MSWI process-oriented working condition drift detection and quantitative analysis method

Abstract

The invention provides a working condition drift detection and quantitative analysis method for an MSWI process. The method relates to the technical field of the MSWI process, and comprises the following steps: acquiring a flame image and analyzing to obtain first local working condition drift data; obtaining temperature data in the furnace and analyzing the temperature data to obtain second local working condition drift data, wherein the second local working condition drift data comprises real local working condition drift data and virtual local working condition drift data; acquiring process data, and fusing the process data, the flame image and the temperature data in the furnace to obtain fused data; obtaining multi-mode working condition drift data according to the fusion data; and determining optimal working condition drift data according to the first local working condition drift data, the second local working condition drift data and the multi-mode working condition drift data. According to the method, the problem of how to realize global working condition drift detection and quantitative analysis in the combustion process to support subsequent control in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of MSWI process, and in particular to a method for detecting and quantitatively analyzing operating condition drift of MSWI process. Background Art

[0002] MSW incineration (MSWI) is an important part of the global urban renewable energy recycling process. At present, the operation of MSWI process mainly relies on domain experts to use their experience and knowledge to manually control the MSWI process to achieve stable control. Based on the actual operation of MSWI, it can be seen that the incinerator adopts an operation mode combining intermittent feeding and continuous pushing, which makes MSW have the nature of different levels of uncertainty changes, which in turn leads to operating condition drifts of different ranges and amplitudes during operation.

[0003] Obviously, although the process of the above experts perceiving the occurrence of drift and reasoning about the cause of drift cannot support the realization of intelligent control, its concept can be used as a reference. From the perspective of actual process parameter control, the furnace temperature field is an important characterization variable of the combustion process, but the limited thermocouple layout points on site make it difficult to reflect the changes in the furnace temperature field. For the entire combustion process, the flame and temperature field are both data of a certain stage of a single mode, and their respective drifts can only characterize local information and may be contradictory. At present, there is no unified conclusion in the industry and academia on how to characterize the global operating conditions of the combustion process. Therefore, how to learn from the brain's cognitive mechanism of parallel information to fuse multimodal data and realize the global condition drift detection and quantitative analysis of the combustion process to support subsequent control remains to be solved. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a method for operating condition drift detection and quantitative analysis of the MSWI process. The present invention solves the problem in the prior art of how to realize global operating condition drift detection and quantitative analysis of the combustion process to support subsequent control.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for detecting and quantifying the operating drift of the MSWI process, comprising:

[0007] Acquire a flame image and analyze it to obtain drift data of a first local working condition;

[0008] Acquire the temperature data in the furnace and analyze it to obtain the second local operating condition drift data, wherein the second local operating condition drift data includes: real local operating condition drift data and virtual local operating condition drift data;

[0009] Acquire process data and fuse the process data with the flame image and the temperature data in the furnace to obtain fused data;

[0010] Obtaining multi-modal working condition drift data according to the fused data;

[0011] The optimal operating condition drift data is determined according to the first local operating condition drift data, the second local operating condition drift data and the multi-modal operating condition drift data.

[0012] Preferably, the step of acquiring the flame image and analyzing it to obtain the first local operating condition drift data includes:

[0013] Construct a flame combustion condition information template library;

[0014] Extracting the flame image data frame by frame to obtain preset image data within a current time scale;

[0015] The preset image is input into the operating condition information template library for matching and analogy reasoning based on the twin network to obtain the first local operating condition drift data.

[0016] Preferably, the construction of the flame combustion condition information template library includes:

[0017] Obtain the flame combustion video collected on-site during the MSWI process, perform time-series sampling, and set the video sampling frequency to obtain a set of flame combustion condition images to be labeled;

[0018] Manually annotating the flame combustion condition image set to be annotated to obtain a typical flame combustion condition category image library;

[0019] Automatically calibrating an uncalibrated flame combustion condition image library according to the typical flame combustion condition category image library to obtain automatically calibrated image data;

[0020] A flame combustion condition information template library is constructed according to the automatic calibration image data and the typical flame combustion condition category image library.

[0021] Preferably, the working conditions of the flame combustion working condition image set to be annotated include:

[0022] Normal, overheating, partial overheating and stewing.

[0023] Preferably, the step of acquiring the furnace temperature data and analyzing it to obtain the second local operating condition drift data includes:

[0024] Different types of real temperature drift detection models are constructed using historical control variable data to form a parallel ensemble learner;

[0025] Inputting the furnace temperature data into the parallel integrated learner to obtain a first drift result;

[0026] Establish a learning model with control variable input and real temperature field output;

[0027] Acquire window samples according to the furnace temperature data;

[0028] Inputting the window sample into a learning model to obtain a second drift result;

[0029] Determine second local operating condition drift data according to the first drift result and the second drift result.

[0030] Preferably, the acquiring process data and fusing the process data with the flame image and the temperature data in the furnace to obtain fused data includes:

[0031] a time scale for acquiring the process data;

[0032] The flame image and the temperature data in the furnace are subjected to feature extraction and fusion according to the time scale to obtain fused data.

[0033] Preferably, determining the optimal operating condition drift data according to the first local operating condition drift data, the second local operating condition drift data and the multi-modal operating condition drift data includes:

[0034] Using Bayesian reasoning to fuse the first local operating condition drift data, the second local operating condition drift data, and the multimodal operating condition drift data to obtain fused drift data;

[0035] Performing interval credibility detection on the first local operating condition drift data, the second local operating condition drift data, and the multi-modal operating condition drift data to obtain a first detection result, a second detection result, and a third detection result;

[0036] Optimal operating condition drift data is determined according to the fused drift data and the first detection result, the second detection result and the third detection result.

[0037] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0038] The present invention provides a method for detecting and quantitatively analyzing operating condition drifts for MSWI processes, including: acquiring a flame image and analyzing it to obtain first local operating condition drift data; acquiring furnace temperature data and analyzing it to obtain second local operating condition drift data, wherein the second local operating condition drift data includes: real local operating condition drift data and virtual local operating condition drift data; acquiring process data and fusing the process data with the flame image and its furnace temperature data to obtain fused data; obtaining multimodal operating condition drift data according to the fused data; and determining optimal operating condition drift data according to the first local operating condition drift data, the second local operating condition drift data and the multimodal operating condition drift data. By integrating multiple data sources, the present invention can make full use of various existing monitoring technologies and data, improve resource utilization efficiency, help form a comprehensive monitoring system, and realize global operating condition drift detection and quantitative analysis of the combustion process to support subsequent control. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 A flow chart of a method for detecting and quantitatively analyzing operating condition drift for MSWI process provided by the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, the present invention provides a method for detecting and quantifying the operating condition drift of the MSWI process, including:

[0044] Step 100: Acquire a flame image and analyze it to obtain first local operating condition drift data;

[0045] Step 200: Acquire the temperature data in the furnace and analyze it to obtain the second local operating condition drift data, wherein the second local operating condition drift data includes: real local operating condition drift data and virtual local operating condition drift data;

[0046] Step 300: Acquire process data and fuse the process data with the flame image and the temperature data in the furnace to obtain fused data;

[0047] Specifically, the process data include: grate speed, primary air flow, secondary air flow, and CO concentration.

[0048] Step 400: Obtain multi-modal working condition drift data according to the fusion data;

[0049] Step 500: Determine optimal operating condition drift data according to the first local operating condition drift data, the second local operating condition drift data and the multi-modal operating condition drift data.

[0050] Specifically, the flame combustion images are qualitatively analyzed based on the expert experience of the operation engineers. The known and unknown classes are first manually labeled, and then the convolutional neural network and deep forest are used to build a combustion state classification model for automatic calibration. The causes of each class and the operational rules to be taken are summarized to form an information template library. The acquired combustion state information template library is used to detect known classes, and the adversarial generative network and mechanism knowledge are combined to detect unknown classes. The twin network is improved and expert knowledge is mined to perform drift detection and quantitative analysis to obtain the causes and operational rules. Based on the MSW solid-phase and gas-phase coupling strategy, a numerical simulation model for the furnace temperature field is studied. Based on the actual thermocouple sensor and control variable The furnace temperature field is perceived online, and the real temperature field drift is detected in the joint space of the joint control variables (input space) and the real temperature thermocouple detection values ​​(output space) based on the robust latent variable model and error information entropy. Furthermore, the semi-supervised drift detection and quantitative analysis of the virtual temperature field in the joint space are performed on the basis of the joint control variables (input space) and the virtual temperature pseudo-marker values. Multimodal data encoding space feature fusion and correlation analysis are performed for multimodal data of the same time scale. The drift detection and quantitative analysis results of the working conditions of process data, image data and multimodal data are integrated to perform drift detection and quantitative analysis of Bayesian reasoning fusion multiple models.

[0051] Furthermore, the flame image is acquired and analyzed to obtain the first local operating condition drift data, including:

[0052] Construct a flame combustion condition information template library;

[0053] Extracting the flame image data frame by frame to obtain preset image data within a current time scale;

[0054] The preset image is input into the operating condition information template library for matching and analogy reasoning based on the twin network to obtain the first local operating condition drift data.

[0055] Specifically, for flame combustion drift, operation engineers mainly rely on video recognition transmitted to the control room by observation cameras, supplemented by direct observation and recognition from artificial fire viewing holes. They conduct flame combustion drift detection and quantitative analysis based on machine vision, imitating the perspective of experts who use experience to match drift recognition.

[0056] Furthermore, the construction of the flame combustion condition information template library includes:

[0057] Obtain the flame combustion video collected on-site during the MSWI process, perform time-series sampling, and set the video sampling frequency to obtain a set of flame combustion condition images to be labeled;

[0058] Manually annotating the flame combustion condition image set to be annotated to obtain a typical flame combustion condition category image library;

[0059] Automatically calibrating an uncalibrated flame combustion condition image library according to the typical flame combustion condition category image library to obtain automatically calibrated image data;

[0060] A flame combustion condition information template library is constructed according to the automatic calibration image data and the typical flame combustion condition category image library.

[0061] Specifically, in the acquisition of combustion image data, the flame combustion video collected at the actual MSWI process site is sampled in time series, the combustion conditions are judged and the video sampling frequency is set based on the experience of domain experts, and a flame combustion condition image set for manual annotation is obtained; in the annotation of typical combustion condition images, the visual perception information of the flame combustion image based on the knowledge of domain experts and operation experience is divided into four typical combustion states: normal, cross-fire, partial burning and smoldering, and the causes of each type of condition and the operating rules to be taken are summarized, and then the obtained combustion condition image set is manually annotated to obtain a typical flame combustion condition category image library; in the automatic annotation of combustion image conditions, the obtained typical flame combustion condition category image library is used to train a combustion condition classification model integrating convolutional neural network and deep forest, and then the unlabeled flame combustion images are automatically calibrated to obtain a calibrated flame combustion condition image library; in the construction of the flame combustion condition category information template library, the typical combustion condition image data and the automatically calibrated image data are merged, and the complete flame combustion condition information template library is obtained in combination with the summarized operating rules corresponding to each type of combustion condition.

[0062] More specifically, the flame combustion drift detection and quantitative analysis framework, aimed at the needs of flame combustion drift detection and quantitative analysis in the furnace, takes real-time flame combustion video as input, and designs a framework consisting of four parts: flame combustion image data acquisition, flame image analogy reasoning and combustion drift quantitative analysis, adversarial generation of new working condition flame image data and new working condition flame image quantitative analysis; flame combustion image data acquisition, studying the time-triggered sampling mechanism to obtain the flame combustion video data within the current time scale, extracting the flame combustion video frame by frame to obtain the image data within the current time scale, and performing multi-frame fusion based on the operation process time scale to obtain image data; flame image analogy reasoning and combustion drift quantitative analysis, inputting the obtained image data into the flame combustion working condition information template library for matching and analogy reasoning based on the twin network, if the current flame If the flame image belongs to a known category, the corresponding category, formation cause and operation rules are output; on the contrary, the current flame combustion image represents an unknown combustion condition, that is, a new category of flame combustion drift has occurred, then go to the next stage; generate flame image data for new conditions adversarially. When a new category of condition drift occurs in flame combustion, an adversarial generation strategy is used based on mechanism knowledge to generate flame combustion image data for unknown conditions; quantitative analysis of flame images for new conditions is performed. The similarity between the flame combustion image of unknown conditions and the flame combustion image of known conditions is measured based on the twin network and expert knowledge. The unknown flame combustion condition is defined according to the similarity with the known category conditions, and the probability value belonging to the known category, the possible cause of the new category drift and the operation rules are given. The flame combustion image of the new category condition is stored in the flame combustion condition information template library, and it is used as a new category after confirmation by experts.

[0063] Furthermore, the acquiring of the furnace temperature data and analyzing to obtain the second local operating condition drift data includes:

[0064] Different types of real temperature drift detection models are constructed using historical control variable data to form a parallel ensemble learner;

[0065] Inputting the furnace temperature data into the parallel integrated learner to obtain a first drift result;

[0066] Establish a learning model with control variable input and real temperature field output;

[0067] Specifically, the learning models include: principal component analysis (PCA) model, robust principal component analysis (RPCA) model, and kernel principal component analysis (KPCA) model.

[0068] Acquire window samples according to the furnace temperature data;

[0069] Inputting the window sample into a learning model to obtain a second drift result;

[0070] Determine second local operating condition drift data according to the first drift result and the second drift result.

[0071] Specifically, the furnace temperature field is an important process parameter of the combustion process, but the limited thermocouple layout points on site make it difficult to perceive changes in the temperature field. It is necessary to combine the input and output space to conduct semi-supervised drift detection and quantitative analysis of the furnace temperature field.

[0072] Furthermore, the acquiring process data and fusing the process data with the flame image and the temperature data in the furnace to obtain fused data includes:

[0073] a time scale for acquiring the process data;

[0074] The flame image and the temperature data in the furnace are subjected to feature extraction and fusion according to the time scale to obtain fused data.

[0075] Specifically, the key temperature points of the furnace temperature field are designed. According to the grate furnace process design, the combustion of MSW on the grate can be divided into three stages and four parts (drying stage, combustion stage 1, combustion stage 2 and combustion stage) of drying, combustion and burning out. The furnace temperature field is decomposed into virtual temperature values ​​and actual temperature values. The numerical simulation model of the furnace temperature field is constructed. The temperature field distribution in the incinerator is formed by the thermal radiation generated by the combustion of MSW (including solid phase and gas phase). The strategy of coupling solid phase and gas phase is used to construct the numerical model of the furnace temperature field. The numerical simulation experiment design of the furnace temperature field is based on the obtained The obtained MSW calorific value, industrial analysis and elemental analysis data are combined with expert knowledge to design numerical simulation experiments for the ideal furnace temperature field. Then, the actual thermocouple temperature data is used to correct the experimental results, and finally multiple sets of furnace temperature field simulation data that can be used for identification are obtained. A mapping model between control variables and real thermocouple temperatures and virtual temperature points is constructed, and a mapping network structure with temperature transfer characteristics is designed. Then, based on the obtained simulation data, the mapping relationship between thermocouple temperatures and temperature points is identified one by one to obtain a temperature field mapping model, which provides support for drift detection and control of the furnace temperature field.

[0076] Furthermore, based on the historical control variable data that can be collected, several different types of real temperature field drift detection are constructed, and then the input space working condition drift detection model is obtained through a parallel integrated learner method. At the same time, a learning model with control variables as input and real temperature field as output is constructed as an output space working condition drift detection model; adaptive time window, according to the operating characteristics of the MSWI process, an adaptive sample acquisition window for the operating process control variable data is studied to realize the adaptive triggering of the real temperature field drift detection; current window sample working condition drift evaluation, based on the statistical indicators of the input space to evaluate the drift of the current window sample, and at the same time, the error entropy of the output space is used to evaluate the drift of the actual real temperature, the above two drift results are combined to calculate the current working condition drift degree, and the drift sample under the current time window is obtained; temperature drift quantitative analysis, based on the manual operation experience of engineers in the actual MSWI process field, the drift risk level is divided according to the statistical indicator characteristics of the process data, the main control variables that cause the drift are reconstructed, and the corresponding operation rules are obtained based on expert experience; statistical indicator update, based on the obtained working condition drift samples, the statistical indicators of the detection model are updated.

[0077] More specifically, the input space drift detection model is constructed in the same way as the previous one, and at the same time, a model with a virtual temperature field as the output is constructed as the working condition drift detection model of the output space; an adaptive time window is used to adaptively obtain the input space samples of the virtual temperature field drift detection; the pseudo-true value of the virtual temperature field of the furnace is obtained, and the real-time real thermocouple temperature and control variables are used as inputs, and the temperature field mapping model is used to obtain 5 furnace temperature values ​​(pseudo-true values) in real time; the input space drift detection of the virtual temperature field of the furnace is carried out, and the control variables are input into the input space drift detection model to obtain the drift detection results of the input space of the virtual temperature field of the furnace; the current window Evaluation of sample operating condition drift: Based on statistical indicators, the operating condition drift detection results of the virtual temperature field in the current window are evaluated. At the same time, the time difference algorithm and hypothesis test are combined to perform output space detection. The drift degree of the current operating condition is calculated by combining the above two drift results to obtain drift samples. Quantitative analysis of virtual temperature field drift: Based on the operating experience of field engineers, the drift risk level is divided according to the comprehensive statistical indicator characteristics of the virtual temperature field, the main control variables that cause the drift are found, and corresponding operating rules are given based on experience. Statistical indicator update: The statistical indicators of the virtual furnace temperature field drift detection model are updated based on the obtained pseudo-true value drift samples.

[0078] Further, determining the optimal operating condition drift data according to the first local operating condition drift data, the second local operating condition drift data and the multi-modal operating condition drift data includes:

[0079] Using Bayesian reasoning to fuse the first local operating condition drift data, the second local operating condition drift data, and the multimodal operating condition drift data to obtain fused drift data;

[0080] Performing interval credibility detection on the first local operating condition drift data, the second local operating condition drift data, and the multi-modal operating condition drift data to obtain a first detection result, a second detection result, and a third detection result;

[0081] Optimal operating condition drift data is determined according to the fused drift data and the first detection result, the second detection result and the third detection result.

[0082] Specifically, the real-time operation data of the MSWI process includes two modes: high-dimensional process data and flame combustion data. The drift occurring in each mode can only characterize local information. Therefore, global drift detection and quantitative analysis of operating conditions that comprehensively consider multiple local information are carried out.

[0083] More specifically, the time-triggered sampling mechanism of the operating condition drift detection data. The combustion reaction in the furnace is complex, and it is necessary to analyze the combustion mechanism and process characteristics, study the adaptive data sampling mechanism based on time triggering, and obtain multimodal data running in real time; image data spatiotemporal depth feature extraction, combine the selective attention mechanism and expert cognitive experience to select Transform visual features, extract multi-level temporal and spatial features for the construction of a robust latent variable detection model; multimodal data time scale matching, extract and fuse the flame video stream data based on the process data time scale, and obtain multimodal data after the current time scale is aligned; multimodal deep coding feature fusion and selection, based on the obtained multimodal data, study image and Process multimodal data has multiple types of feature preprocessing mechanisms such as parallel, non-parallel and hybrid. The data features of the two are further deeply encoded, fused and selected to obtain multimodal deep features of the same dimension; correlation and contradiction analysis between multimodal data, based on the process mechanism and the time scale of the data, the correlation between multimodal data is analyzed to explore the potential correlation between the two data, and the contradictions existing in the multimodal data characterization of working conditions are studied to determine the most critical information and sub-critical information in the contradiction, providing an analytical basis for drift quantification; correlation analysis between image features and multimodal features and control variables requires studying and analyzing the correlation between feature quantities and control quantities, providing support for the cause analysis of drift detection results and the extraction of operating rules.

[0084] Furthermore, the multimodal data working condition drift detection and quantitative analysis strategy includes three parts: multimodal data acquisition, multimodal working condition drift detection and quantification, and multimodal working condition drift detection model update; multimodal data acquisition, process data, image data and multimodal fusion data involve two time scales, among which process data and image data determine the time scale to obtain operation data according to their respective characteristics, and multimodal fusion data obtains fusion feature data according to the process data time scale; multimodal working condition drift detection and quantification, based on the above-obtained process, image and multimodal fusion features, are respectively input into the robust latent variable drift detection constructed based on historical process data, image data and multimodal data. The model is used to obtain three types of drift detection results, including two single modes and one multimodal. The drift detection quantitative analysis of Bayesian reasoning fusion of multiple models is based on prior and a posteriori knowledge. The three detection results are fused by Bayesian reasoning. The interval credibility of the detection results is considered at the same time. The three quantitative results are tested for confidence and the operation rules are determined to provide support for subsequent control. The drift detection model of multimodal working conditions is updated. Based on the obtained working condition drift detection model update mechanism, a drift sample library for multimodal data mixing is constructed. The multimodal data representing the new working condition is obtained in real time according to the drift evaluation results. Then, the drift detection model and the prior and a posteriori probabilities are updated based on the drift sample library.

[0085] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0086] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for detecting and quantifying operating drift of MSWI process, characterized in that: include: Acquire a flame image and analyze it to obtain drift data of a first local working condition; Acquire the temperature data in the furnace and analyze it to obtain the drift data of the second local working condition; Acquire process data and fuse the process data with the flame image and the temperature data in the furnace to obtain fused data; Obtaining multi-modal working condition drift data according to the fused data; The optimal operating condition drift data is determined according to the first local operating condition drift data, the second local operating condition drift data and the multi-modal operating condition drift data.

2. The method for detecting and quantifying the operating drift of the MSWI process according to claim 1 is characterized in that: The step of acquiring the flame image and analyzing it to obtain the first local operating condition drift data includes: Construct a flame combustion condition information template library; Extracting the flame image data frame by frame to obtain preset image data within a current time scale; The preset image is input into the operating condition information template library for matching and analogy reasoning based on the twin network to obtain the first local operating condition drift data.

3. The method for detecting and quantifying the operating condition drift of the MSWI process according to claim 1 is characterized in that: The flame combustion condition information template library is constructed, including: Obtain the flame combustion video collected on-site during the MSWI process, perform time-series sampling, and set the video sampling frequency to obtain a set of flame combustion condition images to be labeled; Manually annotating the flame combustion condition image set to be annotated to obtain a typical flame combustion condition category image library; Automatically calibrating an uncalibrated flame combustion condition image library according to the typical flame combustion condition category image library to obtain automatically calibrated image data; A flame combustion condition information template library is constructed according to the automatic calibration image data and the typical flame combustion condition category image library.

4. The method for detecting and quantifying the operating condition drift of the MSWI process according to claim 1 is characterized in that: The working conditions of the flame combustion working condition image set to be labeled include: Normal, overheating, partial overheating and stewing.

5. The method for detecting and quantifying the operating condition drift of the MSWI process according to claim 1 is characterized in that: The step of acquiring the furnace temperature data and analyzing the second local operating condition drift data includes: Different types of real temperature drift detection models are constructed using historical control variable data to form a parallel ensemble learner; Inputting the furnace temperature data into the parallel integrated learner to obtain a first drift result; Establish a learning model with control variable input and real temperature field output; Acquire window samples according to the furnace temperature data; Inputting the window sample into a learning model to obtain a second drift result; Determine second local operating condition drift data according to the first drift result and the second drift result.

6. The method for detecting and quantifying the operating condition drift of the MSWI process according to claim 1, characterized in that: The acquiring process data and fusing the process data with the flame image and the temperature data in the furnace to obtain fused data includes: a time scale for acquiring the process data; The flame image and the temperature data in the furnace are subjected to feature extraction and fusion according to the time scale to obtain fused data.

7. The method for detecting and quantifying the operating condition drift of the MSWI process according to claim 1, characterized in that: The determining the optimal operating condition drift data according to the first local operating condition drift data, the second local operating condition drift data and the multi-modal operating condition drift data includes: Using Bayesian reasoning to fuse the first local operating condition drift data, the second local operating condition drift data, and the multimodal operating condition drift data to obtain fused drift data; Performing interval credibility detection on the first local operating condition drift data, the second local operating condition drift data, and the multi-modal operating condition drift data to obtain a first detection result, a second detection result, and a third detection result; Optimal operating condition drift data is determined according to the fused drift data and the first detection result, the second detection result and the third detection result.

Images