A method and system for intelligent monitoring and quantification of emissions based on tank pressure systems

Abstract

This invention discloses an intelligent monitoring and emission quantification method and system based on a storage tank pressure system, relating to the field of environmental monitoring and control technology. The method includes acquiring the pressure and temperature sequences corresponding to a single breathing event; forming a monitoring input sequence based on the pressure and temperature sequences and the effective time series length of the single breathing event; inputting the monitoring input sequence into a preset probabilistic time series emission quantification model to form an emission distribution representation of the single breathing event, and outputting the mean and variance parameters of the emission amount corresponding to the emission distribution representation. The method of this invention quantifies emissions from a single breathing event based solely on pressure and temperature, adapts to variable-length time series inputs, and outputs the mean and variance parameters of the emission amount.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and control technology, specifically to an intelligent monitoring and emission quantification method and system based on a storage tank pressure system. Background Technology

[0002] Currently, there are still many technical limitations in the monitoring and quantification of breather valve emissions. The problems mainly focus on sensor deployment methods, data acquisition conditions, and dynamic process characterization capabilities. To improve monitoring accuracy, existing solutions typically deploy flow meters, analyzers, and various auxiliary sensors around the breather valve or storage tank to simultaneously collect multi-dimensional parameters such as flow rate, pressure, temperature, composition, or valve position. However, such solutions are highly dependent on hardware conditions, resulting in high equipment costs, complex installation and maintenance, and high requirements for on-site power supply conditions, explosion-proof ratings, structural modification space, and environmental adaptability. Especially for numerous and dispersed in-service storage tanks, continuous monitoring using multiple types and quantities of sensors is often only easily achieved under pilot-scale platforms or laboratory conditions. In actual storage tank sites, it is often difficult to promote and apply due to factors such as limited power supply, strict explosion-proof requirements, difficult wiring, complex long-term exposure environments, and limited maintenance frequency. As a result, existing high-dimensional sensor monitoring solutions lack feasibility in most storage tank scenarios.

[0003] Besides limitations in hardware deployment, existing emission calculation methods also have shortcomings in data processing and result representation. One type of method mainly relies on empirical coefficients or steady-state physical calculations, which can only provide estimation results in the sense of annual, quarterly or operating condition averages, and is difficult to reflect the transient emission process corresponding to a single breathing event. Another type of method introduces data-driven models, but mostly uses point estimation to output a single predicted value, lacking the characterization of the degree of dispersion and fluctuation range of the prediction, and is difficult to adapt to the needs of risk assessment and state identification.

[0004] In terms of time series data processing, since the opening duration of the breathing valve is significantly inconsistent and different samples often have variable length characteristics, traditional truncation, splicing or simple alignment methods are prone to destroying the original dynamic process, causing the loss of effective information, which in turn affects the stability and reliability of subsequent calculation results. Relying solely on empirical models is difficult to adapt to complex working conditions, while relying solely on black-box data models lacks clear physical constraints. As a result, existing technologies generally suffer from insufficient field adaptability, inadequate utilization of dynamic processes and weak ability to interpret results. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is that existing technologies for monitoring the emission of breathing valves rely on complex hardware equipment and high maintenance costs, resulting in cumbersome sensor configuration and high equipment failure rates. In terms of data processing, traditional methods use point estimation, which lacks effective quantification of uncertainty, leading to low accuracy and reliability of prediction results. Furthermore, traditional methods cannot adapt to dynamic changes in different working conditions, resulting in slow response speed and poor flexibility. This invention aims to solve the problems of excessive hardware dependence, insufficient data processing accuracy, and poor dynamic adaptability in traditional methods by simplifying hardware configuration and adopting minimalist sensors and deep learning models.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent monitoring and emission quantification method based on a tank pressure system, comprising: acquiring the pressure sequence and temperature sequence corresponding to a single breathing event; forming a monitoring input sequence based on the pressure sequence and temperature sequence, combined with the effective time series length of the single breathing event; inputting the monitoring input sequence into a preset probabilistic time series emission quantification model to form an emission distribution representation of the single breathing event, and outputting the mean and variance parameters of the emission amount corresponding to the emission distribution representation.

[0008] As a preferred embodiment of the intelligent monitoring and emission quantification method based on the tank pressure system described in this invention, the single breathing event includes: acquiring the breather valve position height sequence, determining the opening range of the breather valve based on the breather valve position height sequence, and determining the pressure sequence and temperature sequence corresponding to the opening range as the pressure sequence and temperature sequence corresponding to the single breathing event.

[0009] As a preferred embodiment of the intelligent monitoring and emission quantification method based on the tank pressure system described in this invention, the breather valve position height sequence includes real-time monitoring of the breather valve's opening displacement by a non-contact infrared displacement sensor disposed above the breather valve core.

[0010] As a preferred embodiment of the intelligent monitoring and emission quantification method based on the tank pressure system described in this invention, the monitoring input sequence includes: performing Z-score normalization on the pressure sequence and temperature sequence respectively, filling short sequences with a preset constant to a uniform length, and generating a binary mask matrix based on the original effective length.

[0011] As a preferred embodiment of the intelligent monitoring emission quantification method based on the tank pressure system described in this invention, the emission label corresponding to the preset probabilistic time-series emission quantification model is the total emission volume of a single breathing event; the total emission volume is the integral result of the volume emission flow rate during the valve opening phase; the probabilistic time-series emission quantification model is a probabilistic one-dimensional convolutional neural network model; the emission label includes: treating the gas inside the tank as a control volume, determining the gas density inside the tank based on the pressure and temperature sequences within the opening interval, and characterizing the gas state changes inside the tank during the single breathing event by combining the mass balance equation and the ideal gas state equation; The mass inflow rate is determined based on the volumetric flow meter reading, and the gas density at the inlet is taken to be equal to the instantaneous density inside the tank. The mass inflow rate is then substituted into the mass balance equation to solve for the mass discharge rate. The volumetric flow meter reading is monitored by a thermal mass flow meter installed between the centrifugal fan and the storage tank, and the volumetric inflow rate output by the thermal mass flow meter is synchronously collected by a programmable logic controller at a fixed frequency to form an intake flow sequence. The mass discharge rate is divided by the instantaneous gas density inside the tank to obtain the volumetric discharge flow rate during a single breathing event. The total discharge volume is obtained by integrating the volumetric discharge flow rate over time using the trapezoidal rule within the opening interval.

[0012] As a preferred embodiment of the intelligent monitoring emission quantification method based on the tank pressure system described in this invention, the method comprises: inputting the monitoring input sequence into a probabilistic time-series emission quantification model; the probabilistic time-series emission quantification model directly establishing a mapping relationship between the pressure sequence and temperature sequence within the opening interval and the emission label; performing convolution operations on the pressure sequence and temperature sequence through a one-dimensional convolutional layer to extract the pressure rise slope feature and pressure-temperature coupling feature, and outputting the mean emission amount and logarithmic variance parameters based on the pressure rise slope feature and the pressure-temperature coupling feature; constructing a sample set based on operating condition samples, and extracting the corresponding time-series feature patterns under different operating conditions based on the sample set; the operating condition samples include fixed-frequency injection operating condition samples, fluctuating injection operating condition samples, and mixed air volume operating condition samples; using the same model structure for different operating conditions, feature extraction is performed on the pressure sequence and temperature sequence, and the mean emission amount and logarithmic variance parameters corresponding to a single breathing event are output; the time-series feature patterns include high-frequency oscillation features under fluctuating injection operating conditions, continuous opening features under continuous injection operating conditions, and variable-length time-series features corresponding to different opening durations.

[0013] As a preferred embodiment of the intelligent monitoring emission quantification method based on a tank pressure system described in this invention, the probabilistic time-series emission quantification model is trained using a Gaussian negative log-likelihood loss function to jointly constrain the mean and log-variance, thereby constructing a confidence interval; the Gaussian negative log-likelihood loss function is expressed as: , in, This represents the loss value of the Gaussian negative log-likelihood loss function. This indicates the number of samples used in the loss calculation. Indicates the sample index. Indicates the first The prediction variance for each sample Indicates the first The actual emissions of each sample Indicates the first Predicted emissions for each sample.

[0014] Another objective of this invention is to provide an intelligent monitoring and emission quantification system based on a tank pressure system. This system solves the problems of high hardware costs and complex installation and maintenance in current traditional technologies by adopting a minimalist sensor configuration, requiring only two conventional sensors: pressure and temperature. By simplifying the hardware configuration and combining it with a deep learning model for data processing, the system avoids the dependence on precision flow meters and multiple complex sensors found in traditional methods. This reduces the overall system operating costs and failure rate, and improves the system's flexibility and adaptability.

[0015] As a preferred embodiment of the intelligent monitoring and emission quantification system based on a storage tank pressure system according to the present invention, the system includes: an event acquisition module, a characterization construction module, and a quantization output module. The event acquisition module acquires the breather valve position height sequence using a non-contact infrared displacement sensor positioned above the breather valve core, determines the breather valve opening interval based on the breather valve position height sequence, and identifies the pressure and temperature sequences corresponding to the opening interval as the pressure and temperature sequences corresponding to a single breathing event. The characterization construction module treats the gas inside the storage tank as a control volume, determines the volumetric emission flow rate based on the pressure and temperature sequences within the opening interval, combined with the mass balance equation and the ideal gas law equation, and performs time integration on the volumetric emission flow rate using the trapezoidal rule within the opening interval to obtain the total emission volume. The pressure and temperature sequences are then subjected to Z-score standardization, uniform length padding, and binary mask matrix generation based on the original effective length to form a monitoring input sequence. The quantization output module inputs the monitoring input sequence into a probabilistic time-series emission quantification model, uses a Gaussian negative log-likelihood loss function to jointly constrain the mean and log-variance, and outputs the mean and log-variance parameters of the emission amount.

[0016] Another object of the present invention is to provide an intelligent monitoring and emission quantification device based on a tank pressure system, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of an intelligent monitoring and emission quantification method based on a tank pressure system.

[0017] Another object of the present invention is to provide an intelligent monitoring emission quantification storage medium based on a tank pressure system, wherein a computer program is stored thereon, and when the computer program is executed by a processor, the steps of the intelligent monitoring emission quantification method based on the tank pressure system are implemented.

[0018] The beneficial effects of this invention are as follows: The intelligent monitoring emission quantification method based on tank pressure system provided by this invention accurately defines the opening interval of each breathing event, calculates emission tags based on pressure and temperature sequences, and uses a deep learning model for data adaptation and prediction. This invention achieves significant results in reducing hardware configuration costs, improving data processing accuracy, and enhancing model adaptability. The step of defining the opening interval ensures accurate quantification of emissions and avoids data errors in traditional methods. The emission tags generated based on the physical model improve the reliability of the calculation and avoid the black box problem of data-driven methods. The application of the deep learning model not only improves computational efficiency and prediction accuracy but also enables real-time response to dynamic changes and adaptation to edge computing deployment. In summary, this invention effectively reduces costs, improves system flexibility and real-time performance, and provides more reliable technical support, especially in environmental compliance and equipment health management. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The above is an overall flowchart of an intelligent monitoring and emission quantification method based on a storage tank pressure system provided in Embodiment 1 of the present invention.

[0021] Figure 2 The location diagram of a non-contact infrared displacement sensor for an intelligent monitoring and emission quantification method based on a storage tank pressure system provided in Embodiment 1 of the present invention.

[0022] Figure 3 This is a continuous feed linear graph for an intelligent monitoring and emission quantification method based on a storage tank pressure system provided in Embodiment 1 of the present invention.

[0023] Figure 4 This is a discontinuous feed linear graph for an intelligent monitoring and emission quantification method based on a storage tank pressure system provided in Embodiment 1 of the present invention.

[0024] Figure 5This is a flowchart of the training and verification process for a probabilistic time-series emission quantification model of an intelligent monitoring emission quantification method based on a storage tank pressure system, provided in Embodiment 2 of the present invention.

[0025] Figure 6 This is a network structure diagram of a probabilistic time-series emission quantification model for an intelligent monitoring emission quantification method based on a storage tank pressure system, provided in Embodiment 2 of the present invention.

[0026] Figure 7(a) is a graph showing the changes in model training loss and validation loss of an intelligent monitoring emission quantification method based on a storage tank pressure system provided in Embodiment 3 of the present invention.

[0027] Figure 7(b) is a graph showing the root mean square error of a model for an intelligent monitoring and emission quantification method based on a tank pressure system provided in Embodiment 3 of the present invention.

[0028] Figure 7(c) shows the emission prediction range coverage of an intelligent monitoring emission quantification method based on a tank pressure system provided in Embodiment 3 of the present invention.

[0029] Figure 7(d) is a scatter plot showing the correlation between the predicted and actual emission values ​​of an intelligent monitoring emission quantification method based on a tank pressure system provided in Embodiment 3 of the present invention. Detailed Implementation

[0030] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0031] Example 1, referring to Figures 1-4 As an embodiment of the present invention, a method for intelligent monitoring and emission quantification based on a storage tank pressure system is provided, comprising: S1: Obtain the pressure sequence 101 and temperature sequence 102 corresponding to a single respiratory event.

[0032] Furthermore, a single breathing event includes acquiring the valve position height sequence of the breathing valve, determining the opening range 100 of the breathing valve based on the valve position height sequence, and determining the pressure sequence 101 and temperature sequence 102 corresponding to the opening range 100 as the pressure sequence 101 and temperature sequence 102 corresponding to the single breathing event.

[0033] The breathing valve position height sequence includes real-time monitoring of the opening displacement of the breathing valve by a non-contact infrared displacement sensor F located above the breathing valve core.

[0034] Among them, pressure sequence 101 refers to the data sequence formed by continuously recording the internal pressure of the storage tank by the pressure sensor and arranging it in the order of sampling time; temperature sequence 102 refers to the data sequence formed by continuously recording the internal temperature of the storage tank by the temperature sensor and arranging it in the order of sampling time.

[0035] It should be noted that the effective period of valve opening for the current single breathing event is determined by the valve opening displacement corresponding to the valve position height sequence of the breathing valve. The internal pressure and temperature are continuously recorded. The recording of a single breathing event is determined by the butterfly valve opening and closing signal. During the process of the fan supplying air, the butterfly valve opens synchronously, the pressure inside the tank increases, the breathing valve does not open, and the height sensor does not display a reading. That is, the pressure and temperature data are recorded at the same time as the butterfly valve opens, and the data is stopped after a 3-second delay after the butterfly valve closes. This process completes the recording of the effective breathing valve emission sample.

[0036] It should also be noted that, such as Figure 2 As shown, a non-contact infrared displacement sensor F is installed above the valve core to monitor the valve displacement data in real time. When the detected opening height is greater than or equal to a preset height threshold, the preset height threshold is determined based on the minimum detectable displacement corresponding to the valve core leaving the seat, and is set to calculate the emission amount of the current data point in the future.

[0037] It should also be noted that, firstly, an experimental platform including a storage tank, a breather valve, an air intake system, and a sensor network was constructed to simulate the operation of the breather valve under different feeding conditions and to synchronously collect the operating parameters during the opening process of the breather valve. The experimental platform was uniformly controlled and data collected by a programmable logic controller (PLC). The PLC synchronously reads the signals of each channel at a preset fixed frequency to ensure that pressure, temperature, air intake flow, breather valve position height, and butterfly valve status form corresponding data sequences at the same time reference, thereby providing basic data for the subsequent delineation of the opening range 100 of a single breathing event and the construction of the emission label 200.

[0038] The opening of the breather valve is driven entirely by the pressure difference between the inside and outside of the tank (the set opening pressure is about 12 mbar). The pressure-time curve directly records the complete dynamic process of "valve opening → discharge → pressure drop → valve closing". The pressure change rate (dP / dt) has a direct physical coupling relationship with the discharge flow rate. According to the ideal gas law, gas density and temperature directly affect the density term in the mass conservation calculation. In actual working conditions, the diurnal temperature difference, solar radiation, and changes in feed temperature will cause the gas inside the tank to expand and contract thermally. If temperature is ignored and only pressure is used, the estimation error can reach more than 30%.

[0039] The valve position sensor is only used for event triggering (determining the opening / closing time). Although its displacement data has a physical mapping relationship with the emission, it requires complex physical modeling. Establishing the relationship between the emission and the displacement of the breather valve core is very difficult. In short, pressure and temperature are the minimum sufficient statistics to describe the "thermodynamic state of the gas inside the tank". They are sufficient to infer the emission through the law of conservation of mass, which is in line with the technical design principle of "minimalist sensor".

[0040] Specifically, the experimental platform uses a 5m³ steel storage tank as its core. The top and sides of the tank are reserved for the installation of breather valves to accommodate the installation and testing of breather valves of different sizes. Temperature and pressure sensors are installed at the connection between the breather valve and the tank to continuously record the internal temperature and pressure of the tank. The temperature sensor has an accuracy of ±0.5℃, and the pressure sensor has a range of 0 to 10 kPa. The temperature and pressure data collected by the temperature and pressure sensors constitute the main input data of the probabilistic time-series emission quantification model 300, which is used to characterize the dynamic changes in the internal thermal state of the tank during the opening of the breather valve.

[0041] To obtain information about the opening and closing process of the breathing valve, a non-contact infrared displacement sensor F is installed above the valve core to monitor the valve opening displacement in real time and form a breathing valve position height sequence. The structural measurement range of the non-contact infrared displacement sensor F is defined as 0 to 35 mm, which can continuously reflect the displacement change process of the valve core from the closed state to the open state and from the open state back to the closed state. Based on the breathing valve position height sequence, the opening interval 100 corresponding to a single breathing event can be further defined, so that the subsequent emission calculation is limited to the actual opening stage of the valve.

[0042] During the data acquisition phase, the programmable logic controller synchronously acquires the internal pressure, internal temperature, airflow, breather valve position height, and butterfly valve status signal of the storage tank at a fixed frequency. Since all parameters are recorded synchronously under a unified clock, it can ensure that the pressure, temperature, flow rate, and valve position height values ​​at the same sampling time correspond one-to-one. Based on this synchronous acquisition mechanism, not only can the opening interval 100 of a single breathing event be accurately extracted, but also a complete pressure sequence 101, temperature sequence 102, and airflow sequence can be formed within the opening interval 100. This provides a unified data foundation for calculating the total emission volume and constructing the input data for the subsequent probabilistic time-series emission quantification model 300.

[0043] S2: Based on the pressure sequence 101 and temperature sequence 102, combined with the effective time length of a single respiratory event, a monitoring input sequence 201 is formed.

[0044] Furthermore, the emission label 200 includes treating the gas inside the storage tank as a control volume, determining the gas density inside the storage tank based on the pressure sequence 101 and temperature sequence 102 within the opening interval 100, and characterizing the gas state changes inside the storage tank during a single breathing event by combining the mass balance equation and the ideal gas state equation.

[0045] The mass inflow rate is determined based on the volumetric flow meter reading, and the gas density at the inlet is taken to be equal to the instantaneous density inside the tank. The mass inflow rate is then substituted into the mass balance equation to solve for the mass discharge rate.

[0046] The mass balance equation is expressed as: ,

[0047] in, Indicates the mass of the gas inside the storage tank. This indicates the mass inflow rate from the fan. This indicates the mass emission rate through the pilot-operated breather valve (PRVR).

[0048] Gas mass, expressed as: , in, Indicates the mass of the gas. Indicates gas density, This indicates a constant tank volume.

[0049] Assuming air behaves like an ideal gas, its density can be calculated from the measured pressure (P) and temperature (T), expressed as: , The mass inflow rate is derived from the volumetric flow meter reading: , in, Indicates the density of the inlet gas. This indicates the intake airflow rate.

[0050] Substituting the above into the mass balance equation and solving for the mass emission rate, we get: , in, Indicates time mass emission rate, The instantaneous density inside the tank. Indicates time The rate of change of instantaneous gas density inside the tank with respect to time.

[0051] Divide the mass emission rate by the instantaneous gas density: , in, Indicates time Volumetric emission flow rate.

[0052] During the valve opening phase of the sample, the total discharge volume is obtained by integrating the volumetric discharge flow rate over time: , in, Indicates the total emission volume. This indicates the start time of the sample valve opening phase. The end time of the sample valve opening phase.

[0053] The volumetric flow meter readings are monitored by a thermal mass flow meter installed between the centrifugal fan and the storage tank. The programmable logic controller synchronously collects the volumetric inflow rate output by the thermal mass flow meter at a fixed frequency to form an intake flow sequence. The mass discharge rate is divided by the instantaneous gas density in the tank to obtain the volumetric discharge flow rate during a single breathing event. The total discharge volume is obtained by integrating the volumetric discharge flow rate over time using the trapezoidal rule within the open interval of 100.

[0054] Regarding the air intake system, a centrifugal fan is connected to the bottom of the storage tank via a pipeline to provide an adjustable air intake to the tank. A thermal mass flow meter is connected in series between the centrifugal fan and the storage tank to measure the air intake volume inflow rate. The thermal mass flow meter has a measurement accuracy of ±0.1 m³ / h. During the experiment, the air intake flow data output by the thermal mass flow meter is recorded as an air intake flow sequence. This air intake flow sequence is used to derive the mass inflow rate during the subsequent emission label 200 generation process, rather than as a direct input to the probabilistic time-series emission quantification model 300. By associating the air intake flow sequence with the pressure and temperature sequences 102, a mass balance relationship can be established within the opening interval 100 corresponding to a single breathing event.

[0055] To construct experimental conditions under different operating states, the experimental platform is also equipped with a pneumatic butterfly valve controlled by a switch signal. The pneumatic butterfly valve opens and closes synchronously with the centrifugal fan to establish the switching process between pressurized and normal pressure conditions on the storage tank. When the pneumatic butterfly valve is open and runs in conjunction with the centrifugal fan, the storage tank enters the pressurization process; when the pneumatic butterfly valve is closed, the storage tank stops air intake or enters a pressure holding state. By controlling the fan status, various feeding processes under different operating conditions can be simulated to obtain experimental data under different opening durations, different pressure change amplitudes, and different air intake flow fluctuation characteristics.

[0056] The monitoring input sequence 201 includes performing Z-score normalization on the pressure sequence 101 and the temperature sequence 102 respectively, filling the short sequence with a preset constant to a uniform length, and generating a binary mask matrix based on the original effective length.

[0057] It should be noted that when the breathing valve opening time is different, the pressure sequence 101 and temperature sequence 102 are Z-score standardized respectively, the short sequences are padded to a uniform length by a constant, and a binary mask matrix is ​​generated based on the original effective length. The pressure sequence 101 and temperature sequence 102 constitute the input parameter set characterizing the thermodynamic state of the gas in the tank, and the mean and logarithmic variance parameters of the emission amount of a single breathing event are output based on the correspondence between the input parameter set and the emission label 200. The binary mask matrix is ​​applied in the global pooling layer of the probabilistic time-series emission quantification model 300 to perform mask-weighted averaging on the temporal dimension of the convolutional feature map, so that the pooling result depends only on the effective data area and eliminates the interference of the padded area.

[0058] Because the duration of the breathing valve opening varies under different operating conditions, each single breathing event sample naturally exhibits a variable length characteristic. Specifically, the lengths of the pressure sequence 101 and temperature sequence 102 corresponding to different samples are not entirely consistent. This difference stems from the variation in the actual duration of the breathing valve opening process, rather than from length changes caused by artificial truncation or resampling. To adapt to the requirement of the deep learning model for a fixed shape of the input tensor, while preserving the original dynamic information of the variable-length sequences, a padding mask adaptation is performed. The pressure sequence 101, temperature sequence 102, and the corresponding binary mask matrix together constitute the model input data for the current single breathing event.

[0059] The binary mask matrix plays its role successively in the input layer and the pooling layer. First, in the input layer, the data in the padding region is masked as zero to prevent it from participating in the convolution operation. Then, in the global pooling layer, only the effective data region is averaged, completely eliminating the interference of the padding region on feature extraction.

[0060] It should also be noted that after the experimental platform is built, experimental samples under different operating conditions are formed through the linkage control of centrifugal fan and pneumatic butterfly valve. In one embodiment, the centrifugal fan is controlled by frequency converter to supply air to the storage tank at a constant frequency, forming a continuous feeding process with relatively stable air intake flow. In another embodiment, the air intake flow is fluctuated by adjusting the frequency of centrifugal fan, forming a discontinuous feeding process. Since the change trajectories of the breather valve position height sequence, pressure sequence 101 and temperature sequence 102 are different under different operating conditions, multi-condition samples with different opening durations, different pressure fluctuation amplitudes and different air intake flow change patterns can be obtained through the above control method, providing a differentiated event data basis for the subsequent calculation of emission label 200.

[0061] like Figure 3As shown, during continuous feeding, the flow sensor output increases and remains at a high level, while the breather valve height sensor output correspondingly forms a continuously non-zero height segment, indicating that the breather valve remains continuously open after reaching the opening condition; as Figure 4 As shown, the correspondence between the output of the flow sensor and the output of the breather valve height sensor is shown under discontinuous feeding conditions. The output of the flow sensor fluctuates, while the output of the breather valve height sensor shows multiple pulse opening and closing. This indicates that the breather valve forms multiple separate single breathing events within the same feeding stage. Therefore, the opening interval 100 corresponding to a single breathing event can be defined based on the non-zero segment in the output of the breather valve height sensor.

[0062] The raw time-series data stream acquired synchronously is not directly input into subsequent processing. Instead, single breathing events are extracted based on the breather valve position height sequence. When the breather valve position height changes from the closed state to the rising stage, the corresponding time is marked as the candidate start time of the current single breathing event. When the breather valve position height changes from the open state back to the closed state, the corresponding time is marked as the candidate end time of the current single breathing event. Subsequently, the synchronous data segments between the candidate start time and the candidate end time are extracted to form the pressure sequence 101, temperature sequence 102, air flow rate sequence, and breather valve position height sequence corresponding to the breather valve breathing event caused by the current single tank injection. For multiple opening and closing situations that occur during discontinuous feeding, the data segments corresponding to each opening process are extracted separately, so that each data segment corresponds to an independent single breathing event.

[0063] After the event data is generated, each single breathing event is further organized into sample recording units. In each recording unit, at least the sample number, operating condition type, opening interval 100 length, pressure sequence 101, temperature sequence 102, airflow sequence, and breathing valve position height sequence are saved. Pressure sequence 101 and temperature sequence 102 are retained as the basic input data for the subsequent probabilistic time-series emission quantification model 300, airflow sequence is retained as the data source for the subsequent emission label 200 calculation, and breathing valve position height sequence is retained as the basis for defining the opening interval 100.

[0064] With this sample organization method, the output data is no longer just a simple continuous monitoring result, but forms an event-level sample structure that strictly corresponds to a single respiratory event, so that the calculation of the total emission volume and the formation of the emission label 200 can be directly correlated to a single sample.

[0065] S3: Input the monitoring input sequence 201 into the preset probabilistic time-series emission quantification model 300 to form an emission distribution representation of a single respiratory event, and output the mean and variance parameters of the emission amount corresponding to the emission distribution representation.

[0066] During the real-time monitoring phase, this method uses a pre-trained probabilistic emission quantification model. This model can be deployed in the monitoring device after offline training using the model training method described in the embodiments of the present invention. During the use of the model, there is no need to train it again, nor is it necessary to rely on devices such as flow meters. Only the pressure and temperature sequences collected in real time need to be input to quickly obtain the mean and variance estimates of emissions.

[0067] The probabilistic time-series emission quantification model is a probabilistic one-dimensional convolutional neural network model. The probabilistic time-series emission quantification model 300 directly establishes a mapping relationship between the pressure sequence 101 and the temperature sequence 102 within the open interval 100 and the emission label 200. The pressure sequence 101 and the temperature sequence 102 are convolved by a one-dimensional convolutional layer to extract the pressure rise slope feature and the pressure-temperature coupling feature. Based on the pressure rise slope feature and the pressure-temperature coupling feature, the mean and logarithmic variance parameters of the emission are output. A sample set is constructed based on the operating condition samples, and the corresponding time-series feature patterns under different operating conditions are extracted based on the sample set.

[0068] The probabilistic temporal emission quantization model 300 includes an input layer, a one-dimensional convolutional layer, an adaptive masked pooling layer, a fully connected layer, and a probabilistic output layer. The input layer accepts the pressure sequence 101 and temperature sequence 102 tensors and applies an input mask. The one-dimensional convolutional layer comprises a multi-layer convolution-batch normalization-ReLU-Dropout structure to extract temporal local features. The adaptive masked pooling layer performs masked weighted summation and effective region normalization on the temporal dimension of the convolutional feature map using a binary mask before global average pooling, eliminating interference from filled regions on the pooling results. The fully connected layer maps the convolutional features to the output space. The probabilistic output layer outputs the mean and logarithmic variance in parallel to represent the predicted distribution. The fully connected layer and the probabilistic output layer form a dual-head output structure.

[0069] The operating condition samples include fixed-frequency injection operating condition samples, fluctuating injection operating condition samples, and mixed air volume operating condition samples.

[0070] For different operating conditions, the same model structure is used to extract features from pressure sequence 101 and temperature sequence 102, and the mean emission and logarithmic variance parameters of the corresponding single breathing event are output. The time series feature patterns include high-frequency oscillation features under fluctuating injection conditions, continuous opening features under continuous injection conditions, and variable-length time series features corresponding to different opening durations.

[0071] If a purely physical method (such as orifice plate flow equation or CFD simulation) is used, it must be assumed in advance that the gas is an ideal gas. However, the actual gas may not meet the requirements of an ideal gas. It is also necessary to assume that the flow is steady-state / quasi-steady-state, that the valve opening is linearly related to the flow coefficient, etc. Traditional models need to establish correction coefficients for each operating condition, and it is difficult to exhaust all the complex factors of the actual operating conditions in the actual production process.

[0072] A preferred method using deep learning is represented as follows: 1D-CNN directly establishes a mapping from pressure-temperature time series curves to emissions without explicitly deriving complex intermediate physical equations. The network automatically learns the nonlinear relationship between the "pressure rise slope + temperature coupling mode" and emissions through convolutional kernels, without the need for manual analysis of which factors need to be considered. If any key factors are not considered, then the physical modeling will have significant biases.

[0073] Training with multi-condition data (fixed frequency / fluctuation / mixed air volume) allows the model to automatically extract feature patterns under different conditions (such as high-frequency oscillation features under fluctuation conditions). If it were physical or mathematical modeling, a physical model would need to be designed separately for each condition, and physical formulas would need to be derived. In the breathing valve emission scenario, where the conditions are complex and nonlinear factors are difficult to exhaustively list, deep learning provides a technical path of "replacing manual mechanism modeling with data-driven approaches".

[0074] The training of the probabilistic time-series emission quantification model 300 uses a Gaussian negative log-likelihood loss function to jointly constrain the mean and log-variance, and constructs confidence intervals.

[0075] The Gaussian negative log-likelihood loss function is expressed as: , in, This represents the loss value of the Gaussian negative log-likelihood loss function. This indicates the number of samples used in the loss calculation. Indicates the sample index. Indicates the first The prediction variance for each sample Indicates the first The actual emissions of each sample Indicates the first Predicted emissions for each sample.

[0076] The total emission volume is used as emission label 200, and a sample correspondence is established with the corresponding pressure sequence 101 and temperature sequence 102 within the open interval 100. The mean and log-variance are jointly constrained by the Gaussian negative log-likelihood loss function. For each sample, emission label 200 is used as the actual emission amount, the mean emission amount is used as the predicted center value, and the log-variance parameter corresponding to the log-variance is used as the distribution width characterization. The mean, log-variance, and actual emission amount are substituted into the Gaussian negative log-likelihood loss function to complete the parameter update. After the parameter convergence is completed, the pressure sequence 101 and temperature sequence 102 corresponding to the single respiratory event to be measured are subjected to filling mask adaptation in the same way and input into the probabilistic time series emission quantification model 300. The mean emission amount and log-variance parameter of the single respiratory event are output, and a confidence interval is constructed based on the mean emission amount and log-variance parameter.

[0077] Example 2, refer to Figures 5-6 As an embodiment of the present invention, a specific process of an intelligent monitoring and emission quantification method based on a storage tank pressure system is provided, which is expressed as follows: The blower is controlled by a frequency converter to supply air to the storage tank at a constant frequency (frequency range 0~50Hz) to simulate a stable loading and unloading process (about 15~25s). Under this condition, the air intake flow rate is relatively stable, but there are still pressure fluctuations after the breather valve is opened. The PLC records the pressure, temperature, valve position, and flow rate at a frequency of 5Hz and calculates the discharge volume of this loading and unloading. The pressure and temperature data are extracted, the data are filled to a uniform length (depending on the overall feature length of all samples), and a corresponding mask is generated. The model outputs the mean μ and standard deviation σ.

[0078] The fluctuating injection condition simulates the unstable steam recovery or feeding process in actual production by randomly adjusting the fan frequency (e.g., alternating between 10-30Hz). Under this condition, the air intake flow rate is unstable, fluctuating greatly, and the breather valve opens and closes multiple times. The PLC records pressure, temperature, valve position, and flow rate at a frequency of 5Hz and calculates the instantaneous discharge flow rate. The pressure and temperature data are extracted, normalized, filled to a uniform length, and a corresponding mask is generated. The model outputs the mean μ and standard deviation σ.

[0079] like Figure 5 As shown, the original sample size is 221, which is divided into 221 * 0.8 = 177 samples in the dataset. By using the data augmentation factor factor = 3, the number of training samples is tripled to 177 * 3 = 531, and then the test set is 221 - 177 = 44 samples.

[0080] The specific configuration is as follows: input layer, convolutional layer (3 layers of one-dimensional convolution), pooling configuration, mask-aware global pooling, fully connected layer, and probabilistic output layer. Gaussian negative log-likelihood (NLL) loss is used, with additional interval width penalties. Evaluation metrics include accuracy, uncertainty quality, and R². 2 .

[0081] like Figure 6 As shown, Time series data refers to time series data. By inputting pressure and temperature sensor sequence data, the model realizes the emission inversion of the input sample and gives the point prediction value μ and standard deviation prediction σ. Then, according to the application requirements, different confidence level intervals are constructed, such as the 95% confidence interval [μ-1.96σ,μ+1.96σ].

[0082] I. Prediction of Emissions Based on Fixed-Frequency Sampling

[0083] In a certain fixed-frequency injection, the model extracted temperature and pressure sensor sequence data and calculated the predicted value μ = 0.67 m³, with a standard deviation σ = 0.07 m³, and output a 95% confidence interval prediction range of [0.53~0.81]. This result indicates that the model estimated the emission amount for this sample to be 0.67 m³. However, due to sensor noise and the inherent uncertainty of the model estimation, the actual emission amount may be around 0.67. Therefore, an estimated standard deviation σ is given, and it is assumed that the actual value falls within the interval [μ-1.96σ, μ+1.96σ] with a 95% confidence level / probability. The actual emission amount for this sample was 0.66 m³, indicating that the model prediction is relatively accurate (0.67 vs 0.66), the actual emission amount falls within the 95% confidence interval, the model prediction accuracy is high, and the uncertainty estimation is reasonable.

[0084] II. Prediction of Fluctuating Sampling Emissions

[0085] In a particular fluctuating sample injection, the model extracted temperature and pressure sensor sequence data and calculated the predicted value μ = 2.53 m³, with a standard deviation σ = 0.87 m³. The predicted range for the 95% confidence interval was [0.82~4.23]. This result indicates that the model estimated the emission amount of this sample to be 2.53 m³. However, due to factors such as sensor noise and the uncertainty of the model estimation itself, the actual emission amount may be around 2.53. Therefore, an estimated value for the standard deviation σ is given, and it is assumed that there is a 95% confidence level / probability that the actual value falls within the interval [μ-1.96σ, μ+1.96σ]. The actual emission amount of this sample was 2.68 m³. The result shows that the model's point prediction for the sample is relatively close to the actual value (2.53 vs 2.68), but the estimate for the 95% confidence interval range is too conservative. Ultimately, the actual emission amount falls within the 95% confidence interval.

[0086] Due to the unstable intake flow rate and repeated opening and closing of the breathing valve under fluctuating operating conditions, the model's prediction confidence for this sample decreased, which was reflected in a significant increase in the standard deviation σ. This demonstrates the advantage of probabilistic prediction: when the operating conditions are complex, the model avoids risks by expanding the confidence interval, rather than giving a certain value that may mislead decision-making (such as a point prediction value that is too high or too low, or a large deviation from the true value of 2.68m3).

[0087] Example 3, referring to Figures 7(a)-7(d), is an embodiment of the present invention, which provides an intelligent monitoring and emission quantification method based on a tank pressure system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0088] First, the probabilistic time-series emissions quantification model 300 was comprehensively evaluated using the following core metrics: Mean Absolute Error (MAE), used to quantify the average magnitude of prediction errors; Coefficient of Determination (R²), measuring the proportion of the target variable variance explained by the model; and Mean Prediction Interval Width (MPIW), serving as the primary indicator for assessing the accuracy of uncertainty estimation. The Probability of Prediction Interval Coverage (PICP) measures the empirical proportion of actual values ​​falling within the estimated confidence interval.

[0089] Table 1 shows the comprehensive evaluation results of the probabilistic regression model in ten independent runs. The model showed good prediction accuracy, with an average R² score of 0.86±0.06 and a mean squared error (MSE) of 0.13±0.08.

[0090] In Table 1, Round represents the evaluation result round of the model in each round of training or cross-validation, and Mean represents the average and standard deviation of the evaluation metrics for all Round rounds.

[0091] These metrics indicate that the model explains approximately 86% of the variance in emissions values ​​with relatively low prediction errors. In terms of uncertainty quantification, the model achieves a mean PICP of 0.91 ± 0.06, close to but slightly below the target confidence level of 95%. The mean MPIW is 0.84 ± 0.18 cubic meters, indicating that for typical emissions, i.e., the mean emissions across all samples of 0.82 cubic meters, the model's estimation range is [0.40, 1.24] cubic meters.

[0092] Table 1 Comprehensive Evaluation Results

[0093] As shown in Figure 7(a), the loss curve has a good shape, indicating that the training convergence is stable. As shown in Figure 7(b), the root mean square error (RMSE) of the dataset and the validation set decreases synchronously throughout the iteration process, confirming that the model learns effectively and does not overfit. The model performs well in predicting 20% ​​of the test set (about 44 samples). As shown in Figure 7(c), the ordered dot plot of the predicted values ​​with uncertainty intervals shows that the interval estimates for extremely low emissions (≤~0.2m³) and extremely high emissions (≥~1.8m³) samples are significantly wider. This may be due to higher data noise or greater quantization difficulty in these regions. For most samples in the range of 0.2–1.8m³, the interval remains relatively compact. As shown in Figure 7(d), the overall fitting accuracy of the point prediction is shown. Only four samples show obvious underestimation or overestimation. Although the prediction interval is relatively wide, the accuracy of the point prediction is quite good.

[0094] Example 4 is an embodiment of the present invention, which provides an intelligent monitoring and emission quantification system based on a tank pressure system, including an event acquisition module, a characterization construction module, and a quantification output module.

[0095] The event acquisition module is used to acquire the valve position height sequence of the breathing valve through a non-contact infrared displacement sensor F set above the valve core of the breathing valve, and to determine the opening range 100 of the breathing valve based on the valve position height sequence. The pressure sequence 101 and temperature sequence 102 corresponding to the opening range 100 are determined as the pressure sequence 101 and temperature sequence 102 corresponding to a single breathing event.

[0096] The characterization construction module is used to treat the gas in the storage tank as the control volume. Based on the pressure sequence 101 and temperature sequence 102 within the open interval 100, the volumetric emission flow rate is determined by combining the mass balance equation and the ideal gas law. The total emission volume is obtained by time integration of the volumetric emission flow rate using the trapezoidal rule within the open interval 100. The pressure sequence 101 and temperature sequence 102 are respectively subjected to Z-score normalization, uniform length padding, and binary mask matrix generation based on the original effective length to form the monitoring input sequence 201.

[0097] The quantization output module is used to input the monitoring input sequence 201 into the probabilistic time series emission quantification model 300, and uses the Gaussian negative log-likelihood loss function to jointly constrain the mean and log-variance, and outputs the emission mean and log-variance parameters.

[0098] This embodiment also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements an intelligent monitoring and emission quantification system based on a tank pressure system as proposed in the above embodiment.

[0099] This embodiment also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements an intelligent monitoring and emission quantification system based on a tank pressure system as proposed in the above embodiment.

[0100] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0102] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0103] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0104] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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. A method for intelligent monitoring and quantification of emissions based on a tank pressure system, characterized in that, include: Obtain the pressure sequence (101) and temperature sequence (102) corresponding to a single respiratory event. Based on the pressure sequence (101) and temperature sequence (102), and combined with the effective time length of the single respiratory event, a monitoring input sequence (201) is formed. The monitoring input sequence (201) is input into a preset probabilistic time-series emission quantification model (300) to form an emission distribution characterization of a single respiratory event, and the mean and variance parameters of the emission amount corresponding to the emission distribution characterization are output.

2. The intelligent monitoring and emission quantification method based on a storage tank pressure system as described in claim 1, characterized in that: The single respiratory event includes, Obtain the valve position height sequence of the breathing valve, and determine the opening range (100) of the breathing valve according to the valve position height sequence of the breathing valve. Then, determine the pressure sequence (101) and temperature sequence (102) corresponding to the opening range (100) as the pressure sequence (101) and temperature sequence (102) corresponding to the single breathing event.

3. The intelligent monitoring and emission quantification method based on a storage tank pressure system as described in claim 2, characterized in that: The breather valve position height sequence includes: The opening displacement of the breathing valve is monitored in real time by a non-contact infrared displacement sensor (F) located above the valve core.

4. The intelligent monitoring and emission quantification method based on a storage tank pressure system as described in claim 3, characterized in that: The monitoring input sequence (201) includes, The pressure sequence (101) and temperature sequence (102) are Z-score normalized respectively, the short sequences are filled to a uniform length by a preset constant, and a binary mask matrix is ​​generated based on the original effective length.

5. The intelligent monitoring and emission quantification method based on a storage tank pressure system as described in claim 2, characterized in that: The emission label (200) corresponding to the preset probabilistic time-series emission quantification model is the total emission volume of a single respiratory event; The total discharge volume is the integral result of the volume discharge flow rate during the valve opening phase; The probabilistic time-series emission quantification model is a probabilistic one-dimensional convolutional neural network model; The emission label (200) includes treating the gas inside the storage tank as a control volume, determining the gas density inside the storage tank based on the pressure sequence (101) and temperature sequence (102) within the opening interval (100), and characterizing the gas state change inside the storage tank during the single breathing event by combining the mass balance equation and the ideal gas state equation. The mass inflow rate is determined based on the volumetric flow meter reading, and the gas density at the inlet is taken to be equal to the instantaneous density inside the tank. The mass inflow rate is then substituted into the mass balance equation to solve for the mass discharge rate. The volumetric flow meter readings are monitored by a thermal mass flow meter installed between the centrifugal fan and the storage tank, and the volumetric inflow rate output by the thermal mass flow meter is synchronously collected by a programmable logic controller at a fixed frequency to form an intake flow sequence. Divide the mass emission rate by the instantaneous gas density inside the tank to obtain the volumetric emission flow rate during a single breathing event. In the open interval (100), the volumetric emission flow rate is integrated over time using the trapezoidal rule to obtain the total emission volume.

6. The intelligent monitoring and emission quantification method based on a storage tank pressure system as described in claim 5, characterized in that: The monitoring input sequence (201) is input into a preset probabilistic time-series emission quantification model (300), including: The probabilistic time-series emission quantification model (300) directly establishes a mapping relationship between the pressure sequence (101) and temperature sequence (102) within the open interval (100) and the emission label (200); The pressure sequence (101) and temperature sequence (102) are convolved by a one-dimensional convolutional layer to extract the pressure rise slope feature and the pressure-temperature coupling feature, and the mean emission and logarithmic variance parameters are output based on the pressure rise slope feature and the pressure-temperature coupling feature. A sample set is constructed based on the working condition samples, and the corresponding time-series feature patterns under different working conditions are extracted based on the sample set; The operating condition samples include fixed-frequency injection operating condition samples, fluctuating injection operating condition samples, and mixed air volume operating condition samples; For different operating conditions, the same model structure is used to extract features from the pressure sequence (101) and temperature sequence (102), and the mean emission and logarithmic variance parameters of the corresponding single breathing event are output. The time-series feature patterns include high-frequency oscillation features under fluctuating injection conditions, continuous opening features under continuous injection conditions, and variable-length time-series features corresponding to different opening durations.

7. The intelligent monitoring and emission quantification method based on a storage tank pressure system as described in claim 6, characterized in that: The training of the probabilistic time-series emission quantification model (300) employs a Gaussian negative log-likelihood loss function to jointly constrain the mean and log-variance, thereby constructing a confidence interval. The Gaussian negative log-likelihood loss function is expressed as: ， in, This represents the loss value of the Gaussian negative log-likelihood loss function. This indicates the number of samples used in the loss calculation. Indicates the sample index. Indicates the first The prediction variance for each sample Indicates the first The actual emissions of each sample Indicates the first Predicted emissions for each sample.

8. An intelligent monitoring and emission quantification system based on a tank pressure system, employing the intelligent monitoring and emission quantification method based on a tank pressure system as described in any one of claims 1 to 7, characterized in that: Event acquisition module, representation construction module, and quantization output module; The event acquisition module is used to acquire the valve position height sequence of the breathing valve through a non-contact infrared displacement sensor (F) set above the valve core of the breathing valve, and determine the opening range (100) of the breathing valve according to the valve position height sequence, and determine the pressure sequence (101) and temperature sequence (102) corresponding to the opening range (100) as the pressure sequence (101) and temperature sequence (102) corresponding to a single breathing event. The characterization construction module is used to treat the gas in the storage tank as a control volume. Based on the pressure sequence (101) and temperature sequence (102) in the opening interval (100), the volumetric discharge flow rate is determined by combining the mass balance equation and the ideal gas state equation. The total discharge volume is obtained by time integration of the volumetric discharge flow rate using the trapezoidal rule in the opening interval (100). The pressure sequence (101) and temperature sequence (102) are respectively subjected to Z-score normalization, uniform length padding, and binary mask matrix generation based on the original effective length to form the monitoring input sequence (201). The quantization output module is used to input the monitoring input sequence (201) into the probabilistic time series emission quantification model (300), and use the Gaussian negative log-likelihood loss function to jointly constrain the mean and log-variance, and output the emission mean and log-variance parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent monitoring and emission quantification method based on the tank pressure system as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent monitoring and emission quantification method based on the tank pressure system as described in any one of claims 1 to 7.

Images