Industrial internet of things monitoring system for inferior oil device constant residue heat exchange process

Abstract

The industrial internet-of-things monitoring system for the inferior oil device constant residue heat exchange process belongs to the technical field of industrial process monitoring and fault prediction, and comprises a distributed multi-modal sensing unit, an edge computing and data preprocessing unit, a critical state prediction unit and a self-adaptive sensor migration control unit; the distributed multi-modal sensing unit is used for collecting original signals of the heat exchange process pipeline; the edge computing and data preprocessing unit is used for processing the original signals to extract key physical characteristic parameters; the critical state prediction unit is used for calculating a phase space collapse risk index based on the key physical characteristic parameters, and performing discriminant processing on the phase space collapse risk index to generate a coking early warning signal or an activation instruction; the present application provides a state cognition for the system which is much more profound and comprehensive than conventional monitoring, so that it can capture subtle changes driven by multi-field coupling in the coking budding stage.

Description

Technical Field

[0001] This invention relates to the field of industrial process monitoring and fault prediction, specifically to an industrial Internet of Things (IoT) monitoring system for the atmospheric residue heat exchange process of a substandard oil processing plant. Background Technology

[0002] In the processing of low-quality heavy oil in petrochemical industries, traditional safety monitoring methods mainly rely on fixed sensor networks deployed on heat exchange equipment. These networks typically only monitor conventional macroscopic process parameters such as temperature and pressure, and are severely inadequate in addressing dynamic and complex coking problems. Because low-quality heavy oil is a highly complex fluid with multiple phases and components, the emergence and development of coking risks exhibit strong localization and dynamic migration characteristics, resulting in the constantly changing location of risk sources, i.e., monitoring blind spots.

[0003] This situation has led to the following core problems:

[0004] The high false negative rate means that fixed sensors cannot track dynamically migrating risk sources. When signs of coking appear in the blind zone between two sensors, the system cannot capture the key signal, resulting in missed reports and false negative warnings, which pose a huge threat to production safety.

[0005] Selective blindness occurs because traditional methods rely on apparent parameters such as temperature and pressure, which are insensitive to deep physicochemical changes driven by multi-field coupling in the early stages of coking, such as micelle aggregation and micro-oscillations. This leaves the monitoring system selectively blind before coking actually occurs, resulting in the loss of valuable early warning time.

[0006] Lacking foresight, the existing static monitoring model is a passive and lagging response mechanism. It cannot predict the future development direction and location of risks based on the current state, and cannot provide decision support for preventive maintenance and proactive control.

[0007] The root cause of these problems lies in the limitations of monitoring strategies and sensing technologies. On the one hand, static monitoring architectures cannot match dynamic fault processes; on the other hand, single, macroscopic sensing dimensions cannot reveal the deep physical precursors behind complex processes. The ultimate result is that in the face of sudden coking faults, managers cannot obtain timely and accurate early warnings, severely impacting the safety and economic efficiency of the deep processing of substandard oil.

[0008] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0009] The purpose of this invention is to provide an industrial Internet of Things (IoT) monitoring system for the atmospheric residue heat exchange process of a substandard oil plant, in order to solve the problems mentioned in the background art.

[0010] The technical solution of the present invention includes a distributed multimodal sensing unit, an edge computing and data preprocessing unit, a critical state prediction unit, and an adaptive sensor migration control unit.

[0011] Distributed multimodal sensing units are used to acquire raw signals from heat exchange process pipelines;

[0012] Edge computing and data preprocessing unit is used to process the raw signal to extract key physical feature parameters;

[0013] The critical state prediction unit is used to calculate the phase space collapse risk index based on key physical characteristic parameters, and to perform discrimination processing on the phase space collapse risk index in order to generate a coking early warning signal or activation command.

[0014] An adaptive sensor migration control unit is used to determine the optimal sensor position at the next moment in response to an activation command, and to generate migration control commands based on the optimal sensor position.

[0015] Preferably, the process of extracting key physical feature parameters by the edge computing and data preprocessing unit includes:

[0016] The high-frequency dynamic pressure time series data is analyzed to calculate the Marangoni number oscillation amplitude;

[0017] The power spectral density of the micro-vibration signal and the infrasound signal was analyzed to calculate the Lyapunov exponent of the focal precursor generation rate.

[0018] The echo signal of the ultrasonic transducer array was inverted and calculated to obtain the acoustic impedance gradient of the coking layer.

[0019] The nanoscale electrochemical probe data were processed to calculate the standard deviation of the Debye length fluctuation.

[0020] Preferably, the critical state prediction unit is calculated using a non-Markov model, which combines a memory kernel function that describes the weight decay of historical state influences, and the oscillation amplitudes of the Lyapunov exponent and Marangoni number output by the edge computing and data preprocessing unit.

[0021] Preferably, the non-Markov model includes a denominator term that characterizes the chaotic synchronization-desynchronization critical transition between the Lyapunov exponent and the Marangoni number oscillation amplitude.

[0022] When the Lyapunov exponent and the Marangoni number oscillate in a near-synchronous manner, the denominator approaches a minimum, causing the risk of phase space collapse to increase dramatically.

[0023] Preferably, the non-Markov model also incorporates a Heaviside step trigger function to perform logical judgments;

[0024] When the standard deviation of the Debye length fluctuation is within the preset critical aggregation range, and the amplitude of the third harmonic component of the coking layer acoustic impedance gradient exceeds the preset acoustic harmonic threshold, a trigger signal with a value of 1 is generated; otherwise, a trigger signal with a value of 0 is generated.

[0025] Preferably, the critical state prediction unit performs the following process to discriminate the phase space collapse risk index:

[0026] The phase space collapse risk index is compared and analyzed with the preset risk threshold.

[0027] When the phase space collapse risk index exceeds the risk threshold, an activation command and the highest level coking warning signal are generated.

[0028] No activation command is generated when the phase space collapse risk index does not exceed the risk threshold.

[0029] Preferably, the adaptive sensor migration control unit determines the optimal sensor position at the next moment, and the process includes:

[0030] Obtain the phase space collapse risk index calculated by each sensor node;

[0031] The risk index of each sensor node and its corresponding fixed spatial location vector are weighted and averaged to generate the spatial distribution centroid of the risk field.

[0032] Preferably, the process of determining the optimal sensor position at the next moment further includes:

[0033] Based on the risk index values ​​of neighboring sensor nodes around the centroid of spatial distribution, the local gradient of the risk field is calculated using a numerical approximation method.

[0034] Preferably, the process of determining the optimal sensor position at the next moment further includes:

[0035] By combining the spatially distributed centroid and the local gradient of the risk field, the optimal sensor position for the next moment is generated through extrapolation calculation.

[0036] Based on the optimal sensor position at the next moment, a migration control command is generated.

[0037] This invention provides an improved industrial IoT monitoring system for the atmospheric residue heat exchange process in a low-quality oil processing plant. Compared with existing technologies, it has the following improvements and advantages:

[0038] 1. The system's distributed multimodal sensing unit can comprehensively acquire multi-dimensional physical signals directly related to the coking process, surpassing traditional methods that rely solely on apparent process parameters. The edge computing and data preprocessing unit efficiently transforms these high-dimensional, heterogeneous raw signals into key physical characteristic parameters with clear physical meaning, including the Marangoni number oscillation amplitude, the Lyapunov exponent of the coking precursor formation rate, the acoustic impedance gradient of the coking layer, and the standard deviation of the Debye length fluctuation. This multimodal feature fusion based on deep physical precursors provides the system with a far deeper and more comprehensive state understanding than conventional monitoring, enabling it to capture subtle changes driven by multi-field coupling in the coking budding stage.

[0039] 2. By introducing a memory kernel function that describes the weight decay of historical state influences, the cumulative effect of the coking process can be quantified. This model can accurately characterize the chaotic synchronization-desynchronization critical transition between the Lyapunov exponent and the Marangoni number oscillation amplitude, enabling the system to issue early warnings by observing the critical behavior of deep physical mechanisms even when the normal signal appears stable. At the same time, the model combines the Heaviside step trigger function, which only activates the complex risk integral operation when the system enters the critical pre-aggregation state defined by the Debye length fluctuation and acoustic harmonics, significantly improving the computational economy of real-time monitoring.

[0040] 3. The system's adaptive sensor migration control unit, upon receiving the activation command generated by the critical state prediction unit, can achieve closed-loop adaptive tracking of risk sources. By calculating the weighted average of the risk indices of each sensor node, a robust risk field spatial distribution centroid is generated, effectively filtering out single-point noise. Furthermore, the local gradient of the risk field is calculated using a numerical approximation method, giving the system predictability in predicting the direction of risk migration. Combining the centroid and gradient, extrapolation calculations are performed to generate the optimal sensor position for the next moment. This centroid positioning-gradient prediction-extrapolation control mechanism ensures that sensor resources are always accurately deployed in the area where the risk is most likely to occur, solving the problem of missed detection caused by the dynamic migration of monitoring blind spots. Attached Figure Description

[0041] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0042] Figure 1 This is a flowchart of an industrial Internet of Things (IoT) monitoring system for the atmospheric residue heat exchange process of a substandard oil processing plant, according to the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0044] Example 1

[0045] Please see Figure 1 The present invention provides an industrial Internet of Things monitoring system for the atmospheric residue heat exchange process of a low-quality oil plant, comprising a distributed multimodal sensing unit, an edge computing and data preprocessing unit, a critical state prediction unit, and an adaptive sensor migration control unit.

[0046] Distributed multimodal sensing units are used to acquire raw signals from heat exchange process pipelines;

[0047] Edge computing and data preprocessing unit is used to process the raw signal to extract key physical feature parameters;

[0048] The critical state prediction unit is used to calculate the phase space collapse risk index based on key physical characteristic parameters, and to perform discrimination processing on the phase space collapse risk index in order to generate a coking early warning signal or activation command.

[0049] An adaptive sensor migration control unit is used to determine the optimal sensor position at the next moment in response to an activation command, and to generate migration control commands based on the optimal sensor position.

[0050] This invention discloses an industrial Internet of Things (IoT) monitoring system for the atmospheric residue heat exchange process in a low-quality oil processing plant. The system aims to solve the problem of high false negative rates in coking early warning caused by the dynamic migration of monitoring blind spots when existing fixed sensor networks monitor complex fluids such as low-quality heavy oil, which are multiphase and multicomponent. The system is constructed as a closed-loop perception-prediction-control architecture, which achieves accurate monitoring of the coking process by actively tracking fault precursors.

[0051] In this embodiment, the monitoring system includes four core units: a distributed multimodal sensing unit, an edge computing and data preprocessing unit, a critical state prediction unit, and an adaptive sensor migration control unit.

[0052] The distributed multimodal sensing unit aims to comprehensively and systematically collect various physical signals within the heat exchange process pipeline, providing a raw data foundation for subsequent accurate predictions. Deployed in key areas of the heat exchange process pipeline, the unit comprises not only conventional temperature and pressure sensors, but more importantly, a series of unconventional sensors to capture physicochemical events directly related to the coking process. These sensors include: a nanoscale electrochemical probe for in-situ measurement of Debye length fluctuations in the fluid double layer near the inner wall of the pipeline; an ultrasonic transducer array for emitting and receiving ultrasonic waves, with the data used to invert and calculate the acoustic impedance gradient of the coking layer; a high-frequency dynamic pressure sensor for capturing minute pressure oscillations caused by the Marangoni effect; and a micro-vibration sensor for monitoring the 0.1-20Hz infrasonic signals released during the coking phase transition process.

[0053] The edge computing and data preprocessing unit aims to perform real-time localized processing on the acquired multimodal raw signals to efficiently extract key physical feature parameters that can characterize the evolution of the system state. The unit is deployed close to the sensing unit and performs noise reduction, transformation and calculation on the raw signals to transform high-dimensional and heterogeneous data into low-dimensional and standardized feature vectors, thereby reducing the computational burden on the central server and reducing data transmission latency.

[0054] The critical state prediction unit aims to quantitatively assess and warn of the risk of the system evolving towards the coking critical state based on extracted key physical characteristic parameters. As the system's decision-making module, the unit adopts a non-Markov model that can reflect the system's memory effect. It predicts the precursors of system instability caused by multi-field coupling by calculating a comprehensive phase space collapse risk index. When the risk index reaches a specific condition, the unit will generate a coking warning signal or activate a sensor migration command.

[0055] The adaptive sensor migration control unit aims to respond to the activation command issued by the critical state prediction unit, dynamically adjust the spatial layout of the sensor network, and realize the active tracking of risk sources. After being activated, the unit calculates the optimal location where the risk is most likely to occur at the next moment through an algorithm based on the centroid and gradient of the risk field, and generates migration control command to drive the movable sensor or activate the standby backup sensor, thereby realizing closed-loop adaptive coverage of the monitoring blind spot.

[0056] This embodiment overcomes the selective blindness and monitoring blind zone migration problems of traditional fixed monitoring systems when facing dynamic and nonlinear processes by constructing a complete closed-loop system that includes sensing, preprocessing, prediction and control. The system can proactively and predictively track coking risk sources and significantly improve the accuracy and lead time of early warning for sudden coking failures by capturing deep physical precursor information rather than apparent process parameters, thus ensuring the safe and efficient operation of the deep processing of inferior oil.

[0057] This technical solution provides an industrial IoT monitoring system for the atmospheric residue heat exchange process of a low-quality oil plant. By constructing a closed-loop sensing-prediction-control architecture consisting of a distributed multimodal sensing unit, an edge computing and data preprocessing unit, a critical state prediction unit, and an adaptive sensor migration control unit, it overcomes the shortcomings of existing technologies in the face of dynamic and nonlinear processes, such as selective blindness and fixed monitoring blind spots, and transforms static monitoring into proactive and predictive risk tracking.

[0058] The process of extracting key physical feature parameters by the edge computing and data preprocessing unit includes:

[0059] The high-frequency dynamic pressure time series data is analyzed to calculate the Marangoni number oscillation amplitude;

[0060] The power spectral density of the micro-vibration signal and the infrasound signal was analyzed to calculate the Lyapunov exponent of the focal precursor generation rate.

[0061] The echo signal of the ultrasonic transducer array was inverted and calculated to obtain the acoustic impedance gradient of the coking layer.

[0062] The nanoscale electrochemical probe data were processed to calculate the standard deviation of the Debye length fluctuations.

[0063] This embodiment describes the process of extracting key physical feature parameters by the edge computing and data preprocessing unit; accurate parameter extraction is the foundation for the accuracy of subsequent critical state prediction.

[0064] The process includes analyzing high-frequency dynamic pressure time-series data to calculate the Marangoni number oscillation amplitude; in this technical solution, the Marangoni number oscillation amplitude is represented by the Marangoni number oscillation amplitude used to quantify the instability of the surface tension gradient-driven flow induced by the Marangoni effect. , is a dimensionless parameter; its value is obtained by performing fast Fourier transform and wavelet analysis on data collected by a high-frequency dynamic pressure sensor.

[0065] The process also includes analyzing the power spectral density of the micro-vibration signal and the infrasound signal to calculate the Lyapunov exponent of the focal precursor formation rate; the Lyapunov exponent is denoted as... , is a parameter used to characterize the degree of chaos in the coking process, and its dimension is the reciprocal of time, for example. The numerical values ​​are derived by analyzing the changes in the power spectral density of signals collected by micro-vibration sensors and infrasound sensors, in order to reveal the dynamic transition path of the system from order to chaos.

[0066] The process further includes inverting the echo signal from the ultrasonic transducer array to obtain the acoustic impedance gradient of the coma layer; the acoustic impedance gradient of the coma layer is denoted as... , is a parameter describing the spatial variation of the physical properties of the coking layer, with dimensions of . The numerical data is obtained by inverting the echo signals emitted and received by the ultrasonic transducer array through an acoustic model, which is used to non-invasively detect the formation and hardening degree of the focal layer.

[0067] The process also includes processing the nanoscale electrochemical probe data to calculate the standard deviation of the Debye length fluctuations; the standard deviation of the Debye length fluctuations is denoted as... It is a parameter characterizing the stability of electrostatic interactions between micelle particles in the fluid near the inner wall of the pipe; the value is derived from statistical calculations of in-situ measurement data of nanoscale electrochemical probes, which is used to capture the precursors of critical aggregation of asphaltenes micelles.

[0068] This embodiment defines extraction methods for four key physical feature parameters, transforming raw signals from different physical dimensions, encompassing fluid dynamics, chaotic dynamics, acoustics, and electrochemistry, into quantitative indicators with clear physical meaning. This multimodal feature fusion provides the system with a much deeper and more comprehensive perspective than traditional process parameters such as temperature and pressure, enabling it to capture subtle changes driven by multi-field coupling in the early stages of the coking process, thus laying a solid data foundation for achieving high-precision early warning.

[0069] The system's distributed multimodal sensing unit can comprehensively acquire multidimensional physical signals directly related to the coking process, surpassing traditional methods that rely solely on apparent process parameters. The edge computing and data preprocessing unit efficiently transforms high-dimensional, heterogeneous raw signals into key physical characteristic parameters with clear physical meaning, including the Marangoni number oscillation amplitude, the Lyapunov exponent of the coking precursor generation rate, the acoustic impedance gradient of the coking layer, and the standard deviation of the Debye length fluctuation. This multimodal feature fusion based on deep physical precursors provides the system with a far deeper and more comprehensive state understanding than conventional monitoring, enabling it to capture subtle changes driven by multi-field coupling in the coking bud stage.

[0070] Example 2

[0071] The critical state prediction unit uses a non-Markov model for calculation, which combines a memory kernel function that describes the weight decay of historical state influences, as well as the oscillation amplitude of the Lyapunov exponent and Marangoni number output by the edge computing and data preprocessing unit.

[0072] The non-Markov model includes a denominator term, which is used to characterize the chaotic synchronization-desynchronization critical transition between the Lyapunov exponent and the Marangoni number oscillation amplitude.

[0073] When the Lyapunov exponent and the Marangoni number oscillate in a near-synchronous manner, the denominator term approaches a minimum value, causing the phase space collapse risk index to increase dramatically.

[0074] The non-Markov model also incorporates the Helvetica step trigger function to perform logical judgments;

[0075] When the standard deviation of the Debye length fluctuation is within the preset critical aggregation range, and the amplitude of the third harmonic component of the acoustic impedance gradient of the coking layer exceeds the preset acoustic harmonic threshold, a trigger signal with a value of 1 is generated; otherwise, a trigger signal with a value of 0 is generated.

[0076] This embodiment elaborates on the core algorithm model used in the critical state prediction unit; to overcome the shortcomings of traditional models in describing historical cumulative effects, this unit adopts a non-Markov model.

[0077] Non-Markov models aim to calculate a phase space collapse risk index that can accurately quantify the risk of system failure and coking abrupt changes by introducing a dependence on the system's state history, i.e., the memory effect. The model's computation combines a memory kernel function that describes the decay of weights influenced by historical states, with the oscillation amplitudes of the Lyapunov exponent and Marangoni number output from edge computing and data preprocessing units; the mathematical expression is as follows:

[0078]

[0079] The parameters in the formula are explained as follows:

[0080] : Phase space collapse risk index at time t, a dimensionless scalar, which is the final calculation output of the element; t: the current time; The starting time of integration;

[0081] : Memory kernel function, describing past moments The weight decay of the state's influence on the current time t, with dimensions of In this embodiment, the function form can be set as follows: Among them, the memory time constant It was determined by performing autocorrelation analysis on historical operating data, based on the observed 15-20 minute hysteresis effect; : Past moment; The memory time constant is determined through autocorrelation analysis of historical operating data. : Dimensionless model weight coefficients, determined through machine learning optimization algorithms;

[0082] Past moment The Lyapunov exponent, with dimensions of The values ​​are calculated by the preceding edge computing and data preprocessing units;

[0083] Past moment The Marangoni number oscillation amplitude is dimensionless, and its value is also calculated from the preceding unit.

[0084] Chaotic synchronization coupling coefficient, with dimensions of , characterization and The coupling strength between them; by experimentally determining a characteristic frequency that best represents the coupling effect as... ; The standard deviation of the Debye length fluctuation characterizes the stability of electrostatic interactions between micelle particles in the fluid near the inner wall of the pipe. Acoustic impedance gradient of the coke layer The amplitude of the third harmonic component in the spectrum;

[0085] The controlled experiment was conducted in a laboratory-scale shell-and-tube heat exchanger. The coking process was simulated by precisely controlling the inlet temperature (300-450°C) and flow rate (0.1-0.5 m / s) of the inferior oil sample. Micro-vibration signals and high-frequency dynamic pressure signals were simultaneously acquired during the experiment, and the corresponding Lyapunov exponents were calculated in real time. and Marangoni number oscillation amplitude By calculating the coherence spectra of these two time-series signals, the frequencies at which the coherence coefficient reaches its peak in the 0.1-20Hz frequency range were identified. and Chaotic synchronization coupling coefficient That is, it is defined as the ratio of these two characteristic frequencies, i.e. Repeat the experiment multiple times and take the statistical average of the ratios as the final calibration result;

[0086] : Dimensionless model weight coefficients; their values ​​are determined through machine learning optimization algorithms, aiming to maximize the accuracy of early warning and minimize the false alarm rate, by iteratively optimizing on historical coking accident data and normal operation data;

[0087] This embodiment employs a method combining grid search and random forest classifiers to determine the optimal solution. and Construct a training dataset containing at least 5000 time-series samples, where the features are the Lyapunov index in historical working conditions. Marangoni number oscillation amplitude Debye length fluctuation standard deviation and the amplitude of the third harmonic component of the acoustic impedance gradient of the coking layer The label indicates coking, 1 or normal, 0; settings The search range is [0.1, 10], and the step size is 0.1; set The numerical search range is [0.01, 1], and the dimensions are... Same, for The step size is 0.01; 5-fold cross-validation is used for each group ( , The parameters are combined for evaluation, and the phase space collapse risk index for all time points in the historical dataset is calculated using these parameter combinations. An optimal decision threshold is determined through receiver operating characteristic curve analysis. This ensures that, at this threshold, the classification of historical data results in the highest F1 score for the "foreground / normal" group, and the group that achieves the highest F1 score is selected. , ) as the optimal parameters of the model; for example, in a specific calibration of this embodiment, the optimal parameters obtained are and ;

[0088] : Settings and dimensional The numerical search range is [0.01, 1]. ;

[0089] Heaviside step trigger function, dimensionless, value is 0 or 1, input parameters and All are calculated from the preceding unit;

[0090] The non-Markov model contains a specially designed denominator term. The denominator term is used to characterize the chaotic synchronization-desynchronization critical transition between the Lyapunov exponent and the Marangoni number oscillation amplitude; when and When they approach a state of synchronization, that is, when the values ​​of the two terms are close, the denominator term approaches a minimum value, and is thus... Ensuring that it is not zero will lead to a risk index of phase space collapse. The number of cases increases dramatically; the design enables the model to issue early warnings by capturing the critical behavior of deep physical mechanisms during periods when traditional monitoring signals appear to be stable.

[0091] Furthermore, the non-Markov model incorporates a Heaviside step trigger function to perform logical judgments; the function The working principle consists of dual conditional logic gates; when the standard deviation of the Debye length fluctuates... Within a preset critical aggregation range, for example Furthermore, the acoustic impedance gradient of the coking layer Amplitude of the third harmonic component in the spectrum Exceeding the preset acoustic harmonic threshold When the function generates a trigger signal with a value of 1, it generates a trigger signal with a value of 0; otherwise, it generates a trigger signal with a value of 0. The physical basis of the rule is that the Debye length range corresponds to the critical state where the electrostatic repulsion between asphaltene micelles weakens and they are more prone to aggregation, while the acoustic harmonic threshold... The acoustic fingerprint of the critical oscillatory state of the pseudoplastic-dilatational transition observed in the corresponding experiment is also based on historical experimental data;

[0092] The calibration process is as follows: In the controlled experiment described above, when a pseudoplastic-dilatation transition is observed in the fluid using a rheometer, the corresponding ultrasonic echo signal is recorded; the signal is then subjected to spectral analysis to extract the amplitude of its third harmonic component. Repeat the process multiple times, recording multiple critical states. The amplitude values ​​were statistically analyzed, and the lower limit of the 90% confidence interval was taken as the acoustic harmonic threshold. To ensure high sensitivity. For example, calibrated in this embodiment. 0.85 at a specific sound pressure level;

[0093] The synergistic effect of these mechanisms enables risk prediction to combine historical retrospection, physical depth, and computational economy; the introduction of the memory kernel function allows the model to account for the cumulative effect of coking; the precise characterization of the chaotic synchronization critical point in the denominator term endows the system with the ability to perceive deep risks in the pseudo-normal stage where the normal signal is stable; the application of the Heaviside step trigger function, which acts as a logic gate, ensures that complex risk integral calculations are only activated when the system enters a high-risk pre-sequence state, significantly improving computational efficiency and making real-time monitoring possible;

[0094] The non-Markov model used in its critical state prediction unit quantifies the cumulative effect of the coking process by introducing a memory kernel function that describes the weight decay of historical state influences. The model can accurately characterize the chaotic synchronization-desynchronization critical transition between the Lyapunov exponent and the Marangoni number oscillation amplitude, enabling the system to issue early warnings by observing the critical behavior of deep physical mechanisms even when the normal signal appears stable. At the same time, the model combines the Heaviside step trigger function, which only activates the complex risk integral operation when the system enters the critical aggregation pre-state defined by the Debye length fluctuation and acoustic harmonics, significantly improving the computational economy of real-time monitoring.

[0095] Example 3

[0096] The critical state prediction unit performs the following process to determine the phase space collapse risk index:

[0097] The phase space collapse risk index is compared and analyzed with the preset risk threshold.

[0098] When the phase space collapse risk index exceeds the risk threshold, an activation command and the highest level coking warning signal are generated.

[0099] No activation command is generated when the phase space collapse risk index does not exceed the risk threshold;

[0100] This embodiment describes how the critical state prediction unit performs discrimination processing after calculating the risk index; the function of the process is to convert the continuous risk index into discrete operable instructions.

[0101] The discrimination and processing process is as follows: The phase space collapse risk index calculated in real time is... With preset risk threshold Perform comparative analysis; preset risk thresholds The setting is not fixed, but is optimized based on risk management strategies by analyzing receiver operating characteristic curves of historical data, aiming to achieve the best balance between early warning sensitivity and specificity.

[0102] when The calculation results show that when the phase space collapse risk index exceeds the risk threshold, the system is judged to enter a high-risk state. At this time, the critical state prediction unit will generate two outputs simultaneously: an activation command, which is sent to the adaptive sensor migration control unit to start the active tracking of the sensor; and the highest level coking warning signal, which is sent to the operator through the monitoring system.

[0103] when The calculation results show that when the phase space collapse risk index does not exceed the risk threshold, the system is judged to be in a safe or sub-healthy state. At this time, no activation command is generated, and the adaptive sensor migration control unit remains in standby state. The system can generate different levels of regular warnings or status prompts according to the specific value of the risk index, but does not trigger the highest level alarm and sensor migration.

[0104] This embodiment introduces a clear and dynamically optimizable risk threshold discrimination mechanism, transforming complex model outputs into explicit, binary action commands. This ensures the decisiveness and appropriateness of the system response, avoiding excessively frequent or unstable system responses due to minor fluctuations in the risk index. It establishes a key gating mechanism in the automated decision-making process, ensuring that subsequent high-cost closed-loop control behaviors, such as physical movement of sensors, are only triggered when a high-confidence, impending coking event is identified, thus achieving efficient utilization of monitoring resources.

[0105] Example 4

[0106] The adaptive sensor migration control unit determines the optimal sensor position for the next moment, and the process includes:

[0107] Obtain the phase space collapse risk index calculated by each sensor node;

[0108] The risk index of each sensor node and its corresponding fixed spatial location vector are weighted and averaged to generate the spatial distribution centroid of the risk field.

[0109] The process of determining the optimal sensor position for the next moment also includes:

[0110] Based on the risk index values ​​of neighboring sensor nodes around the centroid of spatial distribution, the local gradient of the risk field is calculated using a numerical approximation method.

[0111] The process of determining the optimal sensor position for the next moment also includes:

[0112] By combining the spatially distributed centroid and the local gradient of the risk field, the optimal sensor position for the next moment is generated through extrapolation calculation.

[0113] And based on the optimal sensor position at the next moment, a migration control command is generated;

[0114] This embodiment describes the entire process of how the adaptive sensor migration control unit determines the optimal sensor position at the next moment after receiving the activation command; in order to achieve robust and predictive control, the unit's execution process includes the following steps;

[0115] The unit obtains the phase space collapse risk index calculated by each sensor node i. To avoid decision instability caused by noise or transient fluctuations from a single sensor, the first step of the algorithm is not to directly track the single point with the highest risk, but to analyze the risk index of each sensor node and its corresponding fixed spatial location vector. A weighted average calculation is performed to generate a smooth and robust spatial distribution centroid of the risk field. The calculation model is as follows:

[0116] ;

[0117] in, : Centroid vector, t: time, N: total number of sensors Node risk index : Node location vector; i: Index of the sensor node; : The phase space collapse risk index of node i at time t;

[0118] The physical meaning of the calculation is that areas with higher risks have greater weight in determining the current center position of the blind spot;

[0119] The center of mass of risk has been determined. Next, the system needs to predict the direction of risk migration; therefore, the next step of the algorithm is to calculate the local gradient of the risk field using numerical approximation methods based on the risk index values ​​of neighboring sensor nodes around the spatially distributed centroid. Numerical approximation methods refer to methods that can estimate the gradient of a function without requiring an analytical expression. In this embodiment, a multi-point finite difference method based on Taylor expansion can be used, according to the centroid. and its neighboring sensors The gradient vector is estimated by comparing the position vector and risk value between them;

[0120] For a sensor network on a two-dimensional pipe cross-section, the risk field gradient Two components It can be approximated using the central difference scheme; assuming it is at the centroid. The nearest sensor nodes in the x and y directions are respectively and The corresponding risk indices are respectively Then the gradient can be approximated as:

[0121] ;

[0122] in, and These are the sensor spacings in the x and y directions, respectively. This method provides stable and relatively accurate gradient estimation; At the center of mass Local gradient of the risk field at location; Risk index value of the sensor near the centroid in the x-direction; Risk index value of sensors near the centroid in the y-direction;

[0123] After obtaining the current center location and migration direction of the risk, the algorithm performs the final extrapolation calculation; by combining the spatial distribution centroid... Local gradient of the risk field Generate the optimal sensor position for the next moment. The model is as follows:

[0124] ;

[0125] in, Optimal position Predict the time step. Blind zone migration feature speed; The next moment Optimal sensor position; t: current time; The center of mass of risk at the current moment t; Local gradient of the risk field; The magnitude of the local gradient in the risk field, used to normalize the gradient vector to a unit vector;

[0126] The logic of the formula lies in setting the current risk centroid. As a benchmark, along the direction of the fastest risk growth, i.e., the gradient unit vector. Extrapolate the direction; the extrapolated distance is determined by the characteristic velocity of the pre-calibrated blind zone migration. and prediction time step The product determines the speed; It can be determined based on the statistical average rate of risk field migration in historical data;

[0127] To determine The study analyzed data from 10 complete historical coking accidents. For each accident, the centroid of the risk field was calculated every 5 minutes within 6 hours prior to the coking event. By calculating the change in the position of the centroid at consecutive time points. With time interval The ratio was used to obtain a series of instantaneous migration velocities; outliers were removed, and the 95th percentile of all instantaneous velocities in all accidents was calculated. This value was then used as the characteristic velocity of blind zone migration. The 95th percentile, rather than the average, is used to ensure that the sensor's migration speed is fast enough to provide a degree of predictability and robustness when risks develop rapidly.

[0128] Calculated optimal sensor position for the next moment It is used to generate migration control commands, which are transmitted to the physical actuators of the sensor, such as robotic arms or rails, to drive it to the target location, or to activate standby sensors waiting near the predicted location.

[0129] This embodiment achieves a precise adaptive adjustment of sensor network layout through a three-step method of centroid localization, gradient prediction, and extrapolation control. The first step, centroid calculation, ensures the robustness of target tracking and effectively filters out single-point noise. The second step, gradient calculation, gives the system the predictability of risk migration direction. The third step, extrapolation control, transforms the prediction into specific physical actions, realizing closed-loop control. The synergistic effect of this series of processes enables the sensor network to actively track risk sources smoothly and accurately, solving the problem of missed detection caused by the dynamic migration of monitoring blind spots, and ensuring that sensing resources are always deployed where they are most needed.

[0130] The system's adaptive sensor migration control unit, upon receiving the activation command generated by the critical state prediction unit, can achieve closed-loop adaptive tracking of risk sources. By calculating the weighted average of the risk indices of each sensor node, a robust risk field spatial distribution centroid is generated, effectively filtering out single-point noise. Furthermore, the local gradient of the risk field is calculated using a numerical approximation method, giving the system predictive ability to forecast the direction of risk migration. Combining the centroid and gradient, extrapolation calculations are performed to generate the optimal sensor position for the next moment. This centroid positioning-gradient prediction-extrapolation control mechanism ensures that sensing resources are always accurately deployed in the area where the risk is most likely to occur, solving the problem of missed detection caused by the dynamic migration of monitoring blind spots.

[0131] This technical solution, through the synergistic effect of the above-mentioned structure and method, significantly improves the accuracy and lead time of early warning for sudden coking faults, ensuring the safe and efficient operation of the deep processing of inferior oil.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An industrial internet of things monitoring system for a poor oil plant constant residue heat exchange process, characterized in that, The system comprises a distributed multi-modal sensing unit, an edge computing and data preprocessing unit, a critical state prediction unit, and an adaptive sensor migration control unit. The distributed multi-modal sensing unit is configured to collect original signals of the heat exchange pipeline. The edge computing and data preprocessing unit is configured to process the original signals to extract key physical characteristic parameters. The critical state prediction unit is configured to calculate a phase space collapse risk index based on the key physical characteristic parameters, and perform discriminant processing on the phase space collapse risk index to generate a coking early warning signal or an activation instruction. The adaptive sensor migration control unit is configured to determine an optimal sensor position at the next moment in response to the activation instruction, and generate a migration control instruction based on the optimal sensor position. The adaptive sensor migration control unit determines the optimal sensor position at the next moment, and the process comprises: obtaining the phase space collapse risk index calculated by each sensor node; and performing weighted average calculation on the risk index and the corresponding fixed spatial position vector of each sensor node to generate the spatial distribution centroid of the risk field; The process of determining the optimal sensor position at the next moment further comprises: based on the risk index values of the adjacent sensor nodes around the spatial distribution centroid, the local gradient of the risk field is calculated by a numerical approximation method; The process of determining the optimal sensor position at the next moment further comprises: combining the spatial distribution centroid and the local gradient of the risk field, the optimal sensor position at the next moment is generated by extrapolation calculation; and generating a migration control instruction based on the optimal sensor position at the next moment; The mathematical expression form of the phase space collapse risk index is as follows: ； The parameters in the formula are explained as follows: : time instant : phase space collapse risk index, dimensionless scalar, is the final computed output of the unit; : current time instant; : starting time instant of integration; : memory kernel function describing the weight decay of the state at past time t on the current time t ; ; : past time t : dimensionless model weight coefficient determined by a machine learning optimization algorithm; : past time Lyapunov exponent at the past time , dimensionless, calculated by the edge computing and data preprocessing unit of the prologue; : past time : oscillation amplitude of Marangoni number, dimensionless, value calculated by the preceding unit; : chaotic synchronization coupling coefficient, dimensionless , representing the coupling strength between and ; a characteristic frequency most representing the coupling effect is calibrated through experiment as ; : standard deviation of fluctuation of Debye length, representing the stability of electrostatic interaction between micelles in the fluid near the inner wall of the pipe; : acoustic impedance gradient of coking layer : amplitude of third harmonic component in frequency spectrum; : dimensionless model weight coefficient; the value is determined by machine learning optimization algorithm, with the goal of maximizing the early warning accuracy and minimizing the false positive rate, and is iteratively optimized on historical coking accident data and normal operation data; : set the same dimension as : set the same dimension as : set the same dimension as : set the same dimension as : Heaviside step trigger function, dimensionless, value of 0 or 1, input parameter and are calculated by the pre-sequencing unit.

2. The industrial internet of things monitoring system for the poor oil device constant residue heat exchange process flow according to claim 1, characterized in that, The process of extracting key physical characteristic parameters by the edge computing and data preprocessing unit comprises: analyzing high-frequency dynamic pressure time series data to calculate the Marangoni number oscillation amplitude; analyzing the power spectral density of micro-vibration signals and infrasound signals to calculate the Lyapunov exponent of the coking precursor generation rate; inversion calculation is performed on the echo signal of the ultrasonic transducer array to obtain the acoustic impedance gradient of the coking layer; and processing the nanoscale electrochemical probe data to calculate the fluctuation standard deviation of the Debye length.

3. The industrial internet of things monitoring system for the poor oil device constant residue heat exchange process flow of claim 1, wherein, The critical state prediction unit uses a non-Markov model for calculation, which combines a memory kernel function describing the decay of historical state influence weight, and the Lyapunov exponent and Marangoni number oscillation amplitude output by the edge computing and data preprocessing unit.

4. The industrial internet of things monitoring system for the inferior oil device constant residue heat exchange process flow of claim 3, wherein, The non-Markov model contains a denominator term, which is used to describe the chaotic synchronization-desynchronization critical transition between the Lyapunov exponent and the Marangoni number oscillation amplitude; When the Lyapunov exponent and the Marangoni number oscillation amplitude tend to be in a synchronized state, the denominator term tends to a minimum value, causing the phase space collapse risk index to increase dramatically.

5. The industrial internet of things monitoring system for the poor oil device constant residue heat exchange process flow of claim 3, wherein, The non-Markov model also combines a Heaviside step trigger function to perform logical judgment; When the fluctuation standard deviation of the Debye length is within a preset critical aggregation range, and the amplitude of the third harmonic component of the acoustic impedance gradient of the coking layer exceeds a preset acoustic harmonic threshold, a trigger signal with a value of 1 is generated; otherwise, a trigger signal with a value of 0 is generated.

6. The industrial internet of things monitoring system for the constant residue heat exchange process of poor oil device according to claim 1, characterized in that, The process of the critical state prediction unit discriminating the phase space collapse risk index is as follows: The phase space collapse risk index is compared and analyzed with a preset risk threshold value; When the phase space collapse risk index exceeds the risk threshold value, an activation instruction and a highest-level coking warning signal are generated; When the phase space collapse risk index does not exceed the risk threshold value, no activation instruction is generated.

Images