A ship exhaust emission online analysis method and system based on intelligent sensors

Abstract

This invention discloses an online analysis method and system for ship exhaust emissions based on intelligent sensors, belonging to the field of ship emission monitoring technology. The method includes: acquiring ship motion attitude data through a six-axis inertial sensor; compensating for motion disturbances in the raw spectral signals collected by a spectral detection sensor; and performing Gaussian weighted averaging based on the roll angle to obtain a motion-compensated spectral signal; performing attenuation compensation self-calibration using the carbon dioxide response signal as an internal reference benchmark; identifying and correcting fuel types based on the characteristic component concentration ratio to obtain multi-component concentration data; obtaining a residual vector through an inter-component correlation constraint model; obtaining carbon balance deviation through a carbon balance equation; collaboratively determining sensor offset and actual emission changes; and labeling data quality levels; and when data is unavailable, using a physical constraint neural network to infer concentration values. This invention can adapt to the ship's motion environment, achieve continuous self-calibration and multi-fuel adaptation, and improve measurement reliability.

Description

Technical Field

[0001] This invention relates to the field of ship emission monitoring technology, and in particular to an online analysis method and system for ship exhaust emissions based on intelligent sensors. Background Technology

[0002] As a major means of transportation in global trade, ships' exhaust emissions have a significant impact on the marine environment and air quality in coastal areas. The International Maritime Organization (IMO) and various countries have successively introduced strict regulations on ship emissions, requiring real-time monitoring of pollutants such as sulfur dioxide, nitrogen oxides, and particulate matter in ship exhaust.

[0003] Existing online monitoring systems for ship exhaust gases mainly face the following technical challenges: First, during navigation, ships experience six degrees of freedom motion due to waves, wind, and other forces, which disturbs the measurement optical path of the shipborne spectral detection sensor, introducing measurement errors. Existing technologies typically employ mechanical stabilization platforms or simple data smoothing, but mechanical stabilization platforms increase system complexity and cost, while simple data smoothing cannot effectively compensate for systematic deviations caused by motion.

[0004] Secondly, optical components experience performance degradation when operating in high-temperature, high-humidity, and corrosive gas environments for extended periods, leading to a decrease in sensor response sensitivity. Traditional methods employ periodic standard gas calibration, but this requires manual intervention, and measurement accuracy cannot be guaranteed during calibration intervals. Furthermore, the transport and storage of standard gases in the marine environment poses safety hazards.

[0005] Third, ships may use different types of fuel, such as heavy oil, light oil, liquefied natural gas, or blended fuels. The elemental composition and combustion characteristics of different fuels vary significantly, and using fixed calibration parameters will lead to measurement errors. Current technology lacks automatic fuel type identification and adaptive parameter adjustment mechanisms.

[0006] Fourth, sensor malfunctions and actual emission changes share similar data characteristics, making it easy to misjudge using a single threshold method. Current technologies lack effective mechanisms for diagnosing sensor health status and assessing data quality, making it difficult to distinguish between measurement system malfunctions and actual operating condition changes.

[0007] Therefore, there is a need for an online analysis method for ship exhaust emissions that can adapt to the ship's motion environment, achieve continuous self-calibration, support multiple fuel types, and have intelligent fault diagnosis capabilities. Summary of the Invention

[0008] The purpose of this invention is to provide an online analysis method and system for ship exhaust emissions based on intelligent sensors, in order to solve the technical problems in the prior art, such as the accuracy of ship motion interference measurement, the need for manual calibration of sensor attenuation, poor adaptability to multiple fuel types, and difficulty in identifying sensor faults.

[0009] In a first aspect, this invention proposes an online analysis method for ship exhaust emissions based on intelligent sensors, comprising the following steps: S1: Obtain the ship's motion attitude data through a six-axis inertial sensor, perform motion disturbance compensation based on the ship's motion attitude data on the raw spectral signal collected by the spectral detection sensor, and perform weighted averaging based on the ship's roll angle within a preset time window to obtain the motion-compensated spectral signal. S2: Using the response signal of carbon dioxide in the motion-compensated spectral signal as the internal reference benchmark, the response signals of each target component are self-calibrated by attenuation compensation based on the rate of change of the carbon dioxide response signal relative to the initial calibration value, and converted into preliminary multi-component concentration data; the current fuel type is identified based on the characteristic component concentration ratio of the preliminary multi-component concentration data, and the correction parameters and carbon content parameters corresponding to the current fuel type are retrieved from the multi-fuel parameter library to correct the preliminary multi-component concentration data, thereby obtaining multi-component concentration data; S3: Based on the element content parameters corresponding to the current fuel type, the residual vector is obtained through the component correlation constraint model, and the carbon balance deviation is obtained through the carbon balance equation. When the residual vector and the carbon balance deviation deviate from the normal range at the same time within the preset time window, it is determined to be sensor offset. When the residual vector deviates from the normal range but the carbon balance deviation remains within the normal range, it is determined to be actual emission change. The data quality level is marked according to the judgment result. The data quality level includes at least an usable level and an unusable level. S4: When the target component is marked as unavailable in step S3, the concentration value of the target component is inferred using a hybrid model of data-driven and physical equations; the hybrid model of data-driven and physical equations continuously learns and updates online using historical measured data marked as available in step S3 as training samples.

[0010] A second aspect of this invention provides an online analysis system for ship exhaust emissions based on intelligent sensors, comprising: a processor and a memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the online analysis method for ship exhaust emissions based on smart sensors as described in the first aspect.

[0011] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: (1) The motion attitude data of the ship is acquired in real time by a six-axis inertial sensor, the motion disturbance compensation of the spectral signal is performed, and a Gaussian weighted average method based on the roll angle is adopted to allocate more weight to the sampling points when the ship attitude is stable, which effectively reduces the impact of ship motion on measurement accuracy. Compared with the equal weighted average method, the motion compensation spectral signal can be mainly determined by the sampling points when the ship attitude is relatively stable. (2) Using the carbon dioxide response signal that is always present in the exhaust gas as the internal reference benchmark, the decay compensation self-calibration of each target component is achieved by monitoring the rate of change of the carbon dioxide response signal. Compared with the method of periodic calibration using external standard gas, calibration can be performed automatically each time, without manual intervention, thus avoiding the safety hazards of carrying and storing standard gas. (3) Based on the concentration ratio of characteristic components, a fingerprint vector of exhaust gas components is constructed. The current fuel type is automatically identified by the posterior probability method. The corresponding correction parameters and carbon content parameters are retrieved from the multi-fuel parameter library. It can adapt to different fuel types such as heavy oil, light oil, liquefied natural gas and mixed fuels. Compared with the fixed parameter method, it improves the measurement accuracy under multi-fuel conditions. (4) By establishing a collaborative judgment mechanism between residual vector and carbon balance deviation, and using the stoichiometric relationship between components and carbon element conservation for dual verification, it can effectively distinguish between sensor offset and actual emission changes. Compared with the single threshold judgment method, it avoids misjudging the actual emission changes during sudden changes in operating conditions as sensor failures and provides graded data quality labeling. (5) When sensor failure leads to data unavailability, the physical constraint neural network model is used to infer the component concentration. By encoding the stoichiometric relationship of the combustion process into the hard constraint condition of the neural network, the prediction result is ensured to meet the physical conservation law. Compared with the pure data-driven model, it avoids outputting prediction values ​​that violate physical laws and improves the fault tolerance and data continuity of the system. Attached Figure Description

[0012] Figure 1 A flowchart illustrating an online analysis method for ship exhaust emissions based on intelligent sensors, provided as an embodiment of the present invention; Figure 2 This is a schematic diagram of a ship exhaust gas emission online analysis system based on intelligent sensors, provided as an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0014] Reference manual attached Figure 1The diagram shows a flowchart of an online analysis method for ship exhaust emissions based on intelligent sensors, provided by an embodiment of the present invention.

[0015] This invention provides an online analysis method for ship exhaust emissions based on intelligent sensors, which may include the following steps: S1: Adaptive ship motion compensation: The ship motion attitude data is acquired by a six-axis inertial sensor, and the motion disturbance compensation based on the ship motion attitude data is performed on the original spectral signal collected by the spectral detection sensor. The weighted average is then performed according to the ship roll angle within a preset time window to obtain the motion compensation spectral signal.

[0016] Specifically, the six-axis inertial sensor collects the ship's six-degree-of-freedom motion attitude data in real time. This data includes roll angle, pitch angle, bow angle, and three-axis linear acceleration. Let the roll angle be denoted as... The pitch angle is The bow roll angle is The linear accelerations of the three axes are respectively , , ,in, , , The unit is degrees and the value ranges from -180 degrees to +180 degrees. , , The unit is meters per second squared.

[0017] Furthermore, the spectral detection sensor acquires the raw spectral signal at a fixed sampling frequency within a preset time window. Internal collection The sampling point is denoted as the nth sampling point. The original spectral signal of each sampling point is ,in, Indicates the sampling point index and , Indicates the total number of sampling points and , Indicates the first The time of each sampling point is in seconds. This is a spectral response vector containing multiple wavelength channels.

[0018] Furthermore, motion disturbance compensation is performed on the original spectral signal. Motion disturbance compensation includes: sedimentation offset correction of particulate matter measurements based on roll and pitch angles, and velocity fluctuation filtering of gas phase component measurements based on triaxial linear acceleration, to obtain the... Compensated spectral signal at each sampling point .

[0019] Furthermore, a Gaussian weighting function is constructed based on the roll angle, with a preset time window length. With the current oscillation cycle To match the results, the compensated spectral signal is weighted and averaged using a Gaussian weighting function within a preset time window. The formula for calculating the Gaussian weighting coefficients is shown in formula (1): in, Indicates the first The normalized weighted coefficients of each sampling point and and , Indicates the sampling point index and , Indicates the first The roll angle at each sampling point, in degrees. The standard deviation of the Gaussian function is expressed in degrees. , This represents an exponential function with the natural constant as its base. To sum the indices and iterate from 1 to For all sampling points, the sum of the denominators is always greater than 0 to ensure that the division is effective and achieves normalization.

[0020] It should be noted that the roll angle corresponding to the roll balance position in formula (1) is 0 degrees. Therefore, the center of the Gaussian function is set to 0, and the summation of the denominator is used for normalization so that the sum of all weighted coefficients is strictly equal to 1.

[0021] Optionally, The value is 5 degrees. The value is the current roll cycle. It is 1.5 times that. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to the actual situation, and the embodiments of the present invention do not specifically limit them in this regard.

[0022] Furthermore, the formula for calculating the motion-compensated spectral signal is shown in formula (2): in, Represents the motion-compensated spectral signal. Indicates wavelength index. Indicates the total number of sampling points. To sum the indices and iterate from 1 to All sampling points, Indicates the first Normalized weighted coefficients for each sampling point Indicates the first Each sampling point at wavelength The intensity of the compensated spectral signal at the location.

[0023] In this embodiment of the invention, formula (1) adopts a Gaussian weighting function based on the roll angle. When the roll angle is 0 degrees, the weighting coefficient reaches its maximum value. The larger the absolute value of the roll angle, the more the weighting coefficient decays exponentially. By normalizing the denominator, it is ensured that the sum of all weighting coefficients is strictly equal to 1. According to the mathematical properties of the Gaussian distribution, when the roll angle of the sampling point is within... When the range is within a certain range, the cumulative weighting coefficient corresponding to that sampling point accounts for approximately 68.3%. When the roll angle of the sampling point is within a certain range... Within the range, the cumulative proportion is approximately 95.4%, meaning that formula (2) allocates more than 95% of the weight to sampling points whose roll angle deviates from the equilibrium position by no more than two standard deviations. The equal-weighted averaging method assigns the same weight to all sampling points within the time window, including sampling points when the hull is in a large swing, while formulas (1) and (2) assign higher weight to sampling points whose roll angle is close to the equilibrium position, so that the motion compensation spectral signal is mainly determined by sampling points when the hull attitude is relatively stable.

[0024] S2: Continuous Self-calibration and Fuel Identification: Using the carbon dioxide response signal in the motion-compensated spectral signal as the intrinsic reference standard, the response signals of each target component are self-calibrated by attenuation compensation based on the rate of change of the carbon dioxide response signal relative to the initial calibration value, converting it into preliminary multi-component concentration data. The current fuel type is identified based on the characteristic component concentration ratios of the preliminary multi-component concentration data, and the corresponding correction parameters and carbon content parameters are retrieved from the multi-fuel parameter library to correct the preliminary multi-component concentration data, obtaining the final multi-component concentration data.

[0025] In one possible implementation, the response signal of carbon dioxide is the absorption response signal of carbon dioxide in the mid-infrared 4.26-micron band. Since the spectral detection of carbon dioxide, sulfur dioxide, and nitrogen oxides uses the same optical path, the attenuation of the optical elements in each detection channel is correlated.

[0026] Specifically, the wavelength of carbon dioxide is recorded. The initial calibration response intensity at is The currently measured carbon dioxide response intensity is The formula for calculating the attenuation compensation coefficient is shown in formula (3): in, Indicates the target component The attenuation compensation coefficient and , Indicates the target component index and , Indicates the total number of target components and , Indicates the currently measured carbon dioxide response intensity and , Indicates the intensity of the carbon dioxide response at the initial calibration time and , Indicates the pre-calibrated target components The wavelength-dependent attenuation ratio is dimensionless, and the exponential operation requires the base to be greater than 0.

[0027] Optionally, for the sulfur dioxide component, The value is 0.95. For the nitrogen oxide component, The value is 0.92. For the methane component, The value is 0.98. It is understood that those skilled in the art can adjust the value of the above parameter according to the actual situation, and the embodiments of the present invention do not specifically limit this.

[0028] Furthermore, the original measured concentrations of each target component were... After attenuation compensation, preliminary multi-component concentration data are obtained. The calculation formula is as follows: ,in, Indicates the target component index. Indicates components The original measured concentration and the unit is ppm. , Indicates components The initial concentration is given in ppm. This ensures that the division is effective.

[0029] In this embodiment of the invention, formula (3) is obtained by monitoring the rate of change of the carbon dioxide response signal relative to the initial calibration value, combined with the wavelength-related attenuation coefficient. Calculate the attenuation compensation coefficient for each target component. Since carbon dioxide, as an inevitable byproduct of combustion, is always present in the exhaust gas, and its volume fraction is typically in the range of a few percent to over ten percent, its response signal strength is sufficient to support real-time monitoring. Therefore, the carbon dioxide response signal can be used as an internal reference standard to achieve continuous self-calibration. Methods using external standard gases for periodic calibration require manual intervention, and the calibration interval is typically several hours to several days. In contrast, the internal self-calibration method of this invention can automatically perform calibration during each measurement without manual intervention.

[0030] Furthermore, based on preliminary multi-component concentration data, the concentration ratios of characteristic components are calculated, including the ratio of sulfur dioxide concentration to carbon dioxide concentration. The ratio of methane concentration to carbon dioxide concentration The ratio of carbon monoxide concentration to carbon dioxide concentration The ratio of nitrogen oxide concentration to carbon dioxide concentration and the ratio of ammonia concentration to nitrogen oxide concentration At least three of the components are used to perform a moving average of each ratio within a preset time window, and combined to form a fingerprint vector of exhaust gas components. Among them, exhaust gas component fingerprint vector for 3D column vector, The components are as follows: , Indicates the number of ratio types selected and .

[0031] Furthermore, the multi-fuel parameter library stores standard fingerprint templates corresponding to each fuel type. ,in, Indicates fuel type index and , Indicates the total number of fuel types and Standard fingerprint template for Column vector. A pattern recognition method is used to match the exhaust gas composition fingerprint vector with a standard fingerprint template. The calculation formula for fuel type identification is shown in formula (4): in, Indicates fuel type The posterior probability and and , Indicates fuel type index and , Indicates the current fingerprint vector With fuel type Standard fingerprint template Euclidean distance between them and The calculation formula is: , Describing the Euclidean norm, The scale parameter of the similarity function is dimensionless and , This represents an exponential function with the natural constant as its base. To sum the indices and iterate from 1 to For all fuel types, the sum of the denominators is always greater than 0 to ensure the division is valid. The fuel type with the highest posterior probability is selected as the current fuel type. .

[0032] Optionally, The value is 0.1. The value of 5 includes heavy oil, light oil, liquefied natural gas, blended fuel A, and blended fuel B. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual circumstances, and this embodiment of the invention does not specifically limit this.

[0033] Furthermore, based on the identified current fuel type Retrieve the corresponding calibration parameter matrix from the multi-fuel parameter library. and carbon content parameters The preliminary multi-component concentration data were corrected to obtain the multi-component concentration data. Among them, multi-component concentration data for 3D column vector, The components are as follows: , for OK Cross-interference compensation matrix of columns, carbon content parameter Indicates fuel type The carbon mass fraction is dimensionless and ranges from 0 to 1. Indicates the target component The corrected concentration is in ppm and is a volume fraction. Indicates the target component index and .

[0034] It should be noted that in the embodiments of the present invention, the unit of concentration ppm represents volume fraction, that is, the volume number of the target component in one million volumes of gas, which is a dimensionless quantity.

[0035] In this embodiment of the invention, formula (4) uses a Gaussian similarity-based posterior probability calculation method to identify fuel types. By calculating the Euclidean distance between the measured fingerprint vector and each fuel standard template, and converting the distance into a posterior probability, the fuel type with the highest posterior probability is selected as the identification result, providing a quantitative assessment of the similarity of each fuel type. The fixed threshold decision method can only provide a binary decision of whether or not a match exists, while the posterior probability method can simultaneously output the matching probabilities of multiple fuel types, facilitating the identification of transitional states such as fuel mixing or fuel switching. By automatically identifying fuel types and retrieving corresponding correction parameters and carbon content parameters, the system can adapt to differences in the elemental composition of different fuels.

[0036] S3: Intelligent Fault Diagnosis and Quality Assessment: Based on the elemental content parameters corresponding to the current fuel type, a residual vector is obtained through an inter-component correlation constraint model, and the carbon balance deviation is obtained through the carbon balance equation. When both the residual vector and the carbon balance deviation deviate from the normal range within a preset time window, it is determined to be sensor offset. When the residual vector deviates from the normal range but the carbon balance deviation remains within the normal range, it is determined to be a true emission change. The data quality level is marked according to the judgment result, and the data quality level includes at least an usable level and an unusable level.

[0037] In one possible implementation, the inter-component correlation constraint model is a constraint model established based on the stoichiometric relationships between the main measured components. Specifically, theoretical correlations between the concentrations of each component are established based on the stoichiometric relationships of fuel combustion, taking into account the current fuel type. The corresponding elemental content parameters, including carbon content parameters Hydrogen content parameters Sulfur content parameters and nitrogen content parameters Calculate the theoretical component concentration vector ,in, , , , Each represents a different fuel type. The mass fractions of carbon, hydrogen, sulfur, and nitrogen, all dimensionless and ranging from 0 to 1. for 3D column vector, The components are as follows: , Indicates the number of components participating in the association constraint and , Indicates components The theoretical concentration is given in ppm and is expressed as a volume fraction. Indicates component index and The formula for calculating the residual vector is shown in formula (5): in, Describes the residual vector and is 3D column vector, The components are as follows: , Represents the vector of actual measured component concentrations and is 3D column vector, This represents the theoretical component concentration vector calculated based on elemental content parameters, and is... 3D column vector, Indicates components The residual is given in ppm. Indicates component index and .

[0038] Optionally, The value is set to 4, and carbon dioxide, sulfur dioxide, nitrogen oxides, and carbon monoxide are selected as the components participating in the correlation constraint. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to the actual situation, and the embodiments of the present invention do not specifically limit this.

[0039] Furthermore, the carbon balance equation is established based on the principle of carbon mass conservation, and the carbon balance state is determined by comparing the ratio of input carbon molar flow rate to output carbon molar flow rate. The input carbon molar flow rate is determined by the fuel consumption rate. With carbon content parameter Divide by the molar mass of carbon The calculated output carbon molar flow rate is determined by the exhaust flow rate. carbon dioxide concentration and the reciprocal of the molar volume of a gas under standard conditions The carbon balance deviation was calculated as shown in formula (6): in, Indicates the carbon balance deviation and is dimensionless. , Indicates fuel consumption rate in kilograms per hour and , The conversion factor from ppm to volume fraction is dimensionless. This represents the molar mass of carbon and has a value of 0.012 kg per mole. Indicates exhaust flow rate in standard cubic meters per hour and , The concentration of carbon dioxide is expressed in ppm and is a volume fraction. , This represents the reciprocal of the molar volume of a gas under standard conditions, and its value is 44.64 moles per standard cubic meter. This represents absolute value operations.

[0040] Furthermore, fuel consumption rate Engine speed data and exhaust flow rate are obtained in real time from the ship's engine control system via a data interface. It is obtained by measuring the exhaust pipe flow rate sensor or by calculating based on the engine intake air flow rate and fuel consumption rate.

[0041] Furthermore, we define the norm of the residual vector. As a measure of the degree of residual deviation, among which, , To represent the square root operation, Index the components. Set the residual threshold. and carbon balance deviation threshold The preset time window is The decision logic is as follows: When in Continuous satisfaction within the time window and When this occurs, it is determined to be sensor offset.

[0042] When in Continuous satisfaction within the time window but When this occurs, it is determined to be a genuine change in emissions.

[0043] Optionally, The value is 50 ppm. The value is 0.15. The value is 60 seconds. It is understood that those skilled in the art can adjust the value of the above parameter according to the actual situation, and the embodiments of the present invention do not specifically limit this.

[0044] Furthermore, data quality levels are assigned based on the assessment results. Data quality levels include at least an usable level and an unusable level. In one possible implementation, data quality levels are divided into four levels: When all components of the residual vector are within the first preset threshold range and the carbon balance deviation is within the first deviation threshold range, it is marked as the first level.

[0045] When each component of the residual vector exceeds the first preset threshold range but is within the second preset threshold range, it is marked as the second level. When the vector angle between the direction of the residual vector and the feature direction of a certain emission event type in the pre-built emission event feature template library is less than the preset angle threshold, the marking level will be reverted to the first level.

[0046] If a single component in the residual vector continuously exceeds the second preset threshold range within the preset monitoring window, but the carbon balance deviation remains within the normal range, it is marked as the third level and triggers S2 to re-execute the attenuation compensation self-calibration.

[0047] When multiple components in the residual vector simultaneously exceed the second preset threshold range and the carbon balance deviation synchronously exceeds the second deviation threshold range, it is marked as level four and the virtual sensor takeover of S4 is triggered.

[0048] The first, second, and third levels correspond to the available levels, while the fourth level corresponds to the unavailable level.

[0049] Furthermore, when the carbon balance deviation continuously exceeds the preset threshold for a preset duration, the recalibration of the attenuation compensation self-calibration in S2 is triggered.

[0050] It should be noted that those skilled in the art can set the first preset threshold range, the first deviation threshold range, the second preset threshold range, the preset angle threshold, the second deviation threshold range, the preset threshold, and the preset duration according to actual needs, and the present invention does not limit these settings.

[0051] In this embodiment of the invention, by establishing a collaborative determination mechanism for residual vector and carbon balance deviation, and utilizing the stoichiometric relationship between components and the conservation of carbon, dual verification can be performed, distinguishing between sensor offset and actual emission changes. When both the residual vector and carbon balance deviation deviate from the normal range, it indicates a systematic deviation between the measured value and the theoretical expectation based on fuel characteristics, violating the conservation of carbon, and is determined to be sensor offset. When the residual vector deviates but the carbon balance deviation is normal, it indicates a change in component concentration but the conservation of carbon is maintained, and is determined to be an actual emission change. Using a single threshold determination method cannot distinguish between sensor offset and actual emission changes, and it is easy to misjudge actual emission changes during sudden changes in operating conditions as sensor failure. The dual verification mechanism provides an independent verification path through the carbon balance equation, avoiding such misjudgment.

[0052] S4: Virtual Sensor Fault-Tolerant Inference: When the target component is marked as unavailable in step S3, the concentration value of the target component is inferred using a hybrid model of data-driven and physical equations. The hybrid model of data-driven and physical equations continuously learns and updates online using historical measured data marked as available in step S3 as training samples.

[0053] In one possible implementation, the data-driven and physical equation hybrid model is a physical constraint neural network model, which includes a physical constraint layer and a data learning layer.

[0054] Specifically, the physical constraint layer encodes the physical equations of the engine combustion process as hard constraints for a neural network. The physical equations include stoichiometric equations relating oxygen consumption to carbon dioxide production based on the excess air coefficient.

[0055] Furthermore, based on the stoichiometric relationship under the assumption of complete combustion, the excess air coefficient... The relationship between oxygen concentration and oxygen content satisfies an equation, and the formula for calculating the constraint loss term is shown in formula (7): in, Represents the stoichiometric constraint loss term and the unit is ppm. , This indicates the oxygen concentration, expressed in ppm (parts per second) and as a volume fraction. This represents the oxygen concentration corresponding to the theoretical oxygen demand under stoichiometric combustion conditions, expressed in ppm as a volume fraction. Indicates the excess air coefficient and is dimensionless. , This represents absolute value operations.

[0056] It should be noted that formula (7) is based on the assumption of complete combustion, when the excess air coefficient is... At that time, the actual oxygen concentration provided was The stoichiometric concentration of oxygen consumed in combustion Therefore, the excess oxygen concentration is This constraint ensures that the oxygen concentration predicted by the neural network and the excess air coefficient satisfy the stoichiometric relationship.

[0057] Furthermore, the data learning layer takes engine operating parameters as input and the target component concentration as output. The loss function of the neural network includes both data fitting terms and physical constraint terms. The formula for calculating the loss function is shown in formula (8): in, Represents the total loss function. This represents the data fitting loss term and is calculated using the following formula: , Represents the physical constraint loss term and is calculated using the following formula: , The weighting coefficients of the physical constraint terms are dimensionless and , Represents the number of training samples and , Represents the sample index and , Indicates the first The true concentration value of each sample, in ppm. Indicates the first The neural network predicts the concentration value of each sample, in ppm. Indicates the first The stoichiometric constraint loss for each sample, in ppm, denominator This ensures that the division is effective.

[0058] Optionally, The value is 0.1. It is understood that those skilled in the art can adjust the value of the above parameter according to the actual situation, and the embodiments of the present invention do not specifically limit this.

[0059] Furthermore, the neural network outputs the concentration inference value and appends a confidence interval calculated based on the model's prediction variance. The lower bound of the confidence interval is... The upper boundary is ,in, This represents the concentration of the target component predicted by the neural network, expressed in ppm. The coefficients representing the confidence levels are dimensionless and , The standard deviation of the predicted variance is expressed in ppm. .

[0060] Alternatively, for a 95% confidence level, The value is 1.96. It is understood that those skilled in the art can adjust the value of the above parameter according to the actual situation, and the embodiments of the present invention do not specifically limit this.

[0061] Furthermore, the data-driven and physical equation hybrid model continuously learns and updates online using historical measured data labeled as usable levels in step S3 as training samples. Specifically, when step S3 labels the data at a certain moment as level one or level two, the engine operating parameters and corresponding multi-component concentration data at that moment are added to the training dataset. When labeled as level four, the data is not added to the training set.

[0062] In this embodiment of the invention, formula (8) adds the physical equation constraint terms as regularization terms to the loss function of the neural network, so that the model satisfies the physical conservation laws while fitting the data. Purely data-driven neural network models are trained only by minimizing the data fitting error, which may output predicted values ​​that violate physical laws. For example, the predicted oxygen concentration and air excess coefficient may not satisfy the stoichiometric relationship. However, the physical constraint neural network encodes the stoichiometric relationship of the combustion process as a hard constraint condition, and considers both the data fitting error and the degree of violation of physical constraints in the loss function, ensuring the consistency between the prediction results and physical laws.

[0063] In one possible implementation, when the data quality level marked in step S3 is level three, the execution cycle of attenuation compensation self-calibration in S2 is adaptively adjusted based on the duration for which a single component continuously exceeds the threshold. When marked as level four, the deviation pattern characteristics of the residual vector and carbon balance deviation are recorded for subsequent fault type identification.

[0064] It should be noted that those skilled in the art can set the threshold value according to actual needs, and this invention does not limit it.

[0065] Reference manual attached Figure 2 The diagram shows a schematic representation of an online analysis system for ship exhaust emissions based on intelligent sensors, provided by an embodiment of the present invention.

[0066] This invention also provides an online analysis system 20 for ship exhaust emissions based on intelligent sensors, comprising: a processor 201 and a memory 202; The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-described online analysis method for ship exhaust emissions based on smart sensors and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0067] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0068] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0069] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0070] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0071] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0072] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0073] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0074] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0075] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0076] 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 of 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.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online analysis of ship exhaust emissions based on intelligent sensors, characterized in that, Includes the following steps: S1: Obtain the hull motion attitude data through a six-axis inertial sensor, perform motion disturbance compensation based on the hull motion attitude data on the original spectral signal collected by the spectral detection sensor, and perform weighted averaging according to the hull roll angle within a preset time window to obtain the motion-compensated spectral signal. S2: Using the response signal of carbon dioxide in the motion-compensated spectral signal as the intrinsic reference standard, the response signals of each target component are self-calibrated by attenuation compensation based on the rate of change of the carbon dioxide response signal relative to the initial calibration value, and converted into preliminary multi-component concentration data. The current fuel type is identified based on the characteristic component concentration ratio of the preliminary multi-component concentration data, and the correction parameters and carbon content parameters corresponding to the current fuel type are retrieved from the multi-fuel parameter library to correct the preliminary multi-component concentration data, thereby obtaining multi-component concentration data. S3: Based on the element content parameters corresponding to the current fuel type, a residual vector is obtained through the inter-component correlation constraint model, and a carbon balance deviation is obtained through the carbon balance equation; when the residual vector and the carbon balance deviation both deviate from the normal range within a preset time window, it is determined to be a sensor offset; when the residual vector deviates from the normal range but the carbon balance deviation remains within the normal range, it is determined to be a real emission change, and the data quality level is marked according to the judgment result. The data quality level includes at least an usable level and an unusable level. S4: When the target component is marked as unavailable in step S3, the concentration value of the target component is inferred using a data-driven and physical equation hybrid model; the data-driven and physical equation hybrid model continuously learns and updates online using historical measured data marked as available in step S3 as training samples.

2. The online analysis method for ship exhaust emissions based on intelligent sensors according to claim 1, characterized in that, S1 specifically includes: S101: Obtain the six-degree-of-freedom motion attitude data of the hull through the six-axis inertial sensor. The six-degree-of-freedom motion attitude data of the hull includes at least the roll angle, pitch angle and three-axis linear acceleration. S102: Perform motion disturbance compensation on the original spectral signal to obtain a compensated spectral signal. The motion disturbance compensation includes sedimentation offset correction of particulate matter measurement values ​​based on the roll angle and pitch angle, and flow velocity fluctuation filtering of gas phase component measurement values ​​based on the triaxial linear acceleration. S103: Construct a Gaussian weighting function based on the roll angle and determine that the length of the preset time window matches the current roll cycle. Within the preset time window, perform weighted integration on the compensated spectral signal according to the Gaussian weighting function to maximize the sampling weight corresponding to the ship being in a roll equilibrium position, thereby obtaining the motion-compensated spectral signal.

3. The online analysis method for ship exhaust emissions based on intelligent sensors according to claim 1, characterized in that, In step S2, the response signal of carbon dioxide is the absorption response signal of carbon dioxide in the mid-infrared 4.26-micron band; the spectral detection of carbon dioxide, sulfur dioxide and nitrogen oxides uses the same optical path; the attenuation compensation self-calibration calculates the attenuation compensation coefficient of each target component detection channel through a pre-calibrated wavelength-related attenuation ratio coefficient.

4. The online analysis method for ship exhaust emissions based on intelligent sensors according to claim 1, characterized in that, S2 specifically includes: S201: Using the response signal of carbon dioxide in the motion-compensated spectral signal as the intrinsic reference standard, the response signals of each target component are self-calibrated by attenuation compensation according to the rate of change of the carbon dioxide response signal relative to the initial calibration value, and converted into the preliminary multi-component concentration data. S202: Calculate the concentration ratios of characteristic components and combine them into a fingerprint vector of exhaust gas components. The concentration ratios of characteristic components include at least three of the following: the ratio of sulfur dioxide concentration to carbon dioxide concentration, the ratio of methane concentration to carbon dioxide concentration, the ratio of carbon monoxide concentration to carbon dioxide concentration, the ratio of nitrogen oxide concentration to carbon dioxide concentration, and the ratio of ammonia concentration to nitrogen oxide concentration. Perform a moving average on each ratio within the preset time window. S203: The exhaust gas component fingerprint vector is matched with the standard fingerprint templates corresponding to each fuel type in the multi-fuel parameter library using a pattern recognition method, and the fuel type with the highest posterior probability is taken as the identification result of the current fuel type; S204: Based on the current fuel type, retrieve the corresponding correction parameters and carbon content parameters from the multi-fuel parameter library, correct the preliminary multi-component concentration data, and obtain the multi-component concentration data.

5. The online analysis method for ship exhaust emissions based on intelligent sensors according to claim 1 or 4, characterized in that, In S3, the carbon balance equation is established based on the fuel consumption rate, the carbon content parameter, and the exhaust flow rate. The required fuel consumption rate and engine speed data are obtained in real time from the ship engine control system through a data interface. The exhaust flow rate is measured by an exhaust pipe flow velocity sensor or calculated based on the engine intake flow rate and fuel consumption rate. When the carbon balance deviation continuously exceeds a preset threshold for a preset duration, the recalibration of the attenuation compensation self-calibration in S2 is triggered.

6. The online analysis method for ship exhaust emissions based on intelligent sensors according to claim 1, characterized in that, In S3, the component correlation constraint model is a constraint model established based on the stoichiometric relationship between the main measured components. The residual vector is obtained by comparing the actual measured component concentration with the theoretical component concentration calculated based on the element content parameters corresponding to the current fuel type.

7. The online analysis method for ship exhaust emissions based on intelligent sensors according to claim 6, characterized in that, The data quality levels in S3 are divided into four levels: When all components of the residual vector are within a first preset threshold range and the carbon balance deviation is within a first deviation threshold range, it is marked as the first level. When each component of the residual vector exceeds the first preset threshold range but is within the second preset threshold range, it is marked as the second level. When the vector angle between the direction of the residual vector and the feature direction of a certain emission event type in the pre-built emission event feature template library is less than the preset angle threshold, the marking level will be reverted to the first level. When a single component in the residual vector continuously exceeds the second preset threshold range within the preset monitoring window, but the carbon balance deviation remains within the normal range, it is marked as the third level and S2 is triggered to re-execute the attenuation compensation self-calibration. When multiple components in the residual vector simultaneously exceed the second preset threshold range and the carbon balance deviation synchronously exceeds the second deviation threshold range, it is marked as the fourth level and the virtual sensor takeover of S4 is triggered. The first level, the second level, and the third level correspond to the available level, and the fourth level corresponds to the unavailable level.

8. The online analysis method for ship exhaust emissions based on intelligent sensors according to claim 1, characterized in that, In S4, the data-driven and physical equation hybrid model is a physical constraint neural network model, which includes a physical constraint layer and a data learning layer. The physical constraint layer encodes the physical equations of the engine combustion process into hard constraints of a neural network. The physical equations include stoichiometric equations of oxygen consumption and carbon dioxide generation based on the excess air coefficient. The data learning layer takes engine operating parameters as input and target component concentration as output, outputting concentration inference values ​​and adding confidence intervals calculated based on model prediction variance.

9. The online analysis method for ship exhaust emissions based on intelligent sensors according to claim 7, characterized in that, When the data quality level labeled in S3 is the third level, the execution cycle of attenuation compensation self-calibration in S2 is adaptively adjusted according to the duration for which the single component continuously exceeds the threshold; when labeled as the fourth level, the deviation mode characteristics of the residual vector and the carbon balance deviation are recorded for subsequent fault type identification.

10. An online analysis system for ship exhaust emissions based on intelligent sensors, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the online analysis method for ship exhaust emissions based on intelligent sensors as described in any one of claims 1 to 9.

Images