A method and system for real-time control of oxygen content in flue gas during solid waste incineration.

Abstract

This application relates to a method and system for real-time control of oxygen content in flue gas during solid waste incineration. The method includes acquiring setpoints for primary and secondary air volumes, as well as historical flue gas oxygen content data, to obtain a training set; constructing a prediction model based on the training set using a randomized network to predict flue gas oxygen content; optimizing the setpoints using a rolling optimization strategy to obtain optimized air volume setpoints; and combining the event-triggered strategy with the rolling optimization strategy to achieve online optimization control of flue gas oxygen content. This application combines the event-triggered strategy with the rolling optimization strategy to achieve online optimization control of flue gas oxygen content based on the prediction model, thereby reducing the computational load of air volume setting and the frequency of controller updates, improving the solution accuracy of rolling optimization, and rationally setting the air volume to ensure effective control of flue gas oxygen content.

Description

Technical Field

[0001] This application relates to the field of solid waste incineration technology, and in particular to a method and system for real-time control of oxygen content in flue gas during solid waste incineration. Background Technology

[0002] With my country's population growth and accelerated urbanization, the amount of urban solid waste is increasing year by year. Currently, the main methods for treating urban solid waste include landfill, composting, and incineration. Compared with the other two methods, incineration has become the main means of treating urban solid waste due to its significant advantages such as harmlessness, volume reduction, and resource recovery.

[0003] During incineration, the oxygen content of flue gas is a key parameter reflecting incineration efficiency. Currently, in models predicting flue gas oxygen content, unreasonable primary and secondary air volume setpoints can lead to unstable control, resulting in excessive pollutant emission concentrations. Furthermore, due to the nonlinear, multi-coupled, and highly volatile nature of urban solid waste incineration, adjusting air volume based on expert experience is insufficient for effective control of flue gas oxygen content, and obtaining accurate process models hinders the performance of PID controllers. Model predictive control (MPC), as an advanced control strategy, effectively handles multivariable, constrained control problems. MPC focuses on techniques such as predictive model building and rolling optimization strategy design. While neural network models have advantages in learning data and approximating nonlinear mappings, using them to build predictive models has shown some effectiveness, issues such as getting trapped in local optima and slow convergence need to be considered.

[0004] Therefore, researching how to reasonably set the air volume to ensure effective control of flue gas oxygen content and selecting a prediction model to improve the solution accuracy of rolling optimization is of great significance for ensuring the stable operation of urban solid waste incineration. Thus, it is urgent to provide a real-time control method and system for flue gas oxygen content in solid waste incineration to solve the above-mentioned technical problems. Summary of the Invention

[0005] This application provides a method and system for real-time control of oxygen content in flue gas during solid waste incineration to solve the above-mentioned technical problems.

[0006] In a first aspect, this application provides a method for real-time control of oxygen content in flue gas during solid waste incineration. The method includes: acquiring setpoints for primary and secondary air volumes and historical flue gas oxygen content data to obtain a training set; constructing a prediction model based on a randomized network using the training set to predict flue gas oxygen content; optimizing the setpoints using a rolling optimization strategy to obtain optimized air volume setpoints; and combining the event-triggered strategy with the rolling optimization strategy to achieve online optimization control of flue gas oxygen content.

[0007] Optionally, the prediction model based on the randomized network is constructed using the training set to predict the oxygen content in flue gas: parameter settings are performed based on the randomized network, and model parameters are initialized, wherein the model parameters include the maximum number of hidden layer neurons L. max Expected error ε; Maximum number of configurations T max The hidden node weights w and biases b are randomly configured within the interval [-λ, λ]. The training set is input into the randomly configured network for training to construct a prediction model, which is described by the following formula:

[0008] y p (k+1)=f(U(k),y(k))

[0009] U(k)={u1(k)~u5(k)} (1),

[0010] In the formula, u1(k)-u5(k) represent the setpoints of the primary and secondary air volumes in the drying section, combustion section 1, combustion section 2, and combustion-out section, respectively; f represents the objective function; the training set is {(X, Y)}, where X represents the setpoint air volume and historical flue gas oxygen content, i.e., U(K); Y is y(k), representing the predicted flue gas oxygen content. p (k+1) represents the predicted value of flue gas oxygen content at step k+1 of the prediction model.

[0011] Optionally, inputting the training set into a randomly configured network for training to construct a prediction model includes: based on the objective function f:R d →R, and the training set {(X, Y)} are used to construct the hidden layer. The network output with L-1 hidden layer nodes is described by the formula:

[0012]

[0013] In the formula, X represents the air volume setpoint and the historical flue gas oxygen content, β j G represents the output weight of the j-th node in the hidden layer. j (·) represents the activation function of the j-th node in the hidden layer, w j and b jLet represent the input weights and biases of the j-th node in the hidden layer, respectively; based on the network output of the hidden layer nodes, calculate the current residual, which is described by formula (3):

[0014] e L-1 =ff L-1 (3),

[0015] If, according to preset requirements, the number of residuals or hidden layer nodes does not meet the preset requirements, a new node will be generated by random network configuration. The preset requirements are described by formula (4):

[0016]

[0017] In the formula, 0 < ||g|| g 0 <r<1,{μ L} represents a sequence of non-negative real numbers, μ L ≤(1-r) and satisfy lim L→∞ μ L =0; Based on the output weights after adding new nodes, the predicted value is obtained. The output weights are described by formula (5):

[0018]

[0019] In the formula, G L ={g1,g2,...,g L}, Let be the Moore-Penrose generalized inverse of matrix GL.

[0020] Optionally, the step of constructing a prediction model based on the training set using a randomly configured network to predict flue gas oxygen content further includes: setting a setpoint for flue gas oxygen content and setting a reference trajectory y for flue gas oxygen content using a first-order smoothing filter based on the setpoint. r (k+j) is described by formula (6):

[0021]

[0022] In the formula, α is an adjustment factor, 0 < α < 1, y sp This indicates the set value for the oxygen content in the flue gas;

[0023] The deviation e(k) between the current predicted value and the actual output value of flue gas oxygen content is compensated by feedback to make the prediction result at the next moment more accurate. η is the compensation coefficient, which is described by formula (7):

[0024]

[0025] ​In the formula, y(k) represents the actual output value of oxygen content in the flue gas. p (k) represents the oxygen content of the flue gas at step k of the prediction model. This represents the predicted oxygen content in the flue gas at step j of the prediction model after feedback correction.

[0026] Optionally, the optimization of the setpoint based on the rolling optimization strategy to obtain the optimized airflow setpoint includes: based on the flue gas oxygen content reference trajectory, the flue gas oxygen content prediction value after feedback correction, and the setpoint, setting an objective function to optimize the airflow setpoint to obtain the optimized airflow setpoint. The objective function is described by formula (8):

[0027]

[0028] In the formula, λ represents the control weight coefficient, and N p Nu represents the prediction time domain, and Nu represents the control time domain (Nu <Np),U max with U min These represent the upper and lower bounds of U, respectively; based on the airflow setting value and the current flue gas oxygen content output value, the predicted value of flue gas oxygen content is obtained using the prediction model.

[0029] Optionally, based on the flue gas oxygen content reference trajectory, the flue gas oxygen content prediction value after feedback correction, and the air volume setpoint, an objective function is set to optimize the air volume setpoint, resulting in the optimized air volume setpoint, including: initializing the population size NP, the maximum number of iterations G, the search dimension D, and the search interval [lb, ub]; based on the population, each dimension of an individual is randomly generated and described by formula (9):

[0030]

[0031] In the formula, i = 1, 2, ..., NP; j = 1, 2, ..., D. Let be the j-th dimension component of the i-th individual in generation 0, and let r represent a random number between [0,1]. Let x represent the minimum and maximum values ​​of an individual in the j-th dimension, respectively. Based on a mutation strategy, a local search is performed on the current individual while maintaining population diversity, resulting in a mutation vector. An adaptive t-distribution operator is then introduced for the superior individual x. pbest Mutation is performed to obtain mutated superior individuals; if the fitness value of the mutated superior individual is less than that of the original superior individual, the mutated individual will replace the superior individual to continue the mutation of the target individual; otherwise, the superior individual remains unchanged. The mutation vector is described by formula (10), the mutated superior individual is described by formula (11), and the selection method of the superior individual is described by formula (12).

[0032]

[0033] x' pbest =x pbest +x pbest ×trnd(t) (11)

[0034]

[0035] In the formula, g = 1, 2, ..., G. For the mutation vector, Let i be the i-th individual in the g-generation population. For individuals randomly selected from the top p% of individuals in generation g, F i Let r1, r2, i represent the mutation factors, which are distinct integers in the range [1, NP], and x represent the mutation factors. pbest ′ represents x pbest The updated position, trnd(t), represents the t-distribution function for degrees of freedom t, i.e., the current iteration number. The experimental individual is obtained by crossing the mutated vector with the target individual, described by the following formula:

[0036]

[0037] In the formula, j rand CR is a random number in the range [1, D]. i The crossover probability;

[0038] The fitness values ​​of experimental individuals and target individuals are compared. Individuals with better fitness values ​​are selected to be retained for the next generation, guiding the population to evolve towards the optimal position. This is described by the following formula:

[0039]

[0040] After selecting the steps, the control parameters that generate excellent individuals are retained, and the control parameters that generate eliminated individuals are regenerated using a normal distribution, with a variation factor F. i For example, it can be described using the following formula:

[0041]

[0042] Each generation of individuals undergoes continuous mutation, crossover, and selection, executing the above steps. At the end of each generation, μ is recalculated. F and μ CR Both are calculated in the same way, and the update is described by the following formula:

[0043] μ F = (1-c)·μ F +c·meanL (S F (16),

[0044] Where μ F μ represents the mean of the normal distribution of the regenerated variation factor F. CR S represents the mean of the normal distribution of the regenerated crossover probability CR, where c is a number between [0,1]. F The set of all successful variants in each generation, meanL(.), is calculated using the Lehmer mean and described by formula (17):

[0045]

[0046] Based on the above continuous variation and update, when the maximum number of iterations G is reached, the corresponding primary and secondary air volume setting values ​​are obtained based on the objective function.

[0047] Optionally, the online optimization control of flue gas oxygen content based on the combination of the event triggering strategy and the rolling optimization strategy includes: Step 1: Presetting trigger conditions for resetting the airflow setpoint; Step 2: Obtaining a predicted value of flue gas oxygen content based on the prediction model according to the airflow setpoint; Step 3: Setting the flue gas oxygen content setpoint y. sp And calculate the reference trajectory y of oxygen content in the flue gas. r (k+j); Step 4: Perform feedback compensation on the deviation e(k) between the current predicted value and the actual output value of flue gas oxygen content, and calculate the predicted value of flue gas oxygen content after feedback correction. Step 5: If the triggering condition is not met, return to Step 3 with the air volume setting value from the previous moment; if the triggering condition is met, proceed to Step 6; Step 6: Iteratively solve the objective function based on the prediction model to obtain a new air volume setting value, and return to Step 3 again.

[0048] Optionally, the triggering conditions include: a fixed threshold triggering design, whereby if the absolute error between the setpoint for the oxygen content of the flue gas and the actual output reaches a threshold θ, a new air volume setpoint is calculated according to a rolling optimization strategy. The triggering conditions are described by formula (18):

[0049] |y sp -y|≥θ (18),

[0050] In the formula y sp y represents the set value of oxygen content in flue gas, and y represents the actual output value of oxygen content in flue gas.

[0051] Optionally, the triggering conditions include: a fixed event triggering design, where if the maximum allowable time limit is reached between the last triggering time and the current time, a new airflow setting value is calculated based on a rolling optimization strategy. The triggering conditions are described by formula (19):

[0052] kk e ≥k max (19),

[0053] In the formula, k e k is the time when it was last triggered. max This is the maximum allowed time limit.

[0054] Secondly, this application provides a real-time control system for the oxygen content of flue gas in a solid waste incineration process. The system includes: an acquisition module for acquiring setpoints for primary and secondary air volumes, as well as historical flue gas oxygen content data, to obtain a training set; a prediction model construction module for constructing a prediction model based on a randomized network using the training set to predict the oxygen content of the flue gas; an optimization and solution module for the setpoints based on a rolling optimization strategy to optimize and solve the setpoints, obtaining optimized setpoints; and an optimization control module for combining the event-triggered strategy with the rolling optimization strategy to achieve online optimization control of the flue gas oxygen content.

[0055] Thirdly, this application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method described above when executing the computer program.

[0056] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method described above.

[0057] This application has at least the following advantages:

[0058] According to the technical content provided in the embodiments of this application, by acquiring historical flue gas oxygen content data generated during solid waste incineration and setting values ​​for primary and secondary air volumes, training and test sets are obtained. Based on a randomized network algorithm, the training set is used as input data to continuously learn and train to obtain a prediction model. This avoids the problem of gradient descent method easily getting trapped in local optima, resulting in higher prediction accuracy. The air volume setting value is optimized and solved based on a rolling optimization strategy. The differential evolution algorithm improved by parameter adaptation and t-distribution strategies can improve the solution performance of rolling optimization. Compared with the manual setting method, the air volume fluctuation range is smaller, which is conducive to the stable operation of the control system and provides a guarantee for the optimized control of the incineration process. At the same time, an event triggering strategy is introduced, which can effectively reduce the computational load of primary and secondary air volume solutions under the premise of meeting the flue gas oxygen content control range, and avoid the wear and tear of the execution equipment caused by frequent controller actions, thereby improving work efficiency. Thus, the functions of optimized air volume setting and accurate tracking control of flue gas oxygen content are realized. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating a method for real-time control of oxygen content in flue gas during solid waste incineration, as shown in one embodiment.

[0060] Figure 2 This is a flowchart illustrating a method for real-time control of oxygen content in flue gas during solid waste incineration, as shown in one embodiment.

[0061] Figure 3 This is a schematic diagram illustrating the process of building a prediction model based on a randomly configured network in one embodiment;

[0062] Figure 4 In one embodiment, a graph showing the setpoint curves of primary air volume and secondary air volume is provided.

[0063] Figure 5 This is a schematic diagram illustrating the process of constructing the hidden layer of a randomly configured network prediction model in one embodiment;

[0064] Figure 6 This is a schematic diagram illustrating the process of feedback correction of prediction results in one embodiment;

[0065] Figure 7 This is a schematic diagram of the curve for tracking the setpoint of apparent flue gas oxygen content in one embodiment;

[0066] Figure 8 This is a schematic diagram illustrating the process of optimizing the airflow setpoint based on a rolling optimization strategy in one embodiment;

[0067] Figure 9 This is a schematic diagram illustrating the specific process of optimizing the airflow setpoint in one embodiment;

[0068] Figure 10 This is a schematic diagram illustrating the process of combining an event-triggered strategy with a rolling optimization strategy to achieve online optimization control of predicted flue gas oxygen content in one embodiment;

[0069] Figure 11 This is a block diagram illustrating the real-time oxygen content control system for flue gas during solid waste incineration in one embodiment.

[0070] Figure 12 This is a schematic structural diagram of a computer device in one embodiment. Detailed Implementation

[0071] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the scope of the present application.

[0072] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when used in this specification, the words “comprising” and / or “including” indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0073] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the various embodiments of this application to facilitate a better understanding of the application. However, the technical solutions claimed in this application can be implemented even without these technical details and various variations and modifications based on the following embodiments. The division of the various embodiments below is for ease of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0074] Figure 1 This is a flowchart illustrating a method for real-time control of oxygen content in flue gas during solid waste incineration, as provided in an embodiment of this application. Figure 2 This is a flowchart of a method for real-time control of oxygen content in flue gas during solid waste incineration; refer to... Figure 1 , Figure 2 As shown, the method may include the following steps:

[0075] S201. Obtain the set values ​​of primary air volume and secondary air volume, as well as historical flue gas oxygen content data, to obtain the training set;

[0076] S202. A prediction model is constructed based on a randomly configured network and the training set to predict the oxygen content of flue gas.

[0077] S203. Based on the rolling optimization strategy, the setpoint is optimized to obtain the optimized air volume setpoint.

[0078] S204. Based on the combination of the event triggering strategy and the rolling optimization strategy, online optimization control of flue gas oxygen content is achieved.

[0079] The following is a detailed explanation of each step:

[0080] Reference Figure 1 S201. Obtain the set values ​​of primary air volume and secondary air volume, as well as historical flue gas oxygen content data, to obtain the training set;

[0081] In this embodiment, it should be noted that obtaining historical data on the oxygen content of flue gas generated during the solid waste incineration process, as well as the set values ​​of primary and secondary air volumes, are necessary conditions for predicting the oxygen content of flue gas during the solid waste incineration process. Obtaining the set values ​​of primary and secondary air volumes and historical flue gas oxygen content data facilitates subsequent operation and analysis.

[0082] In one example, 1000 sets of data from a solid waste incineration plant can be obtained as sample data to train and test a real-time control method for oxygen content in flue gas during solid waste incineration. The sample data is divided into a training set and a test set. The training set is used to train the constructed prediction model, and the test set is used to test whether the final prediction model can accurately predict and determine its accuracy.

[0083] S202. A prediction model is constructed based on a randomly configured network and the training set to predict the oxygen content of flue gas.

[0084] In this embodiment, it should be noted that the Randomly Configured Network (SCN) algorithm is a multi-layered neural network consisting of an input layer, hidden layers, and an output layer. It has advantages such as fast training speed and general approximation properties. Based on historical data generated during the incineration process, a prediction model built using SCN technology can predict the oxygen content of flue gas during solid waste incineration.

[0085] S203. Based on the rolling optimization strategy, the setpoint is optimized to obtain the optimized air volume setpoint.

[0086] In this embodiment, it should be noted that the rolling optimization strategy refers to determining the future control action by optimizing a certain performance index, rather than using a fixed global optimal objective. Instead, a rolling finite-time domain optimization strategy is employed. At each sampling time, the optimal control rate for a finite period is calculated based on the optimized performance index at that time. Specifically, online rolling means that the calculated control action sequence is only actually executed at the current time, and the optimal control rate is recalculated at the next sampling time. The control action in online rolling only takes effect at the current time, and the control action at the next time is a newly optimized control action. The future control action is determined by optimizing the objective function through optimizing a certain performance index. Specifically, the setpoint is optimized according to the rolling optimization strategy, the training set is updated, and new predicted values ​​are obtained based on the prediction model. During the continuous updating process, the model is continuously compared with the reference trajectory of flue gas oxygen content, constantly corrected, and gradually moved closer to the reference trajectory of flue gas oxygen content, thus controlling the later-stage training and prediction model.

[0087] S204. Based on the combination of the event triggering strategy and the rolling optimization strategy, online optimization control of flue gas oxygen content is achieved.

[0088] In this embodiment, it should be noted that by introducing an event-triggered strategy and combining it with a rolling optimization strategy control, the event-triggered strategy controls the update events of the airflow setpoint, thereby controlling the execution frequency of the prediction model, the action frequency of the controller, and the number of calculations. Specifically, the number of trigger events is 140, far less than the 300 times of the traditional MPC method. While meeting control accuracy requirements, this effectively reduces the computational burden and frequent controller actions, which is beneficial for stabilizing the oxygen content of the flue gas within the process requirements and minimizing fluctuations in the airflow setpoint.

[0089] Reference Figure 1 , Figure 3 as well as Figure 4 As shown, in some embodiments, in S202, a prediction model is constructed based on a randomly configured network according to the training set to predict the oxygen content of flue gas, including:

[0090] S021. Set parameters based on a randomly configured network and initialize model parameters, including the maximum number of hidden layer neurons L. max Expected error ε; Maximum number of configurations T max The hidden node weights w and biases b are randomly configured within the interval [-λ, λ].

[0091] S022. Input the training set into a randomly configured network to train and build a prediction model. The prediction model is described by the following formula:

[0092] y p(k+1)=f(U(k),y(k))

[0093] U(k)={u1(k)~u5(k)} (1),

[0094] In the formula, u1(k)-u4(k) represent the primary air volume setpoints for the drying section, combustion section 1, combustion section 2, and combustion section, respectively; u5(k) represents the secondary air volume setpoint; f represents the objective function; the training set is {(X, Y)}; X represents the air volume setpoint and historical flue gas oxygen content, i.e., U(K); Y is y(k), representing the predicted flue gas oxygen content. p (k+1) represents the predicted value of flue gas oxygen content at step k+1 of the prediction model.

[0095] In this embodiment, it should be noted that the parameters are set according to the random configuration network, and the maximum number of hidden layer neurons L in the SCN algorithm is... max The value is 10; the expected error ε is 0.001; the maximum number of configurations T max The value is 100; the hidden node weights w and bias b are randomly configured in the interval [-λ, λ], where λ = 1:10:100. Based on the model parameter settings, the prediction model is trained using the training set.

[0096] Reference Figure 3 , Figure 5 As shown, in some embodiments, in S2022, the training set is input into a randomly configured network to train and construct a prediction model, including:

[0097] S20221, Based on the objective function f:R d →R, and the training set {(X, Y)} are used to construct the hidden layer. The network output with L-1 hidden layer nodes is described by the formula:

[0098]

[0099] In the formula, X represents the air volume setpoint and the historical flue gas oxygen content, β j G represents the output weight of the j-th node in the hidden layer. j (·) represents the activation function of the j-th node in the hidden layer, w j and b j These represent the input weights and biases of the j-th node in the hidden layer, respectively.

[0100] S20222. Based on the network output of the hidden layer nodes, calculate the current residual, which is described by formula (3):

[0101] e L-1 =ff L-1 (3),

[0102] S20223. According to the preset requirements, if the number of residual or hidden layer nodes does not meet the preset requirements, a new node will be generated by randomly configuring the network. The preset requirements are described by formula (4):

[0103]

[0104] In the formula, 0 < ||g|| g 0 <r<1,{μ L} represents a sequence of non-negative real numbers, μ L ≤(1-r) and satisfy lim L→∞ μ L =0;

[0105] S20224. Based on the output weights after adding new nodes, the predicted values ​​are obtained. The output weights are described by formula (5):

[0106]

[0107] In the formula, G L ={g1,g2,...,g L}, Let be the Moore-Penrose generalized inverse of matrix GL.

[0108] In this embodiment, it should be noted that the neural network includes multiple neuron "layers": an input layer, a hidden layer, and an output layer. The hidden layer is responsible for the required calculations and outputting the results to the output layer. The prediction model is constructed based on the training set input, and the hidden layer performs calculations and outputs the results. The obtained results are continuously compared with preset requirements until the results meet the preset requirements, at which point the results are output, thereby constructing a flue gas oxygen content prediction model.

[0109] The activation function of the neuron in the Lth node of the hidden layer primarily performs a nonlinear transformation on the computation result. This nonlinear transformation is applied to each node in each layer of the network. Based on the activations, weights, and biases of the previous layers, a value is calculated for each activation in the next layer. However, before sending this value to the next layer, the output is scaled using the activation function, thus achieving the nonlinear classification task. Commonly used activation functions include the Sigmoid activation function, the tanh activation function, and the ReLU activation function. In this embodiment, the Sigmoid activation function is used. The Sigmoid function is between 0 and 1, and each input neuron, node, or activation is scaled to a value between 0 and 1.

[0110] Reference Figure 3 , Figure 6 ​As shown, in some embodiments, in S202, the process of constructing a prediction model based on a training set using a randomly configured network to predict the oxygen content in flue gas further includes:

[0111] S2023. Set the setpoint for flue gas oxygen content and set the reference trajectory y for flue gas oxygen content based on the setpoint using a first-order smoothing filter. r (k+j) is described by formula (6):

[0112]

[0113] In the formula, α is an adjustment factor, 0 < α < 1, y sp This indicates the set value for the oxygen content in the flue gas;

[0114] S2024. Feedback compensation is performed on the deviation e(k) between the current predicted value of flue gas oxygen content and the actual output to make the prediction result at the next moment more accurate. η is the compensation coefficient, which is described by formula (7):

[0115]

[0116] In the formula, y(k) represents the actual output value of oxygen content in the flue gas. p (k) represents the predicted value of flue gas oxygen content at step k of the prediction model. This represents the predicted oxygen content in the flue gas at step j of the prediction model after feedback correction.

[0117] In this embodiment, it should be noted that, as Figure 7 As shown, a pre-defined setpoint for flue gas oxygen content is provided. This setpoint represents the desired flue gas oxygen content, specifically, the optimal range for actual solid waste incineration is 6%-9%. A softened trajectory, known as the flue gas oxygen content reference trajectory, is then generated based on this setpoint. By calculating the deviation between the predicted and actual outputs, the prediction error is corrected online, resulting in feedback compensation. This reduces prediction bias caused by system disturbances or model mismatch, making the prediction results more accurate for the next moment and ensuring that the actual flue gas oxygen content output trajectory more closely approximates the reference trajectory. Here, η is the compensation coefficient, specifically set to 0.3.

[0118] In some embodiments, reference Figure 1 , Figure 8 As shown in S203, the setpoint is optimized based on the rolling optimization strategy to obtain the optimized airflow setpoint, including:

[0119] S2031. Based on the reference trajectory of flue gas oxygen content, the predicted value of flue gas oxygen content after feedback correction, and the set value, an objective function is set to optimize the air volume set value, and the optimized air volume set value is obtained. The objective function is described by formula (8):

[0120]

[0121] In the formula, λ represents the control weight coefficient, Np represents the prediction time domain, and Nu represents the control time domain (Nu <Np),y r (k+j) represents the reference trajectory for the oxygen content in the flue gas. U represents the predicted oxygen content in the flue gas at step j of the prediction model after feedback correction; max with U min These represent the upper and lower bounds of U, respectively.

[0122] S2032. Based on the set air volume and the current output value of flue gas oxygen content, the predicted value of flue gas oxygen content is obtained using a prediction model.

[0123] In this embodiment, it should be noted that a rolling optimization strategy is introduced based on the air volume setpoint, and an objective function is set to optimize the objective function, thereby improving the calculation accuracy of the future prediction model. Specifically, a flue gas oxygen content reference trajectory is set. In order to track the set, softened flue gas oxygen content reference trajectory and minimize the tracking error, the air volume setpoint is continuously updated. Then, based on the air volume setpoint and the current flue gas oxygen content output value, the predicted value of flue gas oxygen content is obtained based on the prediction model. A rolling optimization method is adopted. At time k, k+j steps of output prediction are performed, but only the first segment is applied to the controlled object. After executing the first segment, the output at time k+1 is obtained. After optimization, j segments are predicted again at time k+1, and again only the first segment is executed. The rolling optimization gradually approaches the flue gas oxygen content reference trajectory.

[0124] In some embodiments, reference Figure 8 , Figure 9 As shown in S2031, based on the flue gas oxygen content reference trajectory, the predicted flue gas oxygen content after feedback correction, and the setpoint, an objective function is set to optimize the airflow setpoint, resulting in the optimized airflow setpoint, including:

[0125] S20311. Initialize the population size NP, the maximum number of iterations G, the search dimension D, and the search interval [lb, ub].

[0126] S20312. Based on the population, each dimension of an individual is randomly generated and described by formula (9):

[0127]

[0128] In the formula, i = 1, 2, ..., NP; j = 1, 2, ..., D. Let be the j-th dimension component of the i-th individual in generation 0, and let r represent a random number between [0,1]. Let these represent the minimum and maximum values ​​of an individual in the j-th dimension, respectively.

[0129] S20313. Based on a mutation strategy, while maintaining population diversity, a local search is performed on the current individual to obtain a mutation vector; an adaptive t-distribution operator is introduced for the superior individual x. pbest Mutation is performed to obtain superior individuals, which participate in the mutation operation of the individual. When the fitness value of the superior individual after mutation is less than that of the original superior individual, the superior individual will replace the superior individual to continue the mutation of the target individual. Otherwise, the superior individual remains unchanged. The mutation vector is described by formula (10), the superior individual after mutation is described by formula (11), and the selection method of the superior individual is described by formula (12).

[0130]

[0131] x' pbest =x pbest +x pbest ×trnd(t) (11)

[0132]

[0133] In the formula, g = 1, 2, ..., G. For the mutation vector, Let i be the i-th individual in the g-generation population. For individuals randomly selected from the top p% of individuals in generation g, F i Let r1, r2, i represent the mutation factors, which are distinct integers in the range [1, NP], and x represent the mutation factors. pbest ′ represents x pbest The updated position, trnd(t), represents the t-distribution function with degrees of freedom t, i.e., the current iteration number.

[0134] S20314. The experimental individuals are obtained by cross-crossing the vectors generated by the mutation with the target individuals. This process retains the superior traits of the parents while increasing population diversity, as described by the following formula:

[0135]

[0136] In the formula, j rand CR is a random number in the range [1, D]. i The crossover probability;

[0137] S20315. Compare the fitness values ​​of the experimental individuals with those of the target individuals. Individuals with better fitness values ​​are selected and retained for the next generation, guiding the population to evolve towards the optimal position. In other words, selection is based on comparing the mutation factor of the i-th individual with that of the superior individuals to obtain the mutation factor of the next generation, which is described by the following formula:

[0138]

[0139] After the selection step, the control parameters that generate excellent individuals are retained, and the control parameters that generate eliminated individuals are regenerated using a normal distribution. In other words, the mutation factor and crossover probability of the i-th individual in the next generation are calculated, with the mutation factor F... i For example, it can be described using the following formula:

[0140]

[0141] S20316. Each generation of individuals undergoes continuous mutation, crossover, and selection, executing the above steps. μ is recalculated at the end of each generation. F and μ CR Both are calculated in the same way, using μ. F For example, the update is described using the following formula:

[0142] μ F = (1-c)·μ F +c·mean L (S F (16),

[0143] Where μ F μ represents the mean of the normal distribution of the regenerated variation factor F. CR S represents the mean of the normal distribution of the regenerated crossover probability CR, where c is a number between [0,1]. F The set of all successful variants in each generation, meanL(.), is calculated using the Lehmer mean and described by the formula:

[0144]

[0145] S20317. Based on the above continuous variation and update, when the maximum number of iterations G is reached, the corresponding primary and secondary air volume setting values ​​are obtained based on the objective function.

[0146] In this embodiment, it should be noted that the differential evolution algorithm is a highly efficient global optimization algorithm, including mutation, crossover, and selection operations. It guides the optimization search direction through the collective intelligence generated by cooperation and competition among individuals within the population. The main process involves starting with a randomly generated initial population. A new individual is generated by summing the vector difference between any two individuals in the population with that of a third individual. This new individual is then compared with the corresponding individual in the current population. If the fitness of the new individual is better than that of the current individual, the new individual replaces the old individual in the next generation; otherwise, the old individual is retained. Through continuous evolution, superior individuals are retained, inferior individuals are eliminated, and the search is guided towards the optimal solution.

[0147] In this embodiment, the optimal airflow setpoint obtained by solving the objective function is based on the Differential Evolutionary Algorithm (DE). It optimizes the setpoint through mutation, crossover, and selection of the difference vector between parent generations. The evolutionary calculation uses a generation-by-generation optimization process, and then parallel processing is used to select the population in a guided random search to achieve the desired goal, obtaining optimized primary and secondary airflow setpoints. Furthermore, an adaptive t-distribution operator is introduced to perturb the solution location, giving the algorithm better global exploration capabilities in the early stages of iteration and good local exploration capabilities in the later stages, thus improving the algorithm's convergence speed.

[0148] Specifically, in one example, the initial population size NP is 50, the maximum number of iterations G is 100, and the search dimension D is 10. According to the above algorithm strategy, it is continuously updated iteratively. When the maximum number of iterations is 100, the primary and secondary air volume setpoints are obtained according to the objective function. Compared with the manual setting method, the air volume fluctuation range is smaller, which is conducive to the stable operation of the control system.

[0149] In some embodiments, reference Figure 1 , Figure 10 As shown, in S204, based on the combination of the event-triggered strategy and the rolling optimization strategy, online optimization control of flue gas oxygen content is achieved, including:

[0150] S2041, Preset trigger conditions for resetting the airflow setting value;

[0151] S2042. Based on the air volume setpoint and the prediction model, the predicted value of the oxygen content in the flue gas is obtained.

[0152] S2043, Set the set value of oxygen content in flue gas y sp And calculate the reference trajectory y of oxygen content in the flue gas. r (k+j);

[0153] S2044. Perform feedback compensation on the deviation e(k) between the current predicted value and the actual output value of flue gas oxygen content, and calculate the predicted value of flue gas oxygen content after feedback correction.

[0154] S2045. If the trigger condition is not met, continue with the previous air volume setting value, i.e., U(k) = U(k-1), and return to S2043; if the trigger condition is met, proceed to S2046.

[0155] S2046. Based on the prediction model, iteratively solve the objective function to obtain the new air volume setpoint, i.e., U(k), and then return to S2043.

[0156] In this embodiment, it should be noted that by introducing an event-triggered strategy, the number of times the air volume setpoint is updated using a rolling optimization strategy can be controlled under the premise of meeting the control range of flue gas oxygen content. This can effectively reduce the amount of calculation for primary and secondary air volume calculation and avoid wear and tear on the execution equipment caused by frequent controller actions.

[0157] In some embodiments, the triggering condition includes a fixed threshold triggering design. If the absolute error between the setpoint for the oxygen content of the flue gas and the actual output reaches a threshold θ, a new air volume setpoint is solved according to a rolling optimization strategy. The triggering condition is described by formula (18):

[0158] |y sp -y|≥θ (18),

[0159] In the formula y sp y represents the setpoint for the oxygen content in the flue gas, and y represents the actual output of the oxygen content in the flue gas; or a fixed event trigger design, if the maximum allowable time limit is reached from the last trigger time, then a new air volume setpoint is solved according to the rolling optimization strategy, and then the predicted value of the oxygen content in the flue gas is obtained. The triggering condition is described by formula (19):

[0160] kk e ≥k max (19),

[0161] In the formula, k e k is the time when it was last triggered. max This is the maximum allowed time limit.

[0162] In this embodiment, it should be noted that the maximum allowable time limit refers to a safe time to prevent unexpected events such as equipment failure. If this time limit is exceeded, the airflow setpoint must be adjusted. The airflow setpoint, the input data for the predictive model, is updated based on the set trigger conditions, thus enabling the predictive model to make predictions. The trigger conditions are set in parallel; as long as any one of the trigger conditions is met, the airflow setpoint will be optimized based on the rolling optimization strategy. In one example, it was found that the number of calculations required to trigger an event through trigger time control was 140, far fewer than the 300 calculations required by the traditional MPC method. This effectively reduces the computational burden and frequent controller actions while maintaining control accuracy, thereby improving work efficiency.

[0163] In this embodiment, it should be noted that if the triggering condition is not met, the current airflow setting value is set to the previous airflow setting value. The reference trajectory is calculated based on the flue gas oxygen content setting value to obtain the flue gas oxygen content prediction value after feedback correction. The setting of the current airflow setting value to the previous airflow setting value is described by formula (20):

[0164] U(k)=U(k-1)(20),

[0165] The process involves updating the predicted oxygen content in the flue gas based on the previous airflow setpoint, and calculating a reference trajectory based on the setpoint. In one example, the setpoint is set to 7% for the first 150 steps and 8% for the next 150 steps. By repeating these steps, the predicted value obtained from the previous airflow setpoint is continuously compared with the trigger condition. The process continues until the absolute error between the setpoint and the actual output value reaches a threshold θ or the maximum permissible time limit K is exceeded. max At the same time, the new air volume setpoint is solved according to the rolling optimization strategy, and the new flue gas oxygen content prediction value is obtained according to the prediction model. This can effectively reduce the calculation burden and the frequent operation of the controller, and improve work efficiency.

[0166] In some embodiments, reference Figure 1 As shown in S204, based on the combination of the event triggering strategy and the rolling optimization strategy, online optimization control of flue gas oxygen content is achieved. This also includes: inputting the test set into the trained prediction model to obtain the output flue gas oxygen content prediction value and performing correction; and calculating the square integral error, integral absolute error, and maximum absolute error between the predicted and actual flue gas oxygen content values ​​according to the error evaluation method to verify the prediction accuracy of the prediction model.

[0167] In this embodiment, it should be noted that by calculating the integral error of the square between the predicted and actual values ​​of the flue gas oxygen content, the integral absolute error (IAE) is 1.6289, the absolute error of the integral (IAE) is 14.6870, and the maximum absolute error (Devmax) is 0.7135. It is found that the control accuracy is met. Under the premise of meeting the control accuracy, the computational burden and frequent controller actions can be effectively reduced, which is conducive to stabilizing the flue gas oxygen content within the range required by the process and minimizing the fluctuation of the air volume setpoint.

[0168] Therefore, in this embodiment, a prediction model is first established based on the set value of air volume. Then, the event triggering strategy is combined with the rolling optimization strategy for solving the set value of air volume. The event triggering strategy determines whether to recalculate the set value of air volume at the next moment. Finally, the actual oxygen content of flue gas in the urban solid waste incineration process (MSWI) is obtained based on the set value of air volume.

[0169] The implementation principle of this embodiment is as follows: The above steps mainly involve acquiring historical flue gas oxygen content data and primary and secondary air volume setpoints generated during solid waste incineration. These are combined to obtain training and testing sets. Based on a randomized network algorithm, the training set is used as input data to continuously learn and train a prediction model. This avoids the problem of gradient descent methods easily getting trapped in local optima, resulting in higher prediction accuracy. A rolling optimization strategy is used to optimize the air volume setpoints. An improved differential evolution algorithm with parameter adaptation and t-distribution strategies further enhances the performance of the rolling optimization. Compared to manual setting methods, the air volume fluctuation range is smaller, which is beneficial for the stable operation of the control system and ensures optimized control of the incineration process. Simultaneously, an event-triggered strategy is introduced. Under the premise of meeting the flue gas oxygen content control range, this effectively reduces the computational load of primary and secondary air volume calculations and avoids wear and tear on the execution equipment caused by frequent controller actions, improving work efficiency. This achieves optimized air volume setting and accurate tracking control of flue gas oxygen content.

[0170] Reference Figure 11 As shown, this application also provides a real-time control system for the oxygen content of flue gas in a solid waste incineration process. This system may include: an acquisition module, a prediction model construction module, an optimization solution module for the setpoint air volume, and an optimization control module. The main functions of each component module are as follows:

[0171] The acquisition module 301 is used to acquire the set values ​​of primary air volume and secondary air volume, as well as historical flue gas oxygen content data, to obtain a training set.

[0172] The prediction model construction module 302 is used to construct a prediction model based on the training set using a randomly configured network to predict the oxygen content of flue gas.

[0173] The air volume setpoint optimization module 303 is used to optimize the setpoint based on a rolling optimization strategy to obtain the optimized air volume setpoint.

[0174] The optimization control module 304 is used to achieve online optimization control of flue gas oxygen content based on the combination of the event triggering strategy and the rolling optimization strategy.

[0175] According to embodiments of this application, this application also provides a computer device and a computer-readable storage medium.

[0176] like Figure 12 The diagram shown is a block diagram of a computer device according to an embodiment of this application. The term "computer device" is intended to represent various forms of digital computers or mobile devices. The digital computer may include a desktop computer, a portable computer, a workbench, a personal digital assistant, a server, a mainframe computer, and other suitable computers. The mobile device may include a tablet computer, a smartphone, a wearable device, etc.

[0177] like Figure 12 As shown, device 600 includes a computing unit 601, a ROM 602, a RAM 603, a bus 604, and an input / output (I / O) interface 605. The computing unit 601, ROM 602, and RAM 603 are interconnected via the bus 604. The input / output (I / O) interface 605 is also connected to the bus 604.

[0178] The computing unit 601 can execute various processes in the method embodiments of this application according to computer instructions stored in the read-only memory (ROM) 602 or computer instructions loaded from the storage unit 608 into the random access memory (RAM) 603. The computing unit 601 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. The computing unit 601 can include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. In some embodiments, the methods provided in the embodiments of this application can be implemented as computer software programs, which are tangibly contained in a computer-readable storage medium, such as the storage unit 608.

[0179] RAM 603 may also store various programs and data required for the operation of device 600. Part or all of the computer program may be loaded and / or installed on device 600 via ROM 602 and / or communication unit 609.

[0180] The input unit 606, output unit 607, storage unit 608, and communication unit 609 in device 600 can be connected to I / O interface 605. The input unit 606 can be, for example, a keyboard, mouse, touchscreen, or microphone; the output unit 607 can be, for example, a display, speaker, or indicator light. Device 600 can exchange information and data with other devices through the communication unit 609.

[0181] It should be noted that the device may also include other components necessary for normal operation. It may also include only the components necessary for implementing the solution of this application, without necessarily including all the components shown in the figures.

[0182] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), payload programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof.

[0183] The computer instructions used to implement the methods of this application may be written in any combination of one or more programming languages. These computer instructions may be provided to the computing unit 601 such that when executed by the computing unit 601, such as a processor, the computer instructions cause the execution of the steps involved in the embodiments of the methods of this application.

[0184] The computer-readable storage medium provided in this application can be a tangible medium that can contain or store computer instructions for performing the steps involved in the method embodiments of this application. The computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, and other forms of storage media.

[0185] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for real-time control of oxygen content in flue gas during solid waste incineration, characterized in that, Includes the following steps: Obtain the set values ​​of primary and secondary air volume and historical flue gas oxygen content data to obtain the training set; A prediction model is constructed based on the training set using a randomly configured network to predict the oxygen content in flue gas. The setpoint is optimized and solved based on the rolling optimization strategy to obtain the optimized air volume setpoint. Based on the combination of the event-triggered strategy and the rolling optimization strategy, online optimization control of flue gas oxygen content is achieved; The method of constructing a prediction model based on the training set using a randomly configured network to predict the oxygen content in flue gas includes: The model parameters are initialized based on a randomly configured network, including the maximum number of hidden layer neurons. L max Expected error Maximum number of configurations Hidden node weights and bias In the interval Internal random configuration; The training set is input into a randomly configured network for training to construct a prediction model, which is described by the following formula: (1)， Where, u 1 (k)- u 4 (k) These represent the primary air volume setpoints for the drying section, combustion stage 1, combustion stage 2, and combustion stage, respectively. u 5 (k) This indicates the secondary air volume setpoint. f Let the objective function be {( X , Y )}, X This indicates the airflow setpoint and historical flue gas oxygen content, which is... U(K) , Y That is y(k) This indicates the predicted oxygen content in the flue gas. Indicates the prediction model's... k +1 step prediction of flue gas oxygen content; The step of inputting the training set into a randomly configured network for training and constructing a prediction model includes: Based on the objective function f : , and the training set {( X , Y Construct hidden layers containing L The network output of -1 hidden layer node is described by the formula: （2）， Where, X This indicates the airflow setpoint and the historical flue gas oxygen content. β j Indicates the hidden layer number 1 j The output weights of each node. g j (·) indicates the hidden layer. j Activation function for each node, w j and b j They represent the hidden layers respectively. j The input weights and biases of each node; Based on the network output of the hidden layer nodes, the current residual is calculated and described by formula (3): （3）； If, according to the preset requirements, the number of residuals or hidden layer nodes does not meet the preset requirements, a new node will be generated by random network configuration. The preset requirements are described by formula (4): （4） In the formula, 0 < || g || < b g , 0 < r <1, { μ L } represents a sequence of non-negative real numbers. ≤(1 - r And satisfy ; The predicted value is obtained by solving the output weights after adding new nodes. The solution of the output weights is described by formula (5): （5） In the formula, , is a matrix G L Moore-Penrose generalized inverse; The method of constructing a prediction model based on the training set using a randomly configured network to predict the oxygen content in flue gas further includes: Set a set value for the oxygen content of flue gas, and based on the set value, set a reference trajectory for the oxygen content of flue gas using a first-order smoothing filter. y r (k+j) It can be described by formula (6): (6)， Where, α As an adjustment factor, 0 < α < 1, y sp This indicates the set value for the oxygen content in the flue gas; The deviation between the current predicted value and the actual output value of flue gas oxygen content e(k) Feedback compensation is performed to make the prediction results at the next moment more accurate. η is the compensation coefficient, which is described by formula (7): （7）， In the formula, y(k) This indicates the actual output value of oxygen content in the flue gas. y p (k) Indicates the prediction model's... k Predicted oxygen content in flue gas at the next step. This indicates the prediction model after feedback correction. j Predicted value of oxygen content in flue gas at the next step; The optimization of the setpoint based on the rolling optimization strategy to obtain the optimized airflow setpoint includes: Based on the reference trajectory of flue gas oxygen content, the predicted value of flue gas oxygen content after feedback correction, and the set value, an objective function is set to optimize the airflow set value, resulting in the optimized airflow set value. The objective function is described by formula (8): (8)， Where, λ Represents the control weight coefficient. N p Indicates the prediction time domain, Nu Represents the control time domain ( Nu < Np ) ; U max and U min These represent the upper and lower bounds of U, respectively. Based on the air volume setpoint and the current flue gas oxygen content output value, the predicted value of flue gas oxygen content is obtained using the prediction model.

2. The method for real-time control of oxygen content in flue gas during solid waste incineration according to claim 1, characterized in that, Based on the flue gas oxygen content reference trajectory, the flue gas oxygen content prediction value after feedback correction, and the airflow setpoint, an objective function is set to optimize the airflow setpoint, resulting in the optimized airflow setpoint, including: Initialize population size NP Maximum number of iterations G Search dimensions D Search range [ lb,ub ]; Based on the population, each dimension of an individual is described by formula (9) according to random generation: (9)， In the formula, i =1,2, …, NP ; j = 1, 2, …, D , For the 0th generation i The first individual j dimensional components, r Represents a random number between [0,1]. , These represent the individual's position in the first place. j Minimum and maximum values ​​in the dimension; Based on a mutation strategy, a local search is performed on the current individual while maintaining population diversity to obtain a mutation vector; an adaptive t-distribution operator is introduced for superior individuals. x pbest Mutation is performed to obtain mutated superior individuals; if the fitness value of the mutated superior individual is less than that of the original superior individual, the mutated individual will replace the superior individual to continue the mutation of the target individual; otherwise, the superior individual remains unchanged. The mutation vector is described by formula (10), the mutated superior individual is described by formula (11), and the selection method of the superior individual is described by formula (12). (10) (11) (12)， In the formula, g =1,2,…, G , For the mutation vector, for g The first generation of the population i Individual, From g Generations ago p% Individuals randomly selected from among the outstanding individuals, F i Indicates the variable factor. r 1, r 2, i For [1, NP Different integers within the range, x pbest express x pbest The updated location trnd(t) This represents the t-distribution function with degrees of freedom t, i.e., the current iteration number; The experimental individuals are obtained by crossing the vectors generated by the mutation with the target individual, and are described by the following formula: (13)， In the formula ，j rand For [1, D Random numbers within the range of ] CR i The crossover probability; The fitness values ​​of experimental individuals and target individuals are compared. Individuals with better fitness values ​​are selected to be retained for the next generation, guiding the population to evolve towards the optimal position. This is described by the following formula: (14)； After selecting the steps, the control parameters that generate excellent individuals will be retained, and the control parameters that generate eliminated individuals will be regenerated using a normal distribution, with the variation factor as the variability factor. F i For example, it can be described using the following formula: (15)， Each generation of individuals undergoes continuous mutation, crossover, and selection, repeating the above steps. The calculation is restarted at the end of each generation. and Both are calculated in the same way, and the update is described by the following formula: (16)， In the formula, This represents the mean of the normal distribution that regenerates the variation factor F. Indicates the probability of regenerating the crossover. CR The mean of the normal distribution, c Numbers between [0, 1] S F It is the set of all successful variants in each generation. mean L (.)use Lehmer The average value is calculated using formula (17): (17)； Based on the above continuous variation and update, when the maximum number of iterations is reached... G Then, based on the objective function, the corresponding primary and secondary air volume setpoints are obtained.

3. The method for real-time control of oxygen content in flue gas during solid waste incineration according to claim 1, characterized in that, The method of combining the event-triggered strategy with the rolling optimization strategy to achieve online optimization control of flue gas oxygen content includes: Step 1: Preset the trigger conditions for resetting the airflow setting; Step 2: Based on the predicted model and the set air volume, obtain the predicted oxygen content of the flue gas; Step 3: Set the set value for flue gas oxygen content y sp And calculate the reference trajectory of oxygen content in flue gas. y r (k+j) ; Step 4: Calculate the deviation between the predicted and actual oxygen content in the flue gas. e(k) Perform feedback compensation and calculate the predicted value of flue gas oxygen content after feedback correction. ; Step 5: If the triggering condition is not met, return to Step 3 with the airflow setting value from the previous moment; if the triggering condition is met, proceed to Step 6. Step 6: Iteratively solve the objective function based on the prediction model to obtain a new air volume setpoint, and then return to Step 3.

4. The method for real-time control of oxygen content in flue gas during solid waste incineration according to claim 3, characterized in that, The triggering conditions include: The fixed threshold trigger design will activate if the absolute error between the setpoint and the actual output value of the flue gas oxygen content reaches the threshold. Then, the new air volume setpoint is solved according to the rolling optimization strategy, and the triggering condition is described by formula (18): （18）， In the formula y sp This indicates the set value for the oxygen content in the flue gas. y This indicates the actual output value of oxygen content in the flue gas.

5. The method for real-time control of oxygen content in flue gas during solid waste incineration according to claim 3, characterized in that, The triggering conditions include: In the fixed event triggering design, if the maximum allowable time limit is reached since the last triggering time, a new air volume setpoint is calculated according to the rolling optimization strategy. The triggering condition is described by formula (19): （19）， Where, k e The last time it was triggered. k max This is the maximum allowed time limit.

6. A real-time control system for oxygen content in flue gas during solid waste incineration, characterized in that, The system includes: The acquisition module is used to acquire the set values ​​of primary air volume and secondary air volume, as well as historical flue gas oxygen content data, to obtain a training set. A prediction model construction module is used to construct a prediction model based on the training set using a randomly configured network for predicting the oxygen content in flue gas; the construction of the prediction model based on the training set using a randomly configured network for predicting the oxygen content in flue gas includes: The model parameters are initialized based on a randomly configured network, including the maximum number of hidden layer neurons. L max Expected error Maximum number of configurations Hidden node weights and bias In the interval Internal random configuration; The training set is input into a randomly configured network for training to construct a prediction model, which is described by the following formula: (1)， Where, u 1 (k)- u 4 (k) These represent the primary air volume setpoints for the drying section, combustion stage 1, combustion stage 2, and combustion stage, respectively. u 5 (k) This indicates the secondary air volume setpoint. f Let the objective function be {( X , Y )}, X This indicates the airflow setpoint and historical flue gas oxygen content, which is... U(K) , Y That is y(k) This indicates the predicted oxygen content in the flue gas. Indicates the prediction model's... k +1 step prediction of flue gas oxygen content; The step of inputting the training set into a randomly configured network for training and constructing a prediction model includes: Based on the objective function f : , and the training set {( X , Y Construct hidden layers containing L The network output of -1 hidden layer node is described by the formula: （2）， In the formula, X This indicates the airflow setpoint and the historical flue gas oxygen content. β j Indicates the hidden layer number 1 j The output weights of each node. g j (·) indicates the hidden layer. j Activation function for each node, w j and b j They represent the hidden layers respectively. j The input weights and biases of each node; Based on the network output of the hidden layer nodes, the current residual is calculated and described by formula (3): （3）； If, according to the preset requirements, the number of residuals or hidden layer nodes does not meet the preset requirements, a new node will be generated by random network configuration. The preset requirements are described by formula (4): （4） In the formula, 0 < || g || < b g , 0 < r <1, { μ L } represents a sequence of non-negative real numbers. ≤(1 - r And satisfy ; The predicted value is obtained by solving the output weights after adding new nodes. The solution of the output weights is described by formula (5): （5） In the formula, , is a matrix G L Moore-Penrose generalized inverse; The method of constructing a prediction model based on the training set using a randomly configured network to predict the oxygen content in flue gas further includes: Set a set value for the oxygen content of flue gas, and based on the set value, set a reference trajectory for the oxygen content of flue gas using a first-order smoothing filter. y r (k+j) It can be described by formula (6): (6)， In the formula, α As an adjustment factor, 0 < α < 1, y sp This indicates the set value for the oxygen content in the flue gas; The deviation between the current predicted value and the actual output value of flue gas oxygen content e(k) Feedback compensation is performed to make the prediction results at the next moment more accurate. η is the compensation coefficient, which is described by formula (7): （7）， Where, y(k) This indicates the actual output value of oxygen content in the flue gas. y p (k) Indicates the prediction model's... k Predicted oxygen content in flue gas at step [step] This indicates the prediction model after feedback correction. j Predicted value of oxygen content in flue gas at the next step; The module for optimizing the airflow setpoint is used to optimize the setpoint based on a rolling optimization strategy to obtain an optimized airflow setpoint. The optimization of the setpoint based on the rolling optimization strategy to obtain the optimized airflow setpoint includes: Based on the reference trajectory of flue gas oxygen content, the predicted value of flue gas oxygen content after feedback correction, and the set value, an objective function is set to optimize the airflow set value, resulting in the optimized airflow set value. The objective function is described by formula (8): (8)， Where, λ Represents the control weight coefficient. N p Indicates the prediction time domain, Nu Represents the control time domain ( Nu < Np ) ; U max and U min These represent the upper and lower bounds of U, respectively. Based on the air volume setpoint and the current flue gas oxygen content output value, the predicted value of flue gas oxygen content is obtained according to the prediction model. The optimization control module is used to achieve online optimization control of flue gas oxygen content by combining the event-triggered strategy with the rolling optimization strategy.

Images