A method and system for combined heat and power integrated waste heat utilization

Abstract

The application provides a cogeneration comprehensive waste heat utilization method and system, and relates to the technical field of solid waste treatment and energy recycling. The cogeneration comprehensive waste heat utilization method specifically comprises the following steps: collecting smoke plume multi-spectral visual data of a chimney exhaust port in a cogeneration process, operation data of a waste heat recovery system comprising at least an organic Rankine cycle unit, and external environment data in real time to form multi-dimensional data; based on the multi-spectral visual data, extracting a dynamic time sequence feature vector representing dynamic stability of the smoke plume in the time dimension, and a geometric-spectral feature vector representing current geometric and spectral characteristics of the smoke plume; fusing the dynamic time sequence feature vector, the geometric-spectral feature vector and the operation data to obtain an efficiency deviation index and a stability diagnosis result; based on the efficiency deviation index, generating maintenance alarm information; and based on the stability diagnosis result, generating predictive control instructions.

Description

Technical Field

[0001] This invention relates to the field of solid waste treatment and energy recovery and utilization technology. Specifically, it relates to a comprehensive waste heat utilization method and system applied to incineration power generation and combined heat and power processes, which is particularly suitable for scenarios involving the cascaded and efficient recovery of waste heat from medium and low temperature flue gas. Background Technology

[0002] With increasing emphasis on environmental protection and resource recycling, incineration of solid waste to reduce its volume, render it harmless, and recover its energy has become a mainstream waste treatment technology. In this process, combined heat and power (CHP), as an advanced hazardous waste treatment technology, uses high-temperature flue gas generated from incineration to drive a steam turbine generator set. This not only stabilizes waste disposal but also produces electricity and heat, making it a key link in realizing the energy utilization of waste. To further improve overall energy efficiency, the industry is focusing on recovering the residual heat from medium- and low-temperature flue gas after primary utilization, building upon traditional CHP. Among these technologies, the Organic Rankine Cycle (ORC) is widely used due to its unique advantages in power generation from medium- and low-temperature heat sources. It can effectively convert waste heat from flue gas in the 100-300℃ range into high-quality electricity. Simultaneously, combined with technologies such as heat pumps, the low-grade heat energy generated after ORC system power generation can be further upgraded for district heating or hot water production, thus forming a complete energy cascade utilization system.

[0003] However, existing combined heat and power (CHP) waste heat recovery systems still face deep-seated technical bottlenecks in actual operation, mainly in the dynamic maintenance of operating efficiency and the precision of process control. The core equipment of waste heat recovery systems, such as the evaporators and various heat exchangers in ORC systems, highly depend on clean heat exchange surfaces. However, in actual operation, even after purification, flue gas from the incinerator inevitably contains small amounts of fly ash, acidic substances, and unburned particulate matter. These substances gradually deposit and scale on the heat exchanger surfaces, forming a fouling layer with significant thermal resistance. This fouling layer continuously and slowly deteriorates heat exchange performance, causing the actual operating efficiency of the entire waste heat recovery system to decline over time. Existing monitoring methods typically rely on indirect inference based on macroscopic parameters such as inlet and outlet temperature and pressure, lacking the ability to directly, in real-time, and online quantitatively assess the core issue of scaling. Therefore, precise maintenance and operational optimization based on the actual health status of the equipment cannot be achieved.

[0004] The operating conditions of a waste heat recovery system are closely coupled with those of its upstream incineration unit. Solid waste, as fuel, exhibits inherent volatility and non-uniformity in its calorific value, moisture content, and form. This makes it difficult to achieve absolute stability in the combustion process of the incinerator, resulting in frequent fluctuations in key parameters such as temperature and flow rate of the flue gas entering the waste heat recovery system. Existing waste heat recovery systems mostly employ automatic control systems based on feedback control logic, such as PID (Proportional-Integral-Derivative) control. The essence of this type of control strategy is reactive; it only begins adjustment after detecting deviations in operating parameters from the setpoint, resulting in inherent control lag and potential overshoot risks. When upstream flue gas experiences drastic or rapid fluctuations, this lag-driven response mechanism cannot guarantee that the waste heat recovery system always operates at its optimal efficiency point, and may even trigger thermal shock and mechanical stress in critical equipment (such as ORC turbines), affecting the long-term stability and service life of the system. In summary, there is an urgent need for an advanced technical solution that can penetrate the surface of the system, achieve in-depth diagnosis of the internal efficiency state, and anticipate upstream operating condition disturbances, thereby implementing feedforward and predictive control. Summary of the Invention

[0005] This invention provides a method for comprehensive waste heat utilization from combined heat and power (CHP) plants, which specifically includes the following steps:

[0006] Real-time acquisition of multispectral visual data of flue gas plumes from chimney outlets during cogeneration processes, operational data of waste heat recovery systems including at least organic Rankine cycle units, and external environmental data to form multidimensional data;

[0007] Based on the multispectral visual data, a dynamic temporal feature vector characterizing the dynamic stability of the plume in the time dimension is extracted, as well as a geometric-spectral feature vector characterizing the current geometric and spectral properties of the plume.

[0008] By fusing the dynamic time-series feature vector, the geometric-spectral feature vector, and the operational data, an efficiency deviation index and stability diagnosis results are obtained.

[0009] Based on the efficiency deviation index, maintenance alarm information is generated; and based on the stability diagnosis results, predictive control commands are generated.

[0010] This specification also proposes a combined heat and power (CHP) waste heat utilization system, which includes:

[0011] Data Acquisition Module: Real-time acquisition of multispectral visual data of flue gas plumes from chimney outlets during cogeneration processes, operational data of waste heat recovery systems containing at least an organic Rankine cycle unit, and external environmental data to form multidimensional data;

[0012] Feature extraction module: Based on the multispectral visual data, extracts dynamic temporal feature vectors that characterize the dynamic stability of the plume in the time dimension, and geometric-spectral feature vectors that characterize the current geometric and spectral properties of the plume;

[0013] Early warning and control module: Based on the efficiency deviation index, it generates maintenance alarm information; and based on the stability diagnosis results, it generates predictive control commands.

[0014] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned method for integrated waste heat utilization in cogeneration.

[0015] Compared with existing technologies, the beneficial effect of this invention is that it fundamentally solves the deep-seated technical problems of traditional cogeneration waste heat utilization systems, such as single perception dimension, ambiguous fault diagnosis, and lagging control response, by constructing a complete technical closed loop from multi-dimensional perception and deep feature extraction to intelligent fusion diagnosis.

[0016] Existing technologies rely heavily on dispersed, low-dimensional thermal parameter sensors, which cannot form a global and forward-looking understanding of the overall operating status of the system, especially the dynamic stability of the upstream combustion process. As a result, the system can only passively respond to fluctuations in operating conditions and it is difficult to maintain the optimal operating range for a long time.

[0017] The multimodal perception and feature extraction capabilities of this invention, by introducing multispectral visual data acquisition from the system's terminal plume, add a novel and information-rich perception channel to the system. More importantly, this invention proposes a deep learning method capable of transforming high-dimensional visual data into low-dimensional, high-intrinsicity feature vectors. In particular, the dynamic temporal feature vector, through a three-branch network architecture and cross-correlation calculation, quantifies the abstract concept of plume stability into a precise and continuously trackable engineering indicator, providing high-quality input for subsequent accurate diagnosis. This enables the system not only to see the plume but also to understand the deep information it contains about the dynamic characteristics of the entire process chain, achieving a leap from simple measurement to deep perception.

[0018] Building upon this foundation, the dual-stream fusion diagnostic model designed in this invention brings significant technological advancements. The attention fusion gate mechanism dynamically balances the weights of visual information and traditional operational data, enabling diagnostic conclusions to leverage both the high-speed responsiveness of visual information and the precise quantification of operational data. The resulting diagnostic results are far more robust and reliable than those from any single information source. Crucially, this invention explicitly decouples the diagnostic task into an efficiency deviation index for chronic equipment health degradation and a stability diagnostic result for acute process instability. This clear distinction allows the system to accurately identify two completely different problems requiring drastically different response strategies. Ultimately, this high-precision diagnostic capability enables proactive and differentiated closed-loop control: the system can trigger precise maintenance warnings for chronic efficiency degradation problems while initiating near real-time predictive control adjustments for acute instability problems. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the integrated waste heat utilization of cogeneration in this invention. Detailed Implementation

[0021] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0022] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this application, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0024] Additionally, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that practice can be carried out without these specific details.

[0025] This invention provides a method for comprehensive waste heat utilization from combined heat and power (CHP) plants, which specifically includes the following steps:

[0026] Real-time acquisition of multispectral visual data of flue gas plumes from chimney outlets during cogeneration processes, operational data of waste heat recovery systems including at least organic Rankine cycle units, and external environmental data to form multidimensional data;

[0027] Based on the multispectral visual data, a dynamic temporal feature vector characterizing the dynamic stability of the plume in the time dimension is extracted, as well as a geometric-spectral feature vector characterizing the current geometric and spectral properties of the plume.

[0028] By fusing the dynamic time-series feature vector, the geometric-spectral feature vector, and the operational data, an efficiency deviation index and stability diagnosis results are obtained.

[0029] Based on the efficiency deviation index, maintenance alarm information is generated; and based on the stability diagnosis results, predictive control commands are generated.

[0030] This embodiment provides a specific implementation scenario for a combined heat and power (CHP) waste heat utilization method. Solidified hazardous waste from different regions is transported to the incineration plant using collection vehicles with varying load capacities. It is then collected and buffered at collection stations and storage pools within the plant. The solidified hazardous waste is continuously and uniformly fed into a regenerative thermal oxidizer (RTO) via chain conveyors or screw feeders. Preferably, the RTO employs layered combustion or rotary combustion technology, controlling the temperature distribution and oxidation environment within the furnace to ensure complete combustion of the hazardous waste at high temperatures, effectively destroying its organic components. The high-temperature flue gas generated during incineration, typically exceeding 850°C, serves as the primary heat carrier, exiting from the top of the RTO and flowing sequentially through subsequent energy recovery and purification systems. This high-temperature flue gas is first introduced into a waste heat boiler system, where it generates superheated steam through heat exchange with water. The superheated steam is transported to the steam turbine generator set, which drives the steam turbine and generator to rotate, thereby generating electricity and realizing the first stage of high-temperature waste heat recovery and cogeneration.

[0031] The flue gas flowing from the waste heat boiler, while significantly cooled, still retains considerable utilization value. This portion of the medium-temperature flue gas is then introduced into a flue gas purification system, preferably composed of a series of multiple treatment devices, including an acid washing tower, an electrostatic precipitator, and an activated carbon adsorber. Within this purification system, particulate matter, SOx, NOx, heavy metals, and other pollutants are thoroughly removed to meet the safety requirements of subsequent equipment and final emission standards. After purification, the clean, medium-low temperature flue gas (e.g., 150-300°C) serves as the heat source for the second stage of waste heat recovery and is fed into the evaporator of an Organic Rankine Cycle (ORC) power generation unit. Inside the evaporator, the flue gas exchanges heat with the organic working fluid of the ORC system, vaporizing it into high-temperature, high-pressure steam. This organic working fluid steam then drives the ORC turbine (expander), which in turn powers a coaxially connected generator to produce electricity. After performing work, the low-pressure organic working fluid vapor enters the ORC condenser, where it is cooled into a liquid state by a cooling medium (such as circulating water). Then, it is pressurized by the working fluid pump and sent back to the evaporator to complete the entire organic Rankine cycle.

[0032] To achieve cascaded energy utilization, this embodiment further recovers the low-grade heat energy released by the ORC system during condensation. Specifically, the hot water (e.g., 40-60°C) from the outlet of the circulating water loop used to cool the ORC condenser is directly introduced into the evaporator of a heat pump system as the heat source for the third-stage waste heat recovery. In this evaporator, the low-grade heat energy is absorbed by the heat pump working fluid. Subsequently, the heat pump compressor compresses the working fluid that has absorbed the heat energy, significantly increasing its temperature and pressure. The high-temperature, high-pressure heat pump working fluid flows through the heat pump condenser, exchanging heat with the return water of the district heating system to produce high-temperature hot water at 60-80°C to meet production or domestic heating needs. The working fluid, having released heat, returns to the heat pump evaporator after being depressurized by a throttling valve, completing the heat pump cycle. After the above three stages of cascaded energy utilization, the temperature of the flue gas has dropped to a relatively low level and is finally discharged into the atmosphere through a chimney.

[0033] Next, this embodiment will describe in detail the specific implementation method of real-time acquisition of multidimensional data. The data acquisition and preprocessing are mainly divided into three parallel parts: multispectral visual data acquisition, system operation data acquisition, and external environment data acquisition.

[0034] S2.1 Multispectral Visual Data Acquisition and Preprocessing

[0035] To acquire detailed visual information on the physical and chemical composition of the plume from the chimney exhaust, a multispectral industrial vision acquisition device is deployed at a suitable distance from the chimney exhaust with a clear and unobstructed line of sight. This device is preferably a snapshot-type multispectral industrial camera equipped with a global shutter CMOS sensor, with a spatial resolution of at least 1280×1024 pixels and a frame rate of at least 15 frames per second. The camera's spectral response range covers the visible and near-infrared bands and is equipped with multiple narrowband filters to acquire image information at specific wavelengths.

[0036] In this embodiment, preferably, the number of spectral channels acquired is... At least four channels are included, specifically: a wide-band channel (preferably 550 nm) in the visible light range for capturing the geometry and opacity of the plume; a channel (preferably 940 nm) in the strong water vapor absorption band; a channel (preferably 1450 nm) in the carbon dioxide absorption band; and a reference channel (preferably 850 nm) in the atmospheric window region.

[0037] The acquired multispectral data stream can be represented as a sequence of image frames. At any discrete time point The acquired single-frame multispectral image is a data cube, represented as a three-dimensional tensor. :

[0038]

[0039] in, and These represent the height and width of the image frame (in pixels), respectively. Represents the number of multispectral channels. Any pixel in an image frame. At the point of time The original pixel value vector is represented as In its first The components of each spectral channel are .

[0040] To obtain physically meaningful spectral information and eliminate atmospheric interference, the acquired raw multispectral data needs to be preprocessed. This preprocessing includes radiometric calibration and atmospheric correction.

[0041] Unitless raw pixel values Convert to spectral radiance value This transformation is achieved through the following linear model:

[0042]

[0043] in, For pixels In the The spectral radiance value of the channel; and The first The channel's calibration gain and bias are two parameters that are precisely calibrated using a standard light source before the camera leaves the factory.

[0044] To reduce the impact of the atmospheric path between the camera and the plume on the radiative signal, atmospheric correction is performed. It is assumed that some pixels in the image (such as shadowed areas of buildings) have approximately zero surface reflectance, and the received signal for these pixels is entirely contributed by atmospheric path radiation. The minimum value in each channel image is identified as the atmospheric path radiation for that channel. And subtract it from the whole image:

[0045]

[0046] in, ; These are the atmospherically corrected spectral radiance values. After preprocessing, the corrected multispectral image frame is obtained. It contains information that better reflects the physicochemical properties of the plume itself.

[0047] S2.2 System Operation Data Acquisition and Preprocessing

[0048] System operating data is acquired in real time from the distributed control system (DCS) of the cogeneration plant via industrial communication protocols such as OPC UA or Modbus TCP / IP. Key operating parameters collected in this embodiment include: the flue gas inlet temperature of the ORC system evaporator. Flue gas outlet temperature of the ORC system evaporator Mass flow rate of the working fluid in the ORC system Inlet pressure of the ORC system turbine The outlet pressure of the ORC system turbine The rotational speed of the ORC system turbine The output power of the ORC system generator Input electrical power of the heat pump system compressor The supply water temperature of the district heating circuit in the heat pump system and return water temperature .

[0049] Considering the flue gas temperature originating from the incinerator Due to its inherent volatility and the susceptibility of sensor measurements to noise interference, this embodiment employs a Kalman filter to preprocess the raw measured values ​​of this parameter in order to obtain an accurate and smooth estimate of this key parameter.

[0050] Establish a linear state-space model for this temperature parameter. Define the actual temperature of the system as a state variable. ,exist The raw sensor measurements at any given time are defined as the observed variable. The state transition equation and the observation equation are defined as follows:

[0051]

[0052] in, Here is the state transition matrix. For the observation matrix, The process noise has a mean of 0 and a covariance of... Gaussian white noise, i.e. This represents the tiny random fluctuations in the actual temperature itself; The noise is measured, with a mean of 0 and a covariance of . Gaussian white noise, i.e. This represents the error introduced during sensor measurement. The recursive process of the Kalman filter includes two steps: prediction and update. Iterative calculations are performed using the following set of equations to finally obtain a smoother and more accurate estimate of the ORC inlet flue gas temperature, denoted as . .

[0053] Prediction steps:

[0054]

[0055]

[0056] Update steps:

[0057]

[0058]

[0059]

[0060] in, and They are respectively Time and The posterior estimate of the state at time 1; for The prior state estimate at time 1; and They are respectively Time and The posterior estimate of the covariance at time t; for Prior estimate of covariance at time; for Kalman gain at time step; It is an identity matrix.

[0061] S2.3. External Environment Data Acquisition

[0062] To eliminate the interference of environmental factors on the visual characteristics of smoke plumes, external environmental data was collected in real time. Key parameters collected included: ambient temperature. relative humidity of the environment and wind speed These data are collected synchronously with system operation data.

[0063] Finally, at each sampling time point The data acquisition module timestamps the three parts of data and integrates them into a multi-dimensional data vector. , The structure is as follows: .

[0064] Next, this embodiment will explain the specific implementation method of the plume visual dynamic feature extraction step in plume feature quantification and intelligent system state diagnosis.

[0065] The dynamic feature extraction model described in this embodiment mainly includes a three-branch deep feature extraction network, an adaptive reference feature fusion module, and a dynamic state representation module. The three input branches of the model are: stable reference frame, historical reference frame, and current analysis frame.

[0066] Specifically, from historical operational data, a time period is selected where both the combustion condition and the waste heat recovery system are confirmed to be in a highly efficient and stable operating state. From the multispectral image frame sequence within this time period, a stable reference frame is obtained by averaging or selecting the optimal frame. This frame represents the standard visual morphology of the plume under ideal operating conditions. During continuous system operation, for the current point in time requiring analysis... Take the corresponding current analysis frame. And take a preset time step. Previous historical reference frames These three image frames will serve as parallel inputs to the model.

[0067] S3.1. Shared deep spatiotemporal-spectral feature extraction, with three input image frames. , and They are each fed into a deep convolutional neural network feature extractor with shared weights. The extractor preferably uses an input layer modified to adapt to... The ResNet-50 architecture uses channel input. ResNet-50 maps high-dimensional image data to a low-dimensional, highly semantic feature space, resulting in three deep spatiotemporal-spectral feature maps:

[0068]

[0069]

[0070]

[0071] in, . and The height and width of the feature map, The number of channels for deep features. These represent plume depth features for the ideal stable state, recent historical state, and current instantaneous state, respectively.

[0072] S3.2. Adaptive Reference Feature Fusion

[0073] To construct an adaptive reference template that reflects ideal operating conditions while adapting to recent slow changes, this embodiment employs a spectral-temporal cross-attention module to process the stable reference feature map. and historical reference feature map To integrate.

[0074] For the two input feature maps and Channel attention weighting is performed independently.

[0075] With any feature map For example, channel descriptors are obtained through global average pooling. Then, the channel weight vector is calculated using one-dimensional convolution and the Sigmoid activation function. :

[0076]

[0077] in, It is the Sigmoid activation function. This is a one-dimensional convolution operation. The resulting weight vector is used to recalibrate the original feature map, yielding a weighted feature map. .

[0078] This process will be respectively and Generate weighted feature maps and .

[0079] The two weighted feature maps are added element-wise, and then processed through a... The convolutional layers are deeply fused to ultimately generate an adaptive reference feature map. :

[0080] in, This represents a convolutional kernel size of The convolution operation. It contains comprehensive information on both the ideal state and the recent state, forming a robust, dynamically updated reference benchmark.

[0081] S3.3. Dynamic State Representation and Eigenvector Generation

[0082] To quantify the deviation of the current plume state from this adaptive reference baseline, this embodiment calculates the current analysis feature map. With adaptive reference feature map The similarity between the two feature maps is calculated independently on each channel, and the results are summed to generate a two-dimensional response map. :

[0083] in, Represents cross-correlation operations. Response graph. The physical significance lies in the fact that the height of its peak and the sharpness of its shape directly reflect the similarity between the current plume state and the reference state. If the plume state is stable, then... It should present a sharp, prominent single peak; if the plume state changes drastically or becomes unstable, then... The peak value will decrease, the shape will become flattened, or multiple messy secondary peaks will appear.

[0084] To make the response diagram This is transformed into a low-dimensional feature vector suitable for quantitative analysis, from which a set of statistical indicators are extracted in this embodiment. Preferably, the extracted indicators include:

[0085] Peak-related response, which is the maximum value in the response graph. ;

[0086] Peak-side lobe ratio ( The ratio of peak energy to sidelobe energy is used to measure the significance of the peak.

[0087] ,in, and These are the pixel mean and standard deviation of the sidelobe regions excluding the central peak region in the response map, respectively.

[0088] The above statistical indicators are combined into a vector to form the current moment. Dynamic temporal feature vector :

[0089]

[0090] The vector The various components together constitute the effect of the plume at a given time point. A highly condensed quantitative description of dynamic stability can be directly used by subsequent intelligent system state diagnostic models. If and A higher value indicates a stable plume state; conversely, a lower value indicates an unstable plume state.

[0091] The dynamic time-series feature vector extraction method described in this invention is crucial for achieving intelligent diagnosis and predictive control of the entire system. In traditional waste heat recovery system control logic, the perception of system state relies entirely on conventional process parameter sensors such as temperature, pressure, and flow rate. While these sensors can reflect instantaneous physical quantities at specific nodes, they are essentially discrete point measurements with inherent measurement noise and response delays. They can answer the current state of the system but cannot effectively characterize the dynamic processes the system is undergoing or its future trends. Especially when upstream combustion conditions fluctuate due to changes in fuel characteristics, these fluctuations are transmitted downstream in the form of complex flow and thermal fields. Traditional sensors have very limited ability to perceive such process-oriented and systemic dynamic changes, causing the control system to only begin a passive response after the fluctuations have already had a substantial impact on the waste heat recovery unit (such as the ORC system). This not only sacrifices operational efficiency but also threatens the safety of core equipment such as turbines.

[0092] The dynamic temporal feature vector extraction method proposed in this embodiment fundamentally solves the above-mentioned problems. By introducing a three-branch deep network architecture consisting of a stable reference frame, a historical reference frame, and the current analysis frame, and processing it using spatiotemporal-spectral cross-attention and deep cross-correlation, the generated dynamic temporal feature vector is no longer an isolated physical quantity, but a highly condensed and quantified representation of the overall visual stability of the plume relative to an ideal and dynamically updated benchmark within a specific time window. The value of this feature vector directly reflects the stability of the upstream combustion process, becoming a sensitive non-contact indicator of system stability.

[0093] Next, in this embodiment, the dynamic temporal feature vector is extracted. Based on this, the subsequent steps of plume feature quantification and intelligent system state diagnosis are explained in detail. This process first extracts supplementary visual features.

[0094] S4.1 Geometric and Spectral Feature Extraction

[0095] To comprehensively characterize the visual state of the plume, in addition to dynamic features, this embodiment also uses preprocessed single-frame multispectral images. Geometric and spectral features are extracted from them.

[0096] Employ a pre-trained semantic segmentation network The input image frame is processed to obtain a binarized smoke plume region mask. In this mask, pixels with a value of 1 correspond to the plume area, and pixels with a value of 0 correspond to the background.

[0097] Based on this mask Calculate the following characteristics: plume area This is obtained by summing the pixel values ​​in the mask matrix, reflecting the size of the plume. Average visible light radiance. This involves calculating the average spectral radiance value within the masked plume region in a wide visible light band (preferably the 550nm channel), a value related to the plume's opacity and concentration. Water vapor absorption index. The ratio of the average radiance values ​​of the strong absorption band channel (940 nm) and the atmospheric window reference channel (850 nm) is used to characterize the relative water vapor content in the plume.

[0098]

[0099] in, Indicates at a point in time Within the plume area, the first The average spectral radiance value of the spectral channel.

[0100] The above features are combined into a geometric-spectral feature vector. .

[0101] S4.2 Multimodal Fusion and State Diagnosis Model

[0102] To achieve accurate diagnosis of system status, this embodiment constructs a diagnostic model based on a dual-stream gated recurrent unit and an attention fusion mechanism. This model can handle the time dependencies of visual feature sequences and system operating parameter sequences separately, and organically combine the information from both through an adaptive fusion mechanism.

[0103] Construct two parallel input sequences. The visual feature sequence consists of the geometric-spectral feature vector at the current time step. The dynamic time-series feature vector obtained To construct by splicing, that is The system's operational data sequence is composed of a vector consisting of the key operational parameters that have been collected and preprocessed. .

[0104] The model contains two parallel GRU networks, namely the visual flow GRU ( ) and running data flow GRU ( These are used to extract temporal context information from their respective sequences. At time points The outputs of the two GRU networks are their respective hidden state vectors. and :

[0105]

[0106] in, and These are the hidden states from the previous moment. It encodes the evolutionary history of the visual state of the smoke plume, and This encodes the evolution history of the system's internal operating parameters.

[0107] To achieve intelligent fusion of two modalities, this embodiment designs an attention fusion gate. This gate learns an attention weight vector. This allows for a dynamic decision on whether to rely more on visual information or operational data at any given moment.

[0108] The two hidden state vectors are concatenated and then passed through a multilayer perceptron. Calculate attention weights:

[0109]

[0110] in, The Sigmoid activation function is used to ensure that the output weights are in the interval (0, 1). The final fused feature vector... Weighted summation based on attention weights:

[0111]

[0112] in This indicates element-wise multiplication. When the dynamic characteristics of the plume indicate drastic fluctuations in operating conditions, the model may learn to assign... Higher weighting for faster response.

[0113] The fused feature vector The data is fed into two independent diagnostic head networks for direct efficiency diagnosis and indirect stability diagnosis, respectively.

[0114] Direct efficiency diagnosis, this diagnostic head It is a regression network whose objective is based on the current operating conditions (mainly derived from...). (Partial) and environmental parameters are used to estimate a theoretical ORC output power under healthy equipment conditions. By comparing this theoretical value with the actual output power collected... This yields an efficiency deviation index. :

[0115]

[0116] when When the value is significantly and consistently below 1, it can be diagnosed that the ORC system has experienced efficiency degradation due to factors such as scaling.

[0117] Indirect stability diagnosis, this diagnostic head It is a classification network whose goal is based on fused features (especially from...) The network uses dynamic information to determine the current overall stable state of the system. It outputs a probability distribution vector across multiple predefined state categories. Preferably, the state categories are defined as three types: {State 0: Stable operation, State 1: Moderate fluctuation, State 2: Severe instability}. The final stability diagnosis result... The category with the highest probability:

[0118] The diagnosis result and efficiency deviation index This is the final output of this embodiment.

[0119] The model designed in this embodiment, based on a dual-stream gated loop unit and an attention fusion mechanism, offers the advantage of capturing the inherent temporal evolution patterns of both visual feature sequences and system operating parameter sequences through a dual-stream parallel processing architecture, fully respecting the characteristics of different modalities. More importantly, the introduced attention fusion gate endows the model with the ability to dynamically weigh different information sources. For example, in scenarios where combustion conditions are stable but chronic fouling is suspected, the model can learn to rely more on the precise thermal parameters provided by the operating data stream when performing efficiency diagnosis; conversely, in scenarios where combustion conditions fluctuate drastically, the model can quickly increase the weight of the visual information stream to utilize its faster response to dynamic changes. This adaptive intelligent fusion mechanism makes the final diagnostic results more robust and accurate than judgments from any single information source, effectively avoiding misjudgments caused by single sensor failures or transient visual interference.

[0120] Furthermore, this embodiment explicitly decouples the diagnostic task into two parallel outputs: direct efficiency diagnosis and indirect stability diagnosis, which has significant practical application value. It enables the system to clearly distinguish between two completely different types of problems requiring entirely different response strategies: one is chronic, gradual performance degradation caused by equipment scaling; the other is acute, dynamic operational instability caused by combustion instability. This clear distinction allows it to generate maintenance alarms for chronic problems and initiate predictive control for acute problems, achieving precise fault location and accurate matching of response strategies, greatly improving the intelligence level of the entire system's operation and maintenance.

[0121] Next, based on the completion of system status diagnosis, this embodiment will provide a detailed description of the specific implementation method of closed-loop feedback and predictive control. This step is the final execution link for the intelligent closed-loop operation of this invention. Based on the diagnostic results, it performs two parallel tasks: generating diagnostic and alarm information and generating and executing predictive control commands.

[0122] Generate diagnostic and alarm information

[0123] The goal of generating diagnostic and alarm information is to monitor and trigger alarms for the chronic, slowly varying performance degradation of the waste heat recovery system (especially the ORC unit) based on the results of direct efficiency diagnostics. To avoid false alarms caused by instantaneous fluctuations, this embodiment employs a decision-making mechanism based on a sliding time window. A time window length is set. (For example, the number of data points in the past hour) and an efficiency decay alarm threshold. (For example, (This represents a 5% performance loss). At each time point Calculation in the past The efficiency deviation index obtained within each sampling period moving average :

[0124] The moving average is then compared to the alarm threshold to update an alarm flag. :

[0125] When the alarm flag is set When the state changes from 0 to 1, the control and output module will immediately generate a structured maintenance alarm on the human-machine interface. This alarm information preferably includes: alarm time, alarm type (e.g., "ORC system efficiency continues to decline"), and the current moving average efficiency index. And recommended checks (e.g., "Please check for scale on the heat exchange surfaces of the ORC evaporator").

[0126] Predictive control commands are generated and executed. The goal of generating and executing these commands is to rapidly and proactively switch the system's control mode based on the results of indirect stability diagnosis, in order to cope with the impact of upstream combustion condition fluctuations. This embodiment constructs a fuzzy logic controller to transform the discrete, qualitative stability states output by the upper-level diagnostic model into continuous, quantitative control adjustment commands that the lower-level actuators can receive.

[0127] The input variable of the fuzzy logic controller is the stability diagnosis result. Its value is {0: stable operation, 1: moderate fluctuation, 2: severe instability}.

[0128] Dynamic characteristic change rate That is, the obtained peak correlation response The change in adjacent time intervals, i.e. This variable reflects the trend of instability development.

[0129] The controller's fuzzification process transforms the precise input described above into a fuzzy language set. For example, for... Define a fuzzy set {stable, fluctuating, unstable}; for Define a fuzzy set {rapidly deteriorating, slowly deteriorating, basically unchanged, and improving}.

[0130] The controller's fuzzy rule base contains a set of "IF-THEN" rules based on expert knowledge, used to establish the mapping relationship between inputs and outputs. Output variables are the adjustments to the setpoints of key actuators, primarily including: the ORC turbine guide vane opening adjustment. ORC working fluid pump speed adjustment amount and heat pump compressor load adjustment amount These output variables are also fuzzified into fuzzy sets such as {significant reduction, slight reduction, unchanged, slight increase}. Example rules from the fuzzy rule base are as follows:

[0131] IF ( is unstable) AND ( (is rapidly deteriorating) THEN ( (is significantly reduced) AND ( (is slightly reduced);

[0132] IF ( is fluctuation) THEN ( (is slightly reduced);

[0133] IF ( is stable) THEN ( (is unchanged);

[0134] During the defuzzification phase, the fuzzy inference engine calculates the fuzzy output based on all activated rules, and then uses the centroid method defuzzification algorithm to obtain precise, numerical control adjustment amounts. , and .

[0135] The final control command generation and execution process is as follows: The system's basic control layer maintains a conventional controller aimed at maximizing efficiency. It outputs a set of standard settings at all times, such as The predictive control command of this invention is the adjustment amount output by the fuzzy logic controller. This is ultimately sent to the setpoint of the DCS or PLC. It is the sum of the standard setpoint and the predictive adjustment:

[0136]

[0137]

[0138]

[0139] In this way, when the diagnostic result is "stable," the fuzzy logic controller outputs zero adjustment, and the system operates in the normal high-efficiency mode. When the diagnostic result is "fluctuating" or "instable," the fuzzy logic controller outputs a negative adjustment, proactively and temporarily reducing the system load and switching the control objective from "maximum efficiency" to "maximum stability," thus smoothly weathering the impact of upstream operating conditions. When the diagnostic result returns to "stable," the adjustment automatically returns to zero, and the system seamlessly switches back to the normal control mode.

[0140] This specification also proposes a combined heat and power (CHP) waste heat utilization system, which includes:

[0141] Data Acquisition Module: Real-time acquisition of multispectral visual data of flue gas plumes from chimney outlets during cogeneration processes, operational data of waste heat recovery systems containing at least an organic Rankine cycle unit, and external environmental data to form multidimensional data;

[0142] Feature extraction module: Based on the multispectral visual data, extracts dynamic temporal feature vectors that characterize the dynamic stability of the plume in the time dimension, and geometric-spectral feature vectors that characterize the current geometric and spectral properties of the plume;

[0143] Early warning and control module: Based on the efficiency deviation index, it generates maintenance alarm information; and based on the stability diagnosis results, it generates predictive control commands.

[0144] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned method for integrated waste heat utilization in cogeneration.

[0145] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for integrated waste heat utilization in combined heat and power generation.

[0146] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0147] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0148] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for comprehensive waste heat utilization in combined heat and power generation, characterized in that, The method includes: Real-time acquisition of multispectral visual data of flue gas plumes from chimney outlets during cogeneration processes, operational data of waste heat recovery systems including at least organic Rankine cycle units, and external environmental data to form multidimensional data; Based on the multispectral visual data, a dynamic temporal feature vector characterizing the dynamic stability of the plume in the time dimension is extracted, as well as a geometric-spectral feature vector characterizing the current geometric and spectral properties of the plume. By fusing the dynamic time-series feature vector, the geometric-spectral feature vector, and the operational data, an efficiency deviation index and stability diagnosis results are obtained. Based on the efficiency deviation index, maintenance alarm information is generated; and based on the stability diagnosis results, predictive control commands are generated. Extracting dynamic temporal feature vectors includes: calculating the deep feature map of the current analysis frame through deep cross-correlation. With an adaptive reference feature map The similarity between the two feature maps is determined by independently calculating the cross-correlation on each channel of the two feature maps, and then summing the results to generate a response map. ; in, Indicate cross-correlation operation; and extract at least peak-correlated responses from the response graph. Compared with the side lobes of the peak The statistical indicators constitute the dynamic time-series feature vector; the calculation formula for the statistical indicators is: ； ； in, and These are the pixel mean and standard deviation of the sidelobe regions excluding the central peak region in the response image; the statistical indicators are synthesized into a vector to form the current time step. Dynamic temporal feature vector : like and A higher value indicates a stable plume state; conversely, a lower value indicates an unstable plume state. The efficiency deviation index is obtained by estimating the theoretical ORC output power based on the fused characteristics. And calculate the efficiency deviation index. : in, For the time point obtained from the said running data The actual output power of the ORC system generator.

2. The method for comprehensive waste heat utilization in cogeneration according to claim 1, characterized in that, After acquiring the multispectral visual data of the smoke plume in real time, the process further includes preprocessing the multispectral visual data to convert the original pixel values ​​into atmospherically corrected spectral radiance values. in, For the sampling time point, In the first spectral channels, pixels The original pixel value at that location, and The first The scaling gain and bias of the channel, This represents the spectral radiance value. For the first Atmospheric path radiation of the channel, This is the spectral radiance value after atmospheric correction.

3. The method for comprehensive waste heat utilization in cogeneration according to claim 2, characterized in that, The extraction of the geometric-spectral feature vector includes: performing semantic segmentation on the multispectral image frame to be processed to obtain a plume region mask, and calculating at least a water vapor absorption index based on the plume region mask. ; geometric-spectral eigenvectors; in, and At time points respectively The average spectral radiance values ​​of the atmospheric window reference channel and the water vapor strong absorption band channel within the plume region.

4. A method for comprehensive waste heat utilization from cogeneration according to claim 3, characterized in that, The obtained stability diagnosis result includes: outputting a probability distribution vector across multiple predefined stable state categories. The category with the highest probability is then determined as the stability diagnosis result. : 。 5. A method for comprehensive waste heat utilization from cogeneration according to claim 4, characterized in that, The step of generating predictive control instructions based on the stability diagnosis results includes: according to the stability diagnosis results... Generate control adjustment values ​​for at least one key actuator. The control adjustment amount is then compared with the conventional setpoint output by the conventional controller using the following formula. The parameters are superimposed to obtain the final operating parameters sent to the waste heat recovery system. : 。 6. A combined heat and power (CHP) waste heat utilization system, used to implement the CHP waste heat utilization method as described in claim 1, characterized in that, The system includes: Data Acquisition Module: Real-time acquisition of multispectral visual data of flue gas plumes from chimney outlets during cogeneration processes, operational data of waste heat recovery systems containing at least an organic Rankine cycle unit, and external environmental data to form multidimensional data; Feature extraction module: Based on the multispectral visual data, extracts dynamic temporal feature vectors characterizing the dynamic stability of the plume in the time dimension, and geometric-spectral feature vectors characterizing the current geometric and spectral properties of the plume; fuses the dynamic temporal feature vectors, the geometric-spectral feature vectors, and the running data to obtain the efficiency deviation index and stability diagnosis results; Early warning and control module: Based on the efficiency deviation index, it generates maintenance alarm information; and based on the stability diagnosis results, it generates predictive control commands.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements a combined heat and power waste heat utilization method as described in any one of claims 1-5.

8. A computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for integrated waste heat utilization in cogeneration as described in any one of claims 1-5.

Images