A Multi-condition Prediction Method for Desulfurization Systems Based on Extreme Learning Machine

Abstract

This invention provides a multi-condition prediction method for desulfurization systems based on Extreme Learning Machine (ELM). It acquires historical operating parameter data of boilers and desulfurization systems in thermal power plants, classifies operating conditions according to six labeled feature parameters, and determines corresponding training datasets. For the training dataset, a classification model for different operating conditions is established based on a Bayesian algorithm, and a suitable modeling sample set is selected. The delay time of variables is estimated using the Pearson coefficient, and the modeling data is reorganized based on this delay time. Using the reorganized dataset, an SO2 prediction model for different labeled operating conditions is established based on the ELM algorithm. Real-time system data is input into the classification model to determine the real-time operating condition category of the system, and the corresponding prediction model is used to predict the SO2 emission concentration. This invention, while reducing the impact of delay time, determines the current operating condition of the system based on boiler operating data, switches between different operating condition prediction models, and improves the prediction efficiency and accuracy of outlet SO2.

Description

Technical Field

[0001] This invention relates to the field of computer modeling and prediction technology, specifically but not limited to a multi-condition prediction method for desulfurization systems based on extreme learning machines. Background Technology

[0002] With the increasing prominence of environmental issues, and under China's strict environmental protection policy of controlling sulfur dioxide emissions, flue gas desulfurization (FGD) has been widely used in thermal power plants, greatly reducing air pollutant emissions from power plants in my country and improving the environment.

[0003] Because the desulfurization process exhibits multiple operating conditions and significant time delays, the parameter values ​​recorded by the distributed control system (DCS) at the current moment cannot accurately predict the SO2 concentration in the clean flue gas. Therefore, the delay time of key parameters in the system and the operating conditions of the desulfurization process become important factors affecting the establishment of SO2 concentration models. With the continuous improvement of power plant automation, the status and parameters of boiler units and other flue gas treatment equipment can be detected by sensors and recorded in databases; however, this data is not being well utilized. Therefore, this data provides a new approach for optimizing SO2 prediction models.

[0004] In view of this, a new prediction method is needed to solve at least some of the above problems. Summary of the Invention

[0005] To address one or more problems in the existing technology, this invention proposes a multi-condition prediction method for desulfurization systems based on extreme learning machines, which solves the problem of inaccurate prediction of outlet SO2 concentration in wet desulfurization systems of thermal power plants under varying operating conditions and large hysteresis.

[0006] The technical solution to achieve the purpose of this invention is as follows:

[0007] A multi-condition prediction method for desulfurization systems based on extreme learning machines includes:

[0008] Step 1: Obtain historical data of operating parameters of boiler units and desulfurization systems, select and determine the marker feature parameters for different operating conditions of boilers, and classify the historical data of boilers and desulfurization systems of thermal power plants according to the marker feature parameters.

[0009] Step 2: Train historical data under different working conditions based on the Bayesian classification algorithm to establish Naive Bayes classification models under different working conditions;

[0010] Step 3: Use the Pearson coefficient method to estimate the delay time of various flue gas parameters, absorber slurry density, and slurry circulation pump flow rate. Based on the delay time, reconstruct and preprocess the historical data samples to obtain the reconstructed data sample set.

[0011] Step 4: Based on the reconstructed data sample set, the Extreme Learning Machine algorithm is used to establish SO2 prediction models for different working conditions.

[0012] Step 5: Obtain real-time system operation data and input it into the Naive Bayes classification model to obtain the real-time operating condition category of the desulfurization system. Switch to the corresponding SO2 prediction model according to the different operating condition categories.

[0013] Step 6: Input the real-time operating data of the system into the corresponding prediction model to obtain the predicted value of SO2 emission concentration.

[0014] Optionally, the characteristic parameters of the marker bit in step 1 include: boiler load W, gas turbine speed r, oxygen content of boiler outlet flue gas Q, flow rate of boiler outlet flue gas F, temperature of boiler outlet flue gas T, flow rate of boiler outlet flue gas V, primary air fan current I, fuel input M, steam temperature T, and pressure P. The operating condition classification includes six operating conditions: startup, grid connection / power supply, cracking, shutdown, outage, and accident.

[0015] Optionally, the state characteristics of the six operating conditions are as follows:

[0016] Start-up characteristics: When the boiler starts up, the oxygen content Q of the flue gas at the boiler outlet gradually decreases, the flue gas velocity V at the boiler outlet gradually increases and exceeds 2m / s, and the flue gas temperature T at the boiler outlet gradually increases and exceeds 40℃.

[0017] Characteristics of grid-connected / power supply operation: gas turbine speed r≈3000r / min, boiler outlet flue gas temperature T≈540℃, primary air fan current I>0, fan starts;

[0018] Characteristics of the cracking operation: The fuel input M gradually decreases, and the boiler load W gradually decreases to 50%;

[0019] Characteristics of boiler shutdown: Boiler outlet flue gas velocity V, boiler outlet flue gas flow rate F, and boiler outlet flue gas temperature T all gradually decrease, while the oxygen content Q of boiler outlet flue gas gradually increases.

[0020] Characteristics of shutdown conditions: oxygen content Q in boiler outlet flue gas > 18%, flue gas velocity V < 1.8 m / s, flue gas flow rate F less than 15% of rated operating conditions, and flue gas temperature T < 30℃.

[0021] The characteristics of the accident operation were: the automatic monitoring data of the exhaust outlet exceeded the standard, and some equipment parameters fluctuated abnormally.

[0022] Optionally, the specific steps for establishing the Naive Bayes classification model under different working conditions in step 2 include:

[0023] Step 2-1: Select 10 key variables, including boiler load W, gas turbine speed r, oxygen content in boiler outlet flue gas Q, flow rate of boiler outlet flue gas F, temperature of boiler outlet flue gas T, velocity of boiler outlet flue gas V, primary air fan current I, fuel input M, steam temperature T, and pressure P, as feature attributes of the classification model. Use the historical data corresponding to the above feature attributes as training samples for the classification model. At the same time, label the working condition type of each training sample according to the state characteristics of each working condition.

[0024] Step 2-2: Calculate the frequency P(work condition) of each working condition in the training sample. Divide the samples of each feature attribute according to the critical value of the feature attribute when the working condition changes. Calculate the frequency P(variable) of each feature attribute in the training sample after the division. Then calculate the conditional probability P(variable|working condition) of each feature attribute in each working condition after the division.

[0025] Steps 2-3: Based on Naive Bayes theory, calculate the conditional probability of samples appearing under each feature attribute after partitioning for each working condition:

[0026]

[0027] Steps 2-4: Calculate the sum of probabilities of the training samples under each working condition:

[0028] S(operating condition) = ∑P(operating condition | variable)

[0029] The working condition type corresponding to the maximum value of S(working condition) is the working condition type to which the sample belongs, and the working condition type number of the sample is output.

[0030] Optionally, the critical values ​​of the characteristic attributes in step 2-2 when the operating conditions change include:

[0031] Boiler load W: {W≤20%, 20%<W<50%, W≥50%}

[0032] Gas turbine speed r: {r≤2000, 2000 <r<3500,r≥3500}

[0033] Oxygen content Q in boiler outlet flue gas: {Q≤5%, 5%<Q<10%, Q≥10%}

[0034] Boiler outlet flue gas flow rate F: {F<0.15*F, F≥0.15*F}

[0035] Boiler outlet flue gas temperature T: {T≤40℃, 40℃<T<540℃, T≥540℃}

[0036] Boiler outlet flue gas velocity V: {V<1.8m / s, V≥1.8m / s}

[0037] Primary fan current I: {I<0.15*I, I≥0.15*I}

[0038] Fuel input M: {M < 0.15 * M, M ≥ 0.15 * M}

[0039] Steam temperature T′: {T′≤400℃, 400℃<T′<600℃, T′≥600℃}

[0040] Steam pressure P: {P≤10kPa, 10kPa<P<25kPa, P≥25kPa}.

[0041] Optionally, the specific steps for obtaining the reconstructed data sample set in step 3 include:

[0042] Step 3-1: Define the operating parameters in the historical data as input variables X, and the SO2 concentration at the desulfurization system outlet as output variable Y, where X = [x1(t), ..., x m [y(t)], m is the dimension of the input variable, Y = y(t), t = 1…N, N is the number of historical data samples;

[0043] Step 3-2: Determine the maximum delay time k of the i-th dimension operating condition parameter, and within the maximum delay time range, set the i-th dimension input variable x... i (t) The data is reconstructed and segmented according to the acquisition time interval, resulting in k sets of reconstructed working condition data sequences, which are x i (t-1), ..., x i (tk);

[0044] Step 3-3: Calculate the Pearson coefficient between the k sets of reconstructed operating condition data sequences and the SO2 concentration y(t) respectively;

[0045] Steps 3-4: Select the delay time corresponding to the set of reconstructed working condition data sequences with the largest Pearson coefficient as the predictor variable x. i The optimal delay time (t) is determined, and the set of reconstructed operating condition data sequences is used as the input variable x. i Reconstructing the data sequence of (t), with input variable x i (t) has a dimension of 1 after reconstruction;

[0046] Step 3-5: Repeat steps 3-2 to 3-4 to calculate the optimal delay time and reconstructed data sequence for all input variables X. The reconstructed data sequence of all input variables X is combined to obtain the reconstructed data sample, which is used as the input sample for the prediction model.

[0047] Optionally, the formula for calculating the Pearson coefficient in step 3-3 is as follows:

[0048]

[0049] Among them, X i To reconstruct the parameters of the operating condition data sequence, Y i For the SO2 concentration at the outlet, To reconstruct the parameter averages of the operating condition data series, This represents the average SO2 concentration at the outlet.

[0050] Optionally, the specific steps for establishing SO2 prediction models for different operating conditions in step 4 include:

[0051] Step 4-1: Obtain a reconstructed data sample set under a certain working condition, and use the historical data of the feature variables in the reconstructed data sample set as the input sample matrix of the extreme learning machine SO2 prediction model;

[0052] Step 4-2: Establish a single-layer feedforward neural network for the samples of the characteristic variables related to the working condition. The single-layer feedforward neural network prediction model includes 6 input layers, L hidden layer nodes and 1 output layer node.

[0053] Step 4-3: Randomly generate the weight matrices W of the input layer and hidden layer of a single-layer feedforward neural network. i This yields the hidden layer output matrix H (6 rows, L columns) and the output weight matrix β (L rows, 1 column).

[0054] Step 4-4: Introduce an L2 regularization term into the error function of the SO2 prediction model to obtain the minimum error function of the model:

[0055]

[0056] Where C is the regularization coefficient, H is the hidden layer output matrix, T is the training objective, β is the output weight matrix, and || is the Firobenius norm of the matrix elements;

[0057] Steps 4-5: Calculate the weight matrix β of the Extreme Learning Machine prediction model. * :

[0058]

[0059] The SO2 prediction model under this operating condition is as follows:

[0060] Y = XW i β *

[0061] Where X is the input matrix and Y is the predicted value of SO2 emission concentration;

[0062] Steps 4-6: Repeat steps 4-1 to 4-5 to establish SO2 prediction models for different operating conditions.

[0063] Optionally, in step 4-1, the input sample matrices and their corresponding feature variables under different working conditions are as follows:

[0064] The input sample matrix under startup conditions is: X=[Q,V,T,ρ,F′,C];

[0065] The input sample matrix under grid-connected / functional operating conditions is: X=[r,T,I,ρ,F′,C];

[0066] The input sample matrix under the decoupling condition is: X=[Q,M,W,ρ,F′,C];

[0067] The input sample matrix under shutdown conditions is: X=[Q,V,T,ρ,F′,C];

[0068] Where P is the density of the absorber slurry, F′ is the flow rate of the slurry circulation pump, C is the sulfur dioxide concentration at the boiler outlet, W is the boiler load, r is the speed of the gas turbine, Q is the oxygen content of the flue gas at the boiler outlet, T is the temperature of the flue gas at the boiler outlet, V is the flue gas velocity at the boiler outlet, I is the current of the primary air fan, and M is the fuel input.

[0069] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0070] The multi-condition prediction method for desulfurization systems based on extreme learning machine of the present invention establishes a condition classification model by using historical data, estimates the delay time of variables to reorganize the data and establish a prediction model. This method reduces the impact of delay time, determines the current condition of the system based on real-time boiler operation data, and switches between different condition prediction models, thereby effectively improving the prediction efficiency and accuracy of SO2 at the outlet. Attached Figure Description

[0071] The accompanying drawings are provided to further illustrate the invention and, together with the description, serve to explain embodiments of the invention, but do not constitute a limitation thereof. In the drawings:

[0072] Figure 1 The diagram shows the process flow of a thermal power plant desulfurization system to which the multi-condition prediction method for desulfurization systems based on extreme learning machines of this invention is applied.

[0073] Figure 2The flowchart illustrates the multi-condition prediction method for desulfurization systems based on extreme learning machines of this invention, which utilizes the Naive Bayes classification algorithm to establish classification models under different operating conditions.

[0074] Figure 3 The flowchart shows the data preprocessing process using the Pearson coefficient method in the multi-condition prediction method for desulfurization systems based on extreme learning machine according to the present invention.

[0075] Figure 4 The flowchart of the multi-condition modeling method for desulfurization systems based on extreme learning machine is shown below.

[0076] Figure 5 The flowchart of the Extreme Learning Machine (ELM) algorithm modeling process for the multi-condition prediction method of desulfurization system based on the present invention is shown below.

[0077] Figure 6 The figure shows the SO2 prediction results of the multi-condition prediction method for desulfurization system based on extreme learning machine of the present invention under the start-up condition.

[0078] Figure 7 The figure shows the SO2 prediction results of the multi-condition prediction method for desulfurization systems based on extreme learning machine of the present invention under grid-connected / power supply conditions.

[0079] Figure 8 The figure shows the prediction results of the multi-condition prediction method for desulfurization systems based on extreme learning machine of the present invention under the disconnection condition.

[0080] Figure 9 The figure shows the prediction results of the multi-condition prediction method for desulfurization system based on extreme learning machine of the present invention under the shutdown condition.

[0081] Figure 10 This is an overall flowchart of the multi-condition prediction method for desulfurization systems based on extreme learning machine according to the present invention.

[0082] The symbols in the attached diagrams are as follows: 1: Boiler; 2: Superheater; 3: Steam drum; 4: Gas turbine; 5: Desulfurizing agent preparation tank; 6: Slurry preparation tank; 7: Desulfurization tower; 8: Slurry circulation pump. Detailed Implementation

[0083] To further understand the present invention, preferred embodiments of the present invention are described below in conjunction with examples. However, it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, and not for limiting the scope of the claims of the present invention.

[0084] The description in this section pertains to only a few typical embodiments, and the present invention is not limited to the scope of the embodiments described. Combinations of different embodiments, substitution of some technical features in different embodiments, and substitution of similar or identical prior art with some technical features in the embodiments are also within the scope of the description and protection of the present invention.

[0085] like Figure 1 The diagram shows the wet desulfurization process flow chart for thermal power plants to which this technical solution applies. The states of the boiler in a thermal power plant are divided into six categories: startup, grid connection / power supply, decompression, shutdown, outage, and accident. Different operating conditions can be determined based on the characteristics of parameters such as those of the boiler, steam drum, desulfurization facilities, and flue gas parameters.

[0086] According to one aspect of the present invention, a multi-condition prediction method for desulfurization systems based on extreme learning machines is provided, such as... Figure 10 As shown, it includes:

[0087] Step 1: Obtain historical data of the operating parameters of the boiler unit and desulfurization system. The historical data can be obtained from the DCS (distributed control system) system in the enterprise's central control room. The historical dataset will be used as the training set for the model.

[0088] Select the marker position characteristic parameters for different operating conditions of the boiler. The marker position characteristic parameters include: boiler load W, gas turbine speed r, oxygen content Q of boiler outlet flue gas, flow rate F of boiler outlet flue gas, temperature T of boiler outlet flue gas, velocity V of boiler outlet flue gas, primary air fan current I, fuel input M, steam temperature T, and pressure P.

[0089] The historical data of the thermal power plant boiler and desulfurization system are classified according to the marker feature parameters. The classification includes six operating conditions: startup, grid connection / power supply, de-cracking, shutdown, outage, and accident. The operating status of the thermal power plant boiler and desulfurization system is determined based on eight operating variables: boiler load W, gas turbine speed r, oxygen content Q in boiler outlet flue gas, flow rate F in boiler outlet flue gas, temperature T in boiler outlet flue gas, velocity V in boiler outlet flue gas, primary air fan current I, and fuel input M. The status characteristics of the six operating conditions are as follows:

[0090] Start-up characteristics: When the boiler starts up, the oxygen content Q of the flue gas at the boiler outlet gradually decreases, the flue gas velocity V at the boiler outlet gradually increases and exceeds 2m / s, and the flue gas temperature T at the boiler outlet gradually increases and exceeds 40℃.

[0091] Characteristics of grid-connected / power supply operation: gas turbine speed r≈3000r / min, boiler outlet flue gas temperature T≈540℃, primary air fan current I>0, fan starts;

[0092] Characteristics of the cracking operation: The fuel input M gradually decreases, and the boiler load W gradually decreases to 50%;

[0093] Characteristics of boiler shutdown: Boiler outlet flue gas velocity V, boiler outlet flue gas flow rate F, and boiler outlet flue gas temperature T all gradually decrease, while the oxygen content Q of boiler outlet flue gas gradually increases.

[0094] Characteristics of shutdown conditions: oxygen content Q in boiler outlet flue gas > 18%, flue gas velocity V < 1.8 m / s, flue gas flow rate F less than 15% of rated operating conditions, and flue gas temperature T < 30℃.

[0095] The characteristics of the accident operation were: the automatic monitoring data of the exhaust outlet exceeded the standard, and some equipment parameters fluctuated abnormally.

[0096] Step 2: Train historical data under different working conditions using the Bayesian classification algorithm to establish Naive Bayes classification models for different working conditions; for example... Figure 2 As shown, the specific steps include:

[0097] Step 2-1: Select 10 key variables, including boiler load W, gas turbine speed r, oxygen content in boiler outlet flue gas Q, flow rate of boiler outlet flue gas F, temperature of boiler outlet flue gas T, velocity of boiler outlet flue gas V, primary air fan current I, fuel input M, steam temperature T, and pressure P, as feature attributes of the classification model. Use the historical data corresponding to the above feature attributes as training samples for the classification model. At the same time, label the working condition type of each training sample according to the state characteristics of each working condition described in Step 1.

[0098] Step 2-2: Calculate the frequencies P(start-up), P(grid-connected), P(disconnected), and P(shutdown) of samples in the training samples labeled as start-up, grid connection / power supply, disconnection, and shutdown conditions, respectively. Divide the samples of each feature attribute according to the critical value of that feature attribute when the operating condition changes. Calculate the frequency P(variable) of the samples of each feature attribute in the training samples after the division, and then calculate the conditional probability P(variable|operating condition) of the samples of each feature attribute under each operating condition. The specific division of the critical values ​​of each feature attribute when the operating condition changes is as follows:

[0099] Boiler load W: {W≤20%, 20%<W<50%, W≥50%}

[0100] Gas turbine speed r: {r≤2000, 2000 <r<3500,r≥3500}

[0101] Oxygen content Q in boiler outlet flue gas: {Q≤5%, 5%<Q<10%, Q≥10%}

[0102] Boiler outlet flue gas flow rate F: {F<0.15*F, F≥0.15*F}

[0103] Boiler outlet flue gas temperature T: {T≤40℃, 40℃<T<540℃, T≥540℃}

[0104] Boiler outlet flue gas velocity V: {V<1.8m / s, V≥1.8m / s}

[0105] Primary fan current I: {I<0.15*I, I≥0.15*I}

[0106] Fuel input M: {M < 0.15 * M, M ≥ 0.15 * M}

[0107] Steam temperature T′: {T′≤400℃, 400℃<T′<600℃, T′≥600℃}

[0108] Steam pressure P: {P≤10kPa, 10kPa<P<25kPa, P≥25kPa};

[0109] Steps 2-3: Based on Naive Bayes theory, calculate the conditional probability of samples appearing under each feature attribute after partitioning for each working condition:

[0110]

[0111] The probabilities P(start-up|variable), P(grid connection|variable), P(de-station|variable), and P(shutdown|variable) of each characteristic variable belonging to each working condition are obtained;

[0112] Steps 2-4: Calculate the sum of probabilities of the training samples under each working condition:

[0113] S(operating condition) = ∑P(operating condition | variable)

[0114] The operating condition type corresponding to the maximum value of S (operating condition) is the operating condition type to which the sample belongs. The four operating condition types—start-up, grid connection / power supply, disconnection, and shutdown—are numbered 1, 2, 3, and 4, and the operating condition type number of the sample is output. In one embodiment, for 6 months of operating data of the flue gas desulfurization system of a 500MW unit in a thermal power plant, after screening, the training sample for this model is set to 400 groups, including 100 groups each for the four operating conditions: start-up, grid connection / power supply, disconnection, and shutdown. The P (variable|operating condition) data obtained after feature attribute division and calculation are shown in the table below:

[0115] Table 1. Statistical Table of P(Variable|Operating Condition)

[0116]

[0117] Step 3: Estimate the delay times of various flue gas parameters, absorber slurry density, and slurry circulation pump flow rate using the Pearson coefficient method. Based on these delay times, reconstruct and preprocess historical data samples to obtain a reconstructed data sample set; for example... Figure 3 As shown, the specific steps include:

[0118] Step 3-1: Define the operating parameters in the historical data as input variables X, and the SO2 concentration at the desulfurization system outlet as output variable Y, where X = [x1(t), ..., x m [y(t)], m is the dimension of the input variable, Y = y(t), t = 1…N, N is the number of historical data samples;

[0119] Step 3-2: Determine the maximum delay time k of the i-th dimension operating condition parameter, and within the maximum delay time range, set the i-th dimension input variable x... i (t) The data is reconstructed and segmented according to the acquisition time interval, resulting in k sets of reconstructed working condition data sequences, which are x i (t-1), ..., x i (tk);

[0120] Step 3-3: Calculate the Pearson coefficient between the k sets of reconstructed operating condition data sequences and the SO2 concentration y(t); the formula for calculating the Pearson coefficient is:

[0121]

[0122] Among them, X i To reconstruct the parameters of the operating condition data sequence, Y i For the SO2 concentration at the outlet, To reconstruct the parameter averages of the operating condition data series, This represents the average SO2 concentration at the outlet.

[0123] Steps 3-4: The data sequence with the largest Pearson coefficient has the highest correlation with the output y(t), and the corresponding delay time is the predictor variable x. i The optimal delay time for (t) is selected by choosing the delay time corresponding to the set of reconstructed working condition data sequences with the largest Pearson coefficient as the predictor variable x. i The optimal delay time (t) is determined, and the set of reconstructed operating condition data sequences is used as the input variable x. i Reconstructing the data sequence of (t), with input variable x i (t) has a dimension of 1 after reconstruction;

[0124] Step 3-5: Repeat steps 3-2 to 3-4 to calculate the optimal delay time and reconstructed data sequence for all input variables X. The reconstructed data sequence of all input variables X is combined to obtain the reconstructed data sample, which is used as the input sample for the prediction model.

[0125] In one embodiment, the above-mentioned delay time analysis is performed on the operating parameters required for modeling, and the optimal delay time statistics table for each input variable is obtained as follows:

[0126] Table 2. Statistics of Delay Time for Each Variable

[0127] Variable name Q V T ρ F′ C I M W Delay time 440s 120s 110s 50s 20s 170s 60s 55s 200s

[0128] Step 4: Based on the reconstructed data sample set, establish SO2 prediction models for different operating conditions using the Extreme Learning Machine algorithm; for example... Figure 5 The diagram shows the modeling process of the Extreme Learning Machine algorithm, including the following steps:

[0129] Step 4-1: Obtain the reconstructed data sample set under different operating conditions. Use the historical data of characteristic variables such as absorber slurry density ρ, slurry circulation pump flow rate F′, boiler outlet sulfur dioxide concentration C, boiler load W, gas turbine speed r, boiler outlet flue gas oxygen content Q, boiler outlet flue gas temperature T, boiler outlet flue gas velocity V, primary air fan current I, and fuel input M from the reconstructed data sample set as the input sample matrix for the extreme learning machine SO2 prediction model.

[0130] The input sample matrices and their corresponding feature variables under different working conditions are as follows:

[0131] The input sample matrix under startup conditions is: X=[Q,V,T,ρ,F′,C];

[0132] The input sample matrix under grid-connected / functional operating conditions is: X=[r,T,I,ρ,F′,C];

[0133] The input sample matrix under the decoupling condition is: X=[Q,M,W,ρ,F′,C];

[0134] The input sample matrix under shutdown conditions is: X=[Q,V,T,ρ,F′,C];

[0135] like Figure 4 The diagram shows a multi-condition modeling flowchart. In one embodiment, before acquiring historical data of the desulfurization system and corresponding boiler units, the possible operating conditions of the boiler units are determined by understanding aspects such as the fuel used by the boiler units, flue gas composition and generation mechanism, and the desulfurization system process flow. These conditions include startup, grid connection / power supply, desulfurization, shutdown, outage, and accidents. Historical data can be obtained from the enterprise's DCS (distributed control system) in the central control room and will be used as the training set for the model. The historical data of the analyzed feature variables, after preprocessing, will be used as input to the extreme learning machine prediction model.

[0136] Step 4-2: As Figure 5 As shown, a single-layer feedforward neural network is established for samples of feature variables related to different working conditions. The single-layer feedforward neural network prediction model includes 6 input layers, L hidden layer nodes and 1 output layer node.

[0137] Step 4-3: Randomly generate the weight matrix W of a single-layer feedforward neural network. i We obtain the hidden layer output matrix H (6 rows, L columns) and the output weight matrix β (L rows, 1 column). We then solve for the output weights to minimize the model's error function, which is as follows:

[0138]

[0139] Where H is the hidden layer output matrix, T is the training objective, β is the output weight matrix, and || is the Firobenius norm of the matrix elements;

[0140] Step 4-4: Introduce an L2 regularization term into the error function of the SO2 prediction model to obtain the minimum error function of the model:

[0141]

[0142] Where C is the regularization coefficient, H is the hidden layer output matrix, T is the training objective, β is the output weight matrix, and || is the Firobenius norm of the matrix elements. Solving this error function is minimum equivalent to the ridge regression problem.

[0143] Steps 4-5: Calculate the weight matrix β of the Extreme Learning Machine prediction model. * :

[0144]

[0145] The SO2 prediction model under this operating condition is as follows:

[0146] Y = XW i β *

[0147] Where X is the input matrix and Y is the predicted value of SO2 emission concentration;

[0148] Steps 4-6: Repeat steps 4-1 to 4-5 to establish SO2 prediction models for different operating conditions.

[0149] Step 5: Obtain real-time system operation data and input it into the Naive Bayes classification model to obtain the real-time operating condition category of the desulfurization system. Switch to the corresponding SO2 prediction model according to the different operating condition categories.

[0150] Step 6: Input the real-time operating data of the system into the corresponding prediction model to obtain the predicted value of SO2 emission concentration.

[0151] like Figure 6 The predicted SO2 emission concentration under start-up conditions is as follows. Figure 7 The predicted SO2 emission concentration under grid-connected / power supply conditions. Figure 8 To analyze the predicted SO2 emission concentration under the decoupling operating conditions, Figure 9 This is the predicted SO2 emission concentration under shutdown conditions.

[0152] The prediction error of this technical solution and the prediction structure error of the BP neural network under different operating conditions are compared as follows:

[0153] Table 2 Accuracy Comparison Table

[0154] start up Grid connection / energy supply Solution Shutdown This publicly disclosed forecasting method 94.5 91.2 94.3 93.2 BP neural network 83.9 85.5 87.2 89.9

[0155] As shown in the table above, the prediction method of this technical solution can effectively improve the prediction accuracy of SO2 emission concentration.

[0156] In one embodiment, real-time data is input into an established operating condition classification model to obtain the boiler unit operating conditions corresponding to the desulfurization system. The predicted boiler unit operating conditions are then compared with manually marked operating conditions in the distributed control system (DCS) to identify any false markings. In another embodiment, real-time data is input into an established classification model and an outlet SO2 concentration prediction model to obtain the predicted SO2 concentration at the desulfurization system outlet. This predicted value can then be used to determine the amount of lime slurry prepared and the dosage of chemicals used in the desulfurization system, thereby reducing the enterprise's flue gas treatment costs.

[0157] The description and application of the present invention herein are illustrative and not intended to limit the scope of the invention to the embodiments described above. The effects or advantages described in the specification may not be apparent in actual experimental cases due to uncertainties in specific conditions or other factors, and such descriptions are not intended to limit the scope of the invention. Variations and modifications to the embodiments disclosed herein are possible, and various substitutions and equivalents of the components in the embodiments are well known to those skilled in the art. It should be understood by those skilled in the art that the invention can be implemented in other forms, structures, arrangements, proportions, and with other components, materials, and parts without departing from the spirit or essential characteristics of the invention. Other variations and modifications can be made to the embodiments disclosed herein without departing from the scope and spirit of the invention.

Claims

1. A multi-condition prediction method for desulfurization systems based on extreme learning machines, characterized in that, include: Step 1: Obtain historical data of operating parameters of boiler units and desulfurization systems, select and determine the marker feature parameters for different operating conditions of boilers, and classify the historical data of boilers and desulfurization systems of thermal power plants according to the marker feature parameters. Step 2: Train historical data under different working conditions based on the Bayesian classification algorithm to establish Naive Bayes classification models under different working conditions; Step 3: Estimate the delay times of various flue gas parameters, absorber slurry density, and slurry circulation pump flow rate using the Pearson coefficient method. Based on these delay times, reconstruct and preprocess historical data samples to obtain a reconstructed data sample set. Specific steps include: Step 3-1: Define the operating parameters in the historical data as input variables X, and the SO2 concentration at the desulfurization system outlet as output variable Y, where X = [x1(t), ..., x m [y(t)], m is the dimension of the input variable, Y = y(t), t = 1…N, N is the number of historical data samples; Step 3-2: Determine the maximum delay time k of the i-th dimension operating condition parameter, and within the maximum delay time range, set the i-th dimension input variable x... i (t) The data is reconstructed and segmented according to the acquisition time interval, resulting in k sets of reconstructed working condition data sequences, which are x i (t-1), ..., x i (tk); Step 3-3: Calculate the Pearson coefficient between the k sets of reconstructed operating condition data sequences and the SO2 concentration y(t) respectively; Steps 3-4: Select the delay time corresponding to the set of reconstructed working condition data sequences with the largest Pearson coefficient as the predictor variable x. i The optimal delay time (t) is determined, and the set of reconstructed operating condition data sequences is used as the input variable x. i Reconstructing the data sequence of (t), with input variable x i (t) has a dimension of 1 after reconstruction; Step 3-5: Repeat steps 3-2 to 3-4 to calculate the optimal delay time and reconstructed data sequence for all input variables X. The reconstructed data sequence of all input variables X is combined to obtain the reconstructed data sample, which is used as the input sample for the prediction model. Step 4: Based on the reconstructed data sample set, the Extreme Learning Machine algorithm is used to establish SO2 prediction models for different working conditions. Step 5: Obtain real-time system operation data and input it into the Naive Bayes classification model to obtain the real-time operating condition category of the desulfurization system. Switch to the corresponding SO2 prediction model according to the different operating condition categories. Step 6: Input the real-time operating data of the system into the corresponding prediction model to obtain the predicted value of SO2 emission concentration.

2. The multi-condition prediction method for desulfurization systems based on extreme learning machines according to claim 1, characterized in that, The characteristic parameters marked in step 1 include: boiler load W, gas turbine speed r, oxygen content in boiler outlet flue gas Q, flow rate of boiler outlet flue gas F, temperature of boiler outlet flue gas T, flow rate of boiler outlet flue gas V, primary air fan current I, fuel input M, steam temperature T, and pressure P. The operating condition classification includes six operating conditions: startup, grid connection / power supply, cracking, shutdown, outage, and accident.

3. The multi-condition prediction method for desulfurization systems based on extreme learning machines according to claim 2, characterized in that, The state characteristics of the six operating conditions are as follows: Start-up characteristics: When the boiler starts up, the oxygen content Q of the flue gas at the boiler outlet gradually decreases, the flue gas velocity V at the boiler outlet gradually increases and exceeds 2m / s, and the flue gas temperature T at the boiler outlet gradually increases and exceeds 40℃. Characteristics of grid-connected / power supply operation: gas turbine speed r≈3000r / min, boiler outlet flue gas temperature T≈540℃, primary air fan current I>0, fan starts; Characteristics of the cracking operation: The fuel input M gradually decreases, and the boiler load W gradually decreases to 50%; Characteristics of shutdown operation: The boiler outlet flue gas velocity V, boiler outlet flue gas flow rate F, and boiler outlet flue gas temperature T all gradually decrease, while the boiler outlet flue gas oxygen content Q gradually increases. Characteristics of shutdown conditions: oxygen content Q in boiler outlet flue gas > 18%, flue gas velocity V < 1.8 m / s, flue gas flow rate F less than 15% of rated operating conditions, and flue gas temperature T < 30℃. The characteristics of the accident operation were: the automatic monitoring data of the exhaust outlet exceeded the standard, and some equipment parameters fluctuated abnormally.

4. The multi-condition prediction method for desulfurization systems based on extreme learning machines according to claim 1, characterized in that, Step 2, which establishes Naive Bayes classification models under different working conditions, includes the following specific steps: Step 2-1: Select 10 key variables: boiler load W, gas turbine speed r, oxygen content in boiler outlet flue gas Q, flow rate of boiler outlet flue gas F, temperature of boiler outlet flue gas T, velocity of boiler outlet flue gas V, primary air fan current I, fuel input M, steam temperature T, and pressure P as feature attributes of the classification model. Use the historical data corresponding to the above feature attributes as training samples for the classification model. At the same time, label the working condition type of each training sample according to the state characteristics of each working condition. Step 2-2: Calculate the frequency P(work condition) of each working condition in the training sample. Divide the samples of each feature attribute according to the critical value of the feature attribute when the working condition changes. Calculate the frequency P(variable) of each feature attribute in the training sample after the division. Then calculate the conditional probability P(variable|working condition) of each feature attribute in each working condition after the division. Steps 2-3: Based on Naive Bayes theory, calculate the conditional probability of samples appearing under each feature attribute after partitioning for each working condition: Steps 2-4: Calculate the sum of probabilities of the training samples under each working condition: S(operating condition) = ∑P(operating condition | variable) The working condition type corresponding to the maximum value of S(working condition) is the working condition type to which the sample belongs, and the working condition type number of the sample is output.

5. The multi-condition prediction method for desulfurization systems based on extreme learning machines according to claim 4, characterized in that, The critical values ​​of the characteristic attributes in step 2-2 when the operating conditions change include: Boiler load W: {W≤20%, 20%<W<50%, W≥50%} Gas turbine speed r: {r≤2000, 2000 <r<3500,r≥3500} Oxygen content Q in boiler outlet flue gas: {Q≤5%, 5%<Q<10%, Q≥10%} Boiler outlet flue gas flow rate F: {F<0.15*F, F≥0.15*F} Boiler outlet flue gas temperature T: {T≤40℃, 40℃<T<540℃, T≥540℃} Boiler outlet flue gas velocity V: {V<1.8m / s, V≥1.8m / s} Primary fan current I: {I<0.15*I, I≥0.15*I} Fuel input M: {M < 0.15 * M, M ≥ 0.15 * M} Steam temperature T′: {T′≤400℃, 400℃<T′<600℃, T′≥600℃} Steam pressure P: {P≤10kPa, 10kPa<P<25kPa, P≥25kPa}.

6. The multi-condition prediction method for desulfurization systems based on extreme learning machine according to claim 1, characterized in that, The formula for calculating the Pearson coefficient in step 3-3 is: Among them, X i To reconstruct the parameters of the operating condition data sequence, Y i For the SO2 concentration at the outlet, To reconstruct the parameter averages of the operating condition data series, This represents the average SO2 concentration at the outlet.

7. The multi-condition prediction method for desulfurization systems based on extreme learning machine according to claim 1, characterized in that, Step 4, which establishes SO2 prediction models for different operating conditions, includes the following specific steps: Step 4-1: Obtain a reconstructed data sample set under a certain working condition, and use the historical data of the feature variables in the reconstructed data sample set under this working condition as the input sample matrix of the extreme learning machine SO2 prediction model; Step 4-2: Establish a single-layer feedforward neural network for the samples of the characteristic variables related to the working condition. The single-layer feedforward neural network prediction model includes 6 input layers, L hidden layer nodes and 1 output layer node. Step 4-3: Randomly generate the weight matrices W of the input layer and hidden layer of a single-layer feedforward neural network. i This yields the hidden layer output matrix H (6 rows, L columns) and the output weight matrix β (L rows, 1 column). Step 4-4: Introduce an L2 regularization term into the error function of the SO2 prediction model to obtain the minimum error function of the model: Where C is the regularization coefficient, T is the training target, and || is the Firobenius norm of the matrix elements; Steps 4-5: Calculate the weight matrix β of the Extreme Learning Machine prediction model. * : The SO2 prediction model under this operating condition is as follows: Y=XW i β * Where X is the input matrix and Y is the predicted value of SO2 emission concentration; Steps 4-6: Repeat steps 4-1 to 4-5 to establish SO2 prediction models for different operating conditions.

8. The multi-condition prediction method for desulfurization systems based on extreme learning machines according to claim 7, characterized in that, In step 4-1, the input sample matrices and their corresponding feature variables under different working conditions are as follows: The input sample matrix under startup conditions is: X=[Q,V,T,ρ,F′,C]; The input sample matrix under grid-connected / functional operating conditions is: X=[r,T,I,ρ,F′,C]; The input sample matrix under the decoupling condition is: X=[Q,M,W,ρ,F′,C]; The input sample matrix under shutdown conditions is: X=[Q,V,T,ρ,F′,C]; Where ρ is the density of the absorber slurry, F′ is the flow rate of the slurry circulation pump, C is the sulfur dioxide concentration at the boiler outlet, W is the boiler load, r is the speed of the gas turbine, Q is the oxygen content of the flue gas at the boiler outlet, T is the temperature of the flue gas at the boiler outlet, V is the flue gas velocity at the boiler outlet, I is the current of the primary air fan, and M is the fuel input.

Images