Method for processing flue gas detection data of copper smelting plant

Abstract

The present application belongs to the technical field of monitoring, and particularly relates to a copper smelting plant flue gas detection data processing method. The present application obtains the shielding coefficient of each area in the plant based on a three-dimensional model, can estimate the flue gas concentration of the equipment-intensive area and the high-temperature area in combination with the shielding coefficient and the measured data, so as to complete the concentration detection blind area of the area which cannot be accurately detected by the sensor, and improve the precision of the flue gas diffusion path simulation.

Description

Technical Field

[0001] This invention belongs to the field of monitoring technology, and specifically relates to a method for processing flue gas detection data in copper smelting plants. Background Technology

[0002] The flue gas generated during copper smelting has a complex composition, including sulfur dioxide, carbon monoxide, and particulate matter. To ensure the health and safety of workers, the flue gas needs to be discharged from the plant in a timely manner. However, sulfur dioxide and carbon monoxide are colorless and transparent gases, making it impossible to monitor their concentration using image modules. Furthermore, the diffusion of the flue gas is affected by the heat radiation from the high-temperature furnaces within the production plant, and the plant contains a dense concentration of production equipment. This makes it impossible to deploy enough sensors in high-temperature and equipment-intensive areas, resulting in a limited number of sensors for monitoring gas concentrations and uneven deployment. Consequently, the collected data is discontinuous, leading to low accuracy in analyzing flue gas concentrations within the plant and failing to meet production requirements. Summary of the Invention

[0003] The purpose of this invention is to provide a method for processing flue gas detection data in a copper smelting plant, which can estimate the concentration of flue gas at various locations within the plant based on the detection data.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A method for processing flue gas detection data in a copper smelting plant includes the following steps:

[0006] Step A: Install several gas sensors sequentially along the flue gas diffusion path. Three or more gas sensors placed close together are connected to the same gas collection box, and each gas collection box is connected to the data processing server.

[0007] Step B: Construct a 3D model of the factory building, and then generate occlusion coefficients based on the 3D model. The occlusion coefficients are used as confidence weights for the original data.

[0008] The occlusion coefficient is the degree of occlusion caused by the device model on the straight path from the gas sensor installation location to the target point in the monitored area. The value range of the occlusion coefficient is [0, 1], where it is 1 when there is no occlusion and 0 when there is complete occlusion.

[0009] Step C: After the data processing server obtains the raw gas concentration monitoring data uploaded by the gas collection box, it preprocesses the raw gas concentration monitoring data to obtain the preprocessed gas concentration data in the plant.

[0010] Step D: Spatial interpolation is performed on the preprocessed data using the Kriging interpolation method to fill in the blank areas of the data and obtain the interpolated simulation data of the gas concentration in the plant.

[0011] Compared with existing technologies, this invention fully utilizes three-dimensional simulation technology to improve the accuracy of flue gas concentration detection. Based on a three-dimensional model, this invention obtains the occlusion coefficient of each area within the factory building. By combining the occlusion coefficient with measured data, the flue gas concentration in densely populated equipment areas and high-temperature areas can be estimated. This fills the concentration detection blind spots in areas where sensors cannot accurately detect, thus improving the accuracy of flue gas diffusion path simulation. Attached Figure Description

[0012] The following is a brief explanation of the contents depicted in the accompanying drawings and the markings within them:

[0013] Figure 1 This is a schematic diagram of the present invention. Detailed Implementation

[0014] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and through the description of the examples.

[0015] A method for processing flue gas detection data in a copper smelting plant includes the following steps:

[0016] Step A: Install several gas sensors 11 sequentially along the flue gas diffusion path, i.e., from the flue gas outlet to the exhaust outlet in the plant. Among them, three or more gas sensors 11 arranged nearby are communicatively connected to the same gas collection box 12, and each gas collection box 12 is communicatively connected to the data processing server 13.

[0017] Step B: Construct a 3D model of the factory building, and then generate an occlusion coefficient based on the 3D model. The occlusion coefficient is used as the confidence weight of the original data. The occlusion coefficient is the degree of occlusion caused by the equipment model on the straight path from the installation position of the gas sensor 11 to the target point of the monitored area. The value range of the occlusion coefficient is [0, 1], where it is 1 when there is no occlusion and 0 when there is complete occlusion.

[0018] Step C: After the data processing server 13 obtains the raw gas concentration monitoring data uploaded by the gas collection box 12, it preprocesses the raw gas concentration monitoring data to obtain the preprocessed gas concentration data in the plant.

[0019] Step D: Spatial interpolation is performed on the preprocessed data using the Kriging interpolation method to fill in the blank areas of the data and obtain the interpolated simulation data of the gas concentration in the plant.

[0020] This invention combines the three-dimensional model data of the factory to calculate the occlusion coefficient of areas where sensors cannot be placed in the factory. Then, it combines the occlusion coefficient with the actual sensor measurement data to estimate the flue gas concentration in densely populated equipment areas and high-temperature areas. This can accurately fill the concentration detection blind spots in areas where sensors cannot accurately detect, providing a reliable basis for the management of flue gas emissions from the factory.

[0021] Specifically, the occlusion coefficient is calculated using the following method in this embodiment:

[0022] Step B1: Establish three-dimensional spatial coordinates for the i-th sensor Sᵢ and the j-th monitoring grid point Pⱼ. .

[0023] Step B2: Construct a detection ray R emitted from Sᵢ to Pⱼ. The expression for ray R is:

[0024] ,

[0025] Where t is the linear coefficient. .

[0026] Step B3: Inspect X-ray R against all equipment models within the factory. The intersection situation,

[0027] ;

[0028] in, This is an indicator function; it is 1 when they intersect, and 0 otherwise.

[0029] For equipment model The obstruction weight is a constant, determined by the degree to which the equipment obstructs the gas flow. For example, a large-volume high-temperature furnace body not only hinders gas diffusion but also generates strong thermal radiation airflow, which has a significant impact on flue gas diffusion. Other equipment has a relatively small impact on gas diffusion and flow, so the obstruction weight of the high-temperature furnace body is greater than that of other equipment. Pipes are small in volume and the gaps between pipes can also allow gas to pass through, so the obstruction weight of the pipe area is less than that of other equipment.

[0030] For a sensor's monitoring grid points, the occlusion coefficient ,

[0031] For overlapping grid points monitored by multiple sensors, the occlusion coefficient is calculated using a weighted average.

[0032] ,

[0033] in, For sensors To grid point The closer the two are, the greater their weight.

[0034] When applying the shielding coefficient to correct gas concentration values, the following formula can be used:

[0035] ,

[0036] in, This refers to the raw sensor readings or the raw output values ​​predicted by the model. This is the corrected concentration value. Let be the occlusion coefficient of the j-th monitoring grid point. The distance attenuation coefficient is a constant, and in this embodiment, it is set to 1.0-1.2. For example, in step C, the gas concentration correction values ​​of other monitoring grid points can be calculated or verified based on the raw gas concentration monitoring data uploaded by the gas acquisition box 12. In step D, the output values ​​of the blank areas output by the prediction model based on the Kriging interpolation method can be corrected.

[0037] To ensure data consistency and facilitate data processing, all gas sensors 11 have the same sampling frequency. The gas acquisition box 12 has a higher sampling frequency than the gas sensors 11. This ensures accurate data acquisition from the downstream gas sensors 11. However, the data acquisition times of the downstream gas sensors 11 may differ. Therefore, in step C, the raw gas concentration monitoring data needs to be aligned to the reference timestamp of the gas acquisition box 12. This embodiment uses linear interpolation to upsample the raw gas concentration monitoring data from the gas sensors 11.

[0038] ,

[0039] in, , This is the original sampling time. For the target alignment time, < < , , This refers to the original gas concentration monitoring data. To estimate the gas concentration data at the moment of target alignment.

[0040] Specifically, the preprocessing of the raw gas concentration monitoring data in this embodiment includes the following steps:

[0041] Step C1: Filter the raw gas concentration monitoring data to remove noise and avoid data drift caused by sensor aging or changes in ambient temperature and humidity, which could affect the accuracy of the monitoring.

[0042] Step C2: Align the sampling times of the original gas concentration monitoring data. Since the data acquisition times of each gas sensor 11 may vary, it is necessary to align the sampling times for easier analysis and processing.

[0043] Step C3: Identify outliers using the Laida criterion, replacing outliers with the average of the preceding and following time points. The filtered data typically approximates a normal distribution, thus the Laida criterion can be used to remove outlier fluctuations. Specifically, a dataset is constructed using gas concentration values ​​at a specific time or over a period of time, and the mean and standard deviation of this dataset are calculated to determine the original data at a given point. If the following formula is satisfied, the value is determined to be abnormal data and replaced with the average value of the time before and after that point.

[0044] ,

[0045] in, This is the original data. This is the mean of the dataset used in this analysis. denoted as the standard deviation of the dataset.

[0046] Step C4: Data normalization, which maps data of different dimensions to the [0,1] interval to facilitate the identification of abnormal data.

[0047] ,

[0048] in, This is the original data. The maximum value of the dataset. The minimum value of the dataset. For normalized data.

[0049] In order to obtain simulated data for the data gap area in the factory where sensors cannot be placed, this embodiment divides the monitoring area in the factory into a uniform cubic grid, and then performs Kriging interpolation calculations on each grid point.

[0050] Kriging interpolation calculations include the following steps:

[0051] Step D1: Modeling the variation function;

[0052] ,

[0053] in, The autocorrelation of spatial data is described by h, where h is the distance between two points, also known as the lag distance. The gas concentration data collected by the gas sensor 11i denoted as h, representing the number of sample point pairs with a lag distance of h.

[0054] Step D2: Construct the exponential model;

[0055] ,

[0056] in, The value represents the nugget value, indicating random error and measurement noise. The sill value represents the total intensity of spatial variability; The effective radius of spatial influence is related to the size of the factory and the ventilation intensity.

[0057] In practical applications, under the influence of the same airflow, at the same time, the gas concentration values ​​collected by gas sensors 11 that are closer together should be more similar; conversely, the gas concentration values ​​collected by gas sensors 11 that are farther apart will have a lower correlation. Therefore... It is related to the two-point lag distance, and eventually reaches a plateau value C as the lag distance increases.

[0058] Step D3: Estimate interpolation, interpolation points The gas concentration interpolation is a linear combination of the data collected by the gas sensor (11).

[0059] ,

[0060] in, Points to be interpolated Gas concentration interpolation, These are the weighting coefficients. The gas concentration data is collected by the gas sensor (11)i.

[0061] Weighting coefficient This is obtained by solving the following system of equations.

[0062] i=1,2,……,n

[0063] ,

[0064] in, Let i be the semivariogram of known points i and j. Given point i and the point to be interpolated The semi-mutation value, It is a Lagrange multiplier.

[0065] Furthermore, the preferred embodiment also includes step E, which involves training a time-series deep learning neural network model using equipment status data and environmental data as input features and correction amount as output features to optimize and correct the interpolated simulation data, thereby obtaining corrected simulation data.

[0066] ,

[0067] in, To correct the simulated concentration, To interpolate and simulate concentration, This is the correction amount output by the neural network model.

[0068] Unless a leak or process fluctuation occurs, the gas concentration value usually does not change abruptly. Therefore, the gas concentration value and its changing trend can be determined based on the equipment operating status and environmental data, and then the interpolated simulation data can be corrected. The input features selected in this embodiment include: the historical value of the original concentration data collected by the gas sensor (11), the historical value of the original concentration data obtained by the gas collection box (12), the spatial distribution of each gas sensor (11) at the current moment, the current production status, furnace shutdown, charging, blowing, copper tapping, equipment obstruction coefficient, and the historical sequence of Kriging interpolation residuals. In specific implementation, the training of the temporal deep learning neural network model can be adjusted according to the requirements of the input features and parameters.

Claims

1. A method for processing flue gas detection data in a copper smelting plant, comprising the following steps: Step A: Install several gas sensors (11) sequentially along the flue gas diffusion path. Three or more gas sensors (11) arranged nearby are connected to the same gas collection box (12) for communication. Each gas collection box (12) is connected to the data processing server (13) for communication. Step B: Construct a 3D model of the factory building, and then generate occlusion coefficients based on the 3D model. The occlusion coefficients are used as confidence weights for the original data. The occlusion coefficient is the degree of occlusion caused by the device model on the straight path from the gas sensor (11) installation position to the target point of the monitored area. The value range of the occlusion coefficient is [0, 1], where it is 1 when there is no occlusion and 0 when there is complete occlusion. Step C: After the data processing server (13) obtains the raw gas concentration monitoring data uploaded by the gas collection box (12), it preprocesses the raw gas concentration monitoring data to obtain the preprocessed data of the gas concentration in the plant. Step D: Spatial interpolation is performed on the preprocessed data using the Kriging interpolation method to fill in the blank areas of the data and obtain the interpolated simulation data of the gas concentration in the plant.

2. The method for processing flue gas detection data in a copper smelter plant according to claim 1, characterized in that: The shading coefficient is calculated as follows: Step B1: Establish three-dimensional spatial coordinates for the i-th sensor Sᵢ and the j-th monitoring grid point Pⱼ. ； Step B2: Emit detection ray R from Sᵢ to Pⱼ. ， Where t is the linear coefficient. ; Step B3: Inspect X-ray R against all equipment models within the factory. The intersection situation, ； in, This is an indicator function; it is 1 when they intersect, and 0 otherwise. For equipment model The occlusion weight is a constant; For a sensor's monitoring grid points, the occlusion coefficient , For overlapping grid points monitored by multiple sensors, the occlusion coefficient is calculated using a weighted average. ， in, For sensors To grid point The Euclidean distance between the two is such that the closer the two are, the greater their weight.

3. The method for processing flue gas detection data in a copper smelting plant according to claim 2, characterized in that: Apply the shading factor to correct the gas concentration value. ， in, This refers to the raw sensor readings or the raw output values ​​predicted by the model. This is the corrected concentration value. Let be the occlusion coefficient of the j-th monitoring grid point. is the distance attenuation coefficient, and is a constant.

4. The method for processing flue gas detection data in a copper smelting plant according to claim 1, characterized in that: Each gas sensor (11) has the same sampling frequency, while the gas collection box (12) has a higher sampling frequency than the gas sensors (11). In step C, the raw gas concentration monitoring data is aligned to the reference timestamp of the gas collection box (12). The raw gas concentration monitoring data of the gas sensor (11) were upsampled using linear interpolation. ， in, , This is the original sampling time. For the target alignment time, < < , , This refers to the original gas concentration monitoring data. To estimate the gas concentration data at the moment of target alignment.

5. The method for processing flue gas detection data in a copper smelting plant according to claim 1, characterized in that: Step C includes the following steps: Step C1: Filter the raw gas concentration monitoring data to remove noise; Step C2: Align the sampling times of each original gas concentration monitoring data; Step C3: Use the Raida criterion to identify outliers, and replace outliers with the average value of the previous and next time points; Step C4: Data normalization, mapping data of different dimensions to the [0,1] interval. ， in, The original data, The maximum value of the dataset. The minimum value of the dataset. For normalized data.

6. The method for processing flue gas detection data in a copper smelting plant according to claim 1, characterized in that: In step D, the monitoring area within the factory is divided into a uniform cubic grid, and kriging interpolation is performed on each grid point. Kriging interpolation calculations include the following steps: Step D1: Modeling the variation function; ， in, The autocorrelation of spatial data is described by h, where h is the distance between two points, also known as the lag distance. The gas concentration data collected by gas sensor (11)i The number of sample point pairs with a lag distance of h; Step D2: Construct the exponential model; ， in, The value represents the nugget value, indicating random error and measurement noise. The sill value represents the total intensity of spatial variability; The effective radius of spatial influence; Step D3: Estimate interpolation, interpolation points The gas concentration interpolation is a linear combination of the data collected by the gas sensor (11). ， in, Points to be interpolated Gas concentration interpolation, These are the weighting coefficients. This is the gas concentration data collected by the gas sensor (11)i; Weighting coefficient This is obtained by solving the following system of equations. ,i=1,2,……,n, ， in, Let i be the semivariogram of known points i and j. Given point i and the point to be interpolated The semi-mutation value, It is a Lagrange multiplier.

7. The method for processing flue gas detection data in a copper smelting plant according to claim 6, characterized in that: The process also includes step E, which uses device status data and environmental data as input features and correction amount as output features to train a time-series deep learning neural network model to optimize and correct the interpolated simulation data, thereby obtaining corrected simulation data. ， in, To correct the simulated concentration, To interpolate and simulate concentration, This is the correction amount output by the neural network model.

8. The method for processing flue gas detection data in a copper smelter plant according to claim 7, characterized in that: The input features include, Historical values ​​of raw concentration data collected by gas sensor (11) Historical values ​​of the raw concentration data obtained by the gas sampling box (12), The spatial distribution of each gas sensor (11) at the current moment, Current production status: furnace shutdown, charging, blowing, copper tapping. Equipment occlusion coefficient, Kriging interpolation residual history sequence.

Images