A real-time monitoring system for components of waste fed into a waste incinerator

Abstract

The application discloses a kind of real-time monitoring system of garbage composition of garbage incinerator, it is related to domestic waste incineration power generation technical field, including: multi-source data acquisition module obtains the multidimensional characteristic information of garbage and sends to data transmission module, the data transmission module forwards data to intelligent analysis processing module, the intelligent analysis processing module is analyzed and fusion analysis to data, and analysis result is output to data storage and visual module and is shown, while control instruction is sent to incinerator control system by feedback control interface module, by from source on stabilizing combustion temperature, optimizing combustion efficiency, realize inhibiting pollutant generation and reducing activated carbon reagent consumption cost, so as to solve the technical problems of detection lag, passive control and high operating cost in prior art.

Description

Technical Field

[0001] This invention relates to the field of municipal solid waste incineration power generation technology, specifically to a real-time monitoring system for the composition of waste entering a waste incinerator. Background Technology

[0002] Municipal solid waste incineration for power generation has become the mainstream method for urban solid waste treatment. However, the composition of municipal solid waste is complex and highly variable, with unstable proportions of components such as kitchen waste, plastics, and paper, which directly affects the combustion efficiency of incinerators, pollutant emissions, and equipment operational safety.

[0003] Existing methods for detecting waste composition, such as Chinese patent CN116879509A, which discloses a degradation efficiency detection system for activated carbon used in waste-to-energy plants to remove dioxins, have significant drawbacks. First, they lack real-time performance. Traditional manual sampling and testing have long cycles, typically requiring hours to days, which cannot meet the needs of real-time control of incineration conditions. Second, they lack precision. Single-sensor detection has low accuracy, only able to identify broad categories of waste, making it difficult to accurately quantify the proportion of key components. Third, they are limited in function. Existing online monitoring equipment mainly focuses on flue gas pollutants (such as dioxin precursors), lacking the ability to perceive the real-time composition of the waste itself entering the furnace. These problems lead to lag in incineration condition control, easily resulting in excessive loss on ignition and increased dioxin formation. Summary of the Invention

[0004] This invention proposes a real-time monitoring system for the composition of waste entering a waste incinerator. Through multi-source heterogeneous sensing and deep learning technology, it can achieve millisecond-level online detection of the composition of waste entering the incinerator, and perform real-time closed-loop optimization and control of the incineration conditions based on the detection results.

[0005] A further objective of this invention is to address the technical problems of delayed detection, passive control, and high operating costs in the prior art by stabilizing combustion temperature and optimizing combustion efficiency from the source, thereby suppressing pollutant generation and reducing the consumption cost of activated carbon reagents.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a real-time monitoring system for the composition of waste entering a waste incinerator, comprising: a multi-source data acquisition module to acquire multi-dimensional characteristic information of the waste and send it to a data transmission module; the data transmission module forwards the data to an intelligent analysis and processing module; the intelligent analysis and processing module parses and fuses the data, and outputs the analysis results to a data storage and visualization module for display; and simultaneously sends control commands to the incinerator control system through a feedback control interface module.

[0007] Preferably, the multi-source data acquisition module includes a high-definition visual acquisition unit, a near-infrared spectral detection unit, and a gas sensor array. The high-definition visual acquisition unit acquires garbage image information, the near-infrared spectral detection unit is used to acquire the spectral information of organic / inorganic components of the garbage, and the gas sensor array is used to acquire the volatile gas index generated by the volatilization of garbage.

[0008] Preferably, the high-definition visual acquisition unit uses OpenCV image processing components in conjunction with FFmpeg video stream encoding components and is deployed above the waste conveyor belt, enabling full-width coverage of the waste flow on the conveyor belt.

[0009] Preferably, the data transmission module is built on the Apache Kafka message queue to construct a high-throughput data stream pipeline.

[0010] Preferably, the intelligent analysis and processing module is equipped with a YOLOv8 deep learning model trained on the ai53_19 / garbage_datasets dataset, and combined with the PyTorch inference framework to perform fine identification of garbage categories on the images transmitted by the high-definition visual acquisition unit.

[0011] Preferably, the YOLOv8 deep learning model is combined with the spectral data transmitted by the near-infrared spectroscopy detection unit to invert the calorific value, moisture content and chlorine content parameters of the waste.

[0012] Preferably, the intelligent analysis and processing module corrects the visual recognition results and spectral inversion results through a multi-source data fusion algorithm to output the final waste component proportion and key parameter quantification values, with a classification accuracy of not less than 95% and an average reasoning time of less than 80ms.

[0013] Preferably, the data storage and visualization module uses Elasticsearch as the storage engine and is used in conjunction with the Grafana monitoring panel to display the proportion of waste components, the trend of calorific value changes, and the system operating status in real time at a refresh interval of 5 seconds.

[0014] Preferably, the feedback control interface module is equipped with a standard communication protocol interface to convert the waste composition data and key parameters output by the intelligent analysis and processing module into incineration condition control commands.

[0015] Preferably, the control commands include adjusting the feed rate and air distribution ratio of the incinerator to achieve closed-loop optimization of the incineration process.

[0016] The beneficial effects of this invention are: 1) High real-time performance: It enables millisecond-level detection and result output of waste composition throughout the entire process of waste entering the furnace, solving the problem of detection lag in traditional methods and providing timely data support for operation condition control; 2) High detection accuracy: It integrates visual recognition, spectral analysis and gas sensing technologies, combined with deep learning algorithms, to achieve fine classification of waste and accurate quantification of key parameters, meeting the data accuracy requirements of incineration optimization; 3) Good compatibility: The system architecture is modularly designed and can be seamlessly integrated into existing waste-to-energy incineration systems without large-scale equipment modifications, thus reducing application costs; 4) Significant environmental and economic benefits: By precisely controlling the incineration conditions, emissions of pollutants such as dioxins and nitrogen oxides can be reduced, ensuring compliance with national environmental protection standards, while improving combustion efficiency and reducing operation and maintenance costs and operational risks. Attached Figure Description

[0017] Figure 1 This is a diagram of the overall system architecture of the present invention.

[0018] Figure 2 This is a schematic diagram of the material feeding and conveying path deployment for the data acquisition module of this invention.

[0019] Figure 3 This is a flowchart of the intelligent analysis and processing of the present invention.

[0020] Figure 4 This is a schematic diagram of the visual interface of the monitoring panel of the present invention. Detailed Implementation

[0021] Example 1

[0022] This embodiment provides a real-time monitoring system for the composition of waste entering a waste incinerator. It aims to build a complete real-time detection closed-loop system. Through the collaborative deployment of multi-source heterogeneous hardware, it solves the problems of single data acquisition, high transmission delay and lagging control response in the prior art, laying the physical foundation for subsequent intelligent analysis and precise control.

[0023] The overall system architecture consists of five core components: a multi-source data acquisition module, a data transmission module, an intelligent analysis and processing module, a data storage and visualization module, and a feedback control interface module. These modules are seamlessly connected via industrial Ethernet and high-speed communication protocols. Figure 1 As shown.

[0024] At the hardware deployment level, the multi-source data acquisition module is installed directly above the feed conveyor belt of the waste incineration plant, at the last critical detection point before the waste enters the furnace.

[0025] Due to the harsh environment of the garbage pit and conveyor channels, including direct sunlight, dust, large humidity fluctuations, and mechanical vibration, this embodiment imposes strict requirements on the protection level of the acquisition equipment. The high-definition visual acquisition unit uses an industrial-grade CCD camera with an IP67 protection rating. A dustproof and defogging device is installed at the front of the lens, and a high-brightness LED strobe light source is configured to ensure that clear garbage images can be acquired even in low-light or high-dust environments.

[0026] Specifically, the high-definition visual acquisition unit in this embodiment uses OpenCV image processing components in conjunction with FFmpeg video stream encoding components. OpenCV is used to perform real-time image preprocessing, such as grayscale conversion, histogram equalization, and edge enhancement, to improve image quality. FFmpeg is responsible for encoding the processed image data into an H.264 format video stream, which is transmitted via the RTSP protocol, effectively reducing bandwidth consumption. Deployed above the waste conveyor belt, this unit, through reasonable optical design and lens selection, can achieve full-width coverage of the waste flow on the conveyor belt, ensuring no blind spots in detection and avoiding data deviations caused by local sampling.

[0027] The data transmission module is the "blood vessel" connecting field devices and the back-end computing center.

[0028] In this embodiment, the data transmission module is based on the Apache Kafka message queue to build a distributed stream processing platform.

[0029] The Kafka cluster consists of three high-performance servers deployed in a local data center and connected to the front-end data acquisition equipment via gigabit industrial Ethernet.

[0030] Kafka's high throughput can effectively buffer image and sensor data streams of up to 100MB per second, preventing data loss due to network fluctuations or excessive instantaneous data volume.

[0031] Apache Kafka, as a high-throughput distributed publish-subscribe messaging system, has the core advantage of being able to handle large-scale real-time data streams.

[0032] In this system, Kafka acts as the central hub of the data flow, receiving various types of data from multiple data acquisition modules and efficiently distributing them to subsequent processing modules. Kafka's distributed architecture and data replication mechanism ensure high data reliability and system scalability.

[0033] Producers (i.e., the data acquisition module) publish data to specific topics in Kafka, while consumers (i.e., the intelligent analysis and processing module) subscribe to and pull data from these topics for processing. This decoupled design allows data production and consumption to occur independently, greatly improving the system's flexibility and stability.

[0034] At the same time, Kafka's persistence mechanism ensures that data is not lost when the system fails, and recovery can be achieved from the point of failure.

[0035] Meanwhile, Kafka's publish / subscribe model supports parallel reading by multiple consumers, which not only meets the needs of real-time analysis but also provides a data interface for subsequent historical data backtracking and model training.

[0036] The intelligent analysis and processing module, serving as the system's "brain," is deployed on a high-performance computing node equipped with an NVIDIA T4 GPU. This module is responsible for running the deep learning inference engine, parsing, identifying, and fusing the collected raw data.

[0037] To cope with the dynamic changes in waste flow, the computing nodes are deployed using Docker containerization technology, enabling rapid elastic scaling. When a surge in waste flow is detected, the system can automatically call upon cloud computing power for auxiliary calculations, ensuring real-time analysis. In this embodiment, the intelligent analysis and processing module is equipped with a YOLOv8 deep learning model trained on the ai53_19 / garbage_datasets dataset, and combined with the PyTorch inference framework, to perform fine-grained identification of waste categories on the images transmitted by the high-definition visual acquisition unit.

[0038] The ai53_19 / garbage_datasets dataset is a large-scale, diverse dataset of garbage images, containing various types and states of garbage images, providing rich samples for model training.

[0039] YOLOv8, as an advanced real-time object detection algorithm, boasts advantages such as fast detection speed and high accuracy. Through thorough training on this dataset, the YOLOv8 model can learn feature representations of different types of trash, thereby achieving rapid and accurate identification of various types of trash in images.

[0040] The PyTorch inference framework provides an efficient and flexible environment for model deployment and operation, supports GPU acceleration, and can significantly improve the inference speed of the model.

[0041] The data storage and visualization module uses Elasticsearch as the core storage engine, combined with the Grafana monitoring panel to achieve real-time data display. The monitoring panel includes features such as... Figure 4As shown, Elasticsearch's distributed search and analysis capabilities can efficiently process massive amounts of time-series and structured data, supporting millisecond-level data retrieval.

[0042] Operators can visually view the composition, calorific value distribution, moisture content, and concentration trends of various pollutant precursors of the waste on the conveyor belt through a Grafana dashboard. Elasticsearch, as an open-source search and analytics engine, is particularly adept at handling large-scale data.

[0043] In this system, it is responsible for storing various analysis results from the intelligent analysis and processing module, including the type, quantity, location information of waste, and related statistical indicators.

[0044] Its powerful indexing and search functions allow users to quickly query specific time periods and types of junk data, providing support for subsequent data analysis and decision-making.

[0045] Grafana, an open-source analytics and monitoring platform, can seamlessly integrate with Elasticsearch to visually display stored data in the form of charts, dashboards, and other visual aids.

[0046] By configuring different data sources and dashboards, Grafana can refresh and display key information such as the changing trends of waste composition and system operating status in real time, helping operators to fully grasp the operating status of the incinerator.

[0047] The feedback control interface module is a key component in realizing the "perception-decision-execution" closed loop. In this embodiment, the module adopts the standard OPC UA communication protocol, which is a cross-platform, secure, and interoperable protocol that can seamlessly interface with DCS (Distributed Control Systems) of incinerators from different manufacturers.

[0048] The interface module is responsible for converting the component data output by the intelligent analysis and processing module into specific control commands, such as adjusting the feeder's pushing frequency and adjusting the opening of the dampers for primary and secondary air.

[0049] OPC UA, as an industry standard communication protocol, provides a unified data access interface, enabling convenient data exchange between devices from different manufacturers and of different types.

[0050] In this system, the feedback control interface module transmits key parameters such as waste composition analysis results and calorific value prediction to the incinerator's DCS system in real time via the OPC UA protocol.

[0051] Based on this information and its own control logic, the DCS system automatically adjusts the incinerator's operating parameters, such as combustion temperature, air supply, and grate speed, to achieve optimal combustion and minimal pollutant emissions.

[0052] This closed-loop control based on real-time data greatly improves the automation and intelligence level of the waste incineration process.

[0053] The deployment of this architecture ensures that the latency of the entire process, from data acquisition, transmission, analysis to final control execution, is kept within 200 milliseconds.

[0054] Compared to traditional offline manual sampling or end-of-pipe flue gas detection, this edge-cloud collaborative architecture enables true online real-time control, providing solid technical support for the refined management of the waste incineration process.

[0055] The intelligent analysis and processing module uses a multi-source data fusion algorithm to correct the visual recognition results and spectral inversion results, outputting the final waste component proportion and key parameter quantification values. The multi-source data fusion algorithm is crucial for improving the system's detection accuracy.

[0056] Visual recognition alone is easily affected by factors such as lighting, occlusion, and dirt on the surface of garbage, while spectral detection can provide deeper information on chemical composition.

[0057] Therefore, this system combines the two, using a specific algorithm to weight, fuse, and correct the visual recognition results and spectral inversion results.

[0058] For example, if visual recognition determines that an object is plastic, but spectral data shows that it does not contain CH bonds, the system will lower the confidence level of the determination and re-analyze.

[0059] Through this fusion, the system can output more accurate and comprehensive waste composition ratios and key parameter quantification values, with a classification accuracy of no less than 95% and an average inference time of less than 80ms.

[0060] High accuracy ensures the reliability of subsequent control, while low latency ensures that the system can keep up with the speed of waste transportation, enabling real-time monitoring and regulation.

[0061] The data storage and visualization module displays the proportion of waste components, calorific value trends, and system operating status in real time with a refresh interval of 5 seconds. This 5-second refresh interval ensures data real-time performance while avoiding excessive system resource consumption from overly frequent refreshes.

[0062] Operators can monitor the overall status of the waste entering the furnace at any time through the large monitoring screen, and promptly detect abnormalities and take measures.

[0063] The feedback control interface module converts the waste composition data and key parameters output by the intelligent analysis and processing module into incineration operation control commands.

[0064] These control commands include adjusting the feed rate and air distribution ratio of the incinerator to achieve closed-loop optimization of the incineration process. For example, when a high calorific value of the waste is detected, the system may instruct to appropriately reduce the feed rate to maintain stable furnace temperature; when a high amount of combustible components are detected in the waste, the system will instruct to increase the air distribution to ensure complete combustion.

[0065] Through this closed-loop optimization, the system can automatically adjust the incineration strategy according to the actual composition and characteristics of the waste, thereby improving combustion efficiency, reducing energy consumption and pollutant emissions, and achieving more environmentally friendly and economical operation. Example 2

[0066] This embodiment provides a multi-source data fusion method for waste component identification, focusing on solving the technical challenges of low identification accuracy and poor anti-interference ability of single visual recognition technology when dealing with complex, dirty, and overlapping waste materials. By introducing near-infrared spectroscopy detection and a gas sensing array, a multi-source data fusion algorithm is constructed to achieve high-precision and robust identification of waste components.

[0067] As the waste stream passes through the detection area at a constant speed along the conveyor belt, the multi-source data acquisition module is activated synchronously. The conveying path is as follows: Figure 2 As shown.

[0068] First, the high-definition visual acquisition unit continuously captures images of garbage at a frame rate of 30fps.

[0069] After preprocessing (denoising and distortion correction), the image data is input into a deep learning model based on the YOLOv8 architecture.

[0070] The YOLOv8 model has been fully trained offline using the ai53_19 / garbage_datasets dataset, and can quickly identify more than ten types of common household waste such as plastic bottles, paper, textiles, and kitchen waste.

[0071] The model outputs the bounding box and confidence score for each type of waste. However, in actual working conditions, because the surface of garbage is often covered with mud or soup, or different types of garbage may obscure each other, relying solely on visual texture and color characteristics often leads to misjudgment. For example, newspaper covered with oil stains may be misjudged as plastic, and slippery kitchen waste may be misjudged as rubber.

[0072] To address this issue, this embodiment simultaneously activates the near-infrared spectroscopy detection unit. The near-infrared spectrometer emits near-infrared light with a wavelength range of 900-1700 nm, illuminating the surface of the waste and receiving the reflected light signal. Different chemical bonds (such as CH, OH, NH) exhibit unique absorption characteristics in the near-infrared band. By analyzing the spectral reflectance curve, the intrinsic chemical composition of the waste can be deduced. For example, cellulose materials (paper, wood) have significant absorption peaks at 1200 nm and 1450 nm, while polyolefin materials (plastics) have characteristic absorptions at 1150 nm and 1720 nm. This embodiment employs a PLS (partial least squares) regression model to map the collected spectral data into specific physical parameters, such as moisture content. Calorific value and chlorine content .

[0073] To achieve the complementary advantages of visual and spectral information, this embodiment designs a multi-source data fusion algorithm based on confidence weighting. Traditional single-vision methods are easily affected by surface interference, while spectral methods, although able to penetrate surfaces, have low spatial resolution. Therefore, a weight correction factor is introduced. To construct the final comprehensive identification confidence level. : in, Spectral feature matching degree. Weighting factor. It is not a fixed value, but rather dynamically adjusted based on environmental parameters. When spectral data shows high moisture content (the 1450nm absorption peak intensity exceeds the threshold), it adjusts accordingly. When the system determines that there may be liquid obscuring the surface of the garbage in that area, it automatically reduces the visual recognition weight. Improve spectral recognition weight Conversely, when the surface of the garbage is dry and clean, visual identification becomes the primary method.

[0074] In addition, a gas sensor array serves as a third verification method, monitoring volatile organic compounds (VOCs) emitted during waste transportation in real time. For example, food waste releases high concentrations of ammonia and hydrogen sulfide, while waste plastics may release benzene compounds. These gas concentration indicators serve as auxiliary feature vectors, which are further input into the fusion algorithm to correct the final classification results.

[0075] Through the aforementioned multi-source data fusion mechanism, the system significantly reduces the misjudgment rate when facing complex operating conditions. Field tests showed that in a purely visual solution, the misjudgment rate due to surface contamination was approximately 8%; however, after introducing the fusion of spectral and gas data, the misjudgment rate was effectively controlled below 1.5%, and the overall recognition accuracy remained consistently above 95%. This high-precision recognition capability ensures the accuracy and reliability of subsequent combustion control commands. Example 3

[0076] This embodiment provides a combustion optimization and activated carbon reduction control process based on component detection, demonstrating how the present invention achieves closed-loop optimization of the combustion process through source component detection, and in particular how to effectively reduce dependence on end-of-pipe activated carbon adsorption technology, thereby overcoming the limitations of the "end-of-pipe treatment" technical route represented by prior art document 2.

[0077] During system operation, the intelligent analysis and processing module continuously analyzes the composition of the waste on the conveyor belt.

[0078] Suppose the system detects a continuous large amount of polyvinyl chloride (PVC) plastic on the conveyor belt, and the chlorine content (Ccl) output by the model remains above the warning line of 5%. PVC readily generates highly toxic dioxins during incineration, and if left uncontrolled, this will lead to excessive flue gas emissions.

[0079] In response to this specific scenario, the feedback control interface module immediately triggers the preset optimized control strategy, generates corresponding control commands, and sends them to the incinerator's DCS control system. The flowchart is as follows: Figure 3 As shown.

[0080] The specific actions to be performed include: Optimize air distribution strategy: The system commands an increase in the injection volume of secondary air. By increasing the turbulence within the furnace, it ensures that oxygen can fully mix with the combustible gases produced by waste pyrolysis, thereby suppressing incomplete combustion caused by oxygen deficiency. Incomplete combustion is a crucial precursor reaction for dioxin formation, and a sufficient oxygen supply can effectively cut off this formation pathway.

[0081] Increasing furnace temperature: The system commands adjust the burner fuel supply or grate speed to raise the temperature of the main combustion zone from the conventional 850℃ to 900℃, ensuring that the flue gas residence time in this high-temperature zone exceeds 2 seconds. High-temperature incineration is the most effective means of decomposing dioxins; through this active temperature increase, dioxin molecules can be completely oxidized and decomposed before they form.

[0082] To verify the effectiveness of this embodiment, a comparative test was conducted for one month at a large-scale municipal solid waste incineration power plant. During the test, when the system detected high-chlorine waste, the aforementioned optimization strategy was automatically executed. Data recorded by the online monitoring equipment (CEMS) installed at the flue gas purification system outlet revealed that the original concentration of dioxins in the flue gas was significantly reduced after source-optimized combustion.

[0083] Further analysis of the operational data from the end-of-pipe activated carbon injection system revealed that due to more complete combustion at the front end and reduced dioxin formation, the amount of activated carbon required for subsequent adsorption of residual dioxins decreased significantly. Statistical calculations show that after the system of this invention was put into operation, the average daily consumption of activated carbon throughout the plant decreased by approximately 30%. This not only directly reduces operating costs but also reduces the risk of secondary pollution from activated carbon production, transportation, and disposal.

[0084] This embodiment fully demonstrates that the present invention, by real-time detection of waste composition and implementation of targeted combustion optimization, achieves a shift from "passive treatment" to "active prevention." Compared to passively responding to pollutants simply by testing activated carbon adsorption efficiency, the present invention reduces the total amount of pollutants generated at the source, resulting in significant economic and environmental benefits. Example 4

[0085] This embodiment details the training process, data augmentation strategies, and lightweight model processing methods of the core algorithm YOLOv8 deep learning model in the intelligent analysis and processing module, to ensure that the model has extremely fast inference speed while maintaining high accuracy.

[0086] The first step in model training is the construction and annotation of the dataset. This invention collected millions of images of household waste from different regions and seasons, covering various scenarios such as mixed waste, sorted waste, and shredded waste. Using annotation tools such as LabelImg, precise rectangular bounding boxes were used to annotate more than fifty subcategories across six major categories, including plastics, metals, glass, paper, textiles, and kitchen waste. To improve the model's robustness to complex backgrounds, this embodiment employs various data augmentation techniques, including random cropping, color dithering, mosaic augmentation, and adaptive anchor box calculation.

[0087] In terms of model structure, this embodiment uses YOLOv8s (small) as the base network. Compared to earlier versions of YOLO, YOLOv8 adopts an anchor-free detector head and a C2f feature pyramid structure, reducing the difficulty of hyperparameter tuning while improving the detection capability for small objects. The loss function uses CIoU Loss combined with FocalLoss to address the imbalance between positive and negative samples and improve the accuracy of bounding box regression.

[0088] To meet the requirements of real-time detection (inference time less than 80ms), this embodiment performs lightweight processing on the trained model. Specifically, the TensorRT inference engine is used to optimize the PyTorch model. TensorRT significantly improves the model's running efficiency on NVIDIA GPUs by fusing network layers, precision calibration (INT8 quantization), and automatic kernel tuning. After optimization, the model maintains an mAP (mean accuracy) of no less than 0.92, and the inference time per frame is stabilized at around 50ms, fully meeting the 30fps real-time requirement. Example 5

[0089] To ensure the stability of the system during 24 / 7 uninterrupted operation, this embodiment provides a complete set of exception handling and fault tolerance mechanisms.

[0090] When the data transmission module detects a network interruption or a Kafka queue backlog exceeding a threshold, the system automatically activates local caching mode. The front-end acquisition devices temporarily store data on local solid-state drives, automatically resuming transmission once the network is restored to prevent data loss. If a GPU computing node in the intelligent analysis and processing module fails, the system monitoring program immediately triggers an alarm and automatically switches the task to a backup computing node to ensure uninterrupted detection. Furthermore, for abnormal sensor data (such as spectrometer signal drift), the system has a built-in self-diagnostic algorithm that automatically identifies and filters abnormal data or marks it as invalid, preventing erroneous data from misleading combustion control and thus ensuring the safety and stability of the entire waste incineration process.

Claims

1. A real-time monitoring system for the composition of waste entering a waste incinerator, characterized in that, include: The multi-source data acquisition module acquires multi-dimensional feature information of waste and sends it to the data transmission module. The data transmission module forwards the data to the intelligent analysis and processing module. The intelligent analysis and processing module parses and integrates the data, and outputs the analysis results to the data storage and visualization module for display. At the same time, the control commands are sent to the incinerator control system through the feedback control interface module.

2. The real-time monitoring system for the composition of waste entering a waste incinerator according to claim 1, characterized in that, The multi-source data acquisition module includes a high-definition visual acquisition unit, a near-infrared spectral detection unit, and a gas sensor array. The high-definition visual acquisition unit acquires garbage image information, the near-infrared spectral detection unit is used to acquire the spectral information of organic / inorganic components of the garbage, and the gas sensor array is used to acquire the volatile gas indicators generated by garbage volatilization.

3. The real-time monitoring system for the composition of waste entering a waste incinerator according to claim 2, characterized in that, The high-definition visual acquisition unit uses OpenCV image processing components in conjunction with FFmpeg video stream encoding components and is deployed above the waste conveyor belt, enabling full-width coverage of the waste flow on the conveyor belt.

4. The real-time monitoring system for waste composition entering a waste incinerator according to claim 1, characterized in that, The data transmission module is built on the Apache Kafka message queue to create a high-throughput data stream pipeline.

5. A real-time monitoring system for the composition of waste entering a waste incinerator according to claim 2, characterized in that, The intelligent analysis and processing module is equipped with a YOLOv8 deep learning model trained on the ai53_19 / garbage_datasets dataset, and combined with the PyTorch inference framework to perform fine identification of garbage categories on the images transmitted by the high-definition visual acquisition unit.

6. A real-time monitoring system for the composition of waste entering a waste incinerator according to claim 5, characterized in that, The YOLOv8 deep learning model, combined with spectral data transmitted by the near-infrared spectroscopy detection unit, inverts the calorific value, moisture content, and chlorine content parameters of the waste.

7. The real-time monitoring system for waste composition entering a waste incinerator according to claim 1, characterized in that, The intelligent analysis and processing module corrects the visual recognition results and spectral inversion results through a multi-source data fusion algorithm to output the final waste component proportion and key parameter quantification values. Its classification accuracy is not less than 95%, and the average reasoning time is less than 80ms.

8. A real-time monitoring system for the composition of waste entering a waste incinerator according to claim 1, characterized in that, The data storage and visualization module uses Elasticsearch as the storage engine and Grafana monitoring panel to display the proportion of waste components, the trend of calorific value changes, and the system operating status in real time with a refresh interval of 5 seconds.

9. A real-time monitoring system for the composition of waste entering a waste incinerator according to claim 1, characterized in that, The feedback control interface module is equipped with a standard communication protocol interface, which converts the waste composition data and key parameters output by the intelligent analysis and processing module into incineration operation control commands.

10. A real-time monitoring system for the composition of waste entering a waste incinerator according to claim 1 or 9, characterized in that, The control commands include adjusting the feed rate and air distribution ratio of the incinerator to achieve closed-loop optimization of the incineration process.

Images