Gas pollution traceability analysis method and system based on multi-source data

Abstract

The application discloses a gas pollution tracing analysis method and system based on multi-source data and belongs to the technical field of data analysis. The method comprises the following steps: collecting the ammonia injection amount, the inlet nitrogen oxide concentration and the flue gas amount time sequence of each denitration unit of a shared total discharge port, and the ammonia escape concentration time sequence of the total discharge port; calculating the instantaneous molar ratio of each denitration unit, constructing an excessive ammonia injection factor sequence according to the part of the instantaneous molar ratio exceeding the preset stoichiometric ratio, and constructing an ammonia injection diffusion factor sequence according to the ammonia injection amount time sequence and the flue gas amount time sequence; taking the ammonia escape concentration time sequence as a response variable and the two types of factor sequences as regression factors to solve a regression equation, obtaining the excessive ammonia injection response coefficient and the branch valve leakage coefficient of each denitration unit, and determining the ammonia escape abnormal type of each denitration unit. The application can locate the denitration unit with ammonia escape abnormality in the shared total discharge port and determine the abnormal type.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, specifically to a method and system for tracing and analyzing gas pollution sources based on multi-source data. Background Technology

[0002] Flue gas denitrification is a core step in achieving nitrogen oxide emission reduction in industries such as thermal power generation and steel metallurgy. Selective catalytic reduction (SCR) denitrification processes require the injection of reducing agents such as ammonia into the flue gas. However, it is difficult to precisely match the amount of reducing agent used with the flue gas conditions and catalyst activity. Unreacted reducing agent is discharged with the flue gas, resulting in ammonia escape. This escaped ammonia not only causes secondary pollution but also readily reacts with acidic components in the flue gas to form ammonium salts, leading to ash accumulation, blockage, and corrosion in downstream equipment. Therefore, the monitoring and control of ammonia escape has become an important aspect of environmental supervision of stationary pollution sources.

[0003] Currently, due to site layout and flue gas pipeline network constraints, multiple pieces of equipment commonly share the same main exhaust outlet in engineering projects. Online monitoring values ​​for ammonia escape at the exhaust outlet only reflect the overall level of the mixed flue gas from each piece of equipment. Existing online monitoring at exhaust outlets essentially only obtains the combined concentration resulting from the contributions of each piece of equipment. When this concentration becomes abnormal, it's impossible to trace the source to the specific faulty equipment or identify whether the anomaly stems from excessive reductant release or valve leakage. Since both causes exhibit the same appearance in the mixed concentration, they are different causes with the same effect, making it difficult to distinguish them based solely on the monitoring values. Furthermore, the transmission time of flue gas from each piece of equipment to the exhaust outlet varies. Leaked reductant, after vaporization and adsorption on the pipe wall, is slowly released, distorting the temporal shape of the mixed signal. Directly relying on temporal changes in mixed concentration for manual judgment is highly prone to misjudgment. Current troubleshooting methods mainly rely on shutting down and isolating each piece of equipment for testing, which requires production interruption, is time-consuming, costly, and the changes in operating conditions during testing may mask the true anomaly. Therefore, how to locate abnormal equipment and distinguish the causes of abnormalities from mixed monitoring signals without interrupting production is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for tracing and analyzing gas pollution sources based on multi-source data, so as to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides a gas pollution source tracing and analysis method based on multi-source data, comprising:

[0006] The time series of ammonia injection, inlet nitrogen oxide concentration, and flue gas volume of each denitrification unit sharing a common main outlet were collected, as well as the time series of ammonia escape concentration at the main outlet.

[0007] The ammonia injection time series refers to the pure ammonia mass flow rate sequence obtained by multiplying the ammonia volume flow rate by the preset ammonia mass concentration. The ammonia volume flow rate is collected by an electromagnetic flow meter on the ammonia injection pipeline of the denitrification unit at a set time step.

[0008] The inlet nitrogen oxide concentration time series refers to the nitrogen oxide concentration sequence collected by the continuous flue gas monitoring system at the reactor inlet of the denitrification unit at a set time step. This concentration, together with the flue gas volume, characterizes the nitrogen oxide load to be removed from each unit.

[0009] Flue gas flow time series refers to the flue gas flow sequence collected by the flow meter at the outlet flue of the denitrification unit at a set time step.

[0010] Ammonia escape concentration time series refers to the ammonia concentration sequence collected by a tunable semiconductor laser absorption spectrometer at the main discharge port at a set time step. Concentration inversion is achieved based on the characteristic absorption of ammonia by near-infrared laser of a specific wavelength, with a rapid response and no need for chemical reagents.

[0011] The four types of sequences mentioned above come from different instruments, and their sampling starting points and sampling frequencies are not completely consistent. They need to be unified to the same time base before joint calculation can be performed.

[0012] Therefore, each sequence is aligned using linear interpolation according to a set time step. Data from any denitrification unit during its start-up or shutdown period is removed based on the unit's operating status record. The aligned sequence is then truncated to a length of [length missing]. sequence.

[0013] Data from start-up and shutdown periods were removed because parameters changed drastically during these periods, failing to meet the assumption of stable operation. Data from subsequent pollution correlation and regression calculations were retained.

[0014] in, The length of the time window; As time number, and ; This refers to the total number of all denitrification units sharing the main discharge outlet; Number the denitrification unit, and .

[0015] After completing sequence alignment, it is necessary to first clean up outliers in the sequence and estimate the time lag between the unit and the outlet.

[0016] Hampel identification method was used to replace outliers in the time series of ammonia injection rate, inlet nitrogen oxide concentration, flue gas volume, and ammonia slip concentration.

[0017] Preset window length Taking each time point as the center, the length is... The window is defined, and the data at the intersection of the window and the sequence are taken at the time when the window is located at the boundary of the sequence. The moving median and the absolute median difference of the data within the window are calculated, and the values ​​that deviate from the moving median by more than three times the absolute median difference are replaced with the moving median at that time.

[0018] The flue gas from each unit needs to be transported through the flue for a period of time before reaching the main exhaust port. Therefore, the changes in ammonia injection on the unit side are reflected in the exhaust port readings with a lag. In order to correctly match the data on both sides in subsequent calculations, this lag needs to be estimated first.

[0019] Preset maximum delay step ,Will to Each integer in the array is used as a candidate delay step. Calculate the first Time series of ammonia injection rates of the denitrification unit The sequence of values, and the time numbering of each moment in the ammonia escape concentration time series. The Pearson cross-correlation coefficient between sequences formed by the values ​​of .

[0020] Among them, time number satisfy Not greater than The candidate delay step size corresponding to the Pearson cross-correlation coefficient with the largest absolute value. Recorded as the number Transmission delay step of the denitrification unit .

[0021] The cross-correlation coefficient reaches its peak at the candidate delay, indicating that the unit-side sequence best matches the outlet sequence after shifting by this step size. Therefore, it is used as the transmission delay of the unit.

[0022] Get the Taiwan denitrification unit at all times The values ​​of the ammonia injection rate time series The values ​​of the inlet nitrogen oxide concentration time series The values ​​of flue gas volume over time Substitute into the formula to calculate the instantaneous molar ratio :

[0023] ;

[0024] This formula calculates the first... Taiwan denitrification unit at all times The actual ammonia molar ratio, which is the ratio of the amount of ammonia in the reducing agent to the amount of nitrogen oxides to be removed, is the basic quantity for measuring whether the ammonia injection is excessive.

[0025] When the molar ratio is higher than the stoichiometric ratio, theoretically there is excess ammonia that has not participated in the reaction.

[0026] The numerator and denominator were originally both mass flow rates, multiplied by Converting to molar units allows for direct comparison of molar ratios with stoichiometric ratios.

[0027] In the formula, This is the molar mass ratio of nitrogen oxides to ammonia, used to convert the mass ratio to a molar ratio. Similarly, the instantaneous molar ratio for each denitrification unit is calculated separately.

[0028] The escape caused by excessive ammonia injection is only related to the portion of the molar ratio exceeding the stoichiometric ratio; based on this, an excess ammonia injection factor is constructed. According to the... The excess ammonia injection factor is calculated for the portion of the instantaneous molar ratio of the denitrification unit that exceeds the preset stoichiometric ratio. :

[0029] ;

[0030] use Function as nonlinear truncation:

[0031] When the molar ratio does not exceed the stoichiometric ratio, this factor is zero, and there is no escape contribution due to over-injection. When it exceeds the stoichiometric ratio, the excess is taken and converted into the concentration scale of the unit's share of the total exhaust gas.

[0032] In the formula, The preset stoichiometric ratio serves as the theoretical benchmark for the amount of ammonia to be injected, and is used to cut off the threshold.

[0033] The mass flow rate of nitrogen oxides in this unit is given, and the dimensionless excess amplitude is converted back to a concentration scale.

[0034] For a moment The total exhaust gas volume is the direct sum of the exhaust gas volumes of all units at the same moment. Since the total volume of exhaust gas is approximately conserved when it flows through the flue, this summation can serve as a unified benchmark for mixing and dilution at the exhaust outlet, used to calculate the unit's share in the mixed exhaust gas. The calculation formula is:

[0035] ;

[0036] Will Arranged in chronological order as number The excess ammonia injection factor sequence for each denitrification unit is constructed. Similarly, excess ammonia injection factor sequences for each denitrification unit are constructed.

[0037] Unlike overspray, the ammonia leaking from the branch valve does not undergo nozzle atomization and metering regulation. Its release process in the flue exhibits a slow tailing characteristic, which needs to be characterized by a separate factor.

[0038] According to the Taiwan denitrification unit at all times The values ​​of the ammonia injection rate time series And the value of the total exhaust gas volume. Calculate the ammonia diffusion factor The calculation formula is:

[0039] ;

[0040] ;

[0041] The ammonia diffusion factor represents the process by which the total amount of ammonia injected is slowly diffused to the discharge outlet after dilution.

[0042] The current value is a first-order weighted mixture of the previous value and the current instantaneous share. The leaked liquid ammonia must be vaporized and slowly released by the adsorption-desorption effect of the pipe wall. This recursive formula simulates this tailing process with a first-order inertial element.

[0043] The smaller the tail, the longer it is; the closer it is to 1, the more it degenerates into an instantaneous share. It effectively depicts the time scale of ammonia leakage from entering the flue to reaching the discharge outlet.

[0044] Will Arranged in chronological order as number The ammonia diffusion factor sequence for the denitrification unit is constructed. Similarly, the ammonia diffusion factor sequence for each denitrification unit is constructed.

[0045] In the formula, For the first The slow-release coefficient of the denitrification unit satisfies: Specific settings include:

[0046] Since the rate of slow release depends on the flue length, flue temperature, and pipe wall condition of each unit and cannot be predetermined, the following data-driven method is used for unit-by-unit calibration:

[0047] A finite number of candidate values ​​are pre-defined within the interval (0,1] as a candidate set; for each candidate value in the candidate set, the candidate value is used as the mitigation coefficient and substituted into the above recursive formula to calculate the... The ammonia diffusion factor sequence of the denitrification unit.

[0048] For each ammonia diffusion factor sequence, calculate the time number in the sequence. The sequence of values, and the time numbering of each moment in the ammonia escape concentration time series. The Pearson cross-correlation coefficient between sequences formed by the values ​​of .

[0049] Among them, time number satisfy Not greater than The candidate value corresponding to the largest absolute value of the Pearson cross-correlation coefficient in the candidate set is determined as the first candidate. The slow-release coefficient of the denitrification unit That is, the candidate value that best matches the shape of the discharge sequence is used as the slow release coefficient of the unit.

[0050] At this point, both types of regression factors have been constructed. The abnormal leakage of the branch valve is expressed as a regression factor with a smooth tail pattern, which distinguishes the time pattern of the excessive ammonia injection factor with an instantaneous pattern, so that the two types of causes can be separated in the same regression.

[0051] The cause of excessive ammonia injection is expressed as a regression factor with a clear time morphology by using the excessive ammonia injection factor. The morphological difference between this factor and the ammonia diffusion factor is the mathematical basis for the separability of the two types of anomalies. Based on this, the two types of factors can be incorporated into a unified regression framework.

[0052] A regression equation was constructed using the ammonia escape concentration time series as the response variable and the excess ammonia injection factor series and ammonia diffusion factor series as regression factors. Specifically, it included:

[0053] Constructing a design matrix With response vector ,remember for The maximum value in the transmission delay step of the denitrification unit. Setting the starting time ensures that the factor values ​​of each unit have been collected at that time, avoiding the reference of non-existent sequence elements.

[0054] For from to Each time number Design matrix The corresponding row of Listed as number The first of the excess ammonia injection factor sequence of the denitrification unit The value, the first value in this row Listed as number The first of the ammonia diffusion factor sequence of the denitrification unit Values, response vector The corresponding element is the first element in the ammonia escape concentration time series. Values.

[0055] Constructing the regression equation: The ammonia escape concentration at the discharge outlet is equal to the linear superposition of the contributions of the two types of factors to each unit plus the residual.

[0056] Each of the two units contributes one excess ammonia injection factor column and one ammonia injection diffusion factor column, totaling... One regression factor. A single solution simultaneously completes the unit location and cause determination, serving as a direct implementation for uninterrupted decoupling.

[0057] Construct an optimization model and define the objective function. :

[0058] ;

[0059] The first term measures the fitting error of the regression equation to the mixed concentration, while the second term constrains the overall magnitude of the coefficient vector.

[0060] In the coefficient vector Under the constraint that all components are greater than or equal to zero, solve for the objective function. The coefficient vector that takes the minimum value .

[0061] Combined with nonnegativity constraints, a nonnegative ridge regression optimization model is constructed:

[0062] The nonnegativity constraint guarantees the physical interpretation of the coefficients, and the regularization term suppresses coefficient oscillations when there is correlation between factors. A trade-off between fitting accuracy and stability.

[0063] In actual operation, the two types of factor sequences of each unit inevitably have a certain correlation. Without regularization constraints, even small data disturbances can cause large fluctuations in the coefficients.

[0064] In the formula, for OK The response vector of the column, whose elements are the time series values ​​of ammonia slip concentration at the discharge outlet, serves as the interpreted mixed monitoring signal. Among them, .

[0065] Design Matrix for OK A matrix of columns, the first column Listed as number Excessive ammonia injection factor from the unit, the first The columns are ammonia diffusion factors, and the rows are aligned according to the transmission delay of each unit.

[0066] The design matrix carries two types of factors for all units, and the delay misalignment mapping eliminates the transmission time difference.

[0067] for OK The coefficient vector of the column, The The component is the first The response coefficient of excessive ammonia injection in the denitrification unit, the first The component is the first Leakage coefficient of branch valves of the denitrification unit.

[0068] As variables to be determined, the significance and magnitude of each component are directly used for anomaly localization and cause determination.

[0069] This is the residual vector, used to absorb measurement noise and other unmodeled sources. This is a regularization parameter used to control the relative weights of the two terms. Let Euclidean norm be defined as the norm operator.

[0070] Before solving, it is necessary to confirm that the two types of factors for each unit are statistically distinguishable; otherwise, the two types of coefficients for the same unit will interfere with each other and lose their interpretable meaning. Therefore, an identifiability check is performed:

[0071] When the When the absolute value of the Pearson cross-correlation coefficient between the excess ammonia injection factor sequence and the ammonia diffusion factor sequence of the denitrification unit exceeds a set correlation threshold, the [missing information] will be [missing information]. The denitrification unit is marked as a denitrification unit of unidentifiable type.

[0072] When the variance of either of the two sequences is zero, it means that one of the two types of factors for the unit remains constant within the window, the correlation coefficient loses its definition basis, and the Pearson cross-correlation coefficient is not calculated.

[0073] After identifiability is confirmed, the optimization model is solved, and the active set is initialized. Empty set, active set The elements in are coefficient vectors The component numbering, the coefficient vector All components are set to zero. The gradient is then calculated. :

[0074] ;

[0075] The above equation represents the negative half-value of the gradient of the objective function, which is the dual variable in the Lawson-Hanson active set method.

[0076] Its symbolic structure carries KKT optimality information: under non-negativity constraints, the dual variables of the components within the active set are zero, and the components outside the active set must be no greater than zero; based on this, the entry criterion and termination condition are formed.

[0077] In the formula, The purpose is to design the transpose of the matrix to project the residuals back into the coefficient space.

[0078] The current residual is used to fit the gap information; This contributes to the gradient of the regularization term, ensuring consistency with the regularized objective function.

[0079] Active collection The element is the component number where the non-negative constraint takes effect, used to define the set of components currently involved in solving the subproblem.

[0080] The number is not in the active set. In each component, when the gradient When the maximum component is greater than zero, add the index corresponding to the maximum component to the active set. .

[0081] Solving using only the active set The component corresponding to the number in the middle is the independent variable and is based on For the least squares subproblem of regularization parameters, when there are negative components in the solution of the subproblem, the index corresponding to the negative component is removed from the active set. and for active collection The components corresponding to the remaining numbers are solved again until all solutions to the subproblems are non-negative.

[0082] The alternation of removal and resolving ensures that the solutions in the active set satisfy the non-negativity constraint at any time, thus maintaining the physical meaning of the solutions.

[0083] Repeat the above two steps until the number is no longer in the active set. The gradients corresponding to each component in The components are all not greater than zero, and the coefficient vector at this time is... This solution serves as the solution to the optimization model. Since the objective function is convex, this solution is the global optimum.

[0084] The magnitude of the coefficient estimate alone is insufficient to determine anomalies; it is also necessary to assess its stability relative to sampling fluctuations. Therefore, a significance test is performed.

[0085] Design Matrix With response vector Pair by row and divide into lengths according to time sequence. A continuous block of data, when the last block is less than the length The actual number of rows in the data block is retained at that time.

[0086] The reason for dividing the data into time blocks rather than resampling point by point is that flue gas data has strong temporal autocorrelation, and point-by-point resampling would underestimate the sampling fluctuation of the coefficients.

[0087] The number of random samples with replacement is Divide by The data block whose quotient is rounded up is the data block from the design matrix. The row and the response vector The elements are concatenated in the order of extraction to form the resampling design matrix and the resampling response vector.

[0088] For each extraction, the design matrix in the optimization model will be... With response vector The replacement is done with the resampled design matrix and resampled response vector obtained in this extraction. The Lawson-Hanson active set method is used to solve the optimized model after the replacement, resulting in a set of resampled coefficient vectors, which are repeated a total of 10 times. Next. Calculate the significance statistic:

[0089] ;

[0090] The molecule is the first The full sample point estimate of the coefficients, with the denominator being the robust dispersion under the moving block bootstrap method, and 1.349 converting the interquartile range to the equivalent standard deviation.

[0091] Essentially, it is a multiple of the coefficient amplitude based on its own sampling fluctuation. The same significance threshold can be applied fairly to coefficients of different magnitudes and different unit sizes.

[0092] when When the value exceeds the set significance threshold, the determination coefficient vector The Each component is significant; when zero and When greater than zero, the determination coefficient vector The Each component is significant, ensuring that the degradation status can be determined.

[0093] when Pick When, coefficient vector The The component is the first The response coefficient of excessive ammonia injection in the denitrification unit; when Pick When, coefficient vector The The component is the first Leakage coefficient of branch valves of the denitrification unit.

[0094] In the formula, coefficient vector The component number and , To adopt a design matrix The coefficient vector obtained by solving the optimization model for all rows The Each component.

[0095] for The first group of resampling coefficients in the vector The difference between the upper and lower quartiles of each component value This is the conversion constant between the interquartile range and the standard deviation of the standard normal distribution.

[0096] The preset data block length serves as the block scale for the moving block bootstrap method, while preserving temporal autocorrelation. The preset number of resampling times, which serves as the number of resampling groups, determines the stability of the dispersion estimate.

[0097] Significance confirms that the coefficient is not just randomly greater than zero, while threshold determination further requires the coefficient to reach a magnitude that is meaningful in engineering. The combination of the two can avoid misreporting small disturbances as anomalies.

[0098] When the When a denitrification unit is not marked as an indistinguishable type denitrification unit, or when the response coefficient of excessive ammonia injection is significantly greater than or equal to the first preset threshold, the first... The ammonia escape anomaly of the denitrification unit was determined to be due to excessive ammonia injection.

[0099] When the When the denitrification unit is not marked as an indistinguishable type denitrification unit, or when the leakage coefficient of the branch valve is significant and greater than or equal to the second preset threshold, the first... The ammonia escape anomaly of the denitrification unit was determined to be a leak in the ammonia injection branch valve.

[0100] When the When a denitrification unit meets both of the above criteria, the first one will be... The ammonia escape anomaly of the denitrification unit was determined to be a complex anomaly.

[0101] Similarly, the excess ammonia injection response coefficient and branch valve leakage coefficient of each denitrification unit are calculated to determine the ammonia escape anomaly type of each denitrification unit.

[0102] according to A topology diagram of the denitrification unit is drawn to show the connection relationship between the denitrification unit and the main exhaust outlet. The topology diagram is drawn according to the actual flue gas flow path, and operators can directly locate the abnormal unit in the plant area by following the diagram:

[0103] On the topology diagram of the denitrification unit, the first, second, third, and fourth marking symbols are used to identify denitrification units that are characterized by excessive ammonia injection, leakage of ammonia injection branch valves, combined abnormalities, and denitrification units whose types cannot be distinguished.

[0104] The present invention also provides a gas pollution source tracing and analysis system based on multi-source data, including a data acquisition module, a factor construction module, a regression solution module, and an anomaly location module.

[0105] The data acquisition module collects the time series of ammonia injection, inlet nitrogen oxide concentration, flue gas volume, and ammonia escape concentration at the total discharge outlet for each denitrification unit sharing the same main discharge outlet.

[0106] The factor construction module calculates the instantaneous molar ratio of each denitrification unit based on the time series of ammonia injection, inlet nitrogen oxide concentration, and flue gas volume. It constructs the excess ammonia injection factor sequence for each denitrification unit based on the portion of the instantaneous molar ratio exceeding the preset stoichiometric ratio, and constructs the ammonia injection diffusion factor sequence for each denitrification unit based on the time series of ammonia injection and flue gas volume.

[0107] The regression solution module constructs a regression equation using the ammonia slip concentration time series as the response variable and the excess ammonia injection factor series and ammonia diffusion factor series as regression factors. The non-negative least squares method is used to solve the regression equation, obtaining the excess ammonia injection response coefficient and valve leakage coefficient for each denitrification unit.

[0108] The anomaly location module compares the excessive ammonia injection response coefficient with the first preset threshold and the branch valve leakage coefficient with the second preset threshold to determine the ammonia escape anomaly type of each denitrification unit.

[0109] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0110] Existing technologies only output the total ammonia slip concentration at the discharge port, and anomaly investigation relies on manual isolation testing of each unit, which requires production to be interrupted. This solution does not require shutdown or new hardware, and can complete the analysis using only the existing DCS ammonia injection rate, CEMS inlet nitrogen oxides, flue gas volume and discharge port laser spectral ammonia slip data. The entire investigation process does not affect production operation.

[0111] This solution can demix the concentration of ammonia escape at the discharge port down to the unit level, and simultaneously output the abnormality type of each unit (excessive ammonia injection, branch valve leakage, combined abnormality). The investigation target is narrowed down from all units to specific components of specific units, and the abnormality handling can be directly carried out on the responsible unit and its valves and ammonia injection links.

[0112] This scheme constructs a design matrix based on the differentiated delays of each unit, ensuring that the factors of each unit and the discharge port readings strictly correspond to their respective delays. After this alignment, the regression is no longer affected by the systematic misalignment caused by the time difference in flue gas transmission between the units.

[0113] After the slow-release recursive correction, the temporal distortion of the leakage factor is compensated, and the smooth tail formed by slow release no longer causes missed detection. Both chronic leakage and instantaneous over-spray, which are different time scales, can be detected.

[0114] After standardization and significance determination, the judgment conclusion of this scheme does not drift with the group size and the number of resampling times. The same judgment standard can be applied fairly to each unit. When the data is insufficient, no judgment is forced. Instead, the data is output with an indistinguishable type label to avoid misjudgment. Attached Figure Description

[0115] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0116] Figure 1 This is a flowchart illustrating the gas pollution source tracing and analysis method based on multi-source data of the present invention;

[0117] Figure 2 This is a schematic diagram of the calibration curves for the transmission delay step size and the mitigation coefficient of the present invention;

[0118] Figure 3 This is a schematic diagram comparing the temporal morphology of the contribution concentrations of the two types of abnormal factors in this invention.

[0119] Figure 4 This is a schematic diagram of the measured values, fitted values, and residuals of the ammonia escape anomaly component at the discharge outlet of the present invention. Detailed Implementation

[0120] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0121] Please see Figure 1 This invention provides a method for tracing and analyzing gas pollution sources based on multi-source data, including:

[0122] The time series of ammonia injection, inlet nitrogen oxide concentration, and flue gas volume of each denitrification unit sharing a common main outlet were collected, as well as the time series of ammonia escape concentration at the main outlet.

[0123] The ammonia injection time series refers to the pure ammonia mass flow rate sequence obtained by multiplying the ammonia volume flow rate by the preset ammonia mass concentration. The ammonia volume flow rate is collected by an electromagnetic flow meter on the ammonia injection pipeline of the denitrification unit at a set time step.

[0124] The inlet nitrogen oxide concentration time series refers to the nitrogen oxide concentration sequence collected by the continuous flue gas monitoring system at the reactor inlet of the denitrification unit at a set time step. This concentration, together with the flue gas volume, characterizes the nitrogen oxide load to be removed from each unit.

[0125] Flue gas flow time series refers to the flue gas flow sequence collected by the flow meter at the outlet flue of the denitrification unit at a set time step.

[0126] Ammonia escape concentration time series refers to the ammonia concentration sequence collected by a tunable semiconductor laser absorption spectrometer at the main discharge port at a set time step. Concentration inversion is achieved based on the characteristic absorption of ammonia by near-infrared laser at a specific wavelength, with a rapid response and no need for chemical reagents.

[0127] The four types of sequences mentioned above come from different instruments, and their sampling starting points and sampling frequencies are not completely consistent. They need to be unified to the same time base before joint calculation can be performed.

[0128] Therefore, each sequence is aligned using linear interpolation according to a set time step. Data from any denitrification unit during its start-up or shutdown period is removed based on the unit's operating status record. The aligned sequence is then truncated to a length of [length missing]. sequence.

[0129] Data from start-up and shutdown periods were removed because parameters changed drastically during these periods, failing to meet the assumption of stable operation. Data from subsequent pollution correlation and regression calculations were retained.

[0130] in, The length of the time window; As time number, and ; This refers to the total number of all denitrification units sharing the main discharge outlet; Number the denitrification unit, and .

[0131] After completing sequence alignment, it is necessary to first clean up outliers in the sequence and estimate the time lag between the unit and the outlet.

[0132] Hampel identification method was used to replace outliers in the time series of ammonia injection rate, inlet nitrogen oxide concentration, flue gas volume, and ammonia slip concentration.

[0133] Preset window length Taking each time point as the center, the length is... The window is defined, and the data at the intersection of the window and the sequence are taken at the time when the window is located at the boundary of the sequence. The moving median and the absolute median difference of the data within the window are calculated, and the values ​​that deviate from the moving median by more than three times the absolute median difference are replaced with the moving median at that time.

[0134] The flue gas from each unit needs to be transported through the flue for a period of time before reaching the main outlet. Therefore, the changes in ammonia injection on the unit side are reflected in the outlet readings with a lag. In order to correctly match the data on both sides in subsequent calculations, this lag needs to be estimated first.

[0135] Preset maximum delay step ,Will to Each integer in the array is used as a candidate delay step. Calculate the first Time series of ammonia injection rates of the denitrification unit The sequence of values, and the time numbering of each moment in the ammonia escape concentration time series. The Pearson cross-correlation coefficient between sequences formed by the values ​​of .

[0136] Among them, time number satisfy Not greater than The candidate delay step size corresponding to the Pearson cross-correlation coefficient with the largest absolute value. Recorded as the number Transmission delay step of the denitrification unit .

[0137] The cross-correlation coefficient reaches its peak at the candidate delay, indicating that the unit-side sequence best matches the outlet sequence after shifting by this step size. Therefore, it is used as the transmission delay of the unit.

[0138] Get the Taiwan denitrification unit at all times The values ​​of the ammonia injection rate time series The values ​​of the inlet nitrogen oxide concentration time series The values ​​of flue gas volume over time Substitute into the formula to calculate the instantaneous molar ratio :

[0139] ;

[0140] This formula calculates the first... Taiwan denitrification unit at all times The actual ammonia molar ratio, which is the ratio of the amount of ammonia in the reducing agent to the amount of nitrogen oxides to be removed, is the basic quantity for measuring whether the ammonia injection is excessive.

[0141] When the molar ratio is higher than the stoichiometric ratio, theoretically there is excess ammonia that has not participated in the reaction.

[0142] The numerator and denominator were originally both mass flow rates, multiplied by Converting to molar units allows for direct comparison of molar ratios with stoichiometric ratios.

[0143] In the formula, This is the molar mass ratio of nitrogen oxides to ammonia, used to convert the mass ratio to a molar ratio. Similarly, the instantaneous molar ratio for each denitrification unit is calculated separately.

[0144] The escape caused by excessive ammonia injection is only related to the portion of the molar ratio exceeding the stoichiometric ratio; based on this, an excess ammonia injection factor is constructed. According to the... The excess ammonia injection factor is calculated for the portion of the instantaneous molar ratio of the denitrification unit that exceeds the preset stoichiometric ratio. :

[0145] ;

[0146] use Function as nonlinear truncation:

[0147] When the molar ratio does not exceed the stoichiometric ratio (insufficient or just right ammonia injection), this factor is zero, and there is no escape contribution due to over-injection. When it exceeds the stoichiometric ratio, the excess is taken and converted into the concentration scale of the unit's share of the total exhaust gas.

[0148] In the formula, The preset stoichiometric ratio serves as the theoretical benchmark for the amount of ammonia to be injected, and is used to cut off the threshold.

[0149] The mass flow rate of nitrogen oxides in this unit is given, and the dimensionless excess amplitude is converted back to a concentration scale.

[0150] For a moment The total exhaust gas volume is the direct sum of the exhaust gas volumes of all units at the same moment. Since the total volume of exhaust gas is approximately conserved when it flows through the flue, this summation can serve as a unified benchmark for mixing and dilution at the exhaust outlet, used to calculate the unit's share in the mixed exhaust gas. The calculation formula is:

[0151] ;

[0152] Will Arranged in chronological order as number The excess ammonia injection factor sequence for each denitrification unit is constructed. Similarly, excess ammonia injection factor sequences for each denitrification unit are constructed.

[0153] Unlike overspray, the ammonia leaking from the branch valve does not undergo nozzle atomization and metering regulation. Its release process in the flue exhibits a slow tailing characteristic, which needs to be characterized by a separate factor.

[0154] According to the Taiwan denitrification unit at all times The values ​​of the ammonia injection rate time series And the value of the total exhaust gas volume. Calculate the ammonia diffusion factor The calculation formula is:

[0155] ;

[0156] ;

[0157] The ammonia diffusion factor represents the process by which the total amount of ammonia injected is slowly diffused to the discharge outlet after dilution.

[0158] The current value is a first-order weighted mixture of the previous value and the current instantaneous share. The leaked liquid ammonia must be vaporized and slowly released by the adsorption-desorption effect of the pipe wall. This recursive formula simulates this tailing process with a first-order inertial element.

[0159] The smaller the tail, the longer it is; the closer it is to 1, the more it degenerates into an instantaneous share. It effectively depicts the time scale of ammonia leakage from entering the flue to reaching the discharge outlet.

[0160] Will Arranged in chronological order as number The ammonia diffusion factor sequence for the denitrification unit is constructed. Similarly, the ammonia diffusion factor sequence for each denitrification unit is constructed.

[0161] In the formula, For the first The slow-release coefficient of the denitrification unit satisfies: Specific settings include:

[0162] Since the rate of slow release depends on the flue length, flue temperature, and pipe wall condition of each unit and cannot be predetermined, the following data-driven method is used for unit-by-unit calibration:

[0163] A finite number of candidate values ​​are pre-defined within the interval (0,1] as a candidate set; for each candidate value in the candidate set, the candidate value is used as the mitigation coefficient and substituted into the above recursive formula to calculate the... The ammonia diffusion factor sequence of the denitrification unit.

[0164] For each ammonia diffusion factor sequence, calculate the time number in the sequence. The sequence of values, and the time numbering of each moment in the ammonia escape concentration time series. The Pearson cross-correlation coefficient between sequences formed by the values ​​of .

[0165] Among them, time number satisfy Not greater than The candidate value corresponding to the largest absolute value of the Pearson cross-correlation coefficient in the candidate set is determined as the first candidate. The slow-release coefficient of the denitrification unit That is, the candidate value that best matches the shape of the discharge sequence is used as the slow release coefficient of the unit.

[0166] At this point, both types of regression factors have been constructed. The abnormal leakage of the branch valve is expressed as a regression factor with a smooth tail pattern, which distinguishes the time pattern of the excessive ammonia injection factor with an instantaneous pattern, so that the two types of causes can be separated in the same regression.

[0167] The cause of excessive ammonia injection is expressed as a regression factor with a clear temporal form (fluctuating instantaneously with ammonia injection regulation) by using the excessive ammonia injection factor. The difference in form between this factor and the ammonia injection diffusion factor is the mathematical basis for the separability of the two types of anomalies. Based on this, the two types of factors can be incorporated into a unified regression framework.

[0168] A regression equation was constructed using the ammonia escape concentration time series as the response variable and the excess ammonia injection factor series and ammonia diffusion factor series as regression factors. Specifically, it included:

[0169] Constructing a design matrix With response vector ,remember for The maximum value in the transmission delay step of the denitrification unit. Setting the starting time ensures that the factor values ​​of each unit have been collected at that time, avoiding the reference of non-existent sequence elements.

[0170] For from to Each time number Design matrix The corresponding row of Listed as number The first of the excess ammonia injection factor sequence of the denitrification unit The value, the first value in this row Listed as number The first of the ammonia diffusion factor sequence of the denitrification unit Values, response vector The corresponding element is the first element in the ammonia escape concentration time series. Values.

[0171] Constructing the regression equation: The ammonia escape concentration (response) at the discharge outlet is equal to the linear superposition of the contributions of the two types of factors of each unit plus the residual.

[0172] Each of the two units contributes one excess ammonia injection factor column and one ammonia injection diffusion factor column, totaling... One regression factor. A single solution simultaneously completes the unit location and cause determination, serving as a direct implementation for uninterrupted decoupling.

[0173] Construct an optimization model and define the objective function. :

[0174] ;

[0175] The first term measures the fitting error of the regression equation to the mixed concentration, while the second term constrains the overall magnitude of the coefficient vector.

[0176] In the coefficient vector Under the constraint that all components are greater than or equal to zero, solve for the objective function. The coefficient vector that takes the minimum value .

[0177] Matching nonnegativity constraints ( The non-negative ridge regression optimization model is constructed with all components ≥ 0.

[0178] The nonnegativity constraint guarantees the physical interpretation of the coefficients (abnormal contributions are not negative), and the regularization term suppresses coefficient oscillations when there is correlation between factors. A trade-off between fitting accuracy and stability.

[0179] In actual operation, the two types of factor sequences of each unit inevitably have a certain correlation. Without regularization constraints, even small data disturbances can cause large fluctuations in the coefficients.

[0180] In the formula, for OK The response vector of the column, whose elements are the time series values ​​of ammonia slip concentration at the discharge outlet, serves as the interpreted mixed monitoring signal. Among them, .

[0181] Design Matrix for OK A matrix of columns, the first column Listed as number Excessive ammonia injection factor from the unit, the first The columns are ammonia diffusion factors, and the rows are aligned according to the transmission delay of each unit.

[0182] The design matrix carries two types of factors for all units, and the delay misalignment mapping eliminates the transmission time difference.

[0183] for OK The coefficient vector of the column, The The component is the first The response coefficient of excessive ammonia injection in the denitrification unit, the first The component is the first Leakage coefficient of branch valves of the denitrification unit.

[0184] As variables to be determined, the significance and magnitude of each component are directly used for anomaly localization and cause determination.

[0185] This is the residual vector, used to absorb measurement noise and other unmodeled sources. This is a regularization parameter used to control the relative weights of the two terms. Let Euclidean norm be defined as the norm operator.

[0186] Before solving, it is necessary to confirm that the two types of factors for each unit are statistically distinguishable; otherwise, the two types of coefficients for the same unit will interfere with each other and lose their interpretable meaning. Therefore, an identifiability check is performed:

[0187] When the When the absolute value of the Pearson cross-correlation coefficient between the excess ammonia injection factor sequence and the ammonia diffusion factor sequence of the denitrification unit exceeds a set correlation threshold, the [missing information] will be [missing information]. The denitrification unit is marked as a denitrification unit of unidentifiable type.

[0188] When the variance of either of the two sequences is zero, it means that one of the two types of factors for the unit remains constant within the window, the correlation coefficient loses its definition basis, and the Pearson cross-correlation coefficient is not calculated.

[0189] After identifiability is confirmed, the optimization model is solved, and the active set is initialized. Empty set, active set The elements in are coefficient vectors The component numbering, the coefficient vector All components are set to zero. The gradient is then calculated. :

[0190] ;

[0191] The above equation represents the negative half-value of the gradient of the objective function, which is the dual variable in the Lawson-Hanson active set method.

[0192] Its symbolic structure carries KKT optimality information: under non-negativity constraints, the dual variables of the components within the active set are zero, and the components outside the active set must be no greater than zero; based on this, the inclusion criterion is formed (the dual variables of the components outside the active set are zero). If the maximum component is greater than zero, then the number is included in the set) and the termination condition (all components outside the active set are ≤0).

[0193] In the formula, The purpose is to design the transpose of the matrix to project the residuals back into the coefficient space.

[0194] The current residual is used to fit the gap information; This contributes to the gradient of the regularization term, ensuring consistency with the regularized objective function.

[0195] Active collection The element is the component number where the non-negative constraint takes effect, used to define the set of components currently involved in solving the subproblem.

[0196] The number is not in the active set. In each component, when the gradient When the maximum component is greater than zero, add the index corresponding to the maximum component to the active set. .

[0197] Solving using only the active set The component corresponding to the number in the middle is the independent variable and is based on For the least squares subproblem of regularization parameters, when there are negative components in the solution of the subproblem, the index corresponding to the negative component is removed from the active set. and for active collection The components corresponding to the remaining numbers are solved again until all solutions to the subproblems are non-negative.

[0198] The alternation of removal and resolving ensures that the solutions in the active set satisfy the non-negativity constraint at any time, thus maintaining the physical meaning of the solutions.

[0199] Repeat the above two steps until the number is no longer in the active set. The gradients corresponding to each component in The components are all not greater than zero, and the coefficient vector at this time is... This solution serves as the solution to the optimization model. Since the objective function is convex, this solution is the global optimum.

[0200] The magnitude of the coefficient estimate alone is insufficient to determine anomalies; it is also necessary to assess its stability relative to sampling fluctuations. Therefore, a significance test is performed.

[0201] Design Matrix With response vector Pair by row and divide into lengths according to time sequence. A continuous block of data, when the last block is less than the length The actual number of rows in the data block is retained at that time.

[0202] The reason for dividing the data into time blocks rather than resampling point by point is that flue gas data has strong temporal autocorrelation, and point-by-point resampling would underestimate the sampling fluctuation of the coefficients.

[0203] The number of random samples with replacement is Divide by The data block whose quotient is rounded up is the data block from the design matrix. The row and the response vector The elements are concatenated in the order of extraction to form the resampling design matrix and the resampling response vector.

[0204] For each extraction, the design matrix in the optimization model will be... With response vector The replacement is done with the resampled design matrix and resampled response vector obtained in this extraction. The Lawson-Hanson active set method is used to solve the optimized model after the replacement, resulting in a set of resampled coefficient vectors, which are repeated a total of 10 times. Next. Calculate the significance statistic:

[0205] ;

[0206] The molecule is the first The full sample point estimate of each coefficient is given by the denominator, which is the robust dispersion under the moving block bootstrap method (time-block resampling, preserving the temporal correlation structure). 1.349 converts the interquartile range to the equivalent standard deviation.

[0207] Essentially, it is a multiple of the coefficient amplitude based on its own sampling fluctuation. The same significance threshold can be applied fairly to coefficients of different magnitudes and different unit sizes.

[0208] when When the value exceeds the set significance threshold, the determination coefficient vector The Each component is significant; when zero and When greater than zero, the determination coefficient vector The Each component is significant, ensuring that the degradation status can be determined.

[0209] when Pick When, coefficient vector The The component is the first The response coefficient of excessive ammonia injection in the denitrification unit; when Pick When, coefficient vector The The component is the first Leakage coefficient of branch valves of the denitrification unit.

[0210] In the formula, coefficient vector The component number and , To adopt a design matrix The coefficient vector obtained by solving the optimization model for all rows The Each component.

[0211] for The first group of resampling coefficients in the vector The difference between the upper and lower quartiles of each component value This is the conversion constant between the interquartile range and the standard deviation of the standard normal distribution.

[0212] The preset data block length serves as the block scale for the moving block bootstrap method, while preserving temporal autocorrelation. The preset number of resampling times, which serves as the number of resampling groups, determines the stability of the dispersion estimate.

[0213] Significance confirms that the coefficient is not just randomly greater than zero, while threshold determination further requires the coefficient to reach a magnitude that is meaningful in engineering. The combination of the two can avoid misreporting small disturbances as anomalies.

[0214] When the When a denitrification unit is not marked as an indistinguishable type denitrification unit, or when the response coefficient of excessive ammonia injection is significantly greater than or equal to the first preset threshold, the first... The ammonia escape anomaly of the denitrification unit was determined to be due to excessive ammonia injection.

[0215] When the When the denitrification unit is not marked as an indistinguishable type denitrification unit, or when the leakage coefficient of the branch valve is significant and greater than or equal to the second preset threshold, the first... The ammonia escape anomaly of the denitrification unit was determined to be a leak in the ammonia injection branch valve.

[0216] When the When a denitrification unit meets both of the above criteria, the first one will be... The ammonia escape anomaly of the denitrification unit was determined to be a complex anomaly.

[0217] Similarly, the excess ammonia injection response coefficient and branch valve leakage coefficient of each denitrification unit are calculated to determine the ammonia escape anomaly type of each denitrification unit.

[0218] according to A topology diagram of the denitrification unit is drawn to show the connection relationship between the denitrification unit and the main exhaust outlet. The topology diagram is drawn according to the actual flue gas flow path, and operators can directly locate the abnormal unit in the plant area by following the diagram:

[0219] On the topology diagram of the denitrification unit, the first, second, third, and fourth symbols are used to mark the denitrification units identified as having excessive ammonia injection, leaking ammonia injection valve, or compound abnormalities, as well as the denitrification units whose type cannot be distinguished.

[0220] The present invention also provides a gas pollution source tracing and analysis system based on multi-source data, including a data acquisition module, a factor construction module, a regression solution module, and an anomaly location module.

[0221] The data acquisition module collects the time series of ammonia injection, inlet nitrogen oxide concentration, flue gas volume, and ammonia escape concentration at the total discharge outlet for each denitrification unit sharing the same main discharge outlet.

[0222] The factor construction module calculates the instantaneous molar ratio of each denitrification unit based on the time series of ammonia injection, inlet nitrogen oxide concentration, and flue gas volume. It constructs the excess ammonia injection factor sequence for each denitrification unit based on the portion of the instantaneous molar ratio exceeding the preset stoichiometric ratio, and constructs the ammonia injection diffusion factor sequence for each denitrification unit based on the time series of ammonia injection and flue gas volume.

[0223] The regression solution module constructs a regression equation using the ammonia slip concentration time series as the response variable and the excess ammonia injection factor series and ammonia diffusion factor series as regression factors. The non-negative least squares method is used to solve the regression equation, obtaining the excess ammonia injection response coefficient and valve leakage coefficient for each denitrification unit.

[0224] The anomaly location module compares the excessive ammonia injection response coefficient with the first preset threshold and the branch valve leakage coefficient with the second preset threshold to determine the ammonia escape anomaly type of each denitrification unit.

[0225] Example 1:

[0226] A thermal power plant has three 300 MW coal-fired denitrification units sharing the same main exhaust outlet. The flue gas volume of each unit is approximately 980,000, 1,050,000, and 1,010,000 cubic meters per hour, respectively, and the nitrogen oxide concentration at the reactor inlet is approximately 375, 392, and 368 mg / m³, respectively.

[0227] With a time step of 1 minute, 720 minutes of stable operating data were collected, and start-up and shutdown periods were excluded according to the unit's operating status. The ammonia concentration was preset to 20%, the pure ammonia mass flow rate was calculated from the reading of the electromagnetic flowmeter in the ammonia injection pipeline, and the ammonia escape at the total discharge outlet was measured by a tunable semiconductor laser absorption spectrometer. The steady-state baseline was 1.21 mg / m³ for the first 30 minutes of the window, and the mixed concentration rose to a maximum of 5.87 mg / m³ during abnormal periods.

[0228] The molar mass ratio of nitrogen oxides to ammonia was set to 46 / 17, approximately 2.71, and the stoichiometric ratio was set to 1.00. The Hampel identification window consisted of 7 points, with outliers replaced by 3 times the absolute median difference. The maximum delay step size was set to 5 steps. The candidate set of sustained-release coefficients ranged from 0.50 to 0.95 with a step size of 0.05. The identifiable relevance threshold was set to 0.85, the significance threshold was set to 2.00, and the first and second preset thresholds were set to 0.070 and 0.100, respectively. The data block length for the moving block bootstrap method was set to 30, and 200 resamples were performed.

[0229] To avoid the slow-changing common-mode component of the load submerging the cross-correlation peak, the cross-correlation coefficients are calculated on the first-order difference sequence; to eliminate the constant DC component of the factors, each factor sequence and the response sequence are zero-point corrected using the steady-state median of the first 30 minutes of the window.

[0230] A total of 91 outliers were replaced using Hampel identification. The first-order differential cross-correlation peaked at steps 1, 2, and 3, with peak coefficients of 0.188, 0.241, and 0.355, respectively. Based on this, the transmission delay steps of the three units were determined to be 1, 2, and 3 minutes, respectively.

[0231] When calibrating the slow-release coefficient, the candidate curves of Units 1 and 2 had no internal peak or a range less than 0.05, indicating that their slow-release components were not significant. The calibration values ​​of 0.75 and 0.70 from the previous flue gas duct maintenance were adopted. Unit 3 was calibrated within a local window near the three step changes in its ammonia injection rate. The cross-correlation reached a peak of 0.805 at the slow-release coefficient of 0.85, so 0.85 was adopted.

[0232] The absolute values ​​of the correlation coefficients of the two types of factor sequences for each of the three units are 0.063, 0.587, and 0.061, respectively, all less than 0.85, indicating that none of them belong to units of indistinguishable type.

[0233] Please see Figure 2 The left half shows the change of the first-order difference cross-correlation coefficients of the three units with the candidate delay step size. The hollow circles mark their respective peak values, which fall on steps 1, 2, and 3 in turn, which is consistent with the difference in flue length of the three units.

[0234] The right half shows the change of the cross-correlation coefficient of Unit 3 with the candidate slow-release coefficient. The curve reaches a peak at 0.85, indicating that this value is the most sufficient to compensate for the slow release process caused by ammonia vaporization and adsorption-desorption on the pipe wall.

[0235] Please see Figure 3 The red curve represents the contribution of excessive ammonia injection from Unit 2 to the discharge concentration. Only during the over-injection period, a cluster of instantaneous peaks that rise and fall rapidly within a few minutes are formed as the regulating valve operates.

[0236] The blue curve represents the contribution of the ammonia diffusion factor injected by Unit 3. After the leak occurred at the 240th minute, it slowly increased in a step-like manner, and after each step, it tended to stabilize after several tens of minutes and retained a smooth trail.

[0237] Both types of anomalies show an increase in mixed concentration readings, but their temporal patterns differ, one being faster and the other slower. This forms the mathematical basis for the separability of the two causes in the same regression.

[0238] The solution was obtained by combining nonnegative ridge regression with the Lawson-Hanson active set method, and verified by 200 sets of moving block bootstrap methods.

[0239] The response coefficient for excessive ammonia injection in Unit 1 was 0.0484, and the leakage coefficient of the branch valve was 0. The significance statistics were 16.84 and 0, respectively.

[0240] The two coefficients for Unit 2 are 0.0983 and 0.0136, with significance statistics of 41.19 and 5.00.

[0241] The two coefficients for Unit 3 are 0.0432 and 0.2093, and the significance statistics are 19.43 and 141.87.

[0242] The regression goodness of fit was 0.993, and the root mean square error was 0.102 mg / m³. Based on the threshold:

[0243] Unit 1 has no coefficients that simultaneously meet significant and exceed limits, and is operating normally;

[0244] The response coefficient of excessive ammonia injection in Unit 2 is significant and greater than 0.070, and the leakage coefficient of the branch valve is less than 0.1, indicating that the ammonia injection is excessive.

[0245] The leakage coefficient of the branch valve of Unit 3 is significant and greater than 0.100, and the response coefficient of excessive ammonia injection is less than 0.070, which indicates that the branch valve of ammonia injection is leaking.

[0246] The conclusion is consistent with the subsequent shutdown verification: the ammonia injection regulating valve of Unit 2 was misaligned between the 380th and 560th minutes, and the branch valve seal of Unit 3 deteriorated from the 240th minute onwards and fluctuated twice.

[0247] Please see Figure 4 The upper part compares the measured values ​​and regression fitting values ​​of the abnormal ammonia escape component at the total discharge outlet. The two curves basically overlap, and both abnormal periods are completely reproduced.

[0248] The lower part is the time-by-time residuals. The residuals are uniformly distributed with zero as the center and the amplitude does not exceed 0.25 mg / m³. There is no systematic bias, indicating that the two types of factors have fully explained the mixed signal and there are no structural omissions.

[0249] Example 2:

[0250] The comparison between canceling delay alignment and mitigation correction: On the same data, cancel the unit-differentiated delay alignment, unify the zero-step pairing, and degenerate the mitigation recursion into instantaneous share, while the remaining parameters and solution process remain unchanged.

[0251] After resolving, the goodness of fit decreased to 0.873, and the root mean square error increased to 0.430 mg / m³.

[0252] The response coefficient for excessive ammonia injection in Unit 1 was artificially inflated to 0.0543. The abnormal contribution of Unit 2 was split into 0.0868 and 0.0271. The fault contribution crosstalked to the fault-free units, and it was impossible to distinguish between the over-injection of Unit 2 and the leakage of Unit 3 from the slow-varying mixed readings.

[0253] The comparison shows that misalignment based on transmission delay and compensation for leakage tailing using first-order inertia recursion are indispensable steps to achieve accurate demixing without shutting down the unit. When relying solely on manual assessment based on the mixing concentration at the main discharge port, neither the abnormal unit nor the cause of the abnormality can be determined, which is consistent with the limitations of existing technologies that rely on shutting down and isolating each unit for testing.

[0254] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0255] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A gas pollution source tracing analysis method based on multi-source data, characterized in that: The method includes: The time series of ammonia injection, inlet nitrogen oxide concentration and flue gas volume of each denitrification unit sharing a common total outlet were collected, as well as the time series of ammonia escape concentration at the total outlet. The instantaneous molar ratio of each denitrification unit was calculated based on the time series of ammonia injection, inlet nitrogen oxide concentration, and flue gas volume. The excess ammonia injection factor sequence of each denitrification unit was constructed according to the portion of the instantaneous molar ratio exceeding the preset stoichiometric ratio. The ammonia injection diffusion factor sequence of each denitrification unit was constructed based on the time series of ammonia injection and flue gas volume. Regression equations were constructed and solved using the ammonia escape concentration time series as the response variable and the excess ammonia injection factor series and ammonia injection diffusion factor series as regression factors to obtain the excess ammonia injection response coefficient and branch valve leakage coefficient of each denitrification unit. The response coefficient of excessive ammonia injection and the leakage coefficient of branch valves were analyzed to determine the ammonia escape anomaly type of each denitrification unit.

2. The gas pollution source tracing analysis method based on multi-source data according to claim 1, characterized in that: The ammonia injection time series refers to the pure ammonia mass flow rate sequence obtained by multiplying the ammonia volume flow rate by the preset ammonia mass concentration. The volumetric flow rate of ammonia water is collected by the electromagnetic flow meter on the ammonia injection pipeline of the denitrification unit at a set time step; the flue gas flow time series refers to the flue gas flow series collected by the flow meter at the outlet flue of the denitrification unit at a set time step. The inlet nitrogen oxide concentration time series refers to the nitrogen oxide concentration sequence collected by the continuous flue gas monitoring system at the reactor inlet of the denitrification unit at a set time step. The ammonia escape concentration time series refers to the ammonia concentration sequence collected by a tunable semiconductor laser absorption spectrometer at the main discharge outlet at a set time step. Each sequence is aligned using linear interpolation according to a set time step. Data from any denitrification unit during its start-up or shutdown period is removed based on the unit's operating status records. The aligned sequence is then truncated to a length of [length missing]. sequence; in, The length of the time window; As time number, and ; This refers to the total number of all denitrification units sharing the main discharge outlet; Number the denitrification unit, and .

3. The gas pollution source tracing analysis method based on multi-source data according to claim 2, characterized in that: After collecting each time series data and before calculating the instantaneous molar ratio of each denitrification unit, a preprocessing step is also included: Hampel identification was used to replace outliers in the time series of ammonia injection rate, inlet nitrogen oxide concentration, flue gas volume, and ammonia slip concentration. Preset window length Taking each time point as the center, the length is... The window is defined, and the data at the intersection of the window and the sequence are taken at the time when the window is located at the boundary of the sequence. The moving median and absolute median difference of the data within the window are calculated, and the values ​​that deviate from the moving median by more than three times the absolute median difference are replaced with the moving median at that time. Preset maximum delay step ,Will to Each integer in the array is used as a candidate delay step. Calculate the first Time series of ammonia injection rates of the denitrification unit The sequence of values, and the time numbering of each moment in the ammonia escape concentration time series. The Pearson cross-correlation coefficient between sequences formed by the values ​​of ; Among them, time number satisfy Not greater than The candidate delay step size corresponding to the Pearson cross-correlation coefficient with the largest absolute value. Recorded as the number Transmission delay step of the denitrification unit .

4. The gas pollution source tracing analysis method based on multi-source data according to claim 3, characterized in that: The calculation of instantaneous molar ratio, construction of excess ammonia injection factor sequence and ammonia injection diffusion factor sequence include: Get the Taiwan denitrification unit at all times The values ​​of the ammonia injection rate time series The values ​​of the inlet nitrogen oxide concentration time series The values ​​of flue gas volume time series Substitute into the formula to calculate the instantaneous molar ratio : ； In the formula, This is the molar mass ratio of nitrogen oxides to ammonia; Calculate the excess ammonia injection factor : ； In the formula, To preset the stoichiometric ratio, Arranged in chronological order as number The sequence of excessive ammonia injection factors in a denitrification unit; For a moment The total exhaust gas volume is calculated using the following formula: ； Calculate the ammonia diffusion factor ,Will Arranged in chronological order as number The ammonia diffusion factor sequence of the denitrification unit; the calculation formula is: ； ； In the formula, For the first The slow-release coefficient of the denitrification unit satisfies: Specific settings include: A finite number of candidate values ​​are pre-defined within the interval (0,1] as a candidate set; for each candidate value in the candidate set, the candidate value is used as the mitigation coefficient and substituted into the recursive formula to calculate the... Ammonia diffusion factor sequence of a denitrification unit; For each ammonia diffusion factor sequence, calculate the time number in the sequence. The sequence of values, and the time numbering of each moment in the ammonia escape concentration time series. The Pearson cross-correlation coefficient between sequences formed by the values ​​of ; Among them, time number satisfy Not greater than The candidate value corresponding to the largest absolute value of the Pearson cross-correlation coefficient in the candidate set is determined as the first candidate. The slow-release coefficient of the denitrification unit .

5. The gas pollution source tracing analysis method based on multi-source data according to claim 4, characterized in that: Constructing the regression equation includes: Constructing a design matrix With response vector ,remember for The maximum value in the transmission delay step of the denitrification unit; For from to Each time number Design matrix The corresponding row of Listed as number The first of the excess ammonia injection factor sequence of the denitrification unit The value, the first value in this row Listed as number The first of the ammonia diffusion factor sequence of the denitrification unit Values, response vector The corresponding element is the first element in the ammonia escape concentration time series. One value; Constructing the regression equation: ; Construct an optimization model and set the objective function : ； In the coefficient vector Under the constraint that all components are greater than or equal to zero, solve for the objective function. The coefficient vector that takes the minimum value ; In the formula, the design matrix for OK Column matrix, response vector for OK Column vectors, ; for OK The coefficient vector of the column, The The component is the first The response coefficient of excessive ammonia injection in the denitrification unit, the first The component is the first Leakage coefficient of branch valves in the denitrification unit. For the residual vector, For regularization parameters, It is the Euclidean norm.

6. The gas pollution source tracing analysis method based on multi-source data according to claim 5, characterized in that: Solving the optimization model includes: Initialize active set Empty set, active set The elements in are coefficient vectors The component numbering, the coefficient vector Set all components to zero; calculate the gradient. : ； The number is not in the active set. In each component, when the gradient When the maximum component is greater than zero, add the index corresponding to the maximum component to the active set. ; Solving using only the active set The component corresponding to the number in the middle is the independent variable and is based on... For the least squares subproblem of regularization parameters, when there are negative components in the solution of the subproblem, the index corresponding to the negative component is removed from the active set. and for active collection The components corresponding to the remaining numbers are solved again until all solutions to the subproblems are non-negative. Repeat the above two steps until the number is no longer in the active set. The gradients corresponding to each component in The components are all not greater than zero, and the coefficient vector at this time is... This serves as the solution to the optimization model.

7. The gas pollution source tracing analysis method based on multi-source data according to claim 6, characterized in that: Before solving the optimization model, perform an identifiability check; After solving the optimization model, a significance test is performed. Distinguishing features include: When the When the absolute value of the Pearson cross-correlation coefficient between the excess ammonia injection factor sequence and the ammonia diffusion factor sequence of the denitrification unit exceeds a set correlation threshold, the [missing information] will be [missing information]. The denitrification unit is marked as a type-indistinguishable denitrification unit; When the variance of either sequence is zero, the Pearson cross-correlation coefficient is not calculated. Significance tests include: Design Matrix With response vector Pair by row and divide into lengths according to time sequence. A contiguous block of data, when the last block is less than the length. The actual number of rows in the data block should be retained at that time. The number of random samples with replacement is Divide by The data block whose quotient is rounded up is the data block from the design matrix. The row and the response vector The elements are concatenated in the order of extraction to form the resampling design matrix and the resampling response vector; For each extraction, the design matrix in the optimization model will be... With response vector The replacement is done with the resampled design matrix and resampled response vector obtained in this extraction. The Lawson-Hanson active set method is used to solve the optimized model after the replacement, resulting in a set of resampled coefficient vectors, which are repeated a total of 10 times. Calculate the significance statistic: ； when When the value exceeds the set significance threshold, the determination coefficient vector The Each component is significant; when zero and When greater than zero, the determination coefficient vector The Each component is significant; when Pick When, coefficient vector The The component is the first The response coefficient of excessive ammonia injection in the denitrification unit; when Pick When, coefficient vector The The component is the first Leakage coefficient of branch valves in the denitrification unit; In the formula, coefficient vector The component number and , To adopt a design matrix The coefficient vector obtained by solving the optimization model for all rows The One component; for The first group of resampling coefficients in the vector The difference between the upper and lower quartiles of each component value To preset the data block length, This is the preset number of resampling attempts.

8. The gas pollution source tracing analysis method based on multi-source data according to claim 7, characterized in that: The types of ammonia slip anomalies identified for each denitrification unit include: When the When a denitrification unit is not marked as an indistinguishable type denitrification unit, or when the response coefficient of excessive ammonia injection is significantly greater than or equal to the first preset threshold, the first... The ammonia escape anomaly of the denitrification unit was determined to be excessive ammonia injection. When the When the denitrification unit is not marked as an indistinguishable type denitrification unit, or when the leakage coefficient of the branch valve is significant and greater than or equal to the second preset threshold, the first... The ammonia escape anomaly of the denitrification unit was determined to be a leak in the ammonia injection branch valve. When the When a denitrification unit meets both of the above criteria, the first one will be... The ammonia escape anomaly of the denitrification unit was determined to be a complex anomaly. according to A topology diagram of the denitrification unit was drawn to show the connection relationship between the denitrification unit and the main exhaust outlet. The denitrification unit topology diagram was marked with the first, second, third and fourth symbols to identify denitrification units that were identified as having excessive ammonia injection, leaking ammonia injection branch valve, or having a combined abnormality, as well as denitrification units whose type could not be distinguished.

9. A gas pollution source tracing and analysis system based on multi-source data, applied to the gas pollution source tracing and analysis method based on multi-source data as described in claim 1, characterized in that: The system includes a data acquisition module, a factor construction module, a regression solution module, and an anomaly localization module; The data acquisition module collects the time series of ammonia injection, inlet nitrogen oxide concentration, flue gas volume, and ammonia escape concentration at the total discharge outlet for each denitrification unit sharing the same main discharge outlet. The factor construction module calculates the instantaneous molar ratio of each denitrification unit based on the time series of ammonia injection, the time series of inlet nitrogen oxide concentration, and the time series of flue gas volume. It constructs the excess ammonia injection factor sequence of each denitrification unit according to the portion of the instantaneous molar ratio that exceeds the preset stoichiometric ratio. It also constructs the ammonia injection diffusion factor sequence of each denitrification unit based on the time series of ammonia injection and the time series of flue gas volume. The regression solution module uses the ammonia escape concentration time series as the response variable and the excess ammonia injection factor series and ammonia injection diffusion factor series as regression factors to construct a regression equation; the non-negative least squares method is used to solve the regression equation to obtain the excess ammonia injection response coefficient and branch valve leakage coefficient of each denitrification unit. The anomaly location module compares the excessive ammonia injection response coefficient with the first preset threshold and the branch valve leakage coefficient with the second preset threshold to determine the ammonia escape anomaly type of each denitrification unit.

Images