Methods, apparatus and equipment for predicting the amount of hydrated lime input in a two-stage combined desulfurization furnace

Abstract

This invention provides a method, apparatus, and equipment for predicting the amount of hydrated lime input outside the furnace in a two-stage combined desulfurization process, relating to the field of desulfurization control technology. The method includes: acquiring desulfurization operating parameters, including coal quality characteristics, the amount of limestone input in the boiler, the actual amount of hydrated lime input in the desulfurization tower, the actual SO2 concentration at the boiler outlet, and the actual SO2 concentration at the desulfurization tower outlet; using the desulfurization operating parameters as input, determining the desulfurization operating condition category corresponding to the desulfurization operating parameters through a pre-trained desulfurization operating condition prediction model; and calling the hydrated lime input prediction model corresponding to the determined desulfurization operating condition category, using the amount of limestone input in the boiler, the actual SO2 concentration at the boiler outlet, and the pre-determined target SO2 concentration at the desulfurization tower outlet as input, and predicting the target amount of hydrated lime input in the desulfurization tower through the hydrated lime input prediction model. This invention is based on modeling for combined in-furnace and out-of-furnace desulfurization, realizing the prediction of the amount of hydrated lime input outside the furnace.

Description

Technical Field

[0001] This invention relates to the field of desulfurization control technology, specifically to a method for predicting the amount of hydrated lime input in a two-stage combined desulfurization furnace, a device for predicting the amount of hydrated lime input in a two-stage combined desulfurization furnace, and a terminal device. Background Technology

[0002] Traditional two-stage desulfurization for circulating fluidized bed (CFB) boilers involves first injecting calcium into the boiler furnace for in-furnace desulfurization, followed by further desulfurization of the flue gas using a flue gas treatment device such as a desulfurization tower. The in-furnace calcium injection desulfurization + semi-dry flue gas desulfurization technology first involves in-furnace calcium injection, i.e., injecting limestone into the CFB boiler furnace to reduce some SO2 concentration, followed by external semi-dry desulfurization. Currently, external semi-dry desulfurization uses hydrated lime as the absorbent. However, in actual desulfurization processes, the addition of hydrated lime is mostly manual, leading to large fluctuations in SO2 at the desulfurization tower outlet. This can result in a low hourly average SO2 level, inaccurate control of desulfurizing agent usage, and waste of desulfurizing agent. During load changes, the hourly average may exceed the standard, affecting economic operation and leading to energy waste. Summary of the Invention

[0003] The purpose of this invention is to provide a method for predicting the amount of hydrated lime added outside the furnace in a two-stage combined desulfurization system, a device for predicting the amount of hydrated lime added outside the furnace in a two-stage combined desulfurization system, and a terminal device, so as to solve the problem that the existing technology is difficult to accurately predict the amount of hydrated lime added outside the furnace.

[0004] To achieve the above objectives, the first aspect of the present invention provides a method for predicting the amount of hydrated lime input in a two-stage combined desulfurization furnace, characterized in that it includes:

[0005] Real-time acquisition of desulfurization operation parameters characterizing the desulfurization status of the boiler, including at least coal quality characteristic parameters, limestone input in the boiler, actual lime input in the desulfurization tower, actual SO2 concentration at the boiler outlet and actual SO2 concentration at the desulfurization tower outlet.

[0006] Using the desulfurization operating parameters as input, a pre-trained desulfurization condition prediction model determines the desulfurization condition category corresponding to the desulfurization operating parameters.

[0007] The lime input prediction model corresponding to the determined desulfurization operating condition category is called. The lime input in the boiler, the actual SO2 concentration at the boiler outlet, and the predetermined SO2 target concentration at the desulfurization tower outlet are used as inputs. The lime input prediction model predicts the target lime input for the desulfurization tower.

[0008] The desulfurization operating condition prediction model is obtained by training a preset machine learning algorithm with historical desulfurization operating parameters, and the hydrated lime input prediction model is obtained by training a neural network model with historical desulfurization operating parameters belonging to the corresponding desulfurization operating condition category.

[0009] Optionally, the preset machine learning algorithm is a support vector machine algorithm, and the method further includes:

[0010] Obtain the historical desulfurization operation parameter dataset within a specified time interval;

[0011] Feature filtering is performed on the obtained historical desulfurization operation parameter dataset. The filtered historical desulfurization operation parameter data is used as input, and the desulfurization operation condition category is used as output to train an initial desulfurization operation condition prediction model based on the support vector machine algorithm.

[0012] The model parameters of the initial desulfurization condition prediction model based on the support vector machine algorithm are optimized by the particle swarm optimization algorithm to obtain a pre-trained desulfurization condition prediction model.

[0013] Optionally, the model parameters of the initial desulfurization condition prediction model based on the support vector machine algorithm are optimized using a particle swarm optimization algorithm, including:

[0014] Initial particles are randomly generated using the penalty factor C and kernel parameter g of the initial desulfurization condition prediction model based on the support vector machine algorithm. The particle swarm size, initial position, and initial velocity of the initial particles are set.

[0015] The fitness value of each particle is calculated using the K-fold cross-validation algorithm. The fitness value of each particle is compared with the global fitness value of the current particle. The position and velocity of the particles are updated according to the comparison results until the convergence condition is reached. The optimal penalty factor C and the optimal kernel parameter g of the initial desulfurization condition prediction model based on the support vector machine algorithm are obtained.

[0016] Optionally, feature filtering is performed on the acquired historical desulfurization operation parameter dataset, including:

[0017] In the acquired historical desulfurization operation parameter dataset, historical desulfurization operation parameters in which the actual SO2 concentration at the boiler outlet is lower than the first SO2 concentration threshold, or where the actual SO2 concentration at the boiler outlet is lower than the first SO2 concentration threshold and the actual SO2 concentration at the desulfurization tower outlet is lower than the second SO2 concentration threshold, are identified as abnormal historical desulfurization operation parameters.

[0018] Delete all abnormal historical desulfurization operating parameters from the obtained historical desulfurization operating parameter dataset.

[0019] Optionally, the neural network model is a BP neural network model, and the method further includes:

[0020] Obtain the historical desulfurization operation parameter dataset within a specified time interval;

[0021] Using the historical desulfurization operation parameter dataset as input, the pre-trained desulfurization operation condition prediction model determines the desulfurization operation condition category corresponding to each historical desulfurization operation parameter.

[0022] A BP neural network model is trained based on historical desulfurization operating parameters belonging to different desulfurization operating condition categories to obtain a prediction model for the amount of hydrated lime input corresponding to each desulfurization operating condition category.

[0023] Optionally, a BP neural network model is trained based on historical desulfurization operating parameters belonging to different desulfurization operating condition categories to obtain a prediction model for hydrated lime input corresponding to each desulfurization operating condition category, including:

[0024] Obtain the amount of limestone input in the boiler and the actual SO2 concentration at the boiler outlet from all historical desulfurization operation parameters, and construct a training sample set for the BP neural network model. The number of training sample sets corresponds one-to-one with the number of determined desulfurization operating condition categories.

[0025] Using the obtained training sample sets and the predetermined SO2 target concentration at the desulfurization tower outlet as inputs, and the target amount of slaked lime in the desulfurization tower as outputs, the BP neural network model is trained to obtain a slaked lime input prediction model that corresponds to different desulfurization operating conditions.

[0026] In each training sample set, the amount of limestone fed into the boiler and the actual SO2 concentration at the boiler outlet belong to the same desulfurization condition category, while the amount of limestone fed into the boiler and the actual SO2 concentration at the boiler outlet in different training sample sets belong to different desulfurization condition categories.

[0027] Optionally, the desulfurization operating condition categories include: a first desulfurization operating condition category, a second desulfurization operating condition category, and a third desulfurization operating condition category; the method further includes:

[0028] The desulfurization operating parameters corresponding to the desulfurization operating parameters where the actual SO2 concentration at the boiler outlet is higher than the third SO2 concentration threshold and the actual SO2 concentration at the desulfurization tower outlet is lower than the fourth SO2 concentration threshold are defined as the first desulfurization operating condition category.

[0029] The desulfurization operating parameters that are determined to have an actual SO2 concentration at the boiler outlet higher than the third SO2 concentration threshold and an actual SO2 concentration at the desulfurization tower outlet higher than the fifth SO2 concentration threshold are classified as the second desulfurization operating condition category.

[0030] The desulfurization operating parameters corresponding to the desulfurization operating parameters where the actual SO2 concentration at the boiler outlet is lower than the sixth SO2 concentration threshold and the actual SO2 concentration at the desulfurization tower outlet is lower than the fourth SO2 concentration threshold are classified as the third desulfurization operating condition category.

[0031] A second aspect of the present invention provides a device for predicting the amount of hydrated lime input in a two-stage combined desulfurization furnace, comprising:

[0032] The data acquisition module is configured to acquire desulfurization operation parameters that characterize the desulfurization status of the boiler in real time. The desulfurization operation parameters include at least coal quality characteristic parameters, limestone input in the boiler, actual lime input in the desulfurization tower, actual SO2 concentration at the boiler outlet, and actual SO2 concentration at the desulfurization tower outlet.

[0033] The desulfurization operating condition prediction module is configured to use the desulfurization operating parameters as input and a pre-trained desulfurization operating condition prediction model to determine the desulfurization operating condition category corresponding to the desulfurization operating parameters.

[0034] The hydrated lime input prediction module is configured to call the hydrated lime input prediction model corresponding to the determined desulfurization operating condition category, and use the limestone input in the boiler, the actual SO2 concentration at the boiler outlet and the predetermined SO2 target concentration at the desulfurization tower outlet as inputs to predict the target hydrated lime input of the desulfurization tower through the hydrated lime input prediction model.

[0035] The desulfurization operating condition prediction model is obtained by training a preset machine learning algorithm with historical desulfurization operating parameters, and the hydrated lime input prediction model is obtained by training a neural network model with historical desulfurization operating parameters belonging to the corresponding desulfurization operating condition category.

[0036] A third aspect of the present invention provides a terminal device, comprising:

[0037] At least one processor;

[0038] A memory connected to the at least one processor;

[0039] The memory stores instructions that can be executed by the at least one processor, which executes the above-described method for predicting the amount of hydrated lime added outside the two-stage combined desulfurization furnace.

[0040] A fourth aspect of the present invention provides a machine-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the above-described method for predicting the amount of quicklime added outside the two-stage combined desulfurization furnace.

[0041] The embodiments provided by the present invention have the following beneficial effects:

[0042] By modeling based on in-furnace and out-of-furnace co-desulfurization, the desulfurization operating parameters are first identified to determine the desulfurization operating condition category corresponding to the current desulfurization operating parameters. Then, the slaked lime input prediction model corresponding to the current desulfurization operating condition category is called. Taking the parameters with strong correlation to the slaked lime input and the predetermined SO2 target concentration at the desulfurization tower outlet as input, the slaked lime input prediction model predicts the target slaked lime input of the desulfurization tower, thereby enabling the prediction of the input of slaked lime outside the furnace. This solves the problem of inaccurate control of desulfurizing agent dosage in the existing technology. At the same time, by classifying and identifying the operating parameters into operating condition categories and establishing a slaked lime input prediction model corresponding to the operating condition category, the prediction accuracy is further improved.

[0043] Other features and advantages of the embodiments or implementations of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0044] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0045] Figure 1 The schematic diagram illustrates a flowchart of a preferred embodiment of the present invention for predicting the amount of hydrated lime to be added to a two-stage combined desulfurization furnace.

[0046] Figure 2 The diagram illustrates the classification results of a preferred embodiment of the present invention using a soft-margin SVM.

[0047] Figure 3 The schematic diagram illustrates the SVM parameter optimization flowchart of a preferred embodiment of the present invention;

[0048] Figure 4 The schematic diagram illustrates the PSO-SVM model training flowchart of a preferred embodiment of the present invention;

[0049] Figure 5 The BP neural network structure of a preferred embodiment of the present invention is illustrated schematically;

[0050] Figure 6 This diagram illustrates a preferred embodiment of the lime input prediction model of the present invention.

[0051] Figure 7 The diagram illustrates a schematic block diagram of a two-stage combined desulfurization furnace external hydrated lime input prediction device according to a preferred embodiment of the present invention. Detailed Implementation

[0052] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0053] like Figure 1 As shown, the first aspect of this embodiment provides a method for predicting the amount of hydrated lime input in a two-stage combined desulfurization furnace, characterized by comprising:

[0054] S100. Real-time acquisition of desulfurization operation parameters characterizing the desulfurization status of the boiler. The desulfurization operation parameters include at least coal quality characteristic parameters, limestone input in the boiler, actual lime input in the desulfurization tower, actual SO2 concentration at the boiler outlet and actual SO2 concentration at the desulfurization tower outlet.

[0055] S200. Using desulfurization operating parameters as input, the pre-trained desulfurization operating condition prediction model determines the desulfurization operating condition category corresponding to the desulfurization operating parameters.

[0056] S300, call and determine the lime input prediction model corresponding to the desulfurization operating condition category, take the lime input in the boiler, the actual SO2 concentration at the boiler outlet and the predetermined SO2 target concentration at the desulfurization tower outlet as input, and predict the target lime input of the desulfurization tower through the lime input prediction model.

[0057] The desulfurization operating condition prediction model is obtained by training a preset machine learning algorithm with historical desulfurization operating parameters, while the hydrated lime input prediction model is obtained by training a neural network model with historical desulfurization operating parameters belonging to the corresponding desulfurization operating condition category.

[0058] Thus, this embodiment, through modeling based on in-furnace and out-of-furnace coordinated desulfurization, first identifies the desulfurization operating condition category of the acquired desulfurization operating parameters to determine the desulfurization operating condition category corresponding to the current desulfurization operating parameters. Then, it calls the slaked lime input prediction model corresponding to the current desulfurization operating condition category. Taking the parameters in the current desulfurization operating parameters that are strongly correlated with the slaked lime input and the predetermined SO2 target concentration at the desulfurization tower outlet as inputs, the slaked lime input prediction model predicts the target slaked lime input of the desulfurization tower, thereby enabling the prediction of the input of slaked lime outside the furnace. This solves the problem of inaccurate control of desulfurizing agent dosage in the prior art. At the same time, by classifying and identifying the operating parameters into operating condition categories and establishing a slaked lime input prediction model corresponding to the operating condition category, the prediction accuracy is further improved.

[0059] Specifically, in step S100, the desulfurization operating parameters can be directly obtained in real time through the circulating fluidized bed boiler control system. Among these, coal quality characteristic parameters include coal quality and Ca / S ratio. The SO2 concentration at the boiler outlet needs to be determined numerically based on different coal qualities, because different coal qualities result in different Ca / S ratios, thus requiring different standards to ensure that the final SO2 concentration at the desulfurization tower outlet does not exceed the limit.

[0060] In step S200, the desulfurization operating condition prediction model is obtained by training a preset machine learning algorithm using historical desulfurization operating parameters. The historical desulfurization operating parameters can be desulfurization operating parameters within a specified time period.

[0061] In this embodiment, the preset machine learning algorithm is the Support Vector Machine (SVM) algorithm, and the steps for constructing the desulfurization condition prediction model include:

[0062] Obtain the historical desulfurization operation parameter dataset within a specified time interval;

[0063] Feature filtering is performed on the obtained historical desulfurization operation parameter dataset. The filtered historical desulfurization operation parameter data is used as input, and the desulfurization operation condition category is used as output to train an initial desulfurization operation condition prediction model based on the support vector machine algorithm.

[0064] The model parameters of the initial desulfurization condition prediction model based on the support vector machine algorithm are optimized by the particle swarm optimization algorithm to obtain a pre-trained desulfurization condition prediction model.

[0065] To improve the prediction accuracy of desulfurization operating conditions, a historical desulfurization operating parameter dataset can be preferably constructed by acquiring historical desulfurization operating parameters within a recent time interval. This ensures that the selected desulfurization operating parameters better reflect the current operating status of the circulating fluidized bed boiler. Simultaneously, to improve the training effect of the initial desulfurization operating condition prediction model, this embodiment also performs feature filtering on the acquired historical desulfurization operating parameter dataset to exclude outlier data, thus avoiding any impact from abnormal data on the training results. After feature filtering of the historical desulfurization operating parameter data, the filtered historical desulfurization operating parameter data is used as input, and the desulfurization operating condition category is used as output to train the SVM-based initial desulfurization operating condition prediction model.

[0066] The process involves feature filtering of the acquired historical desulfurization operation parameter dataset, including: identifying historical desulfurization operation parameters in the acquired historical desulfurization operation parameter dataset where the actual SO2 concentration at the boiler outlet is lower than the first SO2 concentration threshold, or where the actual SO2 concentration at the boiler outlet is lower than the first SO2 concentration threshold and the actual SO2 concentration at the desulfurization tower outlet is lower than the second SO2 concentration threshold as abnormal historical desulfurization operation parameters; and deleting all abnormal historical desulfurization operation parameters from the acquired historical desulfurization operation parameter dataset.

[0067] To ensure SO2 emissions meet standards, power plant workers often add excessive amounts of desulfurizing agent both inside and outside the furnace. This excessive addition compromises desulfurization efficiency, and the resulting operating parameters are not conducive to system modeling. Therefore, data showing excessively low SO2 concentrations at the boiler outlet need to be discarded. Simultaneously, data showing low SO2 concentrations at both the desulfurization tower inlet and outlet indicates excessive desulfurizing agent addition inside the furnace, preventing the external addition from working in synergy with the internal process. Such data also needs to be discarded. The remaining data represents the SO2 concentrations at both the desulfurization tower inlet (i.e., boiler outlet) and outlet maintained within a suitable range. The first and second SO2 concentration thresholds can be set according to specific circumstances; typically, an SO2 concentration of 10 mg / m³ is suitable. 3 The following can be considered as low emissions, at 25 mg / m³ 3 The above are considered high emissions.

[0068] After feature filtering of historical desulfurization operation parameter data, the filtered historical desulfurization operation parameter data can be used as input, and the desulfurization operation condition category can be used as output to train the initial desulfurization operation condition prediction model based on SVM.

[0069] Specifically, the SVM algorithm can be simply described as a binary classification problem: finding an optimal hyperplane that satisfies the classification requirements. Its core concept is: for non-linearly separable data inputs, a suitable function mapping is chosen to solve the problem, and then the sample space is mapped to a high-dimensional feature space to achieve linear separability. Furthermore, the introduction of a kernel function extends linear separability to non-linear separability.

[0070] The SVM algorithm includes soft-margin and hard-margin algorithms. This implementation preferably uses the soft-margin SVM algorithm, such as... Figure 2 The image shows the classification results of a soft-margin SVM, where H is the optimal classification hyperplane, and H1 and H2 are the points on either side of the optimal classification hyperplane with a distance of 1 / ||ω|| from H. 2 In the plane, the distance between H1 and H2 is the classification margin of the classifier, γ = 2 / ||ω||. 2 For hard-margin SVM, all training samples exist only in the space outside of H1 and H2, and are accurately classified into two classes by H, such as Figure 2 The samples in the figure are not marked with a cross, while soft-margin SVM allows the classification function to make mistakes in classifying a subset of samples. These errors will occur between H1 and H2, and some samples may even appear in the regions where outlier samples should appear, as shown by the samples marked with a cross in the figure.

[0071] In this embodiment, the classification objective formula of the soft-margin SVM algorithm is as follows:

[0072]

[0073] Where ω is the normal vector, C is the penalty parameter, ξ is the slack variable, and b is the bias.

[0074] Equation (1) encompasses two requirements of soft-margin support vector machines: firstly, to maximize the classification margin, which is expressed by 1 / ||ω|| in the formula. 2 Partially reflected, 1 / 2||ω|| 2 The smaller the classification interval γ = 2 / ||ω|| 2 The larger the value, the better; secondly, it is desirable to minimize the number of misclassified points, which is determined by the formula. Partial implementation, each misclassified point requires an investment of Cξ. i The cost. In the above equation, by introducing the Lagrange function and using the duality principle, equation (1) is transformed into the following equation:

[0075]

[0076] Where, α i For Lagrange multipliers, (x i x j ) represents x i With x j The inner product, y i ,y j ∈(1,-1).

[0077] The kernel function K(x) of formula (3) i ,x j ) Replace (x i ,x j The optimization objective of SVM is obtained as shown in formula (4):

[0078]

[0079]

[0080] Formula (4) is the final expression of the SVM algorithm in this embodiment.

[0081] The advantages of introducing kernel functions include: First, kernel functions help SVM project samples into a high-dimensional space, transforming nonlinear classification problems that SVM cannot solve into linear classification problems; second, classification in high-dimensional space requires calculating the inner product of samples, and the existence of kernel functions connects high-dimensional inner product operations with low-dimensional inner product operations, allowing the calculation of high-dimensional inner products to be achieved by calculating low-dimensional inner products, effectively avoiding the "curse of dimensionality" when calculating high-dimensional samples. In this embodiment, the kernel function is preferably a universal Gaussian radial basis function, whose function expression is:

[0082]

[0083] Here, g is the kernel parameter, which has a significant impact on the classification ability of SVM.

[0084] Because the optimal combination of parameters C and g for SVM classification results varies depending on the classification object, and the inherent structure of the SVM algorithm cannot meet its own requirement of finding the most suitable penalty parameter C and kernel parameter g, this implementation method further optimizes the SVM algorithm by using particle swarm optimization (PSO) to optimize the penalty parameter C and kernel parameter g when constructing the classification model using the SVM algorithm. Therefore, the model parameters of the initial desulfurization condition prediction model based on the support vector machine algorithm are optimized using particle swarm optimization, including:

[0085] Initial particles are randomly generated using the penalty factor C and kernel parameter g of the initial desulfurization condition prediction model based on the support vector machine algorithm. The particle swarm size, initial position, and initial velocity of the initial particles are set.

[0086] The fitness value of each particle is calculated using the K-fold cross-validation algorithm. The fitness value of each particle is compared with the global fitness value of the current particle. The position and velocity of the particles are updated according to the comparison results until the convergence condition is reached. The optimal penalty factor C and the optimal kernel parameter g of the initial desulfurization condition prediction model based on the support vector machine algorithm are obtained.

[0087] Specifically, let

[0088] In equation (6), the vector “x” represents the structural information of the SVM. In this embodiment, it is defined as the penalty factor C of the hidden layer and the kernel parameter g.

[0089] Using x as the input to the PSO algorithm, initialize all x values, setting the penalty factor C to a range of 1 to 20, and the kernel parameter g to a range of 0 to 20. Define a 2-dimensional search space consisting of 50 particles, then the position of the i-th particle is defined as: x i =(x i1 ,x i2 ), i = 1, 2, ..., 20; the velocity of the i-th particle is defined as: v i =(v i1 ,v i2 ), i = 1, 2, ..., 20. Let the best position found so far by the i-th particle be the individual extreme value, denoted as p. i The optimal position found so far by the entire particle swarm is the global extremum, denoted as p. g , where p i p g Both are 2-dimensional vectors. Finding these two optimal values ​​p...i p g Then, the particle updates its velocity and position using the following formula:

[0090]

[0091] In formula (7): c1 and c2 are the inertia factor and learning factor, also known as the acceleration constant, which reflect the intensity of information exchange between particles. Usually, c1 = c2 = 2. r1 and r2 are uniform random numbers in the range [0,1]. i p represents the optimal position found so far for the i-th particle, i.e., the individual extreme value. g The optimal position found so far in the entire particle swarm, i.e., the global extremum. w is called the inertia factor, and its value is non-negative. A larger value indicates stronger global optimization capability, while a smaller value indicates weaker global optimization capability. By adjusting the value of w, the global and local optimization performance can be adjusted. Where w = (w... ini -w end (G) k -g) / G k -w end G k The maximum number of iterations is 20, w ini Let w be the initial inertial weights. end The inertial weights at the maximum evolutionary generation are typically represented by the weight w. ini Take 0.9, w end Let's take 0.4. By using the above method, we can finally obtain the optimal value of x, and thus determine the structural parameters of the SVM.

[0092] like Figure 3 As shown, in this embodiment, the specific steps for optimizing SVM parameters using the particle swarm optimization algorithm are as follows:

[0093] S210 and PSO parameter initialization: Initialize particles and set particle swarm parameters. Initial particles are randomly generated by SVM parameters C and g. Set particle swarm size, maximum number of iterations tmax, inertia weight w, and acceleration coefficients c1 and c2.

[0094] S220. Fitness Calculation. The fitness value of each particle is evaluated, i.e., the average classification accuracy of K-fold cross-validation. The optimal point is the particle's global fitness, and the maximum value is the particle's global fitness. For example, the training sample set is divided into k parts. One part is reserved as validation data for the initial desulfurization condition prediction model based on SVM, and the other k-1 parts are used to train the SVM-based desulfurization condition prediction model. Cross-validation is repeated k times, with each training sample set being validated once. The average prediction result from the k iterations is used as the evaluation value of the prediction accuracy of the initial desulfurization condition prediction model based on SVM. The K-fold cross-validation algorithm is existing technology and is not limited here.

[0095] S230, Update. Update the velocity and position of each particle. Compare the best fitness value of each particle with the best fitness value of the current particle, i.e., the global fitness value of the current particle. If the current value is better, use the current value as the particle's individual best point or global best point to update the particle.

[0096] S240. Determine if the termination condition is met. For example, the termination condition can be whether the maximum number of iterations has been reached. When the maximum number of iterations is reached, the evolution process stops. Otherwise, continue to step S230 until the termination condition is met.

[0097] S250. Determine the parameters. When the number of iterations reaches its maximum and the stopping criterion is met, obtain the optimal parameters C and g. End the training and validation process and establish the PSO-SVM model, i.e., the desulfurization condition prediction model.

[0098] For example, such as Figure 4 As shown, based on the established feature vectors (i.e., the filtered historical desulfurization operation parameter data), 1000 sets of data were selected, with 800 sets as the training set and 200 sets as the test set. The data underwent preprocessing such as normalization. During training, the optimal parameters g and C were obtained iteratively using the PSO algorithm. A base PSO-SVM model was then established based on these optimal parameters. The PSO-SVM model was trained using the training set, and the accuracy of the model output was verified using the test set. If the accuracy did not meet the requirements, g and C were reselected until the model output accuracy met the requirements, thus obtaining the final desulfurization operating condition prediction model.

[0099] In this embodiment, the desulfurization operating conditions are divided into three categories: a first desulfurization operating condition category, a second desulfurization operating condition category, and a third desulfurization operating condition category. The method further includes: determining the desulfurization operating parameters whose actual SO2 concentration at the boiler outlet is higher than the third SO2 concentration threshold and whose actual SO2 concentration at the desulfurization tower outlet is lower than the fourth SO2 concentration threshold as the first desulfurization operating condition category; determining the desulfurization operating parameters whose actual SO2 concentration at the boiler outlet is higher than the third SO2 concentration threshold and whose actual SO2 concentration at the desulfurization tower outlet is higher than the fifth SO2 concentration threshold as the second desulfurization operating condition category; and determining the desulfurization operating parameters whose actual SO2 concentration at the boiler outlet is lower than the sixth SO2 concentration threshold and whose actual SO2 concentration at the desulfurization tower outlet is lower than the fourth SO2 concentration threshold as the third desulfurization operating condition category.

[0100] The third SO2 concentration threshold and the fifth SO2 concentration threshold can be the same or different, and the fourth SO2 concentration threshold and the sixth SO2 concentration threshold can be the same or different, and their values ​​can be set according to specific circumstances. In this embodiment, the first desulfurization condition category is the desulfurization condition category where the actual SO2 concentration at the boiler outlet is high and the actual SO2 concentration at the desulfurization tower outlet is low; the second desulfurization condition category is the desulfurization condition category where the actual SO2 concentration at the boiler outlet is high and the actual SO2 concentration at the desulfurization tower outlet is also high; and the third desulfurization condition category is the desulfurization condition category where the actual SO2 concentration at the boiler outlet is low and the actual SO2 concentration at the desulfurization tower outlet is also low.

[0101] In this implementation, the neural network model is a BP neural network model, and the training method for the lime input prediction model includes:

[0102] Obtain a dataset of historical desulfurization operating parameters within a specified time interval; using the dataset as input, determine the desulfurization operating condition category corresponding to each historical desulfurization operating parameter through a pre-trained desulfurization operating condition prediction model; train a BP neural network model based on historical desulfurization operating parameters belonging to different desulfurization operating condition categories to obtain a slaked lime input prediction model that corresponds one-to-one with different desulfurization operating condition categories.

[0103] Among them, the BP neural network is a multi-layer feedforward neural network that propagates the signal forward and the error backward. A three-layer BP neural network model can effectively simulate arbitrary nonlinear problems. Therefore, in practical applications, a three-layer neural network is typically used for training and prediction. For example... Figure 5 The diagram shows a three-layer BP neural network structure, consisting of an input layer, a hidden layer, and an output layer.

[0104] Figure 5 In this model, the number of neurons in the input layer is P, the number of neurons in the hidden layer is L, and the number of neurons in the output layer is M. Figure 6 The model for predicting the amount of hydrated lime added outside the furnace in a CFB boiler is as follows: the main factors affecting the amount of hydrated lime added are the amount of limestone added inside the furnace, the actual SO2 concentration at the boiler outlet, and the target SO2 concentration at the desulfurization tower outlet. Therefore, the number of input neurons in the BP neural network is set to P = 3; the model outputs the amount of hydrated lime added outside the furnace, so the number of output neurons is set to M = 1; the number of hidden layer neurons is set to L = 7 according to the formula L = 2 * P + 1, and the hidden layer structure is set to a single layer.

[0105] The calculation formula from the input layer to the output layer is as follows:

[0106]

[0107] The calculation formula from the hidden layer to the output layer is:

[0108]

[0109] Where b1 and b2 are thresholds; ω pl and ω lm The connection weights are used; the hidden layer output is f1(S). l f1 is the activation function of this layer, usually chosen as the sigmoid function; the output layer output is f2(S m f2 is the output function of the output layer, usually taken as the purelin function. The specific algorithm of the BP neural network is existing technology and will not be elaborated here.

[0110] Since the amount of hydrated lime input varies under different desulfurization conditions, in order to improve the accuracy of hydrated lime input prediction, this implementation method uses historical desulfurization operation parameters after desulfurization condition classification as training samples, and trains BP neural network models for different desulfurization conditions to obtain hydrated lime input prediction models that correspond one-to-one with desulfurization condition types.

[0111] The process involves training a BP neural network model based on historical desulfurization operating parameters belonging to different desulfurization condition categories. This yields a prediction model for slaked lime input corresponding to each desulfurization condition category. The process includes: acquiring the amount of limestone input in the boiler and the actual SO2 concentration at the boiler outlet from all historical desulfurization operating parameters; constructing a training sample set for the BP neural network model, with the number of training sample sets corresponding to the number of determined desulfurization condition categories; training the BP neural network model using each training sample set and the pre-determined target SO2 concentration at the desulfurization tower outlet as input, and the target slaked lime input for the desulfurization tower as output, to obtain a prediction model for slaked lime input corresponding to each desulfurization condition category. In each training sample set, the amount of limestone input in the boiler and the actual SO2 concentration at the boiler outlet belong to the same desulfurization condition category, while in different training sample sets, these amounts belong to different desulfurization condition categories.

[0112] For example, using historical desulfurization operating parameters belonging to the first desulfurization condition as the first training set, historical desulfurization operating parameters belonging to the second desulfurization condition as the second training set, and historical desulfurization operating parameters belonging to the third desulfurization condition as the third training set, the BP neural network model is trained using the first, second, and third training sets respectively, resulting in the first, second, and third slaked lime input prediction models. When predicting the slaked lime input, the desulfurization condition prediction model first identifies the desulfurization condition category of the real-time collected desulfurization operating parameters. For example, if the current desulfurization operating parameters are identified as belonging to the second desulfurization condition, the second slaked lime input prediction model is called. Using the current desulfurization operating parameters, including the amount of limestone input in the boiler, the actual SO2 concentration at the boiler outlet, and the pre-determined target SO2 concentration at the desulfurization tower outlet, as inputs, the second slaked lime input prediction model predicts the amount of slaked lime input in the desulfurization tower under the current condition. The target SO2 concentration at the desulfurization tower outlet is the expected value of the SO2 concentration at the desulfurization tower outlet, which can be set to a fixed value. In this way, by training the BP neural network with historical desulfurization operating parameters belonging to different operating conditions, a prediction model for the slaked lime input amount corresponding to each operating condition can be obtained. When predicting the slaked lime input amount, only the expected value of the SO2 concentration at the desulfurization tower outlet needs to be determined, and the slaked lime input amount for the desulfurization tower under the current operating condition can be obtained through the pre-trained slaked lime input amount prediction model.

[0113] like Figure 7 As shown, a second aspect of the present invention provides a device for predicting the amount of quicklime input in a two-stage combined desulfurization furnace, comprising:

[0114] The data acquisition module is configured to acquire desulfurization operation parameters that characterize the desulfurization status of the boiler in real time. The desulfurization operation parameters include at least coal quality characteristic parameters, limestone input in the boiler, actual lime input in the desulfurization tower, actual SO2 concentration at the boiler outlet, and actual SO2 concentration at the desulfurization tower outlet.

[0115] The desulfurization operating condition prediction module is configured to take the desulfurization operating parameters as input and determine the desulfurization operating condition category corresponding to the desulfurization operating parameters through a pre-trained desulfurization operating condition prediction model.

[0116] The hydrated lime input prediction module is configured to call the hydrated lime input prediction model corresponding to the determined desulfurization operating condition category. The model uses the limestone input in the boiler, the actual SO2 concentration at the boiler outlet, and the predetermined SO2 target concentration at the desulfurization tower outlet as inputs to predict the target hydrated lime input for the desulfurization tower.

[0117] The desulfurization operating condition prediction model is obtained by training a preset machine learning algorithm with historical desulfurization operating parameters, while the hydrated lime input prediction model is obtained by training a neural network model with historical desulfurization operating parameters belonging to the corresponding desulfurization operating condition category.

[0118] The specific limitations of each functional module in the aforementioned two-stage combined desulfurization furnace external slaked lime input prediction device can be found in the limitations of the two-stage combined desulfurization furnace external slaked lime input prediction method described above, and will not be repeated here. Each module in the aforementioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0119] A third aspect of the present invention provides a terminal device, comprising:

[0120] At least one processor;

[0121] Memory, connected to at least one processor;

[0122] The memory stores instructions that can be executed by at least one processor, which executes the above-described method for predicting the amount of quicklime added outside the two-stage combined desulfurization furnace.

[0123] Understandably, the processor here possesses numerical computation and logical operation capabilities, and at least includes a central processing unit (CPU) with data processing capabilities, random access memory (RAM), read-only memory (ROM), various I / O ports, and an interrupt system. The processor contains a kernel that retrieves the corresponding program units from memory. One or more kernels can be configured, and the aforementioned methods can be implemented by adjusting kernel parameters. The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and includes at least one memory chip.

[0124] A fourth aspect of the present invention provides a machine-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the above-described method for predicting the amount of quicklime added outside the two-stage combined desulfurization furnace.

[0125] In summary, this implementation method uses an SVM model to classify desulfurization operating parameters into different desulfurization condition categories, identifies data that best reflects the synergistic desulfurization effect between the furnace and external furnace, and optimizes the C and g values ​​of the SVM model using the PSO algorithm. The determined data is then used to model the synergistic desulfurization method between the furnace and external furnace through a BP neural network. Under the premise of setting the SO2 concentration at the desulfurization tower outlet, the amount of quicklime added outside the furnace is determined, which can effectively solve problems such as inaccurate control of the amount of desulfurizing agent used.

[0126] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0128] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A two-stage combined desulfurization method for predicting the amount of lime to be added outside the furnace, characterized by, The method comprises the following steps: real-time acquisition of desulfurization operation parameters representing the desulfurization state of the boiler, the desulfurization operation parameters comprising at least coal quality characteristic parameters, limestone input quantity in the boiler, actual input quantity of desulfurization tower quicklime, actual concentration of SO2 at the outlet of the boiler, and actual concentration of SO2 at the outlet of the desulfurization tower; determination of a desulfurization working condition category corresponding to the desulfurization operation parameters by a pre-trained desulfurization working condition prediction model taking the desulfurization operation parameters as input; calling a quicklime input quantity prediction model corresponding to the determined desulfurization working condition category, and predicting a desulfurization tower quicklime target input quantity by the quicklime input quantity prediction model taking the limestone input quantity in the boiler, the actual concentration of SO2 at the outlet of the boiler, and a pre-determined SO2 target concentration at the outlet of the desulfurization tower as input; the desulfurization working condition prediction model is obtained by training a preset machine learning algorithm based on historical desulfurization operation parameters, and the quicklime input quantity prediction model is obtained by training a neural network model based on historical desulfurization operation parameters belonging to the corresponding desulfurization working condition category.

2. The two-stage combined desulfurization method according to claim 1, characterized by, The preset machine learning algorithm is a support vector machine algorithm, and the method further comprises: acquiring a historical desulfurization operation parameter data set within a specified time interval; performing feature screening on the acquired historical desulfurization operation parameter data set, training an initial desulfurization working condition prediction model based on a support vector machine algorithm by taking the screened historical desulfurization operation parameter data as input and taking a desulfurization working condition category as output; optimizing model parameters of the initial desulfurization working condition prediction model based on a support vector machine algorithm by a particle swarm optimization algorithm to obtain a pre-trained desulfurization working condition prediction model.

3. The two-stage combined desulfurization method according to claim 2, characterized by, Optimizing model parameters of the initial desulfurization working condition prediction model based on a support vector machine algorithm by a particle swarm optimization algorithm comprises: randomly generating initial particles from a penalty factor C and a kernel parameter g of the initial desulfurization working condition prediction model based on a support vector machine algorithm, and setting the size of the particle swarm and the initial position and initial speed of the initial particles; calculating the fitness value of each particle by a K-fold cross-validation algorithm, comparing the fitness value of each particle with the global fitness value of the current particle, updating the position and speed of the particle according to the comparison result, and obtaining the optimal penalty factor C and the optimal kernel parameter g of the initial desulfurization working condition prediction model based on a support vector machine algorithm until a convergence condition is reached.

4. The two-stage combined desulfurization method according to claim 2, characterized by, Performing feature screening on the acquired historical desulfurization operation parameter data set comprises: determining that historical desulfurization operation parameters in the acquired historical desulfurization operation parameter data set, in which the actual concentration of SO2 at the outlet of the boiler is lower than a first SO2 concentration threshold, or the actual concentration of SO2 at the outlet of the boiler is lower than the first SO2 concentration threshold and the actual concentration of SO2 at the outlet of the desulfurization tower is lower than a second SO2 concentration threshold, are abnormal historical desulfurization operation parameters; deleting all abnormal historical desulfurization operation parameters from the acquired historical desulfurization operation parameter data set.

5. The two-stage combined desulfurization method according to claim 1, characterized by, The neural network model is a BP neural network model, and the method further comprises: acquiring a historical desulfurization operation parameter data set within a specified time interval; determining a desulfurization working condition category corresponding to each historical desulfurization operation parameter by a pre-trained desulfurization working condition prediction model taking the historical desulfurization operation parameter data set as input. The BP neural network model is trained based on historical desulfurization operation parameters belonging to different desulfurization working condition categories, and a lime input amount prediction model corresponding to different desulfurization working condition categories is obtained.

6. The two-stage combined desulfurization method according to claim 5, characterized by, The BP neural network model is trained based on historical desulfurization operation parameters belonging to different desulfurization working condition categories, and a lime input amount prediction model corresponding to different desulfurization working condition categories is obtained. The lime input amount and the actual SO2 concentration at the boiler outlet in all historical desulfurization operation parameters are obtained, and a training sample set of the BP neural network model is constructed, wherein the number of the training sample set corresponds to the number of the determined desulfurization working condition categories; Each training sample set and the predetermined SO2 target concentration at the desulfurization tower outlet are taken as inputs, and the target lime input amount of the desulfurization tower is taken as an output to train the BP neural network model, so that a lime input amount prediction model corresponding to different desulfurization working condition categories is obtained; The lime input amount and the actual SO2 concentration at the boiler outlet in each training sample set belong to the same desulfurization working condition category, and the lime input amount and the actual SO2 concentration at the boiler outlet in different training sample sets belong to different desulfurization working condition categories.

7. The two-stage combined desulfurization method according to claim 1, characterized by, The desulfurization working condition categories include a first desulfurization working condition category, a second desulfurization working condition category and a third desulfurization working condition category; and the method further includes: The desulfurization working condition category corresponding to the desulfurization operation parameters, in which the actual SO2 concentration at the boiler outlet is higher than the third SO2 concentration threshold and the actual SO2 concentration at the desulfurization tower outlet is lower than the fourth SO2 concentration threshold, is determined as the first desulfurization working condition category; The desulfurization working condition category corresponding to the desulfurization operation parameters, in which the actual SO2 concentration at the boiler outlet is higher than the third SO2 concentration threshold and the actual SO2 concentration at the desulfurization tower outlet is higher than the fifth SO2 concentration threshold, is determined as the second desulfurization working condition category; The desulfurization working condition category corresponding to the desulfurization operation parameters, in which the actual SO2 concentration at the boiler outlet is lower than the sixth SO2 concentration threshold and the actual SO2 concentration at the desulfurization tower outlet is lower than the fourth SO2 concentration threshold, is determined as the third desulfurization working condition category.

8. A two-stage combined desulfurization device, characterized by comprising: a first-stage desulfurization device; a second-stage desulfurization device; and a device for predicting the amount of quicklime to be fed to the second-stage desulfurization device. The method includes: The data acquisition module is configured to acquire desulfurization operation parameters representing the state of boiler desulfurization in real time, wherein the desulfurization operation parameters at least include coal quality characteristic parameters, lime input amount in the boiler, actual lime input amount of the desulfurization tower, actual SO2 concentration at the boiler outlet and actual SO2 concentration at the desulfurization tower outlet; The desulfurization working condition prediction module is configured to determine the desulfurization working condition category corresponding to the desulfurization operation parameters by using the desulfurization operation parameters as inputs and the pre-trained desulfurization working condition prediction model; The lime input amount prediction module is configured to call the lime input amount prediction model corresponding to the determined desulfurization working condition category, to use the lime input amount in the boiler, the actual SO2 concentration at the boiler outlet and the predetermined SO2 target concentration at the desulfurization tower outlet as inputs, and to predict the target lime input amount of the desulfurization tower by using the lime input amount prediction model. The desulfurization working condition prediction model is obtained by training a preset machine learning algorithm based on historical desulfurization operation parameters, and the lime input amount prediction model is obtained by training a neural network model based on historical desulfurization operation parameters belonging to a corresponding desulfurization working condition category.

9. A terminal device, comprising: Comprise: At least one processor; Memory, connected with the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the at least one processor realizes the two-stage combined desulfurization lime input amount prediction method in any one of claims 1 to 7 by executing the instructions stored in the memory.

10. A machine-readable storage medium, characterized in that, The machine readable storage medium stores instructions, and the instructions make the processor be configured to execute the two-stage combined desulfurization lime input amount prediction method in any one of claims 1 to 7 when the processor executes the instructions.

Images