Search

CN-122023839-A - Method and system for predicting MSWI process controlled variables driven by multi-mode sharing characteristics

CN122023839ACN 122023839 ACN122023839 ACN 122023839ACN-122023839-A

Abstract

The invention provides a multi-mode shared characteristic driven MSWI process controlled variable prediction method which comprises the steps of extracting and selecting multi-physical characteristics of a flame image to obtain a multi-view characteristic set, carrying out correlation analysis and characteristic screening on the flame image and a controlled variable, combining a result with the multi-view characteristic set to obtain a multi-mode shared characteristic set, constructing a multi-controlled variable model, carrying out model training on the multi-controlled variable model through the multi-mode shared characteristic set to obtain an initial model, carrying out layered incremental learning on the initial model to obtain a final prediction model, and carrying out data prediction on the MSWI process through the final prediction model to obtain a prediction result. According to the method, the physical characteristics of the flame image are extracted in a quantized mode, the process data are fused, the shared characteristics are generated in a redundancy removing mode, the fuzzy width learning is combined, the layered incremental prediction of the model is achieved, the randomness defect of manual operation is overcome, and the prediction efficiency and the intelligent degree are improved.

Inventors

  • TANG JIAN
  • YANG XIAOXIAN
  • SUN MINGJUN
  • QIAN YU

Assignees

  • 北京工业大学

Dates

Publication Date
20260512
Application Date
20260211

Claims (9)

  1. 1. A method for predicting a controlled variable of a MSWI process driven by a multi-mode sharing characteristic is characterized by comprising the following steps: Extracting and selecting multiple physical features of the flame image to obtain a multi-view feature set; Performing correlation analysis and feature screening on the flame image and controlled variables, and sharing and combining a redundancy removal feature set obtained by screening with the multi-view feature set to obtain a multi-mode shared feature set, wherein the controlled variables comprise hearth temperature, boiler steam flow and flue gas oxygen content; Constructing a multi-controlled variable model, and performing model training on the multi-controlled variable model through the multi-mode shared feature set to obtain an initial model, wherein the multi-controlled variable model comprises a width input layer, an IT2FNN layer, a width enhancement layer and an initial width output layer; performing layered incremental learning on the initial model to obtain a final prediction model; And carrying out data prediction on the MSWI process through the final prediction model to obtain a prediction result.
  2. 2. The multi-modal sharing feature driven MSWI process controlled variable prediction method as set forth in claim 1, wherein the multi-physical feature extraction and selection is performed on the flame image to obtain a multi-view feature set, including: Performing image preprocessing operation on the flame image to obtain a preprocessed image; Performing physical feature extraction operation on the preprocessed image to obtain a multi-physical feature set, wherein the multi-physical feature set comprises brightness features, flame features, color features and texture features; And calculating the pearson correlation coefficient of the multi-physical feature set on the controlled variable, and carrying out feature selection on the multi-physical feature set according to the pearson correlation coefficient to obtain the multi-view feature set.
  3. 3. The multi-modal sharing feature driven MSWI process controlled variable prediction method of claim 2, wherein performing an image preprocessing operation on the flame image to obtain a preprocessed image comprises: Defogging normalization processing is carried out on the flame image, so that a clear image is obtained; Performing stripe denoising treatment on the cleaning image through a notch filter to obtain a denoised image; Removing salt and pepper noise and impulse noise in the denoising image by a median filtering method to obtain a filtered image; and collecting the clear image, the denoising image and the filtering image to obtain the preprocessing image.
  4. 4. The multi-modal sharing feature driven MSWI process controlled variable prediction method as claimed in claim 2, wherein performing a physical feature extraction operation on the preprocessed image to obtain a multi-physical feature set includes: converting the preprocessed image into a gray level image, calculating to obtain a brightness value according to the average gray level, gray level value variance and average brightness of the gray level image, and taking the brightness value as the brightness characteristic; extracting pixel points with gray values larger than a preset gray threshold value from the preprocessed image to obtain the flame characteristics; Dividing a window of the preprocessed image through a sliding window with a fixed scale, calculating a first moment, a second moment and a third moment of the window, and splicing calculation results into the color features; And calculating the average value of gray level co-occurrence matrixes of the preprocessed images in four directions of 0 degree, 45 degree, 90 degree and 135 degree, and taking the entropy of the average value as the texture characteristic.
  5. 5. The multi-modal shared feature driven MSWI process controlled variable prediction method of claim 2, wherein calculating pearson correlation coefficients of the multi-physical feature set for the controlled variable and performing feature selection on the multi-physical feature set according to the pearson correlation coefficients to obtain the multi-view feature set includes: Respectively calculating the pearson correlation coefficients of single characteristics in the multi-physical characteristic set and the hearth temperature, the boiler steam flow and the oxygen content of the flue gas; And if the pearson correlation coefficient is greater than or equal to a preset correlation threshold, marking the single feature as a preselected feature, and integrating all the preselected features into the multi-view feature set.
  6. 6. The multi-modal sharing feature driven MSWI process controlled variable prediction method of claim 1, wherein constructing a multi-controlled variable model and model training the multi-controlled variable model through the multi-modal sharing feature set to obtain an initial model includes: inputting the multi-mode shared feature set to the width input layer, and obtaining input layer output without transformation; The IT2FNN layer comprises a front piece network and a back piece network, wherein the front piece network comprises a membership function sub-layer, a fuzzy rule sub-layer, a degradation sub-layer and a feature mapping sub-layer; nonlinear change is carried out on the output of the rear part through the width enhancement layer, and enhanced node output is obtained; And performing ridge regression solving on the output of the enhanced node through the initial width output layer to obtain an initial predicted value and the initial model.
  7. 7. The method for predicting the controlled variable of the multi-modal sharing feature driven MSWI of claim 6, wherein the feature mapping the input layer output through the membership function and fuzzy rule built in the IT2FNN layer to obtain a back-piece output includes: calculating to obtain the membership degree of the input layer output through the Gaussian membership function built in the membership function sub-layer; performing fuzzy reasoning calculation on the membership degree through a fuzzy rule built in the fuzzy rule sub-layer to obtain activation strength; Weight reduction is carried out on the activation intensity through the reduction sub-layer, so that a characteristic lower bound and a characteristic upper bound are obtained; And weighting the feature lower bound and the feature upper bound through the feature mapping sublayer to obtain the back part output.
  8. 8. The multi-modal sharing feature driven MSWI process controlled variable prediction method of claim 1, wherein performing hierarchical incremental learning on the initial model to obtain a final prediction model includes: dividing the initial model into a plurality of learning layers based on feature nodes and enhancement nodes in the initial model; Calculating the mean square error increment of the output of the learning layer; If the mean square error increment is smaller than a preset increment threshold, the enhancement node is added to update the enhancement matrix of the learning layer; and repeatedly calculating the updated mean square error increment output by the learning layer until the mean square error increment is greater than or equal to the increment threshold value, so as to obtain the final prediction model.
  9. 9. A multi-modal shared feature driven MSWI process controlled variable prediction system, comprising: the image feature extraction module is used for extracting and selecting multiple physical features of the flame image to obtain a multiple-view feature set, wherein the multiple-view feature set comprises brightness features, flame features, color features and texture features; The characteristic sharing redundancy removing module is used for carrying out correlation analysis and characteristic screening on the flame image, the hearth temperature characteristic, the boiler steam flow characteristic and the smoke oxygen content characteristic respectively, and carrying out sharing combination on a redundancy removing characteristic set obtained by screening and the multi-view characteristic set to obtain a multi-mode sharing characteristic set; the controlled variable model construction module is used for constructing a plurality of controlled variable models, and carrying out model training on the controlled variable models through membership functions and fuzzy rules to obtain an initial model; The model optimization learning module is used for carrying out layered incremental learning on the initial model to obtain a final prediction model; and the data prediction module is used for carrying out data prediction on the MSWI process through the final prediction model to obtain a prediction result.

Description

Method and system for predicting MSWI process controlled variables driven by multi-mode sharing characteristics Technical Field The invention relates to the technical field of urban solid waste combustion, in particular to a method and a system for predicting a MSWI process controlled variable driven by a multi-mode sharing characteristic. Background With the acceleration of the urban solid waste (MSW) production, conventional landfill and composting methods are difficult to meet due to land resource shortage, environmental risks and technical limitations. MSW incineration (MSWI) is a core means for solid waste treatment in developing countries because of its advantages of reduction, harmlessness and recycling. However, MSWI processes involve multi-stage complex physicochemical reactions requiring precise control of key controlled variables such as Furnace Temperature (FT), boiler Steam Flow (BSF), and Flue Gas Oxygen Content (FGOC) to balance environmental requirements. Currently developed countries employ automatic combustion control systems that automatically adjust manipulated variables by combining controlled variable feedback with expert rules via PID controllers. However, in developing countries, because of low MSW classification level and insufficient equipment operation and maintenance capability, the ACC system cannot be directly applied, and instead, the method relies on manual operation of field experts. However, in the manual operation process, multimode data such as flame images, process variables, on-site inspection voice and the like are needed to be synthesized, and the trend of pollutants such as FT, BSF, FGOC, CO and the like is predicted empirically, so that the output values of operating variables such as materials, wind, water and the like are adjusted manually. However, subjective differences exist in interpretation of the same combustion state by different experts, so that inconsistent output of the manipulated variable is easy to cause, in addition, feature dimensions of physical features and process data in MSWI are high, irrelevant features and redundant features are required to be removed to keep important features, and meanwhile, the existing industrial means lack of a prediction model for fusing expert fuzzy cognitive mechanisms, so that standardized migration and optimization of experience knowledge cannot be realized. Therefore, it is necessary to design a method and a system for predicting the controlled variables of MSWI processes driven by a multi-modal sharing feature. Disclosure of Invention The invention aims to provide a method and a system for predicting a controlled variable of a MSWI process driven by multi-mode sharing characteristics, which are used for quantitatively extracting physical characteristics of flame images, fusing process data and removing redundancy to generate sharing characteristics, and realizing model layered incremental prediction by combining fuzzy width learning so as to solve the randomness defect of manual operation and improve the prediction efficiency and the intelligent degree. In order to achieve the above object, the present invention provides the following solutions: a method for predicting a controlled variable of a MSWI process driven by a multi-mode sharing characteristic comprises the following steps: Extracting and selecting multiple physical features of the flame image to obtain a multi-view feature set; carrying out correlation analysis and feature screening on the flame image and controlled variables, and sharing and combining a redundancy removal feature set obtained by screening with a multi-view feature set to obtain a multi-mode shared feature set, wherein the controlled variables comprise furnace temperature, boiler steam flow and flue gas oxygen content; Constructing a multi-controlled variable model, and carrying out model training on the multi-controlled variable model through a multi-mode shared feature set to obtain an initial model, wherein the multi-controlled variable model comprises a width input layer, an IT2FNN layer, a width enhancement layer and an initial width output layer; performing layered incremental learning on the initial model to obtain a final prediction model; and carrying out data prediction on the MSWI process through a final prediction model to obtain a prediction result. Optionally, extracting and selecting multiple physical features from the flame image to obtain a multiple view feature set, including: Performing image preprocessing operation on the flame image to obtain a preprocessed image; Performing physical feature extraction operation on the preprocessed image to obtain a multi-physical feature set, wherein the multi-physical feature set comprises brightness features, flame features, color features and texture features; And calculating the pearson correlation coefficient of the multiple physical feature sets on the controlled variable, and carrying out feature selection on the multiple ph