Digital-twin-based dynamic simulation and management and control method and system for catering pollution sources and storage medium

Abstract

The present application relates to the technical field of kitchen safety in the catering industry, and in particular to a catering pollution source dynamic simulation and control method and system based on digital twinning and a storage medium. A digital twin of the smoke exhaust duct is constructed, the inlet temperature, outlet temperature, ambient temperature, airflow speed, fan current and purifier loop current waveform are collected, and the macroscopic heat power residual error is calculated; the temperature is used to correct the gas state parameters, the dynamic breakdown voltage threshold is calculated, and the purifier high-voltage power supply is fed forward adjusted; the pre-breakdown partial discharge events are extracted and the discharge frequency is counted, the causal and correlation analysis is carried out on the basis of lag compensation, and the early warning, derating or isolation control signal is output. The present application provides a catering pollution source dynamic simulation and control method and system based on digital twinning and a storage medium, which improves the early hidden heat source identification capability and reduces the false alarm probability.

Description

Technical Field

[0001] This invention relates to the field of kitchen safety technology in the catering industry, and in particular to a method, system, and storage medium for dynamic simulation and control of catering pollution sources based on digital twins.

[0002] The term "food pollution source" as used in this invention is not limited to the emission sources of oily fumes, volatile organic compounds, or odorous gases directly generated during cooking. It also includes complex abnormal sources within food exhaust ducts and electrostatic purification equipment, which are formed by a combination of factors such as oil deposits, heat accumulation, smoldering, pre-breakdown micro-discharge, and arc ignition risks. These complex abnormal sources not only affect the migration, deposition, and secondary release of oily fume pollutants but also induce fire and electrical safety risks within the exhaust system. Therefore, they are considered pollution sources that require dynamic simulation and control during the operation of food exhaust systems. Background Technology

[0003] Restaurant kitchen exhaust systems operate in a complex environment characterized by high temperature, high humidity, and the coexistence of oily fumes, particulate matter, and volatile organic compounds. Oil deposits easily accumulate on the inner walls of the exhaust ducts and on the surfaces of the electrostatic precipitator plates. As operating time increases, these oil deposits not only serve as a carrier for the continuous adhesion and secondary release of pollutants, but may also become points of enhanced pollution release and fire hazard triggers under localized heat accumulation, smoldering, or electric arcing, thus constituting a dynamic source of pollution in restaurant exhaust systems.

[0004] Currently, the mainstream equipment for removing cooking fumes in the catering industry is the high-voltage electrostatic fume purifier, which operates at an electric field voltage of 8 kV to 12 kV. When a hidden heat source appears in the exhaust airflow, the airflow temperature rises and the gas density decreases. According to Pascal's Law, the gas breakdown voltage decreases accordingly. If the operating voltage margin of the electrostatic purifier is insufficient, pre-breakdown partial discharge or even arcing may occur, generating an ignition source that ignites flammable gases and oil particles in the pipeline, forming an accident chain from smoldering to a cascading fire. Existing technologies have the following three deficiencies in protecting against such accident chains.

[0005] Currently, industrial monitoring methods for fire risks in smoke exhaust ducts primarily rely on fixed-threshold temperature detectors installed inside the ducts. These detectors typically have alarm thresholds above 60°C and can only detect stages where smoldering has fully developed to the point of a significant increase in overall duct temperature, failing to identify early-stage smoldering anomalies that only cause a slight increase in localized airflow temperature. For electrostatic purifiers themselves, their protection mechanisms are mainly reactive, monitoring for overcurrent, overvoltage, or arcing events in the output circuit and tripping the power supply after a fault occurs. These solutions lack a correlation model between thermodynamic anomalies and electrical insulation degradation, representing a single-dimensional, open-loop monitoring architecture.

[0006] To improve early detection of smoldering, engineering practices have attempted to lower the temperature alarm threshold or add various sensors such as smoke and gas sensors. Lowering the temperature threshold directly leads to a sharp increase in the false alarm rate, as changes in normal operating conditions such as ambient temperature fluctuations and full-load operation of fans can trigger alarms, compromising system availability. While adding different types of sensors can supplement information, it only achieves parallel acquisition and independent alarm of multi-sensor data, without establishing a causal relationship mechanism between different physical domains. Essentially, it still involves comparing thresholds of each channel, failing to fundamentally distinguish between the cross-domain chain reaction caused by actual smoldering and the occasional disturbances occurring simultaneously by multiple sensors. Furthermore, all air purifier protection schemes based on post-flashover tripping cannot prevent the occurrence of the initial breakdown arc, which itself can become an ignition source, causing the protection action to lag behind the occurrence of a fire.

[0007] At the thermodynamic detection level, the fundamental reason for the discrepancy between false alarms and false alarms is the mismatch between the control volume thermal balance model and the actual physical system. Existing solutions do not perform real-time modeling of the pipeline thermal balance, making it impossible to distinguish between temperature rises caused by normal energy dissipation and those caused by additional heat sources. They can only rely on a single temperature threshold for coarse-grained judgment. For specific exhaust branch pipes, specific measuring point arrangements, and operating conditions within a certain airflow range, early concealed heat sources may only cause a small local temperature rise. While this temperature rise may not trigger a traditional high-temperature alarm threshold, it may alter the local gas state and affect the insulation margin of the electrostatic purifier. Therefore, a joint judgment based on the thermal balance model and electrical state information is necessary. At the control mechanism level, the fundamental reason for post-event protection failure is the open-loop control architecture. The operating voltage of the electrostatic purifier is fixed according to rated operating conditions and does not dynamically adjust with changes in gas density, a key variable determining the breakdown voltage. Paschen's law establishes a monotonic and nonlinear functional relationship between gas density and breakdown voltage. Open-loop control means that the safety voltage margin relies entirely on the initial design allowance. Under conditions of continuously rising temperature and continuously decreasing density, the margin will be unidirectionally consumed until it reaches zero, making breakdown inevitable. At the cross-domain verification level, the lack of a causal inference mechanism makes it impossible for a single sensor solution to distinguish between the true causal chain of an accident and occasional interference. Statistical data shows that the false alarm rate of a solution relying solely on a single temperature or current threshold can reach false alarm levels in industrial field environments. Frequent false alarms will cause maintenance personnel to become desensitized to alarms, ultimately causing the system to lose its protective effectiveness.

[0008] Therefore, a method is needed to model oil deposits, hidden heat sources, gas discharge states, and the risk of secondary release of pollutants as a complex pollution source in a unified manner, so as to achieve dynamic simulation, early identification, and hierarchical control of the evolution process of pollution sources in catering exhaust systems. Summary of the Invention

[0009] To achieve the above-mentioned objectives, this invention provides a method, system, and storage medium for dynamic simulation and control of catering pollution sources based on digital twins. It aims to model and control oil deposits, hidden heat sources, pre-breakdown micro-discharge, and arc ignition risks in catering exhaust systems as composite pollution sources in a unified manner, thereby solving or at least mitigating the problem that traditional fixed threshold temperature detectors cannot detect the small temperature rise in the early stage of smoldering and that the effective alarm time is seriously delayed.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic simulation and control of catering pollution sources based on digital twins, wherein the catering pollution sources include composite pollution sources in the catering exhaust system formed by oil stains, concealed heat sources, pre-breakdown micro-discharge, and arc ignition risks, including:

[0011] A digital twin of the smoke exhaust duct is constructed, which includes a lumped parameter thermal balance model and a gas discharge model based on Paschen's law;

[0012] The system collects inlet airflow temperature, outlet airflow temperature, ambient temperature, airflow velocity, fan current, and purifier circuit current waveforms, drives the lumped parameter thermal balance model to calculate the macroscopic thermal power residual, corrects the gas state parameters according to the real-time temperature, and inputs the gas discharge model to calculate the dynamic breakdown voltage threshold.

[0013] Based on the dynamic breakdown voltage threshold, the pulse width modulation duty cycle of the high-voltage power supply of the electrostatic purifier is adjusted forward.

[0014] Extract pre-breakdown partial discharge events and count the discharge frequency;

[0015] Within the sliding time window, a statistical correlation analysis is performed on the macroscopic thermal power residual and the discharge frequency, and early warning, derating protection or isolation control is triggered based on the analysis results.

[0016] To further realize the present invention, the following technical solutions may be preferred:

[0017] Preferably, the calculation of the macroscopic thermal power residual includes: calculating the current airflow density based on the real-time temperature and preset or measured pressure using the ideal gas state equation under isobaric approximation, and obtaining the mass flow rate based on the current airflow density, pipe cross-sectional area and airflow velocity;

[0018] The macroscopic thermal power is calculated by subtracting the fan heat injection power calculated from the fan current and adding the pipe wall heat dissipation power calculated by online adaptive estimation of the pipe wall comprehensive heat transfer coefficient through Kalman filtering.

[0019] Preferably, the calculation of the dynamic breakdown voltage threshold includes:

[0020] The gas state parameters are corrected based on the real-time temperature to obtain the equivalent gas number density parameter, and the equivalent gas number density parameter and the purifier electrode spacing are substituted into the Paschen's law expression; or, the equivalent pressure-spacing product parameter is obtained based on the real-time temperature and the purifier electrode spacing, and the equivalent pressure-spacing product parameter is substituted into the Paschen's law expression; then, combined with the non-uniform electric field geometric safety margin coefficient, the dynamic breakdown voltage threshold under the current gas state is obtained;

[0021] Specifically, during initialization, the gas discharge model ensures that the equivalent pressure-spacing product corresponding to the initial operating point is greater than the product corresponding to the minimum value of the Paschen curve, so that the system operating point is located in the region to the right of the minimum value of the Paschen curve. This ensures that when the temperature increases and the equivalent pressure-spacing product decreases, the dynamic breakdown voltage threshold shows a decreasing trend.

[0022] Preferably, the value of the non-uniform electric field geometric safety margin coefficient automatically decreases as the degree of oil accumulation on the purifier electrode increases, so as to correct the impact of the shortened equivalent insulation distance caused by oil adhesion on the breakdown voltage threshold.

[0023] Preferably, the extraction of pre-breakdown partial discharge events includes:

[0024] Transient events whose current spike amplitude exceeds the first threshold, whose pulse width falls within a preset range, and whose leading edge rise rate exceeds the second threshold are identified as candidate discharge events.

[0025] For the candidate discharge event, verify whether the phase angle of its occurrence time relative to the turn-on and turn-off times of the switch is stable in multiple consecutive switching cycles, and actively exclude candidate events that fall within the inherent operating phase exclusion window of the switch. The candidate events that pass the verification are counted as one valid pre-breakdown partial discharge event.

[0026] Preferably, the statistical association analysis includes:

[0027] Establish a bidirectional circular queue with a depth of N, and push the macroscopic thermal power residual and discharge frequency data into their respective queues in each control cycle;

[0028] When both the thermal power anomaly and micro-discharge anomaly flags are valid, the average airflow velocity and the physical distance between the sensors are extracted. The physical lag time of heat transfer is calculated based on these two values, and then converted into a candidate lag order range by combining the control cycle or sampling cycle. Within this candidate lag order range, using the physical lag order determined by the average airflow velocity and the physical distance between the sensors as the constraint center, a time-series regression model is established for each candidate lag order for order comparison. This model is then applied based on the Akaike information criterion, the Bayesian information criterion, and at least one of the residual white noise test criteria. A method involves model screening to determine the optimal lag order; wherein the optimal lag order is selected only from the range of candidate lag orders to simultaneously satisfy the physical transfer time constraint and model diagnostic requirements; based on the optimal lag order, a constrained model and an unconstrained model for Granger causality testing are established, and the joint significance test of the regression terms is used to determine whether the macroscopic thermal power residual is a Granger cause of the discharge frequency; after confirming the existence of Granger causality, the time series data are shifted and aligned according to the time offset corresponding to the optimal lag order, and the Spearman rank correlation coefficient is calculated.

[0029] Preferably, triggering early warning, throttling protection, or isolation control based on the analysis results includes:

[0030] When only the thermal power residual is abnormal, or when the Spearman rank correlation coefficient is lower than the first threshold, an early warning signal is output;

[0031] When the Spearman rank correlation coefficient is between the first threshold and the second threshold and remains between them for a first duration, derating protection is activated, and the operating voltage is limited to a preset percentage of the dynamic breakdown voltage threshold.

[0032] When the Spearman rank correlation coefficient exceeds the second threshold and lasts for a second duration, or momentarily exceeds the third threshold, an isolation control signal is output. The isolation control signal is used to shut off the high-voltage power supply of the electrostatic purifier and send a linkage request to the fire linkage control unit for the fan, fire damper, or fire extinguishing device.

[0033] Preferably, after calculating the macroscopic thermal power residual, the method further includes:

[0034] Within a time window without abnormal steady-state conditions, the baseline fluctuation standard deviation of the macroscopic thermal power residual data is extracted, and an adaptive noise base threshold is dynamically configured accordingly. Only when the absolute value of the macroscopic thermal power residual calculated in real time is greater than the adaptive noise base threshold in multiple consecutive sampling periods, or when the positive value of the macroscopic thermal power residual calculated in real time is higher than the adaptive noise base threshold in multiple consecutive sampling periods, is the thermal power abnormality confirmed, and subsequent feedforward adjustment and statistical correlation analysis processes are triggered.

[0035] A digital twin-based dynamic simulation and control system for catering pollution sources is provided to execute the aforementioned method. This system is used to dynamically simulate and control complex pollution sources in catering exhaust systems, which are composed of oil deposits, concealed heat sources, pre-breakdown micro-discharge, and arc ignition risks. The system includes:

[0036] The digital twin model building module is used to create a controlled pipeline digital twin that includes a lumped parameter thermal balance model and a Paschen gas discharge model;

[0037] The data acquisition module is used to acquire inlet airflow temperature, outlet airflow temperature, ambient temperature, airflow velocity, fan current, and purifier circuit current waveforms.

[0038] Thermodynamic state estimation unit is used to drive the lumped parameter thermal balance model based on the collected data and output the macroscopic thermal power residual.

[0039] The gas discharge safety margin management unit is used to correct the gas state parameters according to the temperature and calculate the dynamic breakdown voltage threshold by the Paschen gas discharge model, thereby feeding forward to adjust the duty cycle of the high voltage power supply of the purifier.

[0040] The electromagnetic transient monitoring unit is used to extract and statistically analyze the frequency of pre-breakdown partial discharge events from the current waveform of the purifier circuit.

[0041] The macro-micro cross-domain time-series causal inference unit is used to perform Granger causality tests on thermal power residuals and discharge frequency within a sliding time window, calculate Spearman rank correlation coefficients, and output control commands based on hierarchical logic.

[0042] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the above-described method.

[0043] The beneficial effects of this invention are:

[0044] By using an online adaptive lumped parameter thermal balance model, the physical basis for detecting pollution source anomalies is expanded from a single temperature threshold to macroscopic thermal power residual detection based on energy balance. This enables the identification of hidden heat sources formed by heat accumulation or smoldering of oil deposits in the early stages before the overall pipeline temperature has risen significantly, thereby improving the ability to detect the early thermal anomaly evolution process of complex pollution sources.

[0045] The protection of the electrostatic purifier is extended from reactive tripping to feedforward derating control based on the Paschen gas discharge model, which enables the breakdown voltage safety boundary to be dynamically adjusted according to changes in the gas state. When the temperature rises and the gas density decreases and the insulation margin shrinks, the system can limit the high-voltage power output of the electrostatic purifier in advance, thereby reducing the risk of pre-breakdown micro-discharge, arcing ignition, oil combustion and secondary release of pollutants caused by insufficient insulation margin. Attached Figure Description

[0046] Figure 1 This is the overall system architecture topology diagram of the present invention.

[0047] Figure 2 This is a block diagram of the micro-discharge multi-feature fusion discrimination logic of the present invention.

[0048] Figure 3 This is a flowchart of macro-micro cross-domain causal reasoning and hierarchical decision-making in this invention. Detailed Implementation

[0049] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1

[0052] The hysteresis order is not directly determined by the physical distance between the sensors at a single point, nor is it arbitrarily determined by information criteria within a global scope. Instead, it is determined using a "statistical optimization within physical constraints" method. The system first estimates the physical hysteresis time of heat transfer based on the average airflow velocity and the physical spatial distance between the two sensors, and then converts this physical hysteresis time into a central value for the physical hysteresis order. Considering velocity measurement errors, flow field non-uniformity, and sensor response delays, the system sets a hysteresis time tolerance to obtain a candidate hysteresis time range, which is then converted into a candidate hysteresis order range based on the control cycle or sampling cycle. Subsequently, the system establishes time-series regression models for order comparison only within the candidate hysteresis order range, and performs model screening based on at least one of the Akaike information criterion, Bayesian information criterion, and residual white noise test to determine the optimal hysteresis order. If multiple candidate lag orders meet the model diagnostic requirements, the candidate lag order with the smallest deviation from the central value of the physical lag order and the lowest model complexity is selected first. If none of the candidate lag orders meet the model diagnostic requirements, the Granger causality confirmation process is not initiated; only a warning is output or downgrade protection is maintained. The system establishes constrained and unconstrained models for Granger causality testing based on the optimal lag order, and determines whether the thermal power residual sequence can significantly improve the predictive ability of the micro-discharge frequency sequence through a joint significance test of the regression terms. After the Granger causality test is passed, the system shifts and aligns the thermal power residual sequence and the micro-discharge frequency sequence according to the time offset corresponding to the optimal lag order, and then calculates the Spearman rank correlation coefficient. The system architecture adopts a three-level topology: a field sensing layer (data acquisition module), an edge digital twin layer (digital twin model construction module), and a response execution layer. Figure 1 As shown, the field sensing layer is responsible for sensing the raw physical quantities, the edge digital twin layer runs the digital twin model and inference algorithm, and the response execution layer completes the final control command execution and protection actions.

[0053] The configuration of the on-site sensing layer revolves around a dual-channel approach: "macroscopic thermodynamic quantities" and "microscopic electromagnetic transient quantities." On the controlled exhaust duct, an armored K-type thermocouple is installed as a temperature difference calculation point on both the inlet side (reference section) and the outlet side (approximately 3.0m to 5.0m upstream of the electrostatic purifier inlet; this spatial spacing is designed to ensure that the physical transfer lag time of the thermofluid state evolution under high airflow conditions is sufficiently greater than the discrete sampling period of the digital algorithm). The protective tube is made of 316L stainless steel. Cold junction compensation and analog-to-digital conversion are performed using a MAX31856 thermocouple-to-digital converter. Combined with a digital filtering algorithm, the system achieves an actual temperature resolution of 0.1℃ and a sampling rate of 10 Hz. Airflow velocity measurement uses an L-type Pitot tube with a PTFE oleophobic coating. The dynamic pressure signal is acquired through a differential pressure transmitter with internal temperature compensation, a full-scale range of ±500 Pa, and an accuracy of ±3% of the reading. The ambient temperature sensor is a PT1000 platinum resistance thermometer, attached to the outer side of the pipe wall in an area unaffected by the heat flow inside the pipe. A current transformer is installed on the fan motor power supply line to indirectly measure the fan's heat injection power. A broadband current transformer with a -3dB bandwidth of 10Hz-5MHz is connected in series in the grounding circuit of the electrostatic purifier's high-voltage power supply to couple nanosecond to microsecond level partial discharge pulse currents.

[0054] The edge digital twin layer is built on a heterogeneous computing architecture, including a high-performance microprocessor (such as an ARM Cortex-M7 or a dual-core DSP core) with a main frequency of no less than 400 MHz and a hardware coprocessor (such as an FPGA or CPLD) for processing high-frequency data streams. The microprocessor integrates a hardware floating-point unit and SIMD instruction extensions. Four core functional units operate collaboratively under the scheduling of the FreeRTOS real-time operating system: the thermodynamic state estimation unit executes at 100 Hz and is responsible for sensor data fusion, adaptive thermal balance residual calculation, and anomaly flag setting; the gas discharge safety margin management unit executes at 100 Hz and is responsible for feedforward derating protection; the electromagnetic transient monitoring unit relies on the parallel hardware logic of the hardware coprocessor and high-speed DMA channel to seamlessly process the 20MSPS high-speed ADC data stream at the front end, realizing pure hardware-level real-time micro-discharge multi-dimensional feature extraction and frequency statistics; and the macro-micro cross-domain time-series causal inference unit executes at 10 Hz and is responsible for Granger causality testing, Spearman rank correlation coefficient calculation, and hierarchical decision-making.

[0055] The response execution layer includes the PWM control terminal of the electrostatic purifier high-voltage power supply, the relay for driving the fire damper in the pipeline, and a PWM output channel for controlling the speed of the kitchen make-up air fan.

[0056] Example 2

[0057] In this invention, the dynamic simulation and control of pollution sources includes the joint simulation and control of pollutant deposition carriers, thermal anomaly evolution processes, and electrical induced risks. Specifically, pollutant deposition carriers include oil deposits on the inner walls of exhaust ducts and the surfaces of electrostatic purifier plates; thermal anomaly evolution processes include heat accumulation in the oil deposits, smoldering, and the resulting changes in macroscopic thermal power residuals; and electrical induced risks include pre-breakdown micro-discharge, arcing, and ignition risks generated by the electrostatic purifier under high temperature, oil contamination, and non-uniform electric field conditions.

[0058] Based on the processing system provided in Embodiment 1, this embodiment provides a method for dynamic simulation and control of catering pollution sources based on digital twins. This method uses a lumped-parameter thermal balance model with online self-calibration capabilities to detect hidden heat sources at the macroscopic thermodynamic level; a gas discharge model with its operating point rigorously verified by Paschen's law provides feedforward protection; and a cross-domain causal inference engine composed of Granger causality tests and Spearman rank correlation analysis completes the closed-loop verification and hierarchical decision-making of the "smoldering-pre-breakdown" physical causal chain. Combined with... Figure 3 The cross-domain reasoning process shown below is implemented in detail at each stage.

[0059] (I) Adaptive Thermal Equilibrium Model and Calculation of Macroscopic Thermal Power Residual

[0060] A digital twin model of a controlled flue gas duct is constructed, the core of which is a lumped-parameter heat balance model based on fluid mass flow rate, inlet and outlet temperature rise, fan motor heat injection, and heat dissipation from the pipe wall to the environment. This model is used to calculate the macroscopic thermal power residual and derive the gas density drop coefficient based on the airflow temperature. The digital twin model also includes online calibration and continuous adaptive estimation of the overall heat transfer coefficient of the pipe wall.

[0061] The limitation of traditional fixed-parameter heat balance models is their inability to handle gradually changing heat dissipation boundary conditions caused by grease buildup on the pipe walls. Grease deposits reduce the actual overall heat transfer coefficient of the pipe walls. Continuing to decline. If the model uses a fixed... Heat dissipation power This will be overestimated, causing a negative bias in the thermal balance residual, which gradually erodes the detection sensitivity to low-power smoldering sources. In severe cases, even normal fan operation can trigger a false alarm.

[0062] Will The state is modeled as a slowly changing random walk and incorporated into the state estimation framework for online tracking. A scalar state is defined. The state transition equation is a first-order Gaussian-Markov process:

[0063]

[0064] Process noise , The decision was made to track agility.

[0065] The system first performs a steady-state condition determination. When the mass flow rate... Motor power and ambient temperature When the standard deviation is less than the preset threshold for a continuous 60-second sliding window, and the thermal anomaly flag is false, the observation equation is established:

[0066]

[0067] Observations , A scalar Kalman filter is used for recursive estimation. :

[0068] Time update steps:

[0069]

[0070]

[0071] Measurement update steps: Calculate Kalman gain ,in Update posterior state estimates and posterior error covariance .

[0072] This system uses the Kalman filter algorithm, combined with model predictions and actual observation data, to achieve a dynamic and smooth estimation of the overall heat transfer coefficient of the pipe wall.

[0073] Real-time thermal balance residual calculation uses the current estimate. :

[0074]

[0075] The mass flow rate is calculated based on the ideal gas law under the isobaric approximation. The system first determines the current gas flow density based on the real-time gas flow temperature and preset or measured pressure, and then combines this with the pipe cross-sectional area and gas flow velocity to obtain the mass flow rate. In other words, temperature and pressure are used to determine the gas flow density; the pipe cross-sectional area is not directly involved in the gas flow density calculation, but rather used to convert the gas flow velocity into the gas flow rate through the pipe cross-section, and then further participates in the mass flow rate calculation. This process avoids mistakenly using the pipe cross-sectional area as a density calculation parameter, while ensuring that the calculation of the macroscopic thermal power residual has a reliable physical basis.

[0076] Calculate the normalized information before each measurement update, if This is considered an outlier and updates are skipped. Freezing occurs when Pitot fails. An alarm will be triggered. If the temperature exceeds the limit when the thermocouple is open-circuited, Level 3 physical isolation will be immediately triggered. When the absolute value of the macroscopic thermal power residual is continuously greater than the noise floor threshold, or the positive value of the macroscopic thermal power residual is continuously higher than the noise floor threshold for 30 seconds, the thermal anomaly flag will be set; where the noise floor threshold can be taken as 3 times the standard deviation of the macroscopic thermal power residual background when there are no anomalies during the commissioning period.

[0077] (II) Gas Discharge Operating Point Verification and Feedforward Derating Protection

[0078] The dynamic breakdown voltage threshold is calculated using an input gas discharge model. Based on this, the PWM duty cycle of the high-voltage power supply for the electrostatic purifier is adjusted via feedforward to constrain the operating voltage within a safe range. This model is constructed based on Paschen's law and the operating point position is forcibly verified during initialization.

[0079] The Paschen curve is V-shaped. If the system operating point is on the left side, the temperature rise will lead to an equivalent Decreasing the voltage actually increases the breakdown voltage, causing the "temperature rise-derating" logic to be directionally incorrect. It is essential to verify that the system remains to the right of the minimum value on the Paschen curve.

[0080] Read during power-on self-test and ,calculate Based on Townsend discharge theory, the minimum value of the curve corresponds to:

[0081]

[0082] Only when At that time, the feedforward derating logic is activated.

[0083] During real-time operation, the system corrects the gas state parameters based on the real-time temperature and selects the corresponding input method according to the data format used. If the equivalent gas number density parameter is used, the corrected gas number density parameter and the purifier electrode spacing are input into the gas discharge model together; if the equivalent pressure-spacing product parameter is used, this parameter already contains electrode spacing information and is directly used as the input to the gas discharge model without repeatedly introducing the electrode spacing. The system obtains the dynamic breakdown voltage threshold by using a lookup table generated offline based on the above input index and combining it with linear interpolation. The lookup table values ​​are pre-calculated based on the theoretical relationship of Paschen's Law.

[0084]

[0085] in, and It is a constant related to the dielectric properties of gases. This represents the secondary electron emission coefficient of the electrode surface. In this scheme, when performing gas state correction and gas discharge model calculations, all collected gas flow temperature data are uniformly converted to absolute thermodynamic temperature at the system's underlying layer before calculation to ensure dimensional consistency. The temperature-corrected gas state parameters reflect the trend of decreasing gas density and shrinking insulation margin due to gas flow temperature rise, and serve as the basic input for calculating the dynamic breakdown voltage threshold.

[0086] The calculation of the upper limit of the feedforward PWM duty cycle follows the principle of maintaining the safety margin proportionally:

[0087]

[0088] This embodiment uses temperature-corrected gas state parameters as input conditions for the gas discharge model, providing basic data support for the feedforward derating protection logic. This processing can be achieved by inputting both the equivalent gas number density parameter and the electrode spacing, or by directly inputting the equivalent pressure-spacing product parameter; however, the electrode spacing should not be included in both methods simultaneously.

[0089] The breakdown voltage calculated under ideal conditions of pure air is much higher than the operating voltage of a conventional air purifier (approximately 10kV). However, in the actual operating conditions of a commercial kitchen with heavy oil stains, the non-uniform deposition of grease on the electrode surface leads to a significant reduction in the effective insulation gap and local electric field distortion. To incorporate this key degradation factor of oil accumulation into the dynamic evolution framework of the digital twin, the non-uniform electric field geometric safety margin coefficient in this embodiment... It is not a fixed constant, but an adaptive parameter that automatically decreases as the degree of oil accumulation increases.

[0090] The adaptive reduction is achieved through one or a combination of the following two methods: First, monitoring the average sustaining current of the purifier's high-voltage power supply under constant output voltage; as the oil buildup on the plates thickens, the leakage current on the oil surface increases, and the sustaining current rises accordingly. The system adjusts the current step size downwards based on a pre-calibrated "current-oil buildup degree" mapping relationship. Secondly, based on the cumulative operating time since the most recent electrode cleaning, the value is gradually reduced using a preset decay function (e.g., exponential decay or piecewise linear decay). The value; the parameters of the attenuation function are determined offline by fitting historical maintenance data under similar working conditions.

[0091] when When the attenuation of the gas and the drop in gas density caused by smoldering are coupled, the actual dynamic breakdown voltage threshold will shrink sharply and approach the rated operating voltage of the equipment. Therefore, since feedforward derating control responds to the changing characteristics of gas breakdown, it can effectively limit the high-voltage output of the electrostatic purifier and reduce the risk of breakdown and arcing caused by the superposition of plate oil contamination and gas density drop. When the preset safety lower limit is reached due to continuous decrease, the system will forcibly output an alarm signal, prompting that electrode cleaning and maintenance must be performed, thus forming a complete closed loop of equipment health management of "monitoring-early warning-maintenance".

[0092] (III) Pre-breakdown micro-discharge discrimination based on multi-feature fusion

[0093] The purifier circuit current is collected by a broadband current transformer, and pre-breakdown partial discharge events are extracted by fusing multiple features such as amplitude, pulse width, leading edge steepness, phase stability and narrowband energy, and the discharge frequency per unit time is statistically analyzed.

[0094] In environments with strong electromagnetic interference, switching noise and partial discharge often overlap in the time domain characteristic space, and cannot be effectively distinguished by a single dimension such as amplitude.

[0095] like Figure 2 As shown, the current signal, after being filtered by anti-aliasing and sampled by an ADC of 20 MHz or higher, is sent in parallel to the pure hardware logic engine deployed within the hardware coprocessor to eliminate the bus bandwidth bottleneck of a single microprocessor. Five-fold discrimination is then performed sequentially:

[0096] First, amplitude condition: , Adaptive setting based on the 99.9th percentile of background noise.

[0097] Secondly, pulse width condition: half-peak width .

[0098] Thirdly, steepness conditions: .

[0099] Fourth, phase stabilization and elimination: phase angle within multiple consecutive switching cycles Stable and does not fall within the switching transistor's on / off phase exclusion window. .

[0100] Fifth, frequency domain energy: The Goertzel algorithm is used to extract the narrowband energy of candidate event segments at characteristic frequencies, and confirmation is only made when the energy exceeds the background noise baseline plus a preset threshold. The Goertzel algorithm essentially performs a single-point discrete Fourier transform, and its output amplitude at the target frequency... It directly participates in the fusion decision as the fifth feature variable.

[0101] A valid pre-breakdown partial discharge event is defined as the condition that all five conditions are met. The micro-discharge frequency is obtained by statistical analysis within a 100 ms window. .when If the frequency exceeds the adaptive frequency threshold (the upper quartile of the 1-hour window) continuously for 10 seconds, the micro-discharge abnormality flag will be set.

[0102] The pulse width and phase are determined by interval discrimination rather than point value comparison, tolerating nanosecond-level hardware jitter. The Goertzel algorithm is implemented in 16-bit fixed-point mode with a truncation error of less than 0.1dB.

[0103] (iv) Cross-domain closed-loop verification based on Granger causality test and Spearman rank correlation

[0104] The Spearman rank correlation coefficient between the macroscopic thermal power residual and the discharge frequency is calculated within a sliding time window implemented by a circular queue. When the correlation coefficient continuously meets the grading threshold and duration conditions, derating protection or physical isolation is triggered. This implementation further incorporates a pre-implementation Granger causality test to enhance the physical validity of the decision. Combined with... Figure 3 The process is shown below, and the steps for this part are as follows.

[0105] Correlation is not the same as causation; two unrelated non-stationary series can appear spuriously highly correlated due to a common trend. This should be confirmed before correlation analysis. right Its predictive ability.

[0106] In the microcontroller's memory and Establish circular queues with a depth of N=900 (sliding window of 90 seconds). When the thermal anomaly flag and the micro-discharge anomaly flag are set simultaneously, initiate causal inference.

[0107] First, the macroscopic thermal power residual sequence and the micro-discharge frequency sequence are respectively subjected to first-order difference to stabilize them: , The differential thermal power residual sequence and discharge frequency sequence were obtained, and the model order was determined within the candidate lag order range. After determining the optimal lag order, constrained and unconstrained models for Granger causality testing were constructed based on this optimal lag order.

[0108] Constraint model (no causal assumption):

[0109]

[0110] Unconstrained model (including causal assumptions):

[0111]

[0112] The hysteresis order is not directly determined by the physical distance between the sensors at a single point, nor is it arbitrarily determined by information criteria within a global scope. Instead, it is determined using a "statistical optimization within physical constraints" method. The system first estimates the physical hysteresis time of heat transfer based on the average airflow velocity and the physical spatial distance between the two sensors, and then converts this physical hysteresis time into a central value for the physical hysteresis order. Considering velocity measurement errors, flow field non-uniformity, and sensor response delays, the system sets a hysteresis time tolerance to obtain a candidate hysteresis time range, which is then converted into a candidate hysteresis order range based on the control cycle or sampling cycle. Subsequently, the system establishes time-series regression models for order comparison only within the candidate hysteresis order range, and performs model screening based on at least one of the Akaike information criterion, Bayesian information criterion, and residual white noise test to determine the optimal hysteresis order. If multiple candidate lag orders meet the model diagnostic requirements, the candidate lag order with the smallest deviation from the central value of the physical lag order and the lowest model complexity is selected first. If none of the candidate lag orders meet the model diagnostic requirements, the Granger causality confirmation process is not initiated; only a warning is output or the degradation protection is maintained. The system establishes a constrained and unconstrained model for Granger causality testing based on the optimal lag order, and determines whether the thermal power residual sequence can significantly improve the predictive ability of the micro-discharge frequency sequence through a joint significance test of the regression terms. After the Granger causality test is passed, the system shifts and aligns the thermal power residual sequence and the micro-discharge frequency sequence according to the time offset corresponding to the optimal lag order, and then calculates the Spearman rank correlation coefficient.

[0113] calculate Statistic:

[0114]

[0115] If the test shows that the combined hysteresis terms of the thermal power residuals are significant, then reject the test. no The null hypothesis of the Granger cause.

[0116] Calculate the Spearman rank correlation coefficient only if the Granger causality test passes:

[0117]

[0118]

[0119] The nonparametric properties of the Spearman coefficients accommodate the possible nonlinear monotonic relationships.

[0120] Tiered response decision-making ( Figure 3 The lower half executes the following logic:

[0121] Level 1 (Warning): Single channel anomaly or Granger test failure, or Issue early warnings.

[0122] Level 2 (Alarm + Reduction): Dual-channel anomaly, Granger passes, and continued Periodic activation of feedforward derating, voltage clamping to .

[0123] Level 3 (Physical Isolation): continued When the periodic or instantaneous high voltage is extremely high, the high voltage power supply of the purifier is shut off to eliminate the electrical ignition source, the fire damper in the pipeline is driven to perform physical isolation, and a linkage control request for the fan, fire damper and fixed fire extinguishing device is sent to the fire linkage control unit.

[0124] Degrades to relying solely on data when there is insufficient data. The downgrade decision is made. The grading threshold utilizes periods without abnormalities. Fluctuation statistics periodic calibration.

[0125] Example 3 Verification Test

[0126] (I) Underground observation equipment and simulation data generation

[0127] A hardware-in-the-loop simulation platform with a realistic exhaust pipe network scale was used. A baseline mass flow rate consistent with typical high-volume exhaust conditions in commercial kitchens was set. During testing, a concealed heat source was continuously injected into the pipes to test the sensitivity of early smoldering detection under high-volume, strong convection dilution effects. Simultaneously, a decay function was introduced to simulate the grease buildup process, ensuring a realistic overall heat transfer coefficient for the pipe walls. Dynamic attenuation occurs. In commercial kitchen data acquisition, there are objectively existing thermoelectric potential fluctuations, micro-pressure measurement jitter, power frequency and harmonic crosstalk, and random electromagnetic pulse interference. The system's built-in Kalman filter and multi-feature extraction algorithm can effectively smooth and suppress these multi-source physical noises, ensuring that the data input to the digital twin model has high fidelity and strong engineering robustness.

[0128] (II) Data Preprocessing Link

[0129] Temperature data , and The differential pressure signal is filtered by a second-order Butterworth digital low-pass filter with a cutoff frequency of 1Hz. Filtered by a 5-point moving average. The online estimation is performed within the steady-state window identified by the system as having no abnormal heat sources. The broadband current is processed in real-time by the hardware pulse feature extraction engine, and the micro-discharge discrimination pre-calibration parameters are: , , Phase stability window tolerance Switch tube exclusion window The Goertzel algorithm monitors a frequency of 500kHz and sets the energy threshold to the background noise spectral density plus 12dB.

[0130] (III) Core Algorithm Calculation and Trajectory Derivation

[0131] Under the strong convection dilution effect of typical commercial high-volume smoke exhaust networks, the absolute temperature rise at a single point is often masked by environmental noise. This invention selects three typical stages of smoldering development for trajectory derivation verification:

[0132] In the early stage of smoldering (taking a large exhaust volume of 8 m³ / s and a simulated introduction of a concealed heat source of approximately 500 W as an example): a concealed heat source is generated within the controlled pipe section. Due to the dilution effect of the large air volume, the absolute temperature rise at the outlet is extremely weak (the measured temperature rise is approximately 0.5℃), and the traditional fixed threshold alarm set at 60℃ cannot respond. However, after online stable estimation of the pipe wall heat transfer coefficient through Kalman filtering and subtraction of the known fan heat injection, the macroscopic thermal power residual calculated by the system first deviates significantly from the adaptive noise baseline after about 15 seconds (e.g., it jumps and stabilizes above 150 W, exceeding the standard deviation of the background), and the residual continues to exceed the limit, triggering the thermal power anomaly flag. At this time, the gas density does not drop significantly, the micro-discharge frequency remains at the background noise level (less than 2 times / second), and the system outputs a first-level warning.

[0133] During the smoldering development stage: the power of the concealed heat source continues to increase (e.g., to over 2.5kW), causing an observable rise in airflow temperature that follows a proportional linear law (e.g., a measured temperature rise of approximately 2.5℃), resulting in a slight but substantial drop in gas density. Under this condition, due to severe oil accumulation on the purifier plates from long-term operation, the geometric safety margin of the non-uniform electric field has significantly decreased, and the system's basic breakdown threshold is already extremely close to the danger zone of the equipment's operating voltage (e.g., it has decreased to around 10.8kV due to oil accumulation and degradation). At this point, the gas density drop caused by this tiny 2.5℃ temperature rise directly becomes the final straw that breaks the insulation balance. Under the deep superposition of the dual physical effects of "oil accumulation decay" and "gas rarefaction," the discharge model successfully predicted a penetrating contraction trend in the dynamic breakdown voltage threshold (e.g., the calculated comprehensive dynamic breakdown threshold eventually fell below the 10.5kV safety threshold). The feedforward derating mechanism was activated to actively reduce the purifier's operating voltage. Simultaneously, the multi-feature extraction logic captured real pre-breakdown partial discharge events, with the micro-discharge frequency significantly exceeding background noise (e.g., jumping to over 15 times / second). In the causal inference stage, the Granger causality test obtained by the system was... The value was strictly less than the preset significance level (measured p<0.01), confirming that the thermal power residual has a unidirectional causal relationship with the micro-discharge frequency; the Spearman rank correlation coefficient then steadily crossed the alarm threshold (e.g., ρs rose to 0.82), triggering a level 2 alarm.

[0134] (iv) Demonstration of physical boundary constraints

[0135] Throughout the entire evolution cycle, the evolution trend of the calculated macroscopic thermal power residuals remained highly consistent with the injected hidden heat sources, verifying the engineering credibility of the lumped thermal balance twin model. The feedforward derating logic followed the monotonic region of the Paschen curve and did not exhibit any false triggering. The abnormal micro-discharge frequency occurred after the surge in thermal power residuals, a phenomenon consistent with the sequential evolution of electrical insulation degradation caused by thermodynamic anomalies within the flue gas duct. Furthermore, the F-statistic of the Granger causality test demonstrated good statistical discrimination in the actual causal evolution stage, effectively eliminating spurious correlations caused by random electromagnetic interference or normal thermal fluctuations, verifying the rigor of the model's boundary constraints.

[0136] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for dynamic simulation and control of catering pollution sources based on digital twins, characterized in that, include: A digital twin of the smoke exhaust duct is constructed, which includes a lumped parameter thermal balance model and a gas discharge model based on Paschen's law; The system collects inlet airflow temperature, outlet airflow temperature, ambient temperature, airflow velocity, fan current, and purifier circuit current waveforms, drives the lumped parameter thermal balance model to calculate the macroscopic thermal power residual, corrects the gas state parameters according to the real-time temperature, and inputs the gas discharge model to calculate the dynamic breakdown voltage threshold. Based on the dynamic breakdown voltage threshold, the pulse width modulation duty cycle of the high-voltage power supply of the electrostatic purifier is adjusted forward. Extract pre-breakdown partial discharge events and count the discharge frequency; Within the sliding time window, a statistical correlation analysis is performed on the macroscopic thermal power residual and the discharge frequency, and early warning, derating protection or isolation control is triggered based on the analysis results.

2. The method according to claim 1, characterized in that, The calculation of the macroscopic thermal power residual includes: calculating the current airflow density based on the real-time temperature and preset or measured pressure using the ideal gas state equation under isobaric approximation, and obtaining the mass flow rate based on the current airflow density, pipe cross-sectional area and airflow velocity; The macroscopic thermal power is calculated by subtracting the fan heat injection power calculated from the fan current and adding the pipe wall heat dissipation power calculated by online adaptive estimation of the pipe wall comprehensive heat transfer coefficient through Kalman filtering.

3. The method according to claim 1, characterized in that, The calculation of the dynamic breakdown voltage threshold includes: The gas state parameters are corrected based on the real-time temperature to obtain the equivalent gas number density parameter, and the equivalent gas number density parameter and the purifier electrode spacing are substituted into the Paschen's law expression; or, the equivalent pressure-spacing product parameter is obtained based on the real-time temperature and the purifier electrode spacing, and the equivalent pressure-spacing product parameter is substituted into the Paschen's law expression; then, combined with the non-uniform electric field geometric safety margin coefficient, the dynamic breakdown voltage threshold under the current gas state is obtained; Specifically, during initialization, the gas discharge model ensures that the equivalent pressure-spacing product corresponding to the initial operating point is greater than the product corresponding to the minimum value of the Paschen curve, so that the system operating point is located in the region to the right of the minimum value of the Paschen curve. This ensures that when the temperature increases and the equivalent pressure-spacing product decreases, the dynamic breakdown voltage threshold shows a decreasing trend.

4. The method according to claim 3, characterized in that, The value of the non-uniform electric field geometric safety margin coefficient automatically decreases as the degree of oil accumulation on the purifier electrodes increases, in order to correct the impact of the shortened equivalent insulation distance caused by oil adhesion on the breakdown voltage threshold.

5. The method according to claim 1, characterized in that, The extraction of pre-breakdown partial discharge events includes: Transient events whose current spike amplitude exceeds the first threshold, whose pulse width falls within a preset range, and whose leading edge rise rate exceeds the second threshold are identified as candidate discharge events. For the candidate discharge event, verify whether the phase angle of its occurrence time relative to the turn-on and turn-off times of the switch is stable in multiple consecutive switching cycles, and actively exclude candidate events that fall within the inherent operating phase exclusion window of the switch. The candidate events that pass the verification are counted as one valid pre-breakdown partial discharge event.

6. The method according to claim 1, characterized in that, The statistical association analysis includes: Establish a bidirectional circular queue with a depth of N, and push the macroscopic thermal power residual and discharge frequency data into their respective queues in each control cycle; When both the thermal power anomaly and micro-discharge anomaly flags are valid, the average airflow velocity and the physical distance between the sensors are extracted. The physical lag time of heat transfer is calculated based on the two, and the physical lag time of heat transfer is converted into a candidate lag order range in combination with the control cycle or sampling cycle. Within the candidate lag order range, the physical lag order determined by the average airflow velocity and the physical distance between the sensors is used as the constraint center. For each candidate lag order, a time-series regression model is established for order comparison. The optimal lag order is determined by screening models based on at least one of the Akaike information criterion, the Bayesian information criterion, and the residual white noise test. The optimal lag order is selected only from the candidate lag order range to simultaneously satisfy the physical transfer time constraint and model diagnostic requirements. Based on the optimal lag order, a constrained and unconstrained model for Granger causality testing is established. The joint significance test of the regression terms is used to determine whether the macroscopic thermal power residual is a Granger cause of the discharge frequency. After confirming the existence of a Granger causal relationship, the time-series data is shifted and aligned according to the time offset corresponding to the optimal lag order, and the Spearman rank correlation coefficient is calculated.

7. The method according to claim 6, characterized in that, The triggering of early warnings, throttling protection, or isolation controls based on analysis results includes: When only the thermal power residual is abnormal, or when the Spearman rank correlation coefficient is lower than the first threshold, an early warning signal is output; When the Spearman rank correlation coefficient is between the first threshold and the second threshold and remains between them for a first duration, derating protection is activated, and the operating voltage is limited to a preset percentage of the dynamic breakdown voltage threshold. When the Spearman rank correlation coefficient exceeds the second threshold and lasts for a second duration, or momentarily exceeds the third threshold, an isolation control signal is output. The isolation control signal is used to shut off the high-voltage power supply of the electrostatic purifier and send a linkage request to the fire linkage control unit for the fan, fire damper, or fire extinguishing device.

8. The method according to claim 1, characterized in that, After calculating the macroscopic thermal power residual, the method further includes: Within a time window without abnormal steady-state conditions, the baseline fluctuation standard deviation of the macroscopic thermal power residual data is extracted, and an adaptive noise base threshold is dynamically configured accordingly. Only when the absolute value of the macroscopic thermal power residual calculated in real time is greater than the adaptive noise base threshold in multiple consecutive sampling periods, or when the positive value of the macroscopic thermal power residual calculated in real time is higher than the adaptive noise base threshold in multiple consecutive sampling periods, is the thermal power abnormality confirmed, and subsequent feedforward adjustment and statistical correlation analysis processes are triggered.

9. A dynamic simulation and control system for catering pollution sources based on digital twins, used to execute the method as described in any one of claims 1 to 8, characterized in that, The system is used to dynamically simulate and control the complex pollution sources in catering exhaust systems, which are composed of oil deposits, concealed heat sources, pre-breakdown micro-discharge, and arc ignition risks. The system includes: The digital twin model building module is used to create a controlled pipeline digital twin that includes a lumped parameter thermal balance model and a Paschen gas discharge model; The data acquisition module is used to acquire inlet airflow temperature, outlet airflow temperature, ambient temperature, airflow velocity, fan current, and purifier circuit current waveforms. Thermodynamic state estimation unit is used to drive the lumped parameter thermal balance model based on the collected data and output the macroscopic thermal power residual. The gas discharge safety margin management unit is used to correct the gas state parameters according to the temperature and calculate the dynamic breakdown voltage threshold by the Paschen gas discharge model, thereby feeding forward to adjust the duty cycle of the high voltage power supply of the purifier. The electromagnetic transient monitoring unit is used to extract and statistically analyze the frequency of pre-breakdown partial discharge events from the current waveform of the purifier circuit. The macro-micro cross-domain time-series causal inference unit is used to perform Granger causality tests on thermal power residuals and discharge frequency within a sliding time window, calculate Spearman rank correlation coefficients, and output control commands based on hierarchical logic.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images