A method for predicting co emissions from an mswi process

Abstract

The application provides a MSWI process CO emission prediction method, and relates to the technical field of MSWI. The method comprises the following steps: obtaining a historical data set of CO emission; constructing an initial CO emission prediction model and an initial kernel feature space detection model according to the historical data set; performing CO prediction on the to-be-tested input data by using the initial CO emission prediction model to obtain initial prediction data; identifying concept drift of the feature space of the to-be-tested input data by using an unsupervised kernel principal component analysis method and the initial kernel feature space detection model to obtain drift samples; merging the drift samples and the historical data set to obtain an updated data set; updating the initial CO emission prediction model and the initial kernel feature space detection model by using the updated data set, and performing prediction by using the updated model to obtain final prediction data. The application solves the problem that the CO prediction model for the MSWI process does not consider that the CEMS system fails to provide the true value of the model and the prediction model cannot be updated when concept drift occurs, thereby causing the CO prediction to be inaccurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of MSWI, in particular to a MSWI process CO emission prediction method. BACKGROUND

[0002] MSW refers to various solid wastes generated in daily production, life and consumption of people, and long-term accumulation and lack of management can cause urban living space pollution, occupation and virus transmission hazards. The latest statistics of the World Bank show that without taking action, the global MSW annual output will rise to 3.4 billion tons at a growth rate of 70% in 2050.

[0003] Carbon monoxide (CO) is one of the toxic gases generated in the MSWI process, which is difficult to be perceived by the senses, but when it combines with hemoglobin, it will hinder the oxygen transport of blood, and even cause myocardial infarction in severe cases, so it must be strictly controlled.

[0004] Due to the unknown and uncertain fluctuations of MSW components, the high-frequency intervention of experts in the field during operation and the difference of experience, the maintenance and aging of incineration equipment, and other factors, the working condition drift of the MSWI process has complex dynamic time-varying characteristics. This difficult to overcome and predictable working condition fluctuation in industrial processes is reflected in the distribution change of data flow over time, which is called concept drift in the academic field. Therefore, to build a high-precision CO prediction model, not only the appropriate modeling algorithm needs to be determined, but also how to detect concept drift and update the historical model with samples that can represent the drift needs to be solved. Although conventional pollutants including CO, NOx, SO2, etc. can be continuously detected by the CEMS system located at the tail of the chimney, the system has unavoidable misalignment and failure phenomena. The existing CO prediction models for the MSWI process include methods based on reduced deep features and LSTM optimization, methods based on fixed window drift detection, and virtual-real data-driven heterogeneous integrated prediction model methods, but these methods do not consider the prediction model update when the CEMS system fails to provide model true values and concept drift occurs. Therefore, when updating the CO emission concentration prediction model, attention also needs to be paid to how to realize effective updating of the model in the case of true value missing. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a MSWI process CO emission prediction method, which solves the problem that the CO prediction model for the MSWI process in the prior art does not consider the prediction model update when the CEMS system fails to provide model true values and concept drift occurs, resulting in CO prediction misalignment.

[0006] To achieve the above purpose, the present application provides the following scheme:

[0007] A method for predicting CO emissions of MSWI process, comprising:

[0008] obtaining a historical data set of CO emissions;

[0009] constructing an initial CO emission prediction model and an initial kernel feature space detection model according to the historical data set;

[0010] predicting CO emissions of the to-be-tested input data by using the initial CO emission prediction model to obtain initial prediction data;

[0011] identifying concept drift of a feature space of the to-be-tested input data by using an unsupervised kernel principal component analysis method and the initial kernel feature space detection model, constructing a pseudo-label model and obtaining pseudo-true values if drift occurs, and confirming drift samples based on PH test to obtain drift samples, and predicting CO emissions based on the initial CO emission prediction model if no concept drift occurs;

[0012] merging the drift samples and the historical data set to obtain an updated data set;

[0013] updating the initial CO emission prediction model and the initial kernel feature space detection model by using the updated data set to obtain a final CO emission prediction model and a final kernel feature space detection model;

[0014] predicting CO emissions of the to-be-tested input data by using the final CO emission prediction model to obtain final prediction data.

[0015] Preferably, the step of constructing a pseudo-label model and obtaining pseudo-true values if drift occurs, and confirming drift samples based on PH test to obtain drift samples, comprises:

[0016] obtaining a historical pseudo-label data set;

[0017] determining whether the detection value of the to-be-tested input data exceeds a KPCA control limit according to the historical pseudo-label data set, and if yes, storing the sample in a to-be-labeled cache window, and when the number of samples in the window reaches a preset window capacity, constructing a pseudo-label model in real time and labeling the drift samples with pseudo-true values;

[0018] determining whether the samples drift by using a Page-Hinkley detection method on the labeled pseudo-true values, and if yes, obtaining drift samples.

[0019] Preferably, the method further comprises:

[0020] updating the historical pseudo-label data set to obtain an updated historical pseudo-label data set;

[0021] resetting the to-be-labeled cache window to obtain a reset to-be-labeled cache window;

[0022] According to the reset to-be-labeled cache window, the updated historical pseudo-label data set, the final CO emission prediction model and the final kernel feature space detection model are updated in real time.

[0023] Preferably, the expression of the initial CO emission prediction model is:

[0024]

[0025] wherein, is an initial CO emission prediction model, is an LSTM model constructed based on historical data, and is also a prediction model used when no drift occurs, is the nth to-be-tested input data as an input of the LSTM model.

[0026] Preferably, the expression of the initial kernel feature space detection model is:

[0027]

[0028] wherein, is a drift indicator of samples in the first window, SPE k is a drift indicator of samples in the second window, P t is a nonlinear principal component matrix, is a KPCA model constructed based on historical data, to constitute an initial kernel feature space detection model.

[0029] Preferably, the expression of the final CO emission prediction model is:

[0030]

[0031] wherein, is an updated data set, is an updated LSTM model, is a final CO emission prediction model.

[0032] The present application discloses the following technical effects:

[0033] The application provides a MSWI process CO emission prediction method, including: obtaining a historical data set of CO emission; constructing an initial CO emission prediction model and an initial kernel feature space detection model according to the historical data set; performing CO prediction on to-be-tested input data by using the initial CO emission prediction model to obtain initial prediction data; identifying concept drift of a feature space of the to-be-tested input data by using an unsupervised kernel principal component analysis method and the initial kernel feature space detection model, if drift occurs, constructing a pseudo-label model, obtaining pseudo-true values, and performing drift sample confirmation based on PH inspection to obtain drift samples, if no concept drift occurs, performing CO emission prediction based on the initial CO prediction model; merging the drift samples and the historical data set to obtain an updated data set; updating the initial CO emission prediction model and the initial kernel feature space detection model by using the updated data set to obtain a final CO emission prediction model and a final kernel feature space detection model; and performing CO prediction on the to-be-tested input data by using the final CO emission prediction model to obtain final prediction data. The application can not only perform prediction by using historical data, but also continuously learn and adapt to new data, and the prediction model can continue to be updated when the CEMS system fails to provide model true values and concept drift occurs, so that the effectiveness in actual application is improved BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0035] Figure 1 A MSWI process CO emission prediction method flow chart is provided for the embodiments of the present application.

[0036] Figure 2 A MSWI process CO emission prediction strategy modularization schematic diagram is provided for the embodiments of the present application.

[0037] Figure 3 An LSTM structure schematic diagram is provided for the embodiments of the present application.

[0038] Figure 4 A training set absolute value error curve diagram is provided for the embodiments of the present application.

[0039] Figure 5 A pseudo-label model historical data set true value difference component curve diagram is provided for the embodiments of the present application.

[0040] Figure 6 A T 2Graphs;

[0041] Figure 7 A SPE graph provided for an embodiment of the present application;

[0042] Figure 8 A comparison graph of pseudo true value labeling results and actual true values provided for an embodiment of the present application;

[0043] Figure 9 A pseudo marker drift detection result graph provided for an embodiment of the present application;

[0044] Figure 10 A prediction performance change graph provided for an embodiment of the present application. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0046] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0047] As Figure 1 shown, the present application provides a MSWI process CO emission prediction method, comprising:

[0048] Step 100: obtaining a historical data set of CO emission;

[0049] Step 200: constructing an initial CO emission prediction model and an initial kernel feature space detection model according to the historical data set;

[0050] Step 300: using the initial CO emission prediction model to perform CO prediction on the to-be-tested input data to obtain initial prediction data;

[0051] Step 400: using an unsupervised kernel principal component analysis method and the initial kernel feature space detection model to identify concept drift of the feature space of the to-be-tested input data, if drift occurs, constructing a pseudo marker model, obtaining pseudo true values, and performing drift sample confirmation based on PH test to obtain drift samples, if no concept drift occurs, performing CO emission prediction based on the initial CO prediction model;

[0052] Step 500: merging the drift samples and the historical data set to obtain an updated data set;

[0053] Step 600: updating the initial CO emission prediction model and the initial kernel feature space detection model using the update data set to obtain a final CO emission prediction model and a final kernel feature space detection model;

[0054] Step 700: using the final CO emission prediction model to predict CO of the to-be-detected input data to obtain final prediction data.

[0055] Specifically, MSWI includes storage and transportation fermentation, solid waste combustion, waste heat exchange, steam power generation, flue gas treatment and flue gas emission process stages. For the solid waste combustion stage, due to insufficient oxygen supply (part of the MSW cannot be completely oxidized), uneven combustion (part of the MSW cannot fully contact oxygen) or insufficient temperature (insufficient incineration temperature, reduced combustion reaction rate), which will cause incomplete combustion of MSW, resulting in the generation of CO. Therefore, changes in factors such as furnace temperature and oxygen content will change the CO generation relationship and produce concept drift.

[0056] The traditional MSWI process can detect the CO emission concentration in real time through the continuous emission monitoring system (CEMS), which is usually measured by complete extraction or dilution extraction. However, complete extraction is prone to blockage of the air extraction port in a positive pressure environment or when the air extraction amount is too large, and dilution extraction has a long measurement response and requires high purity of dry compressed air. In addition, CEMS needs to be regularly maintained by qualified technicians. Therefore, pseudo true values are labeled to analyze the concept drift phenomenon in the process when all true values cannot be obtained.

[0057] Specifically, for the prediction model, first, based on historical data, a CO emission prediction model and a kernel feature space detection model are constructed, and a historical data set used for real-time construction of a pseudo-labeled model is obtained; second, based on real-time data, drift detection is performed for the CO emission prediction model, unsupervised kernel principal component analysis (KPCA) is used to identify concept drift in the feature space, if drift occurs, a pseudo-labeled model is constructed, pseudo-true values are obtained, and PH test is used to confirm the drift samples, if no concept drift occurs, CO emission prediction is performed based on the historical prediction model; then, the CO emission prediction model and the update data set of the kernel feature space detection are obtained by combining the historical samples and the drift samples; finally, when the preset condition is met, the historical model is updated using the new data set.

[0058] Further, if drift occurs, a pseudo-labeled model is constructed, pseudo-true values are obtained, and PH test is used to confirm the drift samples, and the drift samples are obtained, including:

[0059] obtaining a historical pseudo-label data set;

[0060] Based on the historical pseudo-label dataset, determine whether the detection value of the input data to be tested exceeds the KPCA control limit. If so, store the sample in the cache window to be labeled. When the number of samples in the window reaches the preset window capacity, build a pseudo-label model in real time and mark the drift samples with pseudo-true values;

[0061] The Page-Hinkley detection method is used to determine whether the sample has drifted on the marked pseudo-true value. If so, a drifted sample is obtained.

[0062] Furthermore, it also includes:

[0063] Updating the historical pseudo-label dataset to obtain an updated historical pseudo-label dataset;

[0064] Resetting the cache window to be marked to obtain the reset cache window to be marked;

[0065] The final CO emission prediction model and the final kernel feature space detection model are updated in real time according to the reset to-be-labeled cache window and the updated historical pseudo-label data set.

[0066] Specifically, such as Figure 2 As shown in the figure, each step is modularly understood, including: 1) Construction of CO prediction model and drift detection model based on historical data: Historical samples are used to build a prediction model and a kernel feature space detection model to obtain a historical data set to support the pseudo-labeling model.

[0067] 2) Drift detection of the CO2 prediction model based on real-time data: KPCA (unsupervised kernel principal component analysis) is used to detect drift in the feature space of new samples. When the detection value exceeds the KPCA control limit, the sample is considered to have the possibility of drift and is stored in a cache window to be labeled. When the number of samples in the window reaches the preset window capacity, a pseudo-labeling model is constructed in real time and drifting samples are labeled with pseudo-true values. The Page-Hinkley test method is used to determine whether the labeled pseudo-true values ​​have drifted. If concept drift has not occurred, CO2 concentration prediction is performed based on the historical prediction model.

[0068] 3) Updated dataset acquisition: The samples confirmed to have drifted in the current cache window are recorded as the drift sample dataset, which is then merged with the historical dataset to determine the dataset used for model update based on experience.

[0069] 4) Historical model update: The CO prediction model and the kernel feature space detection model are retrained using the acquired updated dataset, and the historical dataset of the pseudo-labeling model is updated. The cache window to be labeled is also reset.

[0070] Furthermore, the expression of the initial CO emission prediction model is:

[0071]

[0072] in, is the initial CO emission prediction model, This is an LSTM model built based on historical data, and is also the prediction model used when no drift occurs. The nth input data to be tested is used as the input of the LSTM model.

[0073] Specifically, the goal of model learning is to determine the prediction model with the minimum prediction error The LSTM-based model is constructed as follows Figure 3 shown.

[0074] Figure 3 middle, represents the nth sample input, h n-1 and c n-1 Respectively represent n-1 The LSTM hidden layer output and state information of each sample is used as the input of the next sample, h n and c n Represent the n-th sample LSTM output and state information respectively.

[0075] 1) In the initial sample calculation, the parameters are set to n = 1 and h n-1 =0, the memory unit c n Set to an empty set, forget gate f n Determine memory unit c n The n-1th sample output value is output to c n The ratio of the input to the nth sample, its output The calculation of is as follows:

[0076]

[0077] Among them, U f and W f Respectively represent h n-1 and The weight of b f represents the bias; σ(·) represents the sigmoid activation function; ⊙ represents the Hadamard product.

[0078] Candidate value for the current input state The calculation of is as follows:

[0079]

[0080] Among them, U c and W c Respectively represent h n-1 and the weights of h c represents the bias; tanh(·) represents the tanh activation function.

[0081] input gate i n for controlling the storage of the n-th sample input into c n in proportion to the output of The calculation of

[0082]

[0083] where U i and W i represent the weights of h n-1 and b i represents the bias.

[0084] The value obtained by the above calculation is stored in the memory cell c n , as follows:

[0085]

[0086] In addition, represents the value of the first sample output c n , that is, there is and

[0087] output gate o n for controlling the value stored in the memory cell c n can be output in proportion to the output formula and the n-th sample hidden layer output as follows:

[0088]

[0089] h n = o n ⊙ tanh(c n ) (8)

[0090] where U o and W o represent the weights of h n-1 and b o represents the bias.

[0091] The calculation of the output value of the n-th sample is as follows:

[0092]

[0093] where W out represents the corresponding weight of the hidden layer.

[0094] Similarly, the above process is repeated for the remaining samples. The output corresponding to all samples is represented as follows:

[0095]

[0096] The loss function e' of the LSTM model is calculated as follows:

[0097]

[0098] Where y n And represents the actual value and the predicted value of the nth sample, and e' represents the error of the two.

[0099] When updating the LSTM parameters, there are also hidden states c n And h n , which are defined as And The corresponding calculation formula is as follows:

[0100]

[0101] Further, the weights and biases of the forgetting gate, the input gate and the output gate are calculated. Taking the output gate as an example, the weight learning formula is as follows:

[0102]

[0103] The corresponding weight update formula is as follows:

[0104]

[0105] Where, Represents the updated output gate weight.

[0106] The LSTM-based CO concentration prediction model constructed by offline samples can be represented by the following formula:

[0107]

[0108] Further, the expression of the initial kernel feature space detection model is:

[0109]

[0110] Specifically, for high-dimensional process variables, the KPCA model is used to calculate the drift index control limit for determining whether the MSWI process has changed.

[0111] 1) Obtain training data

[0112] That is, obtain the data set

[0113] 2) Calculate the kernel matrix

[0114] Will The vector x in Htd Mapping to high-dimensional space F (denoted as D dimension):

[0115] φ(x Htd ):R M →R D , D≥M (16)

[0116] Where φ(·) is a nonlinear mapping function and D is the dimension of the data after mapping.

[0117] The high-dimensional mapping space is as follows:

[0118]

[0119] A kernel function needs to be introduced to calculate the kernel matrix K. The present invention adopts the Sigmoid kernel function:

[0120]

[0121] Here, f(·) uses the tanh function. β0 and β1 are user-defined. β1 is usually set to 0 by default.

[0122] 3) Core Matrix Centralization

[0123] Define a 1 matrix 1 N , and centralize it as:

[0124]

[0125] Then the centralized kernel matrix K c The calculation is as follows:

[0126] K c =K N×N -U N×N K N×N -K N×N U N×N +U N×N K N×N U N×N (20)

[0127] Among them, K c After calculation, the size is still N×N.

[0128] 4) Eigenvalue decomposition and eigenvector normalization

[0129] The matrix eigenvalues ​​and eigenvectors are solved by eigenvalue decomposition method, and the calculation is as follows:

[0130]

[0131] where Q is the eigenvector matrix, Σ = diag(λ1, λ2, …, λ N ) is a diagonal matrix, and each diagonal element is an eigenvalue.

[0132] The eigenvalues are normalized by the following equation:

[0133]

[0134] where Q s is the normalized matrix of Q, and λ is the eigenvector.

[0135] 5) Selection of the number of principal components

[0136] The cumulative eigenvalue contribution rate η and the PCA contribution threshold δ PCA are used to reduce the dimension:

[0137]

[0138] where p is the number of selected principal components, M is the number of eigenvalues of X Htd , and p is less than M.

[0139] Accordingly, δ PCA , the drift control limit for monitoring whether the working condition has changed, is determined by p.

[0140] 6) Calculation of nonlinear principal components

[0141] The nonlinear principal components are calculated by the number of principal components obtained, and the calculation is as follows:

[0142] P = K c Q s_p (24)

[0143] where Q s_p is the matrix of size N × p in Q s , and P is the nonlinear principal component matrix of size N × p.

[0144] 7) Calculation of statistical index control limit

[0145] The T 2 and SPE statistical indexes according to Hotelling are used as the control limit for working condition drift identification. The statistical index can be expressed as:

[0146]

[0147] where F α (p, N-p) is the F distribution with p and (N-p) degrees of freedom; c αdenotes the normal deviation to the upper (1-α) percentile; θ i and h0are calculated as follows:

[0148]

[0149] Thus, the historical concept drift detection model is constructed

[0150] Further, the first-order differential components of the output and feature space of each sample in the historical sample set are calculated For example, the calculation is as follows:

[0151]

[0152] wherein, and respectively denote the feature space of the sample; and respectively denote the true value of the sample.

[0153] According to the above formula, the first-order differential component set of the output and feature space of the historical sample is respectively Δy Htd and ΔX Htd .

[0154] Further, 1) test data is obtained

[0155] that is, the data set

[0156] 2) the kernel matrix is calculated

[0157] The vector in is mapped to the high-dimensional space F, and the kernel function is introduced to calculate the kernel matrix K t (size N k × N). The Sigmoid kernel function is adopted:

[0158]

[0159] wherein, f(·) adopts the tanh function.

[0160] 3) kernel matrix centering

[0161] A 1 matrix of size N k × N is defined Centering it as:

[0162]

[0163] Then, the centered kernel matrix K t_c is calculated as follows:

[0164]

[0165] Among them, K t_c The calculated size is N k ×N.

[0166] 4) Calculate nonlinear principal components

[0167] The nonlinear principal component is calculated based on the number of principal components obtained, as follows:

[0168] P t =K t_c Q s_p (34)

[0169] Among them, Q s_p is a matrix of size N×p. t is the size N k ×p matrix.

[0170] 5) Calculate statistical index values

[0171] The drift value of the new sample is calculated based on the following formula:

[0172]

[0173] SPE k =∑(K t_c Q s ) 2 -∑P t (36)

[0174] Based on the samples in the window (drift indicator of samples in the first window) and SPE k Whether drift occurs is determined by determining whether the drift index of the sample in the second window satisfies the following conditions.

[0175]

[0176] Furthermore, after KPCA screening and the cache window to be labeled is filled, the sample set in the window is recorded as Where w is the preset cache window sample capacity, and the specific labeling strategy is:

[0177] Construct a new first-order difference component set as:

[0178]

[0179] in, and are the true value and feature space respectively; and for The first-order difference quantity calculated jointly with the last sample in the current training set.

[0180] From , the first-order difference quantity of the feature space of is calculated as:

[0181]

[0182] Based on the nearest neighbor idea, the ε feature space difference quantities with the smallest Euclidean distance from ΔX Htd ' are selected, and their corresponding output space difference quantities are jointly recorded as:

[0183] The output space difference quantity

[0184] of is measured using Ω nearest :

[0185]

[0186] The pseudo ground truth of is labeled as:

[0187]

[0188] The above process is repeated until the pseudo ground truth of all samples in the window is labeled.

[0189] Further, reasonable analysis of the difference between the sample pseudo ground truth and the measured value is the key to confirming the final concept drift of the sample. In the Page-Hinkley detection method, given a series of observation values [l1, l2, … l m ], the likelihood ratio statistic of the alternative hypothesis (there is a drift point γ in the observation value, i.e., 1 < γ < m) to the original hypothesis (there is no drift in the observation value, i.e., γ > m) is:

[0190]

[0191] where f D (·) represents the distribution density function of the standard normal distribution N(0, 1); δ represents that the drift sample follows a normal distribution with a mathematical expectation of δ.

[0192] The above formula is expressed in logarithmic form as:

[0193]

[0194] ​Accordingly, the log-likelihood ratio statistic of the alternative hypothesis (with drift) to the original hypothesis (without drift) is:

[0195]

[0196] By setting a threshold value and Z m Comparison can be made to determine whether there is a concept drift in the current series of observation values.

[0197] When the samples in the to-be-labeled cache window are all completed pseudo-true value labeling, the Page-Hinkley detection method is used to detect the concept drift of the output space of the samples. Taking the observation value Obs(t) at time t as an example, the detection process is as follows:

[0198] First, the cumulative variable about Obs(t) is calculated

[0199]

[0200] Among them, Indicates the mean of all historical observation values at time t-1; the cumulative variable Indicates the difference between the current observation value Obs(t) and the mean of the historical observation values.

[0201] Then, the change index PH t is calculated to determine whether the current observation value Obs(t) is abnormal:

[0202]

[0203] Among them, φ t Indicates the minimum cumulative variable value recorded from the current all time; PH t Indicates the difference between the cumulative variable At time t and the minimum cumulative variable value. When the condition PH t > λ, it is considered that the observation value Obs(t) is abnormal, where λ is an empirical threshold value.

[0204] Let the sample set in the to-be-labeled cache window be filled in the dth time and the samples are all labeled Calculate the average measurement error Aveero d of the samples in the current window as follows:

[0205]

[0206] Among them, And Indicate the true value and pseudo-true value of the mth sample in the window, respectively.

[0207] On this basis, the observed value Obs(t) is selected as the cumulative average measurement error of the samples in the window when the window is filled for the dth time, that is,

[0208]

[0209] At this time, the cumulative variable represents the difference between the current cumulative average measurement error and the average of the historical cumulative average measurement error; φ t represents the minimum value recorded at the current time.

[0210] In addition, when the cache window is filled for the first time, that is, d = 1, At this time, there is a lack of drift judgment basis in the sample output space, so φ t is expressed as:

[0211]

[0212] wherein φ0 is the reference cumulative average measurement error, which will be obtained according to the average measurement error of the validation sample. At the same time, λ = 0 is set, that is, when represents that when the cumulative average measurement error in the window is significantly higher than the historical sample, it is considered that the sample in the window can represent the concept drift, and it is used to construct a new training set.

[0213] Further, when the samples in the cache window are confirmed to drift, the present application updates the measurement model according to the historical samples and the samples in the current window to construct a new training set. Taking the samples in the window when the cache window is filled for the dth time as an example, the new training set is constructed as follows:

[0214]

[0215] wherein, and are the feature space and the true value set of the updated data set, respectively; and are the feature space and the pseudo true value set of the samples in the current window, respectively.

[0216] Further, the expression of the final CO emission prediction model is:

[0217]

[0218] wherein, is the updated data set, is the updated LSTM model, is the final CO emission prediction model.

[0219] Specifically, the prediction model and the feature space drift detection model are constructed by using the updated data set.

[0220] Based on The new prediction model of CO concentration can be expressed by the following formula:

[0221]

[0222] The KPCA detection model (the final kernel feature space detection model) can be expressed by the following formula:

[0223]

[0224] The new model is used to replace the old model, and the pseudo-label model historical data set is updated as follows

[0225]

[0226] After the above update, the prediction model can better adapt to the dynamic changes of the MSWI process. More specifically, the CO prediction model uses the LSTM model, the number of hidden layer units is 128, the number of training times is 210, the learning rate is 0.022, the learning drop factor is 0.2, and the dropout rate is 0.4.

[0227] As can be seen from Figure 4 , the training samples in the process data set have certain effect when constructed by using the LSTM model, and the error is very small, so the LSTM can effectively construct the prediction model.

[0228] The principal component contribution rate threshold of KPCA is set to 0.91, and 5 principal components are selected as the principal space for running state detection. Then, the calculated T 2 and the SPE control limit value are 9.4707 and 0.4246, respectively.

[0229] The curve of Δy Htd in the pseudo-label model historical data set is shown in Figure 5 ,

[0230] As can be seen from Figure 5 , most of the true value difference components are small, so the method has certain reliability.

[0231] The T 2 and SPE results of KPCA detection are shown in Figure 6 and Figure 7 .

[0232] The window capacity set in the application is 3, the parameters of the LSTM model constructed for the pseudo labeling of the drift samples are the same as the parameters of the LSTM model, the number of historical samples selected by the Euclidean distance is 15, and the comparison of the pseudo true value labeling result and the actual true value based on TD learning is as shown in Figure 8 .

[0233] As can be seen from Figure 8 , the pseudo true value change trend is similar to the sample true value, so the pseudo true value can be used for approximate analysis of the sample output space drift when the sample true value is difficult to completely obtain. After the pseudo true value labeling of the sample in the kernel feature space drift is completed, the pseudo labeled drift detection result is as shown in Figure 9 .

[0234] Figure 9 For each time the to-be-labeled cache window is filled and the samples in the window are labeled with pseudo true values, the change of the cumulative average measurement error of the samples in the window is shown in the table. Among them, the to-be-labeled cache window is filled 10 times, and the cumulative average measurement error of the samples in the window is obviously increased at the time when the drift occurs, and tends to be stable with the continuous updating of the model, which indicates that the algorithm can effectively detect the concept change existing in the sample output space.

[0235] According to the above detection result, after the model is updated by using a new training set composed of concept drift samples and historical samples, the measurement performance change is as shown in Figure 10 .

[0236] The above results show that the prediction error of the prediction model after using the drift detection algorithm is obviously decreased compared with the original model, the algorithm can significantly improve the measurement performance of the model facing the concept drift samples, and can effectively improve the prediction accuracy of the CO concentration prediction model of the MSWI process in the drift environment.

[0237] More specifically, the application adopts the strategy of combining KPCA and Page-Hinkley detection method to fully reflect the concept drift behavior of new samples in the feature space and sample output space; 2) the semi-supervised mechanism based on TD learning is used for pseudo true value labeling of the feature space drift samples, which provides a new method for semi-supervised concept drift detection for industrial regression problems; 3) the real MSWI process data set is used to verify the feasibility of the method in practical application, and it is shown that the method has better performance than the existing method. Further research work includes: improving the pseudo true value labeling algorithm to improve the accuracy of labeling; considering the mechanism of screening historical modeling samples and mixed pseudo labeled modeling samples to improve the accuracy of the prediction model.

[0238] The various embodiments described in this specification are presented for the purpose of illustrating the principles of the present application and its best mode of operation. Each of the embodiments described in this specification has been provided for the purpose of illustration and is not intended to limit the application.

[0239] The principles and implementations of the present application have been described in the above embodiments, which are only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, the specific implementation and application range of the present application can be changed according to the idea of the present application. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for predicting CO emissions from a MSWI process, characterized in that: include: Obtain historical datasets of CO emissions; constructing an initial CO emission prediction model and an initial kernel feature space detection model based on the historical data set; Using the initial CO emission prediction model to perform CO prediction on the input data to be measured to obtain initial prediction data; Using an unsupervised kernel principal component analysis method and the initial kernel feature space detection model to identify concept drift in the feature space of the input data to be tested, if drift occurs, constructing a pseudo-labeling model and obtaining a pseudo-true value, and performing drift sample confirmation based on a pH test to obtain a drift sample, and if concept drift does not occur, performing CO emission prediction based on the initial CO prediction model; Merging the drift sample and the historical data set to obtain an updated data set; Using the updated data set to update the initial CO emission prediction model and the initial kernel feature space detection model to obtain a final CO emission prediction model and a final kernel feature space detection model; Using the final CO emission prediction model to perform CO prediction on the input data to be measured to obtain final prediction data; The expression of the initial kernel feature space detection model is: ； in, is the drift index of the samples in the first window, is the drift index of the sample in the second window, is the nonlinear principal component matrix, is a KPCA model built based on historical data. Historical data for building the initial kernel feature space detection model.

2. The method for predicting CO emissions from a MSWI process according to claim 1, characterized in that: If drift occurs, a pseudo-label model is constructed and a pseudo-true value is obtained, and drift sample confirmation is performed based on a PH test to obtain a drift sample, including: Obtain historical pseudo-label datasets; Based on the historical pseudo-label dataset, determine whether the detection value of the input data to be tested exceeds the KPCA control limit. If so, store the sample in the cache window to be labeled. When the number of samples in the window reaches the preset window capacity, build a pseudo-label model in real time and mark the drift samples with pseudo-true values; The Page-Hinkley detection method is used to determine whether the sample has drifted on the marked pseudo-true value. If so, a drifted sample is obtained.

3. The method for predicting CO emissions from a MSWI process according to claim 2, wherein: Also includes: Updating the historical pseudo-label dataset to obtain an updated historical pseudo-label dataset; Resetting the cache window to be marked to obtain the reset cache window to be marked; The final CO emission prediction model and the final kernel feature space detection model are updated in real time according to the reset to-be-labeled cache window and the updated historical pseudo-label data set.

4. The method for predicting CO emissions from a MSWI process according to claim 2, wherein: The expression of the initial CO emission prediction model is: ； in, is the initial CO emission prediction model, This is an LSTM model built based on historical data, and is also the prediction model used when no drift occurs. The nth input data to be tested is used as the input of the LSTM model.

5. The method for predicting CO emissions from a MSWI process according to claim 2, wherein: The expression of the final CO emission prediction model is: ； in, To update the dataset, is the updated LSTM model, The final CO emission prediction model.

Images