MSWI process characterization method

Abstract

The invention provides an MSWI process characterization method, and relates to the technical field of urban solid waste. Comprising the steps of obtaining to-be-represented MSWI process data; constructing an incineration characterization model, and performing self-updating on the incineration characterization model to obtain an updated incineration characterization model; inputting to-be-represented MSWI process data into the updated incineration representation model to obtain a representation result; the self-updating method of the incineration characterization model comprises the following steps: acquiring historical modeling data and real-time model acquisition data to obtain a correlation analysis result; according to the correlation analysis result and the process data of the DCS system, updating parameters of the incineration representation model to obtain a parameter updating model; and structure self-organization updating is conducted on the parameter updating model according to the distribution difference of the historical scene data and the current scene data and the correlation analysis result, and an updated incineration characterization model is obtained. The invention aims to provide an MSWI process characterization method, and solves the problem of low accuracy of incineration process characterization in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of municipal solid waste, and in particular to a MSWI process characterization method. Background Art

[0002] Municipal solid waste (MSW) continues to increase at an annual rate of 8% to 10%, leading to an increasing number of cities in my country experiencing the "garbage siege" phenomenon. Recycling MSW is a key issue in developing environmentally friendly cities. MSW incineration (MSWI) is a crucial component of renewable energy recycling in cities worldwide and has become an effective solution to the "garbage siege" phenomenon.

[0003] Research has shown that intelligent control of complex industrial processes requires understanding the operational characteristics of the controlled system. Therefore, building an accurate system characterization model for the MSWI process is fundamental. Because the actual operation of the MSWI process involves high-temperature, high-pressure equipment, and toxic and hazardous gases, constructing a physical model in the laboratory is difficult. Therefore, building data-driven incineration characterization models has become a current research hotspot.

[0004] The MSWI process is a complex, multi-stage dynamic system involving numerous mechanical devices. It employs a combination of intermittent feeding and continuous pushing. In addition to the strong interference inherent in the material composition due to factors such as region, season, and holidays, there are also variations in material composition within and between hoppers during feeding. Factors such as equipment wear and frequent manual intervention further complicate operating condition drift. The multi-level and multi-amplitude complex dynamic drift exhibited by the MSWI process makes it difficult to accurately characterize the incineration using offline incineration characterization models. Research is needed to trigger the self-organizing update of the process characterization model based on real-time data to adapt to the dynamic changes in the incineration process. Summary of the Invention

[0005] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a MSWI process characterization method, which solves the problem of low accuracy of incineration process characterization in the prior art.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A MSWI process characterization method, comprising:

[0008] Obtaining the MSWI process data to be characterized;

[0009] constructing an incineration characterization model and self-updating the incineration characterization model to obtain an updated incineration characterization model;

[0010] Inputting the MSWI process data to be characterized into the updated incineration characterization model to obtain a characterization result;

[0011] The self-updating method of the incineration characterization model is:

[0012] Obtain historical modeling data and real-time model acquisition data;

[0013] Performing correlation analysis on the historical modeling data and the real-time model collection data to obtain a correlation analysis result;

[0014] updating the parameters of the incineration characterization model according to the correlation analysis results and the process data of the DCS system to obtain a parameter update model;

[0015] The parameter update model is structurally self-organized and updated according to the distribution differences of historical scene data and current scene data and the correlation analysis results to obtain an updated incineration characterization model.

[0016] Preferably, the performing correlation analysis on the historical modeling data and the real-time model collection data to obtain a correlation analysis result includes:

[0017] Processing the historical modeling data to obtain a non-time-series typical sample library and a time-series historical sample subset with different window lengths;

[0018] Determining the sliding window size of the real-time model to collect data;

[0019] Perform correlation analysis based on the sliding window size of the real-time model collected data and the non-time-series typical sample library and the time-series historical sample subset corresponding to the sliding window size of the model collected data, and calculate the Pearson correlation coefficient, covariance and mutual information;

[0020] The correlation analysis results are determined based on the Pearson correlation coefficient, covariance and mutual information.

[0021] Preferably, the calculation expression of the Pearson correlation coefficient is:

[0022]

[0023] Preferably, the calculation expression of the covariance is:

[0024]

[0025] Preferably, the calculation expression of the mutual information is:

[0026]

[0027] Preferably, the parameters of the incineration characterization model are updated according to the correlation analysis results and the process data of the DCS system to obtain a parameter update model, including:

[0028] Determining expert and data feedback knowledge related to current operating conditions based on process data of the DCS system;

[0029] Set the model parameter update threshold set;

[0030] Determining a parameter update instruction based on the model parameter update threshold set and the correlation analysis result;

[0031] The parameters of the incineration characterization model are updated according to the parameter update instruction, expert and data feedback knowledge, and the inherent logic of the pollution process to obtain a parameter update model.

[0032] Preferably, the parameter update model is updated by structural self-organization based on the distribution difference of the historical scene data and the current scene data and the correlation analysis result to obtain an updated incineration characterization model, including:

[0033] Set the model structure update threshold set;

[0034] Determining a structure update instruction according to the structure update threshold set;

[0035] Determining an evaluation index based on the distribution difference between the historical scene data and the current scene data and the correlation analysis result;

[0036] The parameter update model is subjected to structural self-organization updating according to the evaluation index and the structural update instruction to obtain an updated incineration characterization model.

[0037] Preferably, the calculation expression of the evaluation index is:

[0038]

[0039] Among them, φ(·) is the source data X S and target data X T Mapping function to feature space

[0040] The present invention discloses the following technical effects:

[0041] The present invention provides an MSWI process characterization method, comprising: obtaining MSWI process data to be characterized; constructing an incineration characterization model and self-updating the incineration characterization model to obtain an updated incineration characterization model; inputting the MSWI process data to be characterized into the updated incineration characterization model to obtain a characterization result; the self-updating method of the incineration characterization model is: obtaining historical modeling data and real-time model acquisition data; performing correlation analysis on the historical modeling data and the real-time model acquisition data to obtain a correlation analysis result; updating the parameters of the incineration characterization model according to the correlation analysis result and process data of a DCS system to obtain a parameter update model; performing structural self-organization updating on the parameter update model according to the distribution difference between historical scenario data and current scenario data and the correlation analysis result to obtain an updated incineration characterization model. The present invention considers various dynamic factors in the incineration process in multiple dimensions, and can effectively improve the flexibility and accuracy of the model through the analysis of real-time data, the integration of expert knowledge, and the transfer of knowledge in multiple scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. 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 any creative work.

[0043] Figure 1 A flow chart of a MSWI process characterization method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0045] 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.

[0046] Figure 1 An embodiment of the present invention provides a method for characterizing an MSWI process, including:

[0047] Step 100: Acquire MSWI process data to be characterized;

[0048] Step 200: Constructing an incineration characterization model f using a long short-term memory neural network based on historical data old (·), as shown below,

[0049]

[0050] in, and denote the predicted values ​​of nitrogen oxides, sulfur dioxide, hydrogen chloride, carbon monoxide, and carbon dioxide, respectively; and Respectively represent the predicted values ​​of furnace temperature, steam flow and flue gas oxygen content; u M u represents the input vector composed of the manipulated variables such as feed rate, grate speed, primary air volume, secondary air volume, boiler feed water volume, etc. MC represents the input vector consisting of the manipulated variables of urea, slaked lime, and activated carbon used for flue gas purification,

[0051] The combustion characterization model is self-updated to obtain an updated combustion characterization model f update (·), as shown below,

[0052]

[0053] in, and represent the predicted values ​​of nitrogen oxides, sulfur dioxide, hydrogen chloride, carbon monoxide and carbon dioxide after the updated model respectively; and They represent the predicted values ​​of furnace temperature, steam flow and flue gas oxygen content after the updated model; u New M u represents the input vector consisting of updated manipulated variables such as feed rate, grate speed, primary air volume, secondary air volume, boiler feed water volume, etc.; New MC represents the input vector consisting of the updated manipulated variables of urea, slaked lime, and activated carbon for flue gas purification,

[0054] Step 300: inputting the MSWI process data to be characterized into the updated incineration characterization model to obtain a characterization result;

[0055] The self-updating method of the incineration characterization model is:

[0056] Step 201: Acquire historical modeling data and real-time model acquisition data;

[0057] Step 202: performing correlation analysis on the historical modeling data and the real-time model collection data to obtain a correlation analysis result;

[0058] Step 203: updating the parameters of the incineration characterization model according to the correlation analysis result and the process data of the DCS system to obtain a parameter update model;

[0059] Step 204: performing structural self-organization updating on the parameter update model according to the distribution differences of the historical scene data and the current scene data and the correlation analysis result to obtain an updated incineration characterization model.

[0060] Specifically, a correlation analysis based on sliding multi-windows is used to analyze the real-time collected data and historical modeling data to support the parameter and structure updates of the subsequent incineration characterization model.

[0061] Based on the correlation analysis results of the above-mentioned real-time collected data and historical modeling data, a model parameter replacement learning mechanism driven by multiple dimensions such as real-time data, feedback knowledge and internal logic is used to update the parameters of the incineration characterization model.

[0062] Based on the correlation analysis results between real-time collected data and historical modeling data and the reuse of model knowledge in multiple historical scenarios, the structure of the incineration characterization model is dynamically self-organized to achieve dynamic self-organization of incineration characterization modeling triggered by real-time data.

[0063] Furthermore, the correlation analysis of the historical modeling data and the real-time model collection data to obtain a correlation analysis result includes:

[0064] Processing the historical modeling data to obtain a non-time-series typical sample library and a time-series historical sample subset with different window lengths;

[0065] Determining the sliding window size of the real-time model to collect data;

[0066] Perform correlation analysis based on the sliding window size of the real-time model collected data and the non-time-series typical sample library and the time-series historical sample subset corresponding to the sliding window size of the model collected data, and calculate the Pearson correlation coefficient, covariance and mutual information;

[0067] The correlation analysis results are determined based on the Pearson correlation coefficient, covariance and mutual information.

[0068] Specifically, first, by processing historical data to obtain a non-time-series typical sample library including multiple types of operating condition samples and a time-series historical sample subset corresponding only to a certain typical operating condition, sequential sampling processing is performed in combination with expert experience to obtain a non-time-series typical sample library and a time-series historical sample subset with different window lengths;

[0069] Secondly, the real-time collected data is processed by setting a variable-size sliding window (the value range is the same as above), that is, it is converted into a real-time sample subset that meets a certain length in the historical data set. The sliding window size is determined based on the data envelopment trend;

[0070] Then, we conduct correlation analysis on the non-time-series typical sample library and the time-series historical sample subset that match the size of the real-time sample subset, and calculate the Pearson correlation coefficient (PCC), covariance, and mutual information respectively to explore the correlation between the real-time samples and the historical samples under different window sizes.

[0071] Specifically, the calculation expression of the Pearson correlation coefficient ρ(X, Y) is:

[0072]

[0073] X and Y are the real-time sample and historical sample sets used to calculate the correlation, respectively. t and y t are the tth real-time sample and the tth historical sample respectively, and n represents the number of samples contained in the real-time sample and the historical sample set used to calculate the correlation. and Represents the mean of the real-time sample and the historical sample set used to calculate the correlation.

[0074] Specifically, the calculation expression of the covariance Cov(X,Y) is:

[0075]

[0076] Specifically, the calculation expression of the mutual information I(X, Y) is:

[0077]

[0078] Among them, p(x i ,y i ) represents the joint probability density of real-time samples and historical samples, p(x i ) and p(y i ) represents the independent probability density of real-time samples and historical samples.

[0079] More specifically, PCC measures the linear correlation between data, covariance measures the overall trend of change between data, and mutual information measures the degree of interdependence between data;

[0080] Finally, based on the above three metrics, the values ​​of real-time data and the non-time-series typical sample library are used as the overall correlation value to determine whether there is an offset between real-time data and historical data; the values ​​of real-time data and the time-series historical sample subset are used as the typical operating condition correlation value to determine whether to perform parameter updates or structural updates for the static heterogeneous integration model.

[0081] Furthermore, the parameters of the incineration characterization model are updated according to the correlation analysis results and the process data of the DCS system to obtain a parameter update model, including:

[0082] Determining expert and data feedback knowledge related to current operating conditions based on process data of the DCS system;

[0083] Set the model parameter update threshold set;

[0084] Determining a parameter update instruction based on the model parameter update threshold set and the correlation analysis result;

[0085] The parameters of the incineration characterization model are updated according to the parameter update instruction, expert and data feedback knowledge, and the inherent logic of the pollution process to obtain a parameter update model.

[0086] Specifically, first, based on domain experts from on-site operations and inspections and process data from the DCS system, expert and data feedback knowledge related to the current operating conditions is obtained to support the setting of relevant weight coefficients for the model parameter replacement learning mechanism;

[0087] Then, based on the process data of the DCS system and the output of the full-process heterogeneous integrated characterization model, combined with real-time error information and the real-time correlation analysis results of typical operating conditions obtained above, an update trigger mechanism is adopted as follows:

[0088]

[0089] Among them, ρ, E entropy , Cov and I represent PCC value, error information entropy, covariance and mutual information value respectively; ε1, ε2, ε3, ε4 are the thresholds set for judging model parameter update; θ1, θ2, θ3, θ4 are the trigger conditions for judging parameter update, and the model parameters are updated when any value is 1;

[0090] Finally, combining the knowledge of experts and data feedback with the inherent logic of the pollution process (process mechanism knowledge), a model parameter replacement learning mechanism based on asynchronous updates is implemented. Its regularized loss function is as follows:

[0091]

[0092] Among them, J2 represents the total loss function, J1 represents the loss function with the L2 regularization constraint term added to reduce the model complexity, λ1 and λ2 represent the weighted coefficients related to the expert and data feedback knowledge, and w i is the model parameter weight, L logic is the internal logic constraint term, C(·) is the cost function

[0093] Furthermore, the parameter update model is updated by structural self-organization based on the distribution differences of the historical scene data and the current scene data and the correlation analysis results to obtain an updated incineration characterization model, including:

[0094] Set the model structure update threshold set;

[0095] Determining a structure update instruction according to the structure update threshold set;

[0096] Determining an evaluation index based on the distribution difference between the historical scene data and the current scene data and the correlation analysis result;

[0097] The parameter update model is subjected to structural self-organization updating according to the evaluation index and the structural update instruction to obtain an updated incineration characterization model.

[0098] Specifically, first, based on expert and data feedback knowledge, real-time error information, and the real-time correlation results obtained above, after reaching a preset threshold for determining model structure updates, the model self-attention mechanism is used to determine which model type, among the "serial controlled variable model," "parallel G1 flue gas pollutant model," and "parallel G3 flue gas pollutant model" driven by real process data, to elastically self-organize and update the structure.

[0099] Then, through multi-scenario knowledge transfer learning, the historical model parameters under multiple typical operating conditions are reused. The multi-criteria correlation value of the typical operating condition and the data distribution difference between the historical scenario and the current scenario are used as evaluation indicators of structural self-organization. The formula for the latter is described as follows:

[0100]

[0101] Among them, φ(·) is the source data X S and target data X T The mapping function mapped to the feature space aims to achieve differential measurement of data distribution in the feature space in an interpretable way. Finally, the stability analysis of the self-organizing update algorithm of the full-process heterogeneous integrated representation model structure and the complexity analysis of the model are performed. The source data and target data are historical scene data and current scene data, respectively.

[0102] 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.

[0103] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A MSWI process characterization method, characterized in that: include: Obtaining MSWI process data to be characterized; Using long short-term memory neural network to build incineration characterization model based on historical data old (·), as shown below, in, and denote the predicted values ​​of nitrogen oxides, sulfur dioxide, hydrogen chloride, carbon monoxide, and carbon dioxide, respectively; and Respectively represent the predicted values ​​of furnace temperature, steam flow and flue gas oxygen content; u M u represents the input vector composed of the manipulated variables such as feed rate, grate speed, primary air volume, secondary air volume, boiler feed water volume, etc. MC represents the input vector consisting of the manipulated variables of urea, slaked lime, and activated carbon used for flue gas purification, The combustion characterization model is self-updated to obtain an updated combustion characterization model f update (·), as shown below, in, and represent the predicted values ​​of nitrogen oxides, sulfur dioxide, hydrogen chloride, carbon monoxide and carbon dioxide after the updated model respectively; and They represent the predicted values ​​of furnace temperature, steam flow rate and flue gas oxygen content after the updated model; u New M u represents the input vector consisting of updated manipulated variables such as feed rate, grate speed, primary air volume, secondary air volume, boiler feed water volume, etc. New MC The input vector representing the updated manipulated variables of urea, slaked lime, and activated carbon for flue gas purification; Inputting the MSWI process data to be characterized into the updated incineration characterization model to obtain a characterization result; The self-updating method of the incineration characterization model is: Obtain historical modeling data and real-time model acquisition data; Performing correlation analysis on the historical modeling data and the real-time model collection data to obtain a correlation analysis result; updating the parameters of the incineration characterization model according to the correlation analysis results and the process data of the DCS system to obtain a parameter update model; The parameter update model is structurally self-organized and updated according to the distribution differences of historical scene data and current scene data and the correlation analysis results to obtain an updated incineration characterization model.

2. The MSWI process characterization method according to claim 1, characterized in that: The performing correlation analysis on the historical modeling data and the real-time model collection data to obtain a correlation analysis result includes: Processing the historical modeling data to obtain a non-time-series typical sample library and a time-series historical sample subset with different window lengths; Determining the sliding window size of the real-time model to collect data; Perform correlation analysis based on the sliding window size of the real-time model collected data and the non-time-series typical sample library and the time-series historical sample subset corresponding to the sliding window size of the model collected data, and calculate the Pearson correlation coefficient, covariance and mutual information; The correlation analysis results are determined based on the Pearson correlation coefficient, covariance and mutual information.

3. The MSWI process characterization method according to claim 2, characterized in that: The calculation expression of the Pearson correlation coefficient is: Among them, X and Y are the real-time sample and historical sample sets used to calculate the correlation, respectively. n is the nth sample in X, n=1,...,N, N is the number of samples, is the mean of all samples in X; similarly, y n is the nth sample in Y, is the mean of all samples in Y; ρ(X,Y) is the correlation coefficient between X and Y.

4. The MSWI process characterization method according to claim 3, characterized in that: The calculation expression of the covariance is: Where Cov(X,Y) is the covariance value between X and Y.

5. The MSWI process characterization method according to claim 3, characterized in that: The calculation expression of the mutual information is: Where I(X,Y) is the mutual information value between X and Y.

6. The MSWI process characterization method according to claim 1, characterized in that: The parameters of the incineration characterization model are updated according to the correlation analysis results and the process data of the DCS system to obtain a parameter update model, including: Determining expert and data feedback knowledge related to current operating conditions based on process data of the DCS system; Set the model parameter update threshold set; Determining a parameter update instruction based on the model parameter update threshold set and the correlation analysis result; The parameters of the incineration characterization model are updated according to the parameter update instruction, expert and data feedback knowledge, and the inherent logic of the pollution process to obtain a parameter update model.

7. The MSWI process characterization method according to claim 1, characterized in that: The parameter update model is updated by structural self-organization based on the distribution differences of historical scene data and current scene data and the correlation analysis results to obtain an updated incineration characterization model, including: Set the model structure update threshold set; Determining a structure update instruction according to the structure update threshold set; Determining an evaluation index based on the distribution difference between the historical scene data and the current scene data and the correlation analysis result; The parameter update model is subjected to structural self-organization updating according to the evaluation index and the structural update instruction to obtain an updated incineration characterization model.

8. The MSWI process characterization method according to claim 7, characterized in that: The calculation expression of the evaluation index is: Among them, φ(·) is the source data X S and target data X T Mapping function to feature space, N s and N t is the number of samples of source data and target data, and Represents source data X S and target data X T The nth s and n t There are n samples s =1,...,N s and n t =1,...,N t ,MMD(X S ,X T ) represents the evaluation index value.

Images