A pipe network sludge component detection method, device, medium and equipment

Abstract

The present application relates to the technical field of environmental monitoring and analysis, and specifically discloses a pipe network sludge component detection method, device, medium and equipment. The present application method pre-processes the spectral data of the pipe network sludge after drying and vibration screening. The pre-processed spectral data is subjected to feature selection by using a competitive adaptive reweighted sampling algorithm. The data set after feature selection is subjected to baseline drift enhancement. The data set after baseline drift enhancement is subjected to active learning screening. An optimal neural network model is constructed based on a neural network architecture search framework. The spectral data is predicted by using the optimal neural network model to obtain a component prediction result of the pipe network sludge. The present application realizes rapid and non-destructive detection of pipe network sludge components. Compared with traditional wet chemical analysis, the present application does not require chemical reagents, and the detection efficiency and accuracy are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and analysis technology, specifically to a method, apparatus, medium, and equipment for detecting the composition of sewage sludge in a pipeline network. Background Technology

[0002] Sewage sludge is a common pollutant in urban sewage pipeline systems. Its composition is complex, containing large amounts of organic matter, nitrogen, phosphorus, and other substances. Accurate prediction and analysis of the organic matter, nitrogen, and phosphorus content in sewage sludge is of great significance for optimizing sewage treatment processes, assessing environmental pollution, and promoting resource recycling.

[0003] However, current traditional pipe network sludge analysis systems mainly rely on chemical analysis methods, such as wet chemical analysis and chromatography. While these systems can provide relatively accurate analytical results, they suffer from drawbacks such as complex operation, long processing time, the need for large amounts of chemical reagents, and difficulty in achieving rapid on-site detection. With the development of spectroscopic technology, near-infrared spectroscopy has been widely used in the field of material composition analysis. Near-infrared spectroscopy has advantages such as speed, non-destructive nature, and the ability to achieve online detection. By analyzing the absorption characteristics of substances to light in a certain wavelength range, it can reflect the vibrational information of hydrogen-containing functional groups such as CH, NH, and OH in molecules, thereby achieving rapid analysis of the chemical composition of substances. However, the composition of pipe network sludge fluctuates greatly, containing a large amount of pipe sediment, grease, and microbial metabolites. Its near-infrared spectral signal is easily affected by particle size, moisture, and background interference, resulting in a weakened correlation between spectral characteristics and target components. Existing near-infrared detection systems are mostly designed for stable systems and lack spectral preprocessing and model optimization strategies specific to the characteristics of pipe network sludge, making it difficult to eliminate interference and resulting in problems such as insufficient detection accuracy and poor applicability, failing to meet practical needs.

[0004] Therefore, how to quickly and accurately detect the organic matter, nitrogen, and phosphorus content in sewage sludge in pipeline networks has become a challenge in the field of environmental monitoring and analysis technology. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the shortcomings of the existing technology and provide a convenient and highly accurate method, device, medium and equipment for detecting the composition of sewage sludge in pipe networks. The specific technical solution is as follows: Firstly, a method for detecting the composition of sewage sludge in a pipe network is provided, the method comprising: Collect spectral data of the sludge from the pipeline network after drying and vibrating screening; The spectral data of the pipeline sludge after drying and vibrating screening are preprocessed, including SG smoothing and normalization. A competitive adaptive reweighted sampling algorithm is used to select features from the preprocessed spectral data; Baseline drift enhancement is applied to the feature-selected dataset; The K-means clustering sampling method is used to actively learn and filter the baseline-shifted enhanced dataset. Constructing the optimal neural network model based on a neural network architecture search framework; By using a pre-trained optimal neural network model to predict the composition of real-time spectral data, the composition of the sludge in the pipeline network can be predicted.

[0006] As a further improvement to the above technical solution: The SG smoothing is used to suppress random high-frequency noise in spectral data through local polynomial fitting while preserving spectral characteristic trends; The standardization is used to eliminate systematic errors, including baseline drift and instrument response differences, between spectral data of sludge from different pipe networks, and ultimately remove comprehensive noise introduced by non-uniformity of sludge particles in the pipe network, instrument fluctuations and environmental interference, thereby improving the stability and consistency of spectral data.

[0007] As a further improvement to the above technical solution: The feature selection of the preprocessed spectral data using the competitive adaptive reweighted sampling algorithm includes: Construct the original spectral dataset using the preprocessed spectral data ; Among them, the training set , This represents the 228-dimensional spectral samples of the training set. The labels correspond to the 228-dimensional spectral samples in the training set. Total number of training samples; test set , This represents the 228-dimensional spectral samples of the test set. The labels correspond to the 228-dimensional spectral samples in the test set. Total number of test samples; For the competitive adaptive reweighted sampling algorithm, the number of iterations is set to... The cross-validation fold number is ; Preprocessed training data A competitive adaptive reweighted sampling algorithm is used for feature selection. For the standardized samples, at the first Number of features retained in the next iteration Satisfy the following expression: ; in, This is the floor function; Based on the feature subset retained from the previous iteration Training a partial least squares model It satisfies the following expression: ; in express Fitting function, To standardize the training sample matrix, To train the label vectors, a new feature subset is obtained by filtering through weight coefficients. ; Finally, the optimal feature subset is obtained. It satisfies the following expression: ; in, The model loss function; The training set after feature selection based on the optimal feature subset is: .

[0008] As a further improvement to the above technical solution: The baseline drift enhancement of the feature-selected dataset includes: To enhance baseline drift, set a set of spline interpolation control points. and It satisfies the following expression: ; ; in, Represents the baseline drift enhancement in the first The spectral wavelength coordinates corresponding to each spline interpolation control point Represents the baseline drift enhancement in the first The spectral baseline intensity value corresponding to each spline interpolation control point L represents the number of control points; the interpolation type is... This is a linear interpolation method used to connect any two adjacent control points with a straight line segment, simulating a smooth, linear baseline drift. This is a cubic spline interpolation method used to construct a cubic polynomial curve between any two adjacent control points. Baseline shift enhancement is performed on the feature-selected training set. The baseline shift signal generation function is: ; The baseline drift signal generation function takes spectral wavelength points as input and spline interpolation control point set and interpolation type as parameters, and is generated through the spline interpolation function. The baseline drift signal is calculated; Augmented samples of a single sample Satisfy the following expression: ; in, For the first Spectral data obtained after feature selection of each sample using a competitive adaptive reweighted sampling algorithm; Set the enhancement factor to Enhanced training set Satisfy the following expression:

[0009] in, Indicates the first Samples generated through secondary enhancement This indicates merging multiple sets into one set.

[0010] As a further improvement to the above technical solution: The active learning and filtering of the baseline-shifted augmented dataset using the K-means clustering sampling method includes: Set the number of clusters to Randomly selected from the augmented training set 1 sample was used as the initial cluster center. ; Calculate the Euclidean distance between each sample in the augmented training set and all initial cluster centers. , ; in, yes Any single spectral sample vector in the array, For the first 1. Initial cluster center vectors are generated; each sample in the augmented training set is assigned to the cluster of the nearest cluster center, resulting in the initial cluster partitioning. ; in, Represented as K-means clustering, the th After the nth iteration, the c-th cluster is obtained. The initial cluster partitioning is the result of the first iteration. ,Right now ; For each cluster Recalculate cluster centers , The number of samples within the cluster. This can be expressed as the summation of all sample vectors within the c-th cluster after the t-th iteration; Repeated sample allocation and cluster center updates continue until the change in cluster centers reaches the maximum number of iterations, resulting in the final cluster partitioning. From each cluster Select distance from cluster center The most recent sample is used as the representative sample, and its index satisfy: ; in, It is the L2 norm, i.e., the Euclidean distance; The training set obtained after active learning selection is ,in, This represents the first selected by active learning. Representative enhanced spectral samples of each cluster, To enhance spectral samples The corresponding actual ingredient label.

[0011] As a further improvement to the above technical solution: The construction of the optimal neural network model based on the neural network architecture search framework includes: Sample reshaping: Reshaping samples from the training set after active learning. Reshape to input format ,in The feature dimensions selected by the competitive adaptive reweighted sampling algorithm; Search space definition: Define the search space It includes basic network components and hyperparameter ranges, wherein the basic network components include convolutional layers, fully connected layers, and activation functions; Search algorithm configuration: A search algorithm is used. In search space Iterative optimization, the search algorithm The evaluation metric is the model's prediction loss on the validation set. ; Optimal model generation: through a search algorithm Candidate model structures are sampled, trained, and evaluated to ultimately select the best ones. Minimal optimal model It satisfies the following expression: , in, For search space Candidate model structure in.

[0012] As a further improvement to the above technical solution: The method of using a pre-trained optimal neural network model to predict real-time spectral data includes: Acquire real-time spectral data of sludge from the pipeline network and preprocess the data; Based on the optimal feature subset determined during the training phase, features corresponding to the dimensions are extracted from the standardized real-time spectral data. The extracted features are then input into the pre-trained optimal neural network model, and the model is used for forward propagation calculations to output the predicted results of organic matter, nitrogen, and phosphorus content.

[0013] Secondly, a pipe network sludge composition detection device is provided, employing the pipe network sludge composition detection method described above, the pipe network sludge composition detection device comprising: A drying and sieving device includes a base, an electric furnace, a petri dish, a screen, and a vibration mechanism. The electric furnace is embedded inside the base and has a positioning groove for placing the petri dish. The electric furnace heats the petri dish through the positioning groove. The vibration mechanism is mounted on the base and is connected to the screen for transmission. The detection device includes a container and a near-infrared spectrometer. The container is detachably mounted on a base and located below the screen. The near-infrared spectrometer is slidably mounted inside the base. The container is positioned within the detection range of the near-infrared spectrometer. The data optimization module is used to perform preprocessing, feature selection, baseline drift enhancement, and active learning filtering on the spectral data acquired by the near-infrared spectrometer. The model architecture search module is used to automatically construct the optimal neural network model based on the spectral data filtered by active learning, through a neural network architecture search framework. The inference and prediction module is used to quantitatively calculate the organic matter, nitrogen, and phosphorus content of spectral data using a pre-trained optimal neural network model.

[0014] As a further improvement to the above technical solution: It also includes a smart terminal, which is used to control the drying and screening device, the detection device, the data optimization module, the model architecture search module, and the inference and prediction module, and to present the composition prediction results of the pipeline sludge.

[0015] As a further improvement to the above technical solution: The electric furnace is embedded in the base through a heat-insulating bracket, and the heat-insulating bracket is lined with a ceramic heat-insulating layer.

[0016] As a further improvement to the above technical solution: The culture dish is a high-temperature resistant glass culture dish; The edge of the high-temperature resistant glass petri dish is equipped with anti-scalding handles.

[0017] As a further improvement to the above technical solution: The base is equipped with a sliding panel, which can slide to form a closed drying chamber with the base.

[0018] As a further improvement to the above technical solution: The vibration mechanism includes a vibration motor and a crank rocker arm; The vibration motor is mounted on the base via a shock-absorbing device. One end of the crank rocker arm is connected to the output shaft of the vibration motor, and the other end is connected to the screen via a quick-release screw. When the vibratory motor is running, the crank rocker arm is used to convert the rotational motion of the vibratory motor into the reciprocating vibration of the screen.

[0019] As a further improvement to the above technical solution: The container has an opening at the bottom, a glass gate at the opening, and the opening is positioned opposite to the data acquisition port of the near-infrared spectrometer.

[0020] As a further improvement to the above technical solution: The base is equipped with a sliding linear guide rail, and the near-infrared spectrometer is detachably connected to the linear guide rail. The linear guide rail can drive the near-infrared spectrometer to slide in and out of the base.

[0021] As a further improvement to the above technical solution: The near-infrared spectrometer covers the 900nm-1700nm band, and the spectral features acquired by the near-infrared spectrometer include 228-dimensional high-dimensional data features.

[0022] As a further improvement to the above technical solution: It also includes a lid that fits the base, which is fastened to the base by a push-button resilient latch.

[0023] As a further improvement to the above technical solution: The lid is made of flame-retardant ABS material in one piece, and a silicone heat insulation pad is laid on the inner wall of the lid.

[0024] As a further improvement to the above technical solution: The base is equipped with pulleys, which are universal wheels with braking function.

[0025] Thirdly, a readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the method for detecting the composition of sludge in a pipeline network as described above.

[0026] Fourthly, an electronic device is provided, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the pipeline sludge composition detection method as described above.

[0027] Compared with the prior art, the advantages of the present invention are as follows: (1) A method for detecting the composition of sludge in a pipe network according to the present invention involves collecting spectral data of sludge from a pipe network after drying and vibrating sieving; preprocessing the spectral data of the sludge after drying and vibrating sieving, the preprocessing including SG smoothing and standardization; performing feature selection on the preprocessed spectral data using a competitive adaptive reweighted sampling algorithm; implementing baseline drift enhancement on the feature-selected dataset; actively learning and filtering the baseline drift-enhanced dataset using a K-means clustering sampling method; constructing an optimal neural network model based on a neural network architecture search framework; and using the pre-trained optimal neural network model to predict the composition of the sludge from the real-time spectral data to obtain the predicted composition of the sludge. This method enables rapid and non-destructive detection of organic matter, nitrogen, and phosphorus components in pipe network sludge, overcoming the limitations of traditional chemical analysis. The present invention uses near-infrared spectroscopy technology, which does not require chemical reagents. Compared with traditional wet chemical analysis, the detection efficiency is significantly improved, the detection cycle can be greatly shortened, and the operation process can be simplified, effectively overcoming the limitations of complex and time-consuming sample pretreatment in traditional methods.

[0028] (2) This invention designs a special data optimization strategy for the complexity of sludge in the pipeline network, which significantly improves the detection accuracy. The SG smoothing and standardization preprocessing eliminate interference such as particle inhomogeneity and instrument fluctuations; the CARS algorithm accurately selects features and reduces data redundancy; the baseline drift enhancement simulates the spectral fluctuations in actual detection, and the K-means active learning selects representative samples, which significantly improves the robustness of the model to compositional fluctuations and solves the problem of poor adaptability of existing near-infrared systems to complex matrices.

[0029] (3) This invention improves the system’s versatility and scalability through an adaptive model architecture. By using Neural Architecture Search (NAS) to automatically optimize the network structure, no manual model design is required, and it can adapt to the differences in sludge samples from different regions and seasons. In terms of hardware, it adopts a linear guide rail and quick-release buckle design, supports the replacement of near-infrared, Raman and other spectrometers, realizes multimodal data fusion detection, and provides a foundation for future expansion to the detection of other pollutants. (4) This invention constructs an integrated closed-loop system of "hardware-algorithm-terminal", which is convenient and intelligent to operate. The entire process of drying and screening, spectral acquisition and data processing is automated and controlled. The real-time calculation of edge devices and the visualization display of intelligent terminals are combined, which makes it easy for grassroots environmental protection personnel to master quickly and promotes the upgrade of pipeline sludge detection from laboratory to real-time on-site monitoring. Attached Figure Description

[0030] Figure 1 A flowchart illustrating the pipeline sludge composition detection process, including drying and vibrating screening, performed at the edge of an embodiment of the present invention. Figure 2 This is a flowchart of a method for detecting the composition of sewage sludge in a pipeline network according to an embodiment of the present invention; Figure 3 This is a structural block diagram of a smart terminal according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a pipe network sludge composition detection device according to an embodiment of the present invention; Figure 5 This is a cross-sectional structural diagram of a pipe network sludge composition detection device according to an embodiment of the present invention.

[0031] The attached figures are labeled as follows: 1. Base; 2. Sliding panel; 3. Electric furnace; 4. Petri dish; 5. Sieve; 6. Quick-release screw; 7. Crank rocker arm; 8. Vibration motor; 9. Support frame; 10. Container; 11. Quick-release buckle; 12. Near-infrared spectrometer; 13. Linear guide rail; 14. Edge end; 15. Smart terminal; 16. Lid; 17. Lock; 18. Pulley. Detailed Implementation

[0032] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] Before conducting sludge composition analysis in the pipeline network, it is necessary to collect and pre-treat sludge samples to ensure that the samples meet the input requirements for subsequent equipment drying, sieving, and spectral analysis. Figure 1 As shown, the specific operating procedure is as follows: First, the sludge collected from the pipeline network on-site is placed in a sterile sampling container, covered, and gently shaken horizontally 3-5 times to homogenize the sample composition (to avoid stratification and sampling deviation); then, a special sampling spoon for pipeline networks made of 304 stainless steel (spoon opening size 50mm×30mm, depth 20mm) is used for sampling. When sampling, the surface floating water should be avoided, and the homogeneous sludge in the middle of the sediment layer should be collected first. After sampling, the sample is placed in a sterile polytetrafluoroethylene weighing dish and weighed using an electronic balance with an accuracy of 0.01g. 20±0.5g of sample is weighed—this amount is suitable for the volume of the high-temperature resistant glass culture dish 4 (60mm diameter, 10mm depth), ensuring the sample thickness is controlled at 2-3mm after spreading, guaranteeing uniform heat transfer during heating by the electric furnace 3 and preventing localized overheating that could lead to organic carbonization; it also retains ≥3g of effective sample after sieving, meeting the requirement of the container dish 10 covering the near-infrared spectrometer 12 acquisition port (15mm diameter), ensuring spectral signal intensity and detection accuracy. After weighing, the sample is evenly spread on the bottom of the high-temperature resistant glass culture dish 4 and gently smoothed with a sterile glass rod to ensure no localized accumulation and a thickness deviation ≤0.5mm. The positioning protrusions on the bottom of the culture dish 4 are then checked for integrity to prevent misalignment with the annular positioning grooves on the surface of the electric furnace 3 due to protrusion damage, which could affect heating uniformity.

[0034] like Figure 2As shown, the method for detecting the composition of sludge in the pipeline network in this embodiment includes the following steps: Collect spectral data of the sludge from the pipeline network after drying and vibrating screening; The spectral data of the pipeline sludge after drying and vibrating screening are preprocessed. The preprocessing includes Savitzky-Golay (SG) smoothing and normalization. SG smoothing suppresses random high-frequency noise and preserves spectral trend through local polynomial fitting. Normalization eliminates systematic errors such as baseline drift and instrument response differences between different samples, and ultimately effectively removes the comprehensive noise introduced by sample particle inhomogeneity, instrument fluctuations and environmental interference, thereby improving the stability and consistency of spectral data. The Competitive Adaptive Reweighted Sampling (CARS) algorithm is used to select features from the preprocessed spectral data. The aim is to remove redundant information, reduce data dimensionality, and improve the efficiency and stability of subsequent model training. Specifically: Construct the original spectral dataset using the preprocessed spectral data ; Among them, the training set , This represents the 228-dimensional spectral samples of the training set. The labels correspond to the 228-dimensional spectral samples in the training set. Total number of training samples; test set , This represents the 228-dimensional spectral samples of the test set. The labels correspond to the 228-dimensional spectral samples in the test set. Total number of test samples; For the competitive adaptive reweighted sampling algorithm, the number of iterations is set to... The cross-validation fold number is ; Preprocessed training data The core logic of using a competitive adaptive reweighted sampling algorithm for feature selection is to iteratively optimize the feature subset. For the standardized samples, at the first Number of features retained in the next iteration Satisfy the following expression: ; in, This is the floor function; Based on the feature subset retained from the previous iteration Training a partial least squares model It satisfies the following expression: ; in express Fitting function, To standardize the training sample matrix, To train the label vectors, a new feature subset is obtained by filtering through weight coefficients. ; Finally, the optimal feature subset is obtained. It satisfies the following expression: ; in, The model loss function; The training set after feature selection based on the optimal feature subset is: .

[0035] Baseline drift enhancement is performed on the feature-selected dataset. By simulating spectral baseline drift, sample diversity is increased, and the robustness of the model to spectral drift interference in actual detection is enhanced. Specifically: Set the spline interpolation control point set for baseline drift enhancement and It satisfies the following expression: ; ; in, Represents the baseline drift enhancement in the first The spectral wavelength coordinates corresponding to each spline interpolation control point Represents the baseline drift enhancement in the first The spectral baseline intensity value corresponding to each spline interpolation control point L represents the number of control points; the interpolation type is... This is a linear interpolation method used to connect any two adjacent control points with a straight line segment, simulating a smooth, linear baseline drift. This is a cubic spline interpolation method used to construct a cubic polynomial curve between any two adjacent control points. Baseline shift enhancement is performed on the feature-selected training set. The baseline shift signal generation function is: ; The baseline drift signal generation function takes spectral wavelength points as input and spline interpolation control point set and interpolation type as parameters, and is generated through the spline interpolation function. The baseline drift signal is calculated; Augmented samples of a single sample Satisfy the following expression: ; in, For the first Spectral data obtained after feature selection of each sample using a competitive adaptive reweighted sampling algorithm; Set the enhancement factor to Enhanced training set Satisfy the following expression:

[0036] in, Indicates the first Samples generated through secondary enhancement This indicates merging multiple sets into one set.

[0037] The K-means clustering sampling method is used to actively learn and filter the baseline-shifted augmented dataset, selecting "valuable data" with high information entropy and strong representativeness for training the Neural Architecture Search (NAS) model. Specifically; Set the number of clusters to (preset parameter). Randomly selected from the augmented training set 1 sample was used as the initial cluster center. ; Calculate the Euclidean distance between each sample in the augmented training set and all initial cluster centers. , ; in, yes Any single spectral sample vector in the array, For the first 1. Initial cluster center vectors are generated; each sample in the augmented training set is assigned to the cluster of the nearest cluster center, resulting in the initial cluster partitioning. ; in, Represented as K-means clustering, the th After the nth iteration, the c-th cluster is obtained. The initial cluster partitioning is the result of the first iteration. ,Right now ; For each cluster Recalculate cluster centers , The number of samples within the cluster. This can be expressed as the summation of all sample vectors within the c-th cluster after the t-th iteration; Repeated sample allocation and cluster center updates continue until the change in cluster centers reaches the maximum number of iterations, resulting in the final cluster partitioning. From each cluster Select distance from cluster center The most recent sample is used as the representative sample, and its index satisfy: ; in, It is the L2 norm, i.e., the Euclidean distance; The training set obtained after active learning selection is ,in, This represents the first selected by active learning. Representative enhanced spectral samples of each cluster, To enhance spectral samples The corresponding actual ingredient label.

[0038] The optimal neural network model is constructed based on the Neural Architecture Search (NAS) framework, specifically: Sample reshaping: Reshaping samples from the training set after active learning. Reshape to input format ,in The feature dimensions selected by the competitive adaptive reweighted sampling algorithm; Search space definition: Define the search space It includes basic network components and hyperparameter ranges, wherein the basic network components include convolutional layers, fully connected layers, and activation functions; Search algorithm configuration: A search algorithm is used. In search space Iterative optimization, the search algorithm The evaluation metric is the model's prediction loss on the validation set. ; Optimal model generation: through a search algorithm Candidate model structures are sampled, trained, and evaluated to ultimately select the best ones. Minimal optimal model It satisfies the following expression: , in, For search space Candidate model structure in.

[0039] By using a pre-trained optimal neural network model to predict the composition of real-time spectral data, the following prediction results for the sludge composition in the pipe network are obtained: Acquire real-time spectral data of sludge from the pipeline network and preprocess the data; Based on the optimal feature subset determined during the training phase, features corresponding to the dimensions are extracted from the standardized real-time spectral data. The extracted features are then input into the pre-trained optimal neural network model, and the model is used for forward propagation calculations to output the predicted results of organic matter, nitrogen, and phosphorus content.

[0040] like Figure 4 and Figure 5As shown, this embodiment also provides a pipeline sludge composition detection device, which employs the pipeline sludge composition detection method described above. The pipeline sludge composition detection device includes: The drying and sieving device includes a base 1, an electric furnace 3, a petri dish 4, a sieve 5, and a vibration mechanism. The electric furnace 3 is embedded inside the base 1 and has a 55mm annular positioning groove for placing the petri dish 4. The positioning groove is precisely matched with the 55mm positioning boss at the bottom of the petri dish 4 with a deviation of ≤0.5mm to ensure the centering of the sample and enhance the uniformity of heating. The electric furnace 3 heats the petri dish 4 through the positioning groove. The vibration mechanism is set on the base 1 and is connected to the sieve 5 for transmission. The detection device includes a container 10 and a near-infrared spectrometer 12. The container 10 is detachably mounted on the base 1 and located below the screen 5. The near-infrared spectrometer 12 is slidably mounted inside the base 1. The container 10 is located within the detection range of the near-infrared spectrometer 12. It also includes a smart terminal 15, which can be an electronic device such as a computer or mobile phone. The specific choice depends on the on-site operating space and portability requirements, and there are no strict restrictions. Preferably, the smart terminal 15 is a touch tablet (IP65 protection rating, battery life ≥8h). The smart terminal 15 is used to control the drying and screening device, the detection device, the data optimization module, the model architecture search module, and the inference and prediction module, and presents the composition prediction results of the pipeline sludge.

[0041] like Figure 3 As shown, the intelligent terminal 15 is equipped with an embedded intelligent operating system. The interface is divided into four functional areas: drying operation, vibrating sieving, data acquisition, and result presentation. It achieves command interaction and status feedback through the built-in data processing unit of the edge terminal 14, collaboratively realizing intelligent system control. The edge terminal 14, equipped with the data processing unit, is electrically connected to the electric furnace 3, the vibrating motor 8, the near-infrared spectrometer 12, and the intelligent terminal 15 via external data cables. Drying operation function area 151: Associated with the drying control module 14 at the edge end, it realizes the visual control of the drying temperature (50-120℃, accuracy ±1℃) and duration (1-12h) of the electric furnace 3. Through the logic of "parameter preset → dynamic closed loop", it solves the problem of temperature and time fluctuation in drying and ensures the uniformity of sludge drying. Vibrating Screening Function Area 152: Connects to the edge end 14 vibrating screening control module, configured with vibration amplitude (1-5mm) and frequency (50-200Hz) parameters of vibration motor 8, supports custom vibration mode (continuous vibration, intermittent vibration), and is suitable for multi-particle-size sludge screening scenarios. Data acquisition function area 153: Links the data acquisition control module of edge terminal 14, presets the acquisition parameters of near-infrared spectrometer 12 and triggers detection, and provides real-time feedback on acquisition progress and equipment status; Results Presentation Function Area 154: Receives the organic matter, nitrogen, and phosphorus content results output by the edge-end 14 inference prediction module, and presents them in the form of numerical dashboards, trend curves, and test reports. It supports data export and anomaly backtracking, providing data support for the optimization of the detection model.

[0042] Preferably, the edge terminal 14 and the smart terminal 15 interact and transmit commands, supporting multiple communication methods such as Wi-Fi and Bluetooth to ensure the stability and real-time performance of data transmission. The data processing unit includes: a drying control module, a vibrating screening control module, a data acquisition control module, a data optimization module, a model architecture search module, and an inference and prediction module, wherein: Drying control module: It specifically receives parameter instructions (such as target drying temperature and preset drying time) issued by the drying operation function area of ​​the smart terminal 15, and performs precise control of electric furnace 3 based on the closed-loop logic of "parameter preset → real-time temperature control → deviation correction"; by dynamically monitoring the deviation between the actual temperature of electric furnace 3 and the set value, it automatically adjusts the heating power to effectively solve the problem of temperature and time fluctuation during the drying process, and forms a linkage with the visual control interface of the drying operation function area to jointly ensure the uniformity of sludge drying; Vibration screening control module: Responds to parameter configuration commands (such as vibration amplitude and vibration frequency) in the vibration screening function area of ​​the smart terminal 15, adjusts the output power and operating rhythm of the vibration motor 8 according to the command signal, and drives the crank rocker arm 7 to drive the screen 5 to generate appropriate reciprocating vibration; at the same time, it supports receiving the vibration mode parameters (such as intermittent vibration and continuous vibration) customized in the vibration screening function area to ensure that the particle size of the sample after screening meets the input requirements of subsequent spectral detection. Data acquisition and control module: Based on the trigger command issued by the data acquisition function area in the smart terminal 15, it controls the near-infrared spectrometer 12 to start the spectral acquisition process and matches and configures the underlying parameters of the instrument (such as integration time and number of scans); during the acquisition process, it feeds back the device operating status (such as acquisition progress and whether the instrument is working properly) to the data acquisition function area in real time to ensure the integrity and validity of the spectral data. The inference and prediction module is used to quantitatively calculate the organic matter, nitrogen, and phosphorus content of spectral data using a pre-trained optimal neural network model.

[0043] The data optimization module, model architecture search module, and inference prediction module work together to complete the in-depth processing of spectral data and the generation of detection results. The data optimization module is responsible for the quality purification and feature performance mining of spectral data, including SG smoothing and standardization preprocessing, CARS feature selection, baseline drift enhancement, and screening of high-value samples through active learning algorithms based on diversity sampling, thus laying the foundation for establishing a stable and reliable data model. The model architecture search module is responsible for automatically constructing a predictive model architecture adapted to sludge spectra. Based on a candidate model generation mechanism that combines differentiable architecture search and reinforcement learning, it executes the optimal model screening process and completes the solidification and deployment of lightweight model files, thus providing architectural support for achieving accurate and efficient spectral inference. The inference prediction module performs real-time spectral analysis and prediction of organic matter, nitrogen, and phosphorus content, and finally feeds the generated accurate detection results back to the result presentation function area of ​​the smart terminal 15, providing data support for the visualization of results in the form of numerical dashboards, trend curves, and detection reports.

[0044] The electric furnace 3 is embedded in the base 1 through a heat-insulating bracket. The heat-insulating bracket is lined with a ceramic heat-insulating layer to block heat conduction and protect the structural stability of the base 1. Preferably, in this embodiment, the ceramic heat-insulating layer is made of alumina ceramic material with a thickness of 5-8mm, which has excellent low thermal conductivity and can further enhance the heat barrier effect, preventing the high temperature of the electric furnace 3 from being conducted to the base 1 during operation and causing structural deformation.

[0045] A sliding panel 2 is slidably provided on the base 1, and the sliding panel 2 can slide and enclose the base 1 to form a closed drying chamber. A locking device is provided at the end of the sliding panel 2 or the base 1 to lock the sliding panel 2 when it is closed.

[0046] With the sliding panel 2, the base 1 forms an openable and closable closed cavity structure, which together with the base 1 forms a semi-closed drying chamber, reducing the interference of ambient airflow on the drying temperature and helping to improve the uniformity of sludge drying. The sliding panel 2 can be pushed open, unlocked, and reset, achieving a balance between equipment protection and convenient maintenance.

[0047] The base 1 is equipped with a sliding linear guide rail 13. The near-infrared spectrometer 12 is detachably connected to the linear guide rail 13. The linear guide rail 13 can drive the near-infrared spectrometer 12 to slide in and out of the base 1, thereby facilitating the replacement of the near-infrared spectrometer 12 with other spectroscopic instruments such as Raman spectrometers, and providing a hardware adaptation basis and technical compatibility space for the subsequent expansion of multimodal spectral detection functions.

[0048] The culture dish 4 is a high-temperature resistant glass culture dish made of borosilicate glass, with a temperature resistance of ≥300℃. The edge of the high-temperature resistant glass culture dish 4 is provided with anti-scalding handles, with a temperature resistance of ≥200℃, which facilitates safe loading and unloading of sludge samples under high temperature conditions. Sludge samples can be loaded and unloaded directly by hand without the need for tools, avoiding the operational risks under high temperature conditions.

[0049] The vibration mechanism includes a vibration motor 8 and a crank rocker arm 7. The vibration motor 8 is mounted on the base 1 via a shock-absorbing device, which also includes a support frame 9. The vibration motor 8 is shock-absorbingly mounted on the support frame 9. One end of the crank rocker arm 7 is connected to the output shaft of the vibration motor 8, and the other end is connected to the screen 5 via a quick-release screw 6. The quick-release screw 6 supports quick replacement of the screen 5, adapting to the screening requirements of different particle sizes. The quick-release screw 6 is a hand-tight wing screw, which allows for the disassembly and assembly of the screen 5 without tools, improving operational efficiency.

[0050] When the vibrating motor 8 is running, the crank rocker arm 7 is used to convert the rotational motion of the vibrating motor 8 into the reciprocating vibration of the screen 5, thereby completing the sieving process of the sludge sample. The vibration amplitude can be adjusted within the range of 1-5mm, and the vibration frequency can be controlled from 50-200Hz. For viscous oily sludge, an intermittent vibration mode (vibration for 30s + pause for 10s) can be enabled to avoid sample clumping and clogging of the screen 5, thus efficiently completing the sieving process of the sludge sample.

[0051] The container 10 is detachably mounted on the support frame 9 via quick-release buckles 11, allowing for quick disassembly of the container 10.

[0052] In this embodiment, the screen 5 is made of 304 stainless steel woven mesh and is equipped with 20 mesh (corresponding to a particle size ≤ 0.85 mm) and 100 mesh (corresponding to a particle size ≤ 0.15 mm).

[0053] The container 10 has an opening at its bottom, and a glass gate is installed at the opening. The opening is positioned opposite to the data acquisition port of the near-infrared spectrometer 12. This glass gate is used to precisely control the on / off state of the near-infrared beam during or after near-infrared spectral acquisition. The glass gate can be opened and closed in two ways: it supports remote control via a linkage command issued through the "Data Acquisition Function Area" of the smart terminal 15, and it can also be manually operated via a physical button on the device. This ensures that during the spectral data acquisition phase, the glass gate is fully open, allowing the near-infrared beam to pass through the opening unobstructed and align with the acquisition port of the near-infrared spectrometer 12, ensuring the integrity of the spectral signal transmission. During non-acquisition periods, the glass gate can be closed promptly to isolate stray light from the environment and prevent the introduction of invalid spectral signals. In one alternative implementation, the mechanical drive structure of the aforementioned glass gate can employ existing mature technologies. For example, an opening and closing mechanism based on the iris-like contraction principle can be used, where the glass gate completely blocks or fully opens the opening through the synchronous contraction and expansion of the annular array of blades. Alternatively, a sliding light-blocking plate structure driven by a micro-telescopic rod can be used. By controlling the extension and retraction stroke of the micro-telescopic rod, the light-blocking plate is moved along a preset slide rail at the edge of the opening, thereby precisely controlling whether the light beam passes through. It should be noted that the specific mechanical structure of the aforementioned glass gate is not limited. Whether it is an iris-like, sliding, or other structure capable of controlling the on / off state of the light beam, as long as it meets the core requirement of "no obstruction during the collection period and blocking stray light during non-collection periods," it is acceptable. Those skilled in the art can flexibly select and adapt the corresponding structural design based on the actual equipment's assembly space, control response speed, and other requirements.

[0054] The near-infrared spectrometer 12 covers the 900nm-1700nm band, and the spectral features acquired by the near-infrared spectrometer 12 include 228-dimensional high-dimensional data features.

[0055] It also includes a cover 16 that fits the base 1. The cover 16 has symmetrically distributed push-button elastic latches 17 on both sides. The latches 17 can precisely engage with the corresponding locking grooves on the base 1, achieving a sealed cover between the cover 16 and the base 1. The cover 16 is integrally molded from flame-retardant ABS material. A 5mm thick silicone heat insulation pad is laid on the inner wall of the cover 16, which can prevent external dust, moisture, and debris from entering the equipment, protecting the core components of the testing instrument from contamination or impact damage.

[0056] The base 1 is equipped with casters 18, which are universal casters with braking function. The casters are made of wear-resistant natural rubber, with a diameter of 30mm. Each caster 18 has a load-bearing capacity of no less than 25kg, which can stably support the weight of the entire device (including internal components and the sample to be tested) (the total weight of the testing device is about 5-10kg). Each caster 18 is equipped with an independent press-type brake pad. Pressing the brake pad can quickly lock the caster, preventing accidental slippage of the device during on-site testing operations and ensuring the stability of the sample position and the accuracy of data acquisition during the testing process. The universal casters support 360° flexible rotation. The overall dimensions of the device (including the base 1 and the closed cover 16) are 400mm×300mm×250mm (length×width×height). The compact size design can be adapted to complex sites such as the area around municipal pipeline inspection wells, sewage treatment plant workshops, and field monitoring points, effectively solving the problem that traditional fixed testing equipment is difficult to adapt to on-site mobility needs, and significantly improving the portability and site adaptability of the equipment.

[0057] This embodiment also includes a readable storage medium storing a computer program that, when executed by a processor, implements the pipeline sludge composition detection method as described above.

[0058] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program segments capable of performing a specific function, and these instruction segments describe the execution process of the computer program in the electronic device.

[0059] This embodiment also provides an electronic device, including: a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer program to implement the pipeline sludge composition detection method as described above.

[0060] The electronic device can be a mobile phone, desktop computer, laptop, handheld computer, cloud server, or other computing device. The electronic device may include, but is not limited to, processors and memory. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0061] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. For those skilled in the art, improvements and modifications obtained without departing from the inventive concept should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting the composition of sewage sludge in a pipe network, characterized in that, The method includes: Collect spectral data of the sludge from the pipeline network after drying and vibrating screening; The spectral data of the pipeline sludge after drying and vibrating screening are preprocessed, including SG smoothing and normalization. A competitive adaptive reweighted sampling algorithm is used to select features from the preprocessed spectral data; Baseline drift enhancement is applied to the feature-selected dataset; The K-means clustering sampling method is used to actively learn and filter the baseline-shifted enhanced dataset. Constructing the optimal neural network model based on a neural network architecture search framework; The composition prediction results of the sewage sludge in the pipeline network are obtained by using a pre-trained optimal neural network model to predict real-time spectral data. The feature selection of the preprocessed spectral data using the competitive adaptive reweighted sampling algorithm includes: Construct the original spectral dataset using the preprocessed spectral data ; Among them, the training set , This represents the 228-dimensional spectral samples of the training set. The labels correspond to the 228-dimensional spectral samples in the training set. Total number of training samples; test set , This represents the 228-dimensional spectral samples of the test set. The labels correspond to the 228-dimensional spectral samples in the test set. Total number of test samples; For the competitive adaptive reweighted sampling algorithm, the number of iterations is set to... The cross-validation fold number is ; Preprocessed training data A competitive adaptive reweighted sampling algorithm is used for feature selection. For the standardized samples, at the first Number of features retained in the next iteration Satisfy the following expression: ； in, This is the floor function; Based on the feature subset retained from the previous iteration Training a partial least squares model It satisfies the following expression: ； in express Fitting function, To standardize the training sample matrix, To train the label vectors, a new feature subset is obtained by filtering through weight coefficients. ; Finally, the optimal feature subset is obtained. It satisfies the following expression: ； in, The model loss function; The training set after feature selection based on the optimal feature subset is: ; The baseline drift enhancement of the feature-selected dataset includes: To enhance baseline drift, set a set of spline interpolation control points. and It satisfies the following expression: ； ； in, Represents the baseline drift enhancement in the first The spectral wavelength coordinates corresponding to each spline interpolation control point Represents the baseline drift enhancement in the first The spectral baseline intensity value corresponding to each spline interpolation control point L represents the number of control points; the interpolation type is... This is a linear interpolation method used to connect any two adjacent control points with a straight line segment, simulating a smooth, linear baseline drift. This is a cubic spline interpolation method used to construct a cubic polynomial curve between any two adjacent control points. Baseline drift enhancement is performed on the feature-selected training set. The baseline drift signal generation function is: ； The baseline drift signal generation function takes spectral wavelength points as input and spline interpolation control point set and interpolation type as parameters, and is generated through the spline interpolation function. The baseline drift signal was calculated. Augmented samples of a single sample Satisfy the following expression: ； in, For the first Spectral data obtained after feature selection of each sample using a competitive adaptive reweighted sampling algorithm; Set the enhancement factor to Enhanced training set Satisfy the following expression: ； in, Indicates the first Samples generated through secondary enhancement This means merging multiple sets into one set; The active learning and filtering of the baseline-shifted augmented dataset using the K-means clustering sampling method includes: Set the number of clusters to Randomly selected from the augmented training set 1 sample was used as the initial cluster center. Calculate the Euclidean distance between each sample in the augmented training set and all initial cluster centers. , ; in, yes Any single spectral sample vector in the array, For the first 1. Initial cluster center vectors are generated; each sample in the augmented training set is assigned to the cluster of the nearest cluster center, resulting in the initial cluster partitioning. ; in, Represented as K-means clustering, the th After the nth iteration, the c-th cluster is obtained. The initial cluster partitioning is the result of the first iteration. ,Right now ; For each cluster Recalculate cluster centers , The number of samples within the cluster. This can be expressed as the summation of all sample vectors within the c-th cluster after the t-th iteration; Repeated sample allocation and cluster center updates continue until the change in cluster centers reaches the maximum number of iterations, resulting in the final cluster partitioning. From each cluster Select distance from cluster center The most recent sample is used as the representative sample, and its index satisfy: ; in, It is the L2 norm, i.e., the Euclidean distance; The training set obtained after active learning selection is ,in, This represents the first selected by active learning. Representative enhanced spectral samples of each cluster, To enhance spectral samples The corresponding actual ingredient label; The construction of the optimal neural network model based on the neural network architecture search framework includes: Sample reshaping: Reshaping the samples in the training set after active learning. Reshape to input format ,in The feature dimensions selected by the competitive adaptive reweighted sampling algorithm; Search space definition: Define the search space It includes basic network components and hyperparameter ranges, wherein the basic network components include convolutional layers, fully connected layers, and activation functions; Search algorithm configuration: A search algorithm is used. In search space Iterative optimization, the search algorithm The evaluation metric is the model's prediction loss on the validation set. ; Optimal model generation: through a search algorithm Candidate model structures are sampled, trained, and evaluated to ultimately select the best ones. Minimal optimal model It satisfies the following expression: ， in, For search space Candidate model structure in.

2. The method for detecting the composition of sewage sludge in a pipeline network according to claim 1, characterized in that, The SG smoothing is used to suppress random high-frequency noise in spectral data through local polynomial fitting while preserving spectral characteristic trends; The standardization is used to eliminate systematic errors, including baseline drift and instrument response differences, between sludge spectral data from different pipe networks, and ultimately remove comprehensive noise introduced by pipe network sludge particle inhomogeneity, instrument fluctuations, and environmental interference.

3. The method for detecting the composition of sewage sludge in a pipeline network according to claim 2, characterized in that, The method of using a pre-trained optimal neural network model to predict real-time spectral data includes: Acquire real-time spectral data of sludge from the pipeline network and preprocess the data; Based on the optimal feature subset determined during the training phase, features corresponding to the dimensions are extracted from the standardized real-time spectral data. The extracted features are then input into the pre-trained optimal neural network model, and the model is used for forward propagation calculations to output the predicted results of organic matter, nitrogen, and phosphorus content.

4. A device for detecting the composition of sewage sludge in a pipe network, characterized in that, The method for detecting the composition of sludge in a pipe network according to any one of claims 1-3, wherein the device for detecting the composition of sludge in a pipe network comprises: A drying and sieving device includes a base, an electric furnace, a petri dish, a screen, and a vibration mechanism. The electric furnace is embedded inside the base and has a positioning groove for placing the petri dish. The electric furnace heats the petri dish through the positioning groove. The vibration mechanism is mounted on the base and is connected to the screen for transmission. The detection device includes a container and a near-infrared spectrometer. The container is detachably mounted on a base and located below the screen. The near-infrared spectrometer is slidably mounted inside the base. The container is positioned within the detection range of the near-infrared spectrometer. The data optimization module is used to perform preprocessing, feature selection, baseline drift enhancement, and active learning filtering on the spectral data acquired by the near-infrared spectrometer. The model architecture search module is used to automatically construct the optimal neural network model based on the spectral data filtered by active learning, through a neural network architecture search framework. The inference and prediction module is used to quantitatively calculate the organic matter, nitrogen, and phosphorus content of spectral data using a pre-trained optimal neural network model.

5. The pipe network sludge composition detection device according to claim 4, characterized in that, It also includes a smart terminal, which is used to control the drying and screening device, the detection device, the data optimization module, the model architecture search module, and the inference and prediction module, and to present the composition prediction results of the pipeline sludge.

6. The pipe network sludge composition detection device according to claim 4, characterized in that, The electric furnace is embedded in the base through a heat-insulating bracket, and the heat-insulating bracket is lined with a ceramic heat-insulating layer.

7. The pipe network sludge composition detection device according to claim 4, characterized in that, The culture dish is a high-temperature resistant glass culture dish; The edge of the high-temperature resistant glass petri dish is equipped with anti-scalding handles.

8. The pipe network sludge composition detection device according to claim 4, characterized in that, The base is equipped with a sliding panel, which can slide to form a closed drying chamber with the base.

9. The pipe network sludge composition detection device according to claim 4, characterized in that, The vibration mechanism includes a vibration motor and a crank rocker arm; The vibration motor is mounted on the base via a shock-absorbing device. One end of the crank rocker arm is connected to the output shaft of the vibration motor, and the other end is connected to the screen via a quick-release screw. When the vibratory motor is running, the crank rocker arm is used to convert the rotational motion of the vibratory motor into the reciprocating vibration of the screen.

10. The pipe network sludge composition detection device according to claim 4, characterized in that, The container has an opening at the bottom, a glass gate at the opening, and the opening is positioned opposite to the data acquisition port of the near-infrared spectrometer.

11. The pipe network sludge composition detection device according to claim 4, characterized in that, The base is equipped with a sliding linear guide rail, and the near-infrared spectrometer is detachably connected to the linear guide rail. The linear guide rail can drive the near-infrared spectrometer to slide in and out of the base.

12. The pipe network sludge composition detection device according to claim 4, characterized in that, The near-infrared spectrometer covers the 900nm-1700nm band, and the spectral features acquired by the near-infrared spectrometer include 228-dimensional high-dimensional data features.

13. The pipe network sludge composition detection device according to any one of claims 4-10, characterized in that, It also includes a lid that fits the base, which is fastened to the base by a push-button resilient latch.

14. The pipe network sludge composition detection device according to claim 13, characterized in that, The lid is made of flame-retardant ABS material in one piece, and a silicone heat insulation pad is laid on the inner wall of the lid.

15. The pipe network sludge composition detection device according to claim 4, characterized in that, The base is equipped with pulleys, which are universal wheels with braking function.

16. A readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method for detecting the composition of sludge in a pipeline network as described in any one of claims 1-3.

17. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the pipeline sludge composition detection method as described in any one of claims 1-3.

Images