction. When the fusion weights are updated and the retest results significantly improve, the effectiveness of this action is recorded in the third step, dataset 303, ensuring traceability of strategy optimization.

[0048] Step 4: Define the multi-scale function Then, multi-scale analysis is performed on the corrected differential pressure and deviation. If the value exceeds the preset threshold, a warning or alarm is issued and feedback is given to the previous step; if the abnormality still exists after maintenance, a targeted maintenance suggestion is output or a global recalibration is performed; The step 4 includes the following contents: Step 401: retrieve the data from the third step data set 302 in the time period Real-time corrected micro differential pressure readings continuously recorded and deviation measures , for the same moment The process stage tags (such as batch number or process stage) are associated and searched, and merged to form the fourth step data set 401; In order to more flexibly grasp the evolution characteristics of the error in different time domains, we can introduce a combination of hierarchical cumulative metrics and short-term metrics to define a multi-scale function , for the historical sampling window , conduct a comprehensive assessment: Where: Indicates time The instantaneous error measured at any moment (for example: the micro differential pressure of the system after fusion and theoretical benchmarks ), Indicates the error signal After wavelet transform, the scale (or frequency band) The coefficients on different wavelet scales Corresponding to different time resolutions or frequency resolutions, it can characterize the changing characteristics of errors at multiple levels, such as short-period disturbances, medium-period dynamics, and long-period trends; is an index set containing several scales or frequency bands of interest , different numbers of scales can be set according to application requirements, covering from high frequency to low frequency or from short period to long period; Indicates the scale The integration window length; it can usually match the time support length corresponding to the wavelet, or can be set according to the attention to the signal characteristics at this scale. Integrate the wavelet coefficients within the time interval to obtain the cumulative amount of the error component at this scale within this period of time; (greater than 1) is the power exponent, which is used to amplify the sensitivity to larger errors or sudden anomalies. If takes a larger value, the amplification effect on the error peak is more obvious. (greater than 1) is the aggregation exponent, which is used to perform further non-linear scaling on the integration result; in cooperation with , multi-order cumulative amounts can be formed to perform hierarchical penalties on errors of different magnitudes; is the weight coefficient of scale . The proportion of different scales in the final sum can be determined according to process requirements or the emphasis on different time resolutions; By analyzing the historical curve of , the short-term fluctuation characteristics and long-term cumulative trends of errors can be found. The overall change trend of is extracted by using the adaptive curve fitting or piecewise interpolation method, and the result is recorded in the fourth-step dataset 401, including: The corresponding time range , the maximum value, minimum value and their corresponding moments of , and the initially determined long-term upward or downward trend markers; When in use, the multi-scale function can capture high-frequency disturbances and low-frequency drift characteristics at the same time, and distinguish possible equipment vibrations, frequent valve actions or long-term accumulation problems. Save the maximum and minimum values, trend markers and other information of in the fourth-step dataset 401 to help the factory management quickly grasp the system operation health. When dealing with different production batches, different seasons or different operation teams, combined with the "phase information" or "process phase marker" and

[0049] Step 402. During design, corresponding multi-level thresholds are usually configured for different process sensitivities, such as , , etc.; If , it means that the error is within the normal or acceptable range; If exceeds but does not reach , a warning prompt is generated; If , considered a serious deviation, and feedback or maintenance measures must be taken immediately; When detected Exceeded warning threshold or alarm threshold , a feedback trigger record is made in the fourth step data set 402, and the following notifications are automatically sent to the first and second steps: Inspection of monitoring equipment and sampling links: Prompts to check for possible leaks, blockages, sensor aging, etc. in the gas composition analyzer or sampling circuit; Dynamic gas property model calibration: If the gap between the actual differential pressure and the theoretical reference value accumulates over a long period of time in the third step, the model may need local tuning or global recalibration; At this point, the first and second steps will perform corresponding maintenance processes or model updates based on the feedback information; after completion, the fourth step can obtain a new compensation model version number or sensor calibration result again.

[0050] Each feedback trigger or warning action is recorded in the fourth step data set 402, including: trigger type (warning / alarm), trigger time, trigger time, value, and to which link and what level of feedback request is sent; Automatic feedback is achieved with the first step (hardware and sensor level) and the second step (physical property compensation model level) to ensure that certain large deviations will not be ignored forever. They can also "self-heal" over a long period of time, reducing manual operation and maintenance costs.

[0051] Step 403: Generate a health monitoring report from the long-term archived data, including If the report shows significant deviations in certain time periods and a high correlation with specific process stages, targeted maintenance recommendations can be given, such as cleaning pipelines, calibrating sensors, checking flow control valves, etc., and the fourth step data set 403 can be updated. After maintenance and adjustment, run again for a period of time. The adjustment is recorded as an effective improvement case, which can provide sample data for subsequent abnormal diagnosis. If there is no improvement after multiple maintenance and local tuning, a global recalibration is performed during the shutdown or overhaul to obtain a new version of the physical property compensation model. It is marked and written into the second step data set 203 to facilitate the third step to automatically switch to the new version of the model to perform measurement correction. After completing the maintenance warning and model upgrade, the record in the fourth step data set 403 is marked as processed, and the key conclusions (such as successful correction or still needing attention) are pushed to the management end.

[0052] When the deviation in a certain period is significantly correlated with a specific phase (such as Phase B), the system can call the expert rule base to generate maintenance suggestions targeted at the symptoms, such as cleaning the pipeline, recalibrating the sensor, checking for seal leaks, etc. This avoids the waste of manpower or downtime losses caused by the "comprehensive investigation" in traditional practices and can better trace the effectiveness of maintenance operations. If the problem remains unresolved after multiple targeted maintenance, it further prompts for global calibration or replacement of key equipment to ensure that the system can restore accuracy at a higher dimension. The results of each maintenance and the improvement of MEF(t) are marked in the "Fourth-step dataset 403", forming a long-term data precipitation for continuous optimization to facilitate the in-depth evolution of the intelligent factory.

[0053] I. Trigger mechanism based on deviation pattern and process phase Divide the production / operation process into stages (e.g., Phase A, Phase B, Phase C, etc.), and at each moment Store the information of the current process phase in the Fourth-step dataset 401 or Fourth-step dataset 402, which can be accompanied by the phase label and deviation information at each moment for subsequent targeted correlation analysis in combination with phase-error; Perform local aggregation or statistics of the multi-scale function (or other measurement functions) within different process phase intervals to determine which phases have significant deviations or abnormal jumps. Once the curve of a certain phase (such as Phase B) continuously exceeds the warning line or alarm line, it is marked as a stage anomaly and corresponding records are made in the Fourth-step dataset 401 / 402.

[0054] II. Expert rule base and fault / deviation pattern recognition Collect or summarize common fault / deviation patterns and their possible corresponding causes in advance, for example: Pattern 1: High-frequency deviations gather in Phase A → Possible causes: Frequent valve switching, loose sensor diaphragm. Pattern 2: Low-frequency deviations continuously accumulate in Phase B → Possible causes: Blockage or condensation in the sampling pipeline. Pattern 3: Large deviations throughout the period → Possible causes: Sensor failure, air leakage, serious offset of the flowmeter, etc. For each abnormal pattern, a series of targeted maintenance suggestion templates are given, such as checking or cleaning a certain section of the pipeline, recalibrating a certain type of sensor, correcting the connection of the flow control valve, maintaining or replacing a certain component, etc. These maintenance suggestions have a certain priority or hierarchical processing scheme (e.g., first check simple components, and if the problem persists, conduct a larger-scale overhaul).

[0055] When an anomaly is detected within a certain period and has been located to the corresponding process phase, the most likely deviation pattern can be judged through rule matching or feature-based machine learning classification; For example, if the main manifestation in the Phase B interval is the accumulation of low-frequency deviation and the error curve shows a monotonic increase, it is matched to Pattern 2, and a preliminary suggestion of sampling pipeline blockage or condensation is output; Once a deviation pattern is identified, the system automatically generates a targeted maintenance suggestion entry, which is stored together with the timestamp and phase information in the fourth-step dataset 403 to form a traceable record of faults / deviations and maintenance suggestions; When a significant deviation related to a specific process phase is detected, corresponding maintenance suggestions are automatically output. The more specific content is as follows: Obtain the process phase information Phase(t) from the fourth-step dataset 401 or the fourth-step dataset 402, as well as its aggregated values, trend markers, etc. in each time period (or frequency band). According to the pre-set expert rule base, compare the deviation pattern (high-frequency / low-frequency, sudden / cumulative, etc.) of the current phase with the known pattern items to obtain the most matching pattern ID; if Pattern 2 is matched: for example, low-frequency deviation accumulates in Phase B - the system determines that there may be problems with sampling pipe blockage or condensation.

[0056] According to the pattern ID, the system automatically calls the corresponding maintenance suggestion template. For example, Suggestion 1: Check and clean the specified pipeline; Suggestion 2: Re-calibrate or replace the relevant sensors; Suggestion 3: Observe the valve action of the flow control valve, etc. Write the generated suggestions into the maintenance suggestion record, including: timestamp tgeneratei, the matched pattern ID; the text content of the suggestion (such as please dredge and dry Sampling Pipe No. X during the shutdown period of Phase B), priority label or execution deadline (such as high priority or to be completed within 72 hours). Append this record to the fourth-step dataset 403 in a structured manner to form the current targeted maintenance suggestion entry.

[0057] Through the host computer HMI or the remote maintenance platform, push the current targeted maintenance suggestion to the operator or manager. After the maintenance operation is completed or confirmed, the system operator fills in the execution time and execution result on the interface, and updates the fourth-step dataset 403 again. After maintenance, if subsequent or other error indicators are significantly improved in the same process phase, the system marks that this maintenance operation has effectively solved the problem; otherwise, it may prompt that deeper maintenance or further analysis is required.

[0058] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0059] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0060] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only for some logical function divisions. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be electrical, mechanical, or other forms.

[0061] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0062] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. An intelligent gas differential pressure monitoring system based on multi-sensor fusion, characterized in that: including Collect process status data by a sensor network arranged at the pipeline, verify it, and establish a component fluctuation index Quantify the fluctuation degree of the gas components. If the component fluctuation index is abnormal, mark the corresponding time interval; perform time series alignment on the valid target process data at the same moment, call the dynamic physical property compensation function to output the correction coefficient and differential pressure reference value; if the cumulative deviation between the measured differential pressure and the theoretical differential pressure reference value exceeds the set threshold, perform local optimization or global recalibration and update the model; pair the theoretical differential pressure correction coefficient, the compensated theoretical differential pressure reference value, the micro differential pressure sensor reading, and the target process data, and output the real-time correction value and deviation metric through the fusion algorithm; when the error continues to exceed the standard, dynamically adjust the fusion weight or trigger model calibration; Define the multi-scale function After that, perform multi-scale analysis on the corrected differential pressure and deviation. If the obtained value exceeds the preset threshold, issue a warning or alarm and feedback to the previous step; if there are still abnormalities after maintenance, direct output maintenance suggestions or perform global recalibration.

2. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 1, wherein: Collect the gas component ratio, temperature , pressure , humidity , gas flow and real-time density ; Perform preliminary packaging and validity verification on the data collected at time to form the verified data packet 101, and uniformly store it in the first-step data set 101.

3. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 2, wherein: Constructing a reference benchmark distribution ; Composition data obtained for real-time monitoring Building a component volatility index Quantify, the specific form is as follows: in, For the The weight coefficient of each component, is a power exponent greater than 1, is an additional correction function; if , then it is considered that the current gas composition deviation is in the normal range; if , it is marked as abnormal or warning state.

4. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 3, wherein: Obtain the target process data for the same timestamp from the first-step dataset 101 : gas composition, environmental and process parameters, flow or density parameters. Align the target data in time series according to the timestamp information, and summarize it into the second-step dataset 201; model the influence of the gas composition ratio and process conditions on the differential pressure measurement, and construct a dynamic physical property compensation function : In the formula,[[]]END]] represents the gas composition vector at the current moment , represents the environmental and process conditions, including and flow or density information; is a constant or semi-empirical coefficient related to the th component, is a local characteristic function about the component , temperature and pressure ; is an additional compensation term.

5. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 4, wherein: Introduce a comprehensive correction operator , map the output of the dynamic physical property compensation function to a parameter set that can be directly applied to the correction of differential pressure readings, and it can be defined as: In the formula: represents the real-time theoretical differential pressure correction coefficient; represents the theoretical differential pressure reference value after compensation, where is the mapping relationship obtained through experimental calibration or numerical simulation; Execute the dynamic physical property compensation function on the valid records from the second-step dataset 201 Evaluate to obtain the corresponding to the time, and summarize the result with the original input data into the second-step dataset 202 6. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 5, wherein: Obtain the physical property compensation parameters and correction coefficients in the second-step dataset 202. The sensor network obtains the measured data at the latest moment and targets the same timestamp , pair and with the readings of the differential pressure sensor and the outputs of each auxiliary sensor, and store the qualified data in the third-step dataset 301; Integrate the physical property compensation parameters, correction factors and on-site multi-channel sensor readings to output a relatively more accurate differential pressure value. Select a fusion algorithm and set at the initial stage of the algorithm: State vector: , where represents the differential pressure obtained after fusion; Observation vector and noise term.

7. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 6, wherein: A non-linear cross correction term is introduced to measure the coupling degree between multiple sensor readings, and the formula is as follows Where: is an adaptive fusion coefficient within the range of [0, 1], represents a non-linear cross correction function based on and ; after obtaining the fused differential pressure it is compared with the compensated theoretical differential pressure reference value : This error and the fusion coefficient will be stored in the third-step dataset 302 together at each moment.

8. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 7, wherein: If it is detected within multiple consecutive sampling periods that exceeds a certain threshold , increase the prior weight of the model or the on-site measured weight, and mark the adjustment timestamp and the reason for adjustment; When large deviations continuously occur or the differential pressure reading seriously deviates from the process expectation, trigger the on-line calibration of subsequent steps: send an alarm instruction or a maintenance prompt; or, request to recall the physical property compensation model generated in the second step to check whether recalibration is required 、 or check the gas composition sensor.

9. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 8, wherein: Retrieve the real-time corrected differential pressure readings continuously recorded during the time period and the deviation metric , perform an associative search on the process stage markers at the same moment , and merge them to form the fourth dataset 401; introduce a method that combines hierarchical cumulative metrics and short-cycle metrics to define a multi-scale function , and conduct a comprehensive evaluation for the historical sampling window : In the formula: where: represents the instantaneous error measured at time , represents the coefficient at scale after wavelet transform of the error signal , is the index set, containing several scales or frequency bands of interest , represents the integration window length for scale ; is the power exponent, is the aggregation exponent; is the weight coefficient for scale .

10. The intelligent gas micro differential pressure monitoring system based on multi-sensor fusion according to claim 9, wherein: During design, multiple thresholds are usually configured for different process sensitivities. If , it indicates that the error is within the normal or acceptable range; if exceeds but does not reach , a warning prompt is generated; if , it is regarded as a serious deviation and feedback or maintenance measures must be taken immediately; When it is detected that has exceeded the warning threshold or the alarm threshold , a feedback trigger record is made in the fourth-step dataset 402, and notifications are automatically sent to the first step and the second step: Each feedback trigger or warning action is documented in the fourth-step dataset 402, including: trigger type, trigger time, value at the time of trigger, and what level of feedback request is sent to which link. Value, and what level of feedback request is sent to which link.

Images