A method for collaborative optimization control of boiler combustion and denitration process

Abstract

The present invention relates to a collaborative optimization control method for the boiler combustion-denitration process, including a combustion optimization control module based on CO monitoring, a total reductant control module based on the prediction of parameters such as air volume, and a zone injection control module based on the zone injection quantity distribution table. The present invention establishes a neural network model between the CO concentration and the combustion efficiency, controls the air volume to optimize the boiler combustion efficiency; on this basis, takes the air volume command as a feedforward prediction to overcome the disadvantages of large delay, large inertia and strong nonlinearity of the denitration system, and accurately controls the total amount of reductant injection in real time; further, according to the NOx characteristics in the flue gas under multiple working conditions, establishes a zone injection quantity distribution table, and controls the opening of the zone injection valve in real time to achieve uniform mixing of the reductant and the flue gas, and improve the denitration efficiency; the present invention ensures that the outlet NOx concentration meets the standard, improves the denitration control accuracy, enhances the boiler combustion efficiency, and realizes carbon reduction and emission reduction of the unit under wide-range variable load conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy environment engineering control, and specifically relates to a method for collaborative optimization control of boiler combustion and denitrification process. Background Art

[0002] China consumes about 3.6 billion tons of coal annually (accounting for more than 50% of the global total). Pollutants emitted from coal combustion are an important cause of air pollution in China and also a major component of carbon emissions. The increase in carbon emissions will lead to the greenhouse effect, resulting in extreme weather that affects human production and life. The large emissions of NOx will cause serious harm to human health. The large-scale access of new energy power to the power grid puts higher requirements on the flexible peak regulation ability of coal-fired units. Relevant data show that for every 1% increase in boiler efficiency, the efficiency of the unit can be increased by 0.3%, and the coal consumption for power supply can be reduced by about 0.7%. However, the frequent and large-scale changes in the unit load within a short period of time will cause a decrease in boiler combustion efficiency and a large fluctuation in the concentration of nitrogen oxides (NOx) in the flue gas, resulting in an increase in unit carbon emissions while increasing the difficulty of NOx emission control and an increase in NOx emissions.

[0003] In actual operation, most boilers judge the combustion efficiency in the boiler by monitoring the change of the volume fraction of oxygen in the flue gas, and control the air volume in the boiler based on the volume fraction of oxygen to optimize the boiler combustion process, so as to achieve the purpose of stabilizing the boiler combustion efficiency. However, this method has the problem that the monitoring data of oxygen has relatively large fluctuations and cannot provide good feedback data, resulting in low boiler control quality. At the same time, due to the large lag characteristics of the denitrification system, if the data from the boiler combustion side can be effectively used to predict the changes of some important parameters in advance, the control accuracy can be improved.

[0004] Carbon monoxide (CO) is generated during the boiler combustion process, and research shows that the CO concentration can reflect the combustion efficiency of the boiler. Therefore, by monitoring the CO concentration of the boiler to judge the combustion efficiency and adjusting the air volume in real time, the optimization control of the boiler combustion efficiency based on the CO concentration can be realized. In addition, as an important parameter affecting the generation amount of NOx in the boiler, the air volume also plays an important role in denitrification control, and can provide a feedforward prediction for denitrification control to overcome the large lag problem existing in the denitrification system control.

[0005] For the above reasons, it is necessary to conduct on-line monitoring of the CO concentration, optimize the boiler combustion process, improve the boiler efficiency, and use the air volume parameter as a feedforward prediction, and further cooperate with the total amount control of the reducing agent and the zoning injection control method to improve the denitrification control accuracy, reduce NOx emissions, and achieve the collaborative optimization of the combustion efficiency of the unit and the denitrification process, and operate in a low-carbon and high-efficiency manner. Summary of the Invention

[0006] To overcome the shortcomings and deficiencies of the prior art, the present invention provides a method for collaborative optimization control of boiler combustion and denitration processes. The system aims to predict the impacts of various control and influencing variables on boiler combustion efficiency and outlet NOx concentration control. Based on evolutionary algorithms such as the particle swarm algorithm, neural network models, fuzzy control, model predictive control, and collaborative optimization methods, by continuously monitoring, feedback, simulation calculations, and optimization, it precisely regulates parameters such as air volume, reductant injection volume, and the distribution of reductants in the flue gas duct, ensuring the boiler combustion efficiency while achieving the optimal operation economy of the denitration system and realizing the safe, stable, efficient, and economic operation of the boiler and denitration device.

[0007] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] A collaborative optimization control method for boiler combustion-denitration processes, including a combustion optimization control module based on CO monitoring, a total reductant control module based on parameter prediction including air volume, and a partition injection control module based on a partition injection volume distribution table;

[0009] The combustion optimization control module based on CO monitoring is used to monitor the change in CO concentration and regulate the air volume to improve the boiler combustion efficiency;

[0010] The total reductant control module based on parameter prediction including air volume is used to monitor the key parameters affecting the outlet NOx concentration and predict the total amount of reductant required for the denitration device, and control the frequency of the reductant main pipe pump to achieve the control goal of minimizing the reductant consumption under the condition of meeting the outlet NOx concentration standard;

[0011] The partition injection control module based on the partition injection volume distribution table, according to the partition injection volume distribution table established based on the NOx flow rate and concentration distribution in the flue gas duct under different working conditions, real-time controls the opening degree of each injection valve in the flue gas duct; enables the uniform mixing of the reductant and the flue gas to improve the denitration efficiency and reduce reductant waste;

[0012] The control method includes the following steps:

[0013] (1) Based on historical data, using the Long Short-Term Memory (LSTM) neural network algorithm, establish a neural network model between combustion efficiency and CO concentration;

[0014] (2) Based on historical data analysis, on-site tests, and particle swarm parameter optimization methods, obtain the prediction models of coal amount Coal, total reductant injection amount L, air volume X, and outlet NOx concentration under different working conditions of the boiler, as well as the prediction model of air volume X and CO concentration;

[0015] G Nox = f(Coal, L, X)

[0016] G co = f(X)

[0017] where G Nox is the predicted model of the outlet NOx concentration, which is related to the coal quantity, the total amount of reductant injection, and the air volume. G co is the predicted model of the CO concentration, which is related to the air volume;

[0018] (3) According to the distribution of the partition injection ports of the denitration device, the flue gas cross-section is divided into multiple partitions. By means of on-site tests, the distribution of the NOx concentration and the flue gas flow velocity in the flue gas is determined under multiple working conditions, and the distribution table of the partition injection quantity under multiple working conditions is obtained through analysis;

[0019] (4) The combustion optimization control module controls the air volume based on the real-time CO concentration to optimize the in-furnace combustion efficiency, and takes the air volume control instruction as a feed-forward prediction and inputs it into the total reductant control module;

[0020] (5) According to the real-time working condition data, the intelligent algorithm server judges the real-time boiler working condition through the fuzzy rule base, determines the multi-condition prediction model to be adopted and the weight of each prediction model, calculates and adjusts the frequency of the reductant main pipe pump or the opening degree of the ammonia water regulating main valve to control the total amount of reductant injection;

[0021] (6) According to the real-time working condition data, select the corresponding distribution table of the reductant injection quantity for each partition, finally determine the real-time opening degree of each valve, and adjust the opening degree of the reductant injection valve for each partition in real time.

[0022] In particular, for the combustion optimization control module based on CO monitoring, aiming at the relationship between the change of CO concentration and the combustion efficiency, combining the influence mechanism of parameters such as air volume on the generation of CO in boiler combustion, a neural network model between CO concentration and combustion efficiency is established by using the data in the actual operation process of the boiler by means of machine learning. Further, a prediction model between air volume and CO concentration and outlet NOx concentration is established. The NSGA-II algorithm is used to solve the optimization problem in real time to calculate the optimal air volume, and the air volume control instruction is input into the DCS system to optimize the boiler combustion efficiency. At the same time, the air volume control instruction is used as a prediction and input into the subsequent module.

[0023] In particular, for the total reductant control module based on the prediction of parameters such as air volume, a prediction model is established by exploring the influence mechanism of each key influencing parameter on the outlet NOx concentration under different working conditions according to historical data, combining the air volume prediction to reduce the adverse effects brought by system delay, and judging the prediction model to be used under the current working condition based on fuzzy rules according to real-time data. The stability of the NOx concentration control at the outlet of the denitration device under a wide range of variable load working conditions is ensured by means of membership degree weighted switching under different load working conditions.

[0024] Specifically, the partition injection quantity distribution table is obtained based on both experimental and data analysis methods to determine the influence of each injection valve on its corresponding flue gas area. The required reductant injection quantity for each area is obtained according to the characteristics of NOx concentration and flow velocity in the flue gas under different load conditions, and a set of partition injection quantity distribution values corresponding to different operating conditions is summarized for real-time control of each partition injection valve in the control strategy.

[0025] Steps (1) and (4) of the present invention elaborate on the modeling process and implementation method of the combustion optimization control module based on CO monitoring. Steps (2) and (5) elaborate on the modeling process and implementation method of the total reductant control module based on the prediction of parameters such as air volume. Steps (3) and (6) elaborate on the modeling of the partition injection quantity distribution table in the partition injection control module and the specific implementation of partition injection control.

[0026] Preferably, the system constructed based on the above method includes a boiler, a denitration device and its auxiliary facilities, on-line monitoring equipment, an intelligent algorithm server, and a control device. The intelligent algorithm server realizes real-time communication with the on-line monitoring equipment through object linking and embedding process control services. The control device is connected to the intelligent algorithm server and implements control according to the instructions of the intelligent algorithm server;

[0027] The monitoring indicators of the on-line monitoring equipment include flue gas outlet temperature, NOx concentration, CO concentration, primary air volume, secondary air volume, flue gas oxygen content, coal quantity, reductant main pipe flow pump frequency, reductant main pipe flow, and the opening degree of each area injection valve. The on-line monitoring equipment has the functions of input and output of historical data and real-time data, and realizes information intercommunication with the intelligent algorithm server for real-time feedback;

[0028] The reductant includes ammonia water, urea solution and liquid ammonia with different concentrations;

[0029] The denitration device includes SCR and SNCR;

[0030] The control objects of the denitration device include the opening degree of each partition reductant injection valve and the reductant main pipe pump frequency.

[0031] Preferably, step (1) specifically includes the following steps:

[0032] Step A1: Collect the CO concentration under different load conditions of the boiler, calculate the combustion efficiency by the inverse balance method, preprocess the data, and input it into the long short-term memory neural network model for training to establish a long short-term memory neural network model between the combustion efficiency and the CO concentration;

[0033] Step A2: According to the obtained long short-term memory neural network model between the combustion efficiency and the CO concentration, determine the CO concentration range when the combustion efficiency is higher than the average value, and set the above CO concentration range as the control target.

[0034] Preferably, in step A1, the formula for calculating the combustion efficiency is as follows:

[0035]

[0036] In the formula, η is the combustion efficiency, Q 1 is the output heat, Q r is the output heat, Q 2 is the heat of the exhaust gas, Q 3 is the heat loss due to a part of the combustible gas carried away by the flue gas, Q 4 is the heat loss due to the pulverized coal being doped with fly ash and not participating in combustion, Q 5 is the heat from the furnace wall and pipes to the surrounding environment, Q 6 is the heat carried away by the ash slag and cooling;

[0037] In step A1, the method for preprocessing the data includes four steps: outlier removal, removal of the CEMS (Continuous Emission Monitoring System) purge period, data smoothing, and normalization;

[0038] Outlier removal is carried out by data cleaning, deleting the outliers and replacing them with the average value of adjacent data;

[0039] Removal of the CEMS purge period is carried out by data cleaning, clearing the data for half an hour after the CEMS purge signal is connected to avoid the influence of invalid data on model training;

[0040] Data smoothing adopts a two-dimensional Gaussian filtering method, and the specific formula is as follows:

[0041]

[0042] In the formula, G(x,y) is the value after Gaussian filtering, x,y are the data to be filtered, and σ is a constant that determines the width of the Gaussian filtering function;

[0043] Normalization uses the min-max normalization method to normalize the preprocessed data;

[0044] The method for establishing a long short-term memory neural network model between the combustion efficiency and the CO concentration includes constructing a validation set and a training set and training the neural network model using the training set. The specific process is as follows:

[0045] The processed data is randomly divided into a validation set and a training set. The long short-term memory neural network model is trained using the training set. Each time, a data segment of the training set is input into the long short-term memory layer of the long short-term memory neural network model. The output data is processed by a fully connected layer and then continues to be output to the attention layer, and is further processed by another fully connected layer to obtain the boiler combustion efficiency value calculated based on the CO concentration. Subsequently, this value is compared with the actual boiler combustion efficiency value, and the error is backpropagated to correct the parameters of each fully connected layer. Repeating this process, the required long short-term memory neural network model is obtained.

[0046] Preferably, step (2) specifically includes the following steps:

[0047] Step B1: Analyze the historical operation data of the unit, perform clustering analysis according to the boiler load status and the load rise and fall rate, determine the load rise, flat, and fall working condition characteristics of the boiler at high, slightly high, medium, slightly low, and low loads, and use them for the next experiment. Among them, the load status is high load when it is 90 - 100%, slightly high load when it is 70 - 90%, medium load when it is 50 - 70%, slightly low load when it is 40 - 50%, and low load when it is <40%.

[0048] Step B2: According to the boiler working condition characteristics determined in step B1, conduct experiments with the reductant injection amount unchanged and the reductant injection amount stepping change within each working condition interval to obtain the experimental data of the outlet NOx concentration corresponding to the parameters including air volume, coal volume, and reductant injection amount under different working conditions and the experimental data of the CO concentration corresponding to the air volume.

[0049] Step B3: Use the particle swarm algorithm to optimize the prediction model parameters including the proportional coefficient, lag coefficient, and inertia coefficient to obtain the prediction models between the air volume, coal volume, reductant injection amount parameters and the outlet NOx concentration that best conform to the actual system characteristics under different working conditions and the prediction model between the air volume and the CO concentration. The above prediction models are all represented by a first-order inertia system model with pure lag:

[0050]

[0051] In the formula, K pi represents the proportional link, T pi represents the response link, T di represents the pure lag link, U represents the model input, that is, the influencing parameters including air volume, coal volume, and reductant injection amount, Y represents the model output, that is, the influenced parameters including the outlet NOx concentration and the CO concentration, and G represents the system transfer function.

[0052] Preferably, in step B2, the injection amount amplitude of the reductant injection amount stepping change experiment is 5 L / h to 50 L / h.

[0053] In step B3, the optimization objective of the particle swarm optimization algorithm is to obtain the model parameters with the highest goodness of fit R 2 for the experimental results, which specifically includes the following steps:

[0054] Step B31: Input the data of the influencing parameters (such as air volume) obtained from the experiment into the model, and based on the model with the parameters to be optimized, obtain the step response curves of the corresponding influenced parameters (outlet NOx concentration, CO concentration). Compare this curve with the actual curve of the influenced parameters obtained from the experiment, and calculate the goodness of fit R 2 , and use this value as the fitness of the particle swarm optimization algorithm to participate in the optimization process;

[0055] Step B32: Based on the characteristics of the particle swarm optimization algorithm, optimize by continuously updating the position and velocity of the particles until the convergence condition is reached, and obtain the model parameters with the highest goodness of fit R 2 for the experimental results, so as to obtain the prediction model that best conforms to the characteristics of the current working conditions.

[0056] Preferably, step (3) specifically includes the following steps:

[0057] Step C1: Divide the flue gas cross-section into N regions, each region is numbered i, i ∈ [1, N], and the number of regions is equal to the number of injection valves;

[0058] Step C2: According to the working condition characteristics of the boiler determined in step B1, in each working condition interval, adopt the method of on-site experiment to measure and analyze the NOx concentration and flow velocity in each partition of the flue gas cross-section determined in step C1, and determine the distribution characteristics of the NOx concentration and flow velocity in each partition under different working conditions;

[0059] Step C3: According to the distribution characteristics of the NOx concentration and flow velocity in each partition under the above different working conditions, determine the proportion of the actual required reducing agent in each region to the total reducing agent amount, and establish a distribution table of the injection amount in each partition under each working condition;

[0060] The values in the distribution table of the injection amount in each partition are as follows:

[0061]

[0062] In the formula: K i is the value in the distribution table of the injection amount in each partition (representing the proportion of the reducing agent required in the i-th partition to the total reducing agent amount), C i is the NOx concentration in the i-th partition (the current load partition), and V i is the flue gas flow velocity in the i-th partition (the current load partition).

[0063] Preferably, step (4) specifically includes the following steps:

[0064] Step D1: Collect the CO concentration in the boiler. According to the CO concentration control target area determined in step A2, determine whether the current boiler efficiency reaches the optimum;

[0065] Step D2: If there is room for improvement in the current boiler efficiency, through the intelligent algorithm server, based on the CO and air volume prediction model, use the NSGA-II algorithm to optimize the control of the air volume, aiming to keep the CO concentration within the control target area and achieve higher boiler combustion efficiency without increasing the outlet NOx concentration;

[0066] The optimization objective function of the boiler combustion efficiency is defined as follows:

[0067]

[0068] In the formula, X is the air volume, c co is the CO concentration, X max and X min are the upper and lower limits of the air volume parameters respectively, is the boiler efficiency optimization coefficient, is the function of the outlet NOx concentration changing with the air volume;

[0069] Meanwhile, to ensure that the boiler efficiency and the outlet NOx concentration will not be worse than before control, the following constraints are added to the above optimization problem:

[0070]

[0071] In the formula, X t+1 is the air volume parameter optimized by the NSGA-II algorithm, and X t is the air volume parameter before optimization.

[0072] Preferably, in step (5), it specifically includes the following steps:

[0073] Step E1: According to the operating condition characteristics of the boiler determined in step B1, determine the rules of the fuzzy rule base. The input parameters of the fuzzy rule base are the boiler load state and the load rising and falling rate. The boiler load state is divided into α fuzzy subsets, the load rising and falling rate is divided into β fuzzy subsets, and the output is the current operating condition of the boiler, with a total of α×β operating conditions;

[0074] Step E2: If the operating condition output after defuzzification is high, medium, or low load, only use the prediction model under the corresponding operating condition; if the operating condition output after defuzzification is slightly high or slightly low, then use the prediction models under its adjacent operating conditions at the same time, and perform weighted summation on their calculation results according to their corresponding membership degrees to obtain the final result;

[0075] Step E3: Obtain the measured values of the air volume, coal volume, outlet NOx concentration, and total injection amount of the denitration device in real time. Determine the prediction model and weight required for calculation under the current working condition according to the above fuzzy rules. Use the prediction models corresponding to the air volume and coal volume for feedforward prediction, use the outlet NOx concentration as the feedback correction parameter, and calculate the next total injection amount in real time through model predictive control with feedforward, and adjust the frequency of the reducer main pipe pump.

[0076] Preferably, step (6) specifically includes the following steps:

[0077] Step F1: According to the current working condition determined in step E3, the control strategy selects the corresponding partition injection amount distribution table for the next calculation;

[0078] Step F2: Combine the total reducer injection amount L obtained in step E3 all , and cooperate with the partition injection amount distribution table to calculate the opening degrees corresponding to each valve. The required reducer amount L in each partition i The formula is as follows:

[0079] L i = K i L all

[0080] In the formula, L all is the total reducer injection amount, and L i is the required reducer amount in each partition.

[0081] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0082] 1. From the perspective of combustion process control, aiming at the problems of lag and low accuracy in reflecting combustion efficiency in the strategy of monitoring oxygen content to control air volume in traditional boilers, a neural network model of combustion efficiency and CO concentration is established. By real-time monitoring of CO concentration, the combustion state of the boiler can be quickly and accurately judged, and the air volume can be adjusted to make the boiler operate in an efficient and low-nitrogen combustion state as much as possible, achieving the purpose of reducing pollution and carbon emissions;

[0083] 2. From the perspective of controlling the injection amount of the denitration device, aiming at the problems of low control accuracy and slow response speed in the existing control methods, a prediction model between various influencing parameters and the outlet NOx concentration under multiple working conditions is established, and fuzzy rules are used to determine the boiler working conditions. The model switching under multiple working conditions is realized by means of fuzzy partitioning, and a model predictive control method with feedforward of influencing parameters and prediction of air volume commands is adopted to achieve timely and accurate control of the total amount of reductant;

[0084] 3. In terms of injection implementation, for the existing zonal injection technology that still mainly adjusts the valve opening during furnace shutdown and does not adjust the injection valve opening in real time according to the complex flow field changes under different working conditions, the flow field in the flue under all working conditions is modeled, a zonal injection quantity distribution table is established according to the actual characteristics of the flue, and the method of real-time zonal injection under all working conditions is adopted to improve the mixing uniformity of the reducing agent and the flue gas, improve the denitration efficiency and accuracy, and save costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 is the principle block diagram of the control method of the present invention;

[0086] Figure 2 is the flowchart of the combustion optimization control module based on CO monitoring of the present invention;

[0087] Figure 3 is the flowchart of the total reducing agent control module based on the prediction of parameters such as air volume of the present invention;

[0088] Figure 4 is the flowchart of the zonal injection control module based on the zonal injection quantity distribution table of the present invention;

[0089] Figure 5 is the flowchart of the method proposed by the present invention;

[0090] Figure 6 is the comparison chart of the unit thermal efficiency before and after optimization control;

[0091] Figure 7 is the comparison chart of the control effect under variable load conditions;

[0092] Figure 8 is the comparison chart of the control effect under stable load conditions;

[0093] Figure 9 is the comparison chart of the probability density distribution of the outlet NOx concentration under variable load conditions;

[0094] Figure 10 is the comparison chart of the probability density distribution of the outlet NOx concentration under stable load conditions. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0095] The present invention will be further described below in conjunction with the drawings and specific embodiments, but the scope to be protected by the present invention is not limited thereto. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0096] Embodiment 1

[0097] Refer to Figures 1 - 5, a method for collaborative optimization control of boiler combustion - denitration process. The system constructed by the method of the present invention includes an optimization part of the combustion process based on CO monitoring (combustion process optimization), a total reductant injection control part based on predictive control (total reductant control), and a real - time zonal injection part of the reductant under all operating conditions (zonal injection control). By optimizing the boiler combustion efficiency, accurately controlling the total reductant injection amount, and improving the mixing uniformity of flue gas and reductant, the overall operating efficiency of the boiler system is improved and the operating cost is reduced.

[0098] The system constructed by the method of the present invention includes a boiler, a denitration device and its auxiliary facilities, on - line monitoring equipment, an intelligent algorithm server, and on - line control equipment. The intelligent algorithm server realizes real - time communication with the on - line monitoring equipment through object linking and embedding process control services. The on - line control equipment is connected to the intelligent algorithm server and implements control according to the instructions of the intelligent algorithm server.

[0099] The monitoring indexes of the on - line monitoring equipment include but are not limited to flue gas outlet temperature, NOx concentration, CO concentration, primary air volume, secondary air volume, flue gas oxygen content, coal amount, reductant main pipe flow pump frequency, reductant main pipe flow, opening degrees of injection valves in each area, etc. The on - line monitoring equipment has functions of input and output of historical data and real - time data, and realizes information intercommunication and real - time feedback with the intelligent algorithm server.

[0100] The reductant includes ammonia water, urea solution, and liquid ammonia with different concentrations;

[0101] The denitration device includes SCR and SNCR;

[0102] The control objects of the boiler combustion process include the opening degrees of each damper, and the control objects of the denitration device include the opening degrees of injection valves in each area and the reductant main pipe pump frequency.

[0103] The method for collaborative optimization control of the boiler combustion - denitration process includes the following steps:

[0104] (1) Based on historical data, adopt the long - short - term memory neural network algorithm to establish a neural network model between combustion efficiency and CO concentration;

[0105] For a certain power plant, use the DCS system and the corresponding database to collect data such as air volume, coal amount, temperature, etc. under different load conditions in the past year; use the continuous flue gas detection system to collect data such as CO concentration and outlet NOx concentration; after pre - processing the data, adopt the LSTM algorithm to establish a neural network model between combustion efficiency and CO concentration under multiple operating conditions;

[0106] (2) Based on historical data analysis, on-site tests, and particle swarm parameter optimization methods, prediction models for coal quantity Coal, total reductant injection quantity L, air volume X, and outlet NOx concentration under different boiler operating conditions are obtained, as well as a prediction model for air volume X and CO concentration;

[0107] G Nox = f(Coal, L, X)

[0108] G co = f(X)

[0109] The clustering analysis method is used to classify the data to determine the characteristics of the boiler operating condition intervals for further experiments. Experiments with constant reductant injection quantity and step changes in reductant injection quantity are carried out within each operating condition interval to obtain experimental data on the corresponding outlet NOx concentration for parameters such as air volume, coal quantity, and reductant injection quantity under different operating conditions. The particle swarm algorithm is used to optimize the parameters of the prediction model: proportional coefficient, lag coefficient, and inertia coefficient, so as to obtain the prediction model for the outlet NOx concentration with parameters such as air volume, coal quantity, and reductant injection quantity that best conform to the actual system characteristics under each operating condition, as well as the prediction model for air volume and CO concentration;

[0110] (3) According to the distribution of the partition injection ports of the denitration device, the flue gas duct cross-section is divided into multiple partitions. Through on-site tests, NOx concentration values within the partitions are obtained using NOx probes in each area, and the flue gas flow velocity in different areas is obtained using differential pressure gauges to determine the distribution of NOx concentration and flue gas flow velocity in the flue gas duct under multiple operating conditions, and an analysis is made to obtain the partition injection quantity distribution table under multiple operating conditions;

[0111] In actual situations, the flue gas duct of the power plant denitration equipment is very large. Therefore, the flue gas duct is divided into N areas, each area is numbered i, i ∈ [1, N]. Through on-site tests and historical data, the NOx concentration C i and flue gas flow velocity V i in the flue gas duct under different loads are obtained. Since C i ×V i can obtain the total amount of NOx in the flue gas duct partition. If uniform mixing is to be achieved, the proportion of the reductant in each partition should correspond one-to-one with the NOx amount distribution proportion. Therefore, the actual required reductant amount in each area of the flue gas duct can be calculated. The formula is as follows:

[0112]

[0113] In the formula: K i is the value in the partition injection quantity distribution table, L i is the required reductant amount in each partition (the amount of reductant required for the i-th partition in the partition injection quantity distribution table), L allis the total amount of reductant injection (the total amount of reductant injection calculated based on real-time operating conditions data);

[0114] (4) The combustion optimization control module controls the air volume based on the real-time CO concentration to optimize the in-furnace combustion efficiency, and uses the air volume control instruction as a feedforward prediction for the total amount control module of the reductant;

[0115] (5) According to the real-time operating conditions data, the controller judges the prediction model to be adopted and its weight through the fuzzy rule base, calculates and adjusts the frequency of the reductant main pipe pump to control the total amount of reductant injection;

[0116] (6) According to the real-time operating conditions data, select the corresponding distribution table of injection amounts for each zone, finally determine the real-time opening of each valve, and adjust the opening of the injection valves in each zone in real time.

[0117] Among them, step (1) specifically includes the following steps:

[0118] Step A1: Collect the CO concentration and combustion efficiency parameters under different load conditions of the boiler, preprocess the data, input it into the LSTM neural network model for training, and establish the LSTM neural network model between the combustion efficiency and the CO concentration;

[0119] Step A2: According to the obtained LSTM neural network model of combustion efficiency and CO concentration, determine the CO concentration range when the combustion efficiency is higher than the average value, and set the above CO concentration range as the control target;

[0120] In step A1, the formula for calculating the combustion efficiency is as follows:

[0121]

[0122] In the formula: η is the combustion efficiency Q 1 is the output heat, Q r is the output heat, Q 2 is the heat of the exhaust gas, Q 3 is the heat loss due to a part of the combustible gas carried away by the flue gas, Q 4 is the heat loss due to the mixing of pulverized coal and fly ash without participating in combustion, Q 5 is the heat from the furnace wall and pipes to the surrounding environment, Q 6 is the heat carried away by the ash slag and cooling.

[0123] Preferably, in step A1, the method for preprocessing the data includes four steps: outlier removal, CEMS purge period removal, data smoothing, and standardization processing;

[0124] Outlier removal adopts the method of data cleaning, deletes the outliers, and replaces them with the average value of adjacent data.

[0125] The CEMS purging period is eliminated by means of data cleaning. The data within half an hour after the CEMS purging signal is connected is cleared to avoid the influence of invalid data on model training;

[0126] The smoothing data adopts the two-dimensional Gaussian filtering method, and the specific formula is as follows:

[0127]

[0128] In the formula, G(x,y) is the value after Gaussian filtering, x and y are the data to be filtered, and σ is a constant that determines the width of the Gaussian filtering function;

[0129] The normalization process uses the min-max normalization method to normalize the preprocessed data.

[0130] Preferably, in step A1, the method for establishing the LSTM neural network model between the combustion efficiency and the CO concentration includes constructing a validation set and a training set, and using the training set to train the neural network model. The specific process is as follows:

[0131] The processed data is randomly divided into a validation set and a training set. The training set is used to train the LSTM model. Each time, the training set data segment is input into the LSTM layer of the LSTM model, and the output data is processed by the fully connected layer and then output to the attention layer, and then processed by another fully connected layer to obtain the boiler combustion efficiency value calculated according to the CO concentration; Subsequently, this value is compared with the actual boiler combustion efficiency value, and the error is backpropagated to correct the parameters of each fully connected layer. Repeat this process to obtain the required LSTM neural network model.

[0132] Step (2) specifically includes the following steps:

[0133] Step B1: Analyze the historical operation data of the unit, perform cluster analysis according to the boiler load status and the load increase and decrease rate, and determine the load increase, flat, and decrease working condition characteristics of the boiler at high (90%-100%), slightly high (70%-90%), medium (50%-70%), slightly low (40%-50%), and low load (<40%). Use the following for the next step of the experiment;

[0134] Step B2: According to the boiler working condition characteristics determined in step B1, conduct experiments with the reductant injection amount unchanged and the reductant injection amount step-changing within each working condition interval to obtain the experimental data of the outlet NOx concentration corresponding to the parameters including air volume, coal volume, and reductant injection amount under different working conditions and the experimental data of the CO concentration corresponding to the air volume;

[0135] Step B3: Use the particle swarm optimization algorithm to optimize the prediction model parameters including the proportional coefficient, lag coefficient, and inertia coefficient, and obtain the prediction models between the air volume, coal volume, reductant injection volume parameters that best conform to the actual system characteristics and the outlet NOx concentration and the prediction model between the air volume and the CO concentration under different working conditions.

[0136] The above prediction models are all represented by a first-order inertial system model with pure lag:

[0137]

[0138] In the formula, K pi represents the proportional link, T pi represents the response link, T di represents the pure lag link, U represents the model input, that is, the influencing parameters including the air volume, coal volume, and reductant injection volume, Y represents the model output, that is, the outlet NOx concentration and the CO concentration, and G represents the system transfer function;

[0139] For further optimization, in step B2, the injection volume range of the reductant injection step change test is 5 L / h to 50 L / h;

[0140] The optimization objective of the particle swarm optimization algorithm used in step B3 is to obtain the model parameters with the largest goodness of fit R 2 The specific steps are as follows:

[0141] Step B31: Input the data of the influencing parameters (such as air volume) obtained from the experiment into the model, and based on the model with the parameters to be optimized, obtain the step response curves of the corresponding influenced parameters (outlet NOx concentration, CO concentration). Compare this curve with the actual influenced parameter curve obtained from the experiment, and calculate the goodness of fit R 2 , and use this value as the fitness of the particle swarm optimization algorithm to participate in the optimization process;

[0142] Step B32: Based on the characteristics of the particle swarm optimization algorithm, optimize by continuously updating the positions and velocities of a large number of particles until the convergence condition is reached, and obtain the model parameters with the largest goodness of fit R 2 , so as to obtain the prediction model that best conforms to the characteristics of the current working condition.

[0143] Step (3) specifically includes the following steps:

[0144] Step C1: Divide the flue gas cross-section into N regions, each region is numbered i, i ∈ [1, N], and the number of regions is equal to the number of injection valves;

[0145] Step C2: According to the operating characteristics of the boiler determined in Step B1, in each operating condition range, conduct on-site tests to measure and analyze the NOx concentration and flow rate in each partition of the flue gas cross-section determined in Step C1, and determine the distribution characteristics of the NOx concentration and flow rate in each partition under different operating conditions;

[0146] Step C3: According to the distribution characteristics of the NOx concentration and flow rate in each partition under the above different operating conditions, determine the proportion of the actual required reducing agent in each area to the total reducing agent amount, and establish a distribution table of the injection amount in each partition under each operating condition;

[0147] The values in the distribution table of the injection amount in each partition are calculated by the following formula:

[0148]

[0149] In the formula: K i is the value in the distribution table of the injection amount in each partition (representing the proportion of the reducing agent required in the i-th partition to the total reducing agent amount in the distribution table of the injection amount in each partition), C i is the NOx concentration in the current load partition, and V i is the flue gas flow rate in the current load partition.

[0150] Step (4) specifically includes the following steps:

[0151] Step D1: Collect the CO concentration in the boiler, and determine whether the current boiler efficiency reaches the optimum according to the CO concentration control target area determined in Step A2;

[0152] Step D2: If there is room for improvement in the current boiler efficiency, through the intelligent algorithm server, based on the CO concentration prediction model, use the NSGA-II algorithm to optimize the control of the air volume, with the aim of keeping the CO concentration within the control target area, having a higher boiler combustion efficiency, and not increasing the outlet NOx concentration;

[0153] The optimization objective function of the boiler combustion efficiency is defined as follows:

[0154]

[0155] In the formula, X is the air volume, X max and X min are respectively the upper and lower limits of the air volume parameters, is the boiler efficiency optimization coefficient, is the function of the outlet NOx concentration changing with the air volume;

[0156] At the same time, add the following constraints to the above optimization problem:

[0157]

[0158] In the formula, Xt+1 is the air volume parameter optimized by the NSGA-II algorithm, X t is the air volume parameter before optimization.

[0159] Step (5) specifically includes the following steps:

[0160] Step E1: According to the operating condition characteristics of the boiler determined in Step B1, determine the rules of the fuzzy rule base. The input parameters of the fuzzy rule base are the boiler load state and the load increase / decrease rate. The boiler load state is divided into α fuzzy subsets, the load increase / decrease rate is divided into β fuzzy subsets, and the output is the current operating condition of the boiler, with a total of α×β operating conditions;

[0161] Step E2: If the operating condition output after defuzzification is high, medium, or low load, only use the prediction model corresponding to the operating condition; if the operating condition output after defuzzification is slightly high or slightly low, use the prediction models under its adjacent operating conditions at the same time, and perform weighted summation on their calculation results according to their corresponding membership degrees to obtain the final result;

[0162] Step E3: Obtain the measured values of the air volume, coal quantity, outlet NOx concentration, and injection total amount of the denitration device in real time. According to the above fuzzy rules, determine the prediction model and weight value required for calculation under the current operating condition. Use the prediction models corresponding to the air volume and coal quantity for feedforward prediction, use the outlet NOx concentration as the feedback correction parameter, and calculate the next-step reducing agent injection total amount L all in real time through model predictive control with feedforward, and adjust the frequency of the reducing agent main pipe pump.

[0163] Step (6) specifically includes the following steps:

[0164] Step F1: According to the current operating condition determined in Step E3, the control strategy selects the corresponding partition injection amount distribution table for the next calculation;

[0165] Step F2: Combine the reducing agent injection total amount L all obtained in Step E3, and calculate the opening degrees corresponding to each valve in cooperation with the partition injection amount distribution table. The required reducing agent amount L i in each partition is as follows:

[0166] L i = K i L all

[0167] In the formula, L all is the reducing agent injection total amount, and L i is the required reducing agent amount in each partition.

[0168] Example 2

[0169] Taking the actual control process and control effect of a 1000MW opposed firing pulverized coal boiler as an example, the content of the present invention will be described in detail:

[0170] (1) Select parameters such as air volume, CO concentration, unit load, burner temperature, and NOx at the outlet of the denitration device in the DCS system of the unit in the recent month, calculate the thermal efficiency parameter, and use the LSTM algorithm to obtain the determination coefficient R 2 > 0.94 neural network model of CO concentration and combustion efficiency, and take the CO concentration range when the combustion efficiency is greater than the average value as the control target range;

[0171] (2) First, adopt the method of cluster analysis to divide the boiler operating conditions into 5 types according to the load status: high, slightly high, medium, slightly low, and low. Each type is divided into 3 operating conditions: rising, falling, and stable, for a total of 15 operating conditions; conduct control variable experiments under these 15 operating conditions to obtain a prediction model. For example: when reducing the load under high load conditions, adjust the coal quantity to obtain the change data of the NOx concentration at the outlet, and optimize the prediction model parameters through the particle swarm algorithm to obtain R 2 > 0.9 coal-NOx concentration prediction model at the outlet. Based on the above combination of experimental data and particle swarm optimization, obtain the prediction models of coal quantity, reducing agent dosage, and air volume on the NOx concentration at the outlet under 15 operating conditions, and obtain the prediction model of air volume on CO concentration for subsequent real-time control;

[0172] (3) The flue gas duct area of the denitration device is 200m 2 , divide it into 16 partitions, collect data such as NOx concentration and flue gas flow velocity in the flue gas duct partitions multiple times under the above 15 operating conditions, and calculate the distribution table of partition injection amounts under 15 operating conditions according to the formula;

[0173] (4) According to the proposed method and system, the algorithm server monitors the DCS system data in real time, optimizes and calculates the optimal air volume according to the optimization function, inputs the air volume command as a control parameter into the DCS system, and at the same time, the air volume command data, coal quantity data, boiler operating condition data, etc. will be transmitted to the total reducing agent control module; after the total control module determines the prediction model and weight to be used currently according to the fuzzy rule, calculates the total amount of reducing agent required and the corresponding frequency of the reducing agent main pipe pump, inputs this data as a control parameter into the DCS system, and transmits the current operating condition data and the total reducing agent data to the partition injection control module; the partition injection control module calculates the opening degree of each partition injection valve according to the partition injection amount distribution table and the total reducing agent amount under the current operating condition, inputs this command as a control parameter into the DCS system, and finally realizes the coordinated optimization control of the boiler thermal efficiency and the NOx concentration at the outlet.

[0174] To verify the influence of optimizing and controlling the air volume by this method on the unit thermal efficiency, the changes in the boiler thermal efficiency before and after optimization under different loads were analyzed, and the results are asFigure 6 As shown. The method proposed by the present invention has a better boiler thermal efficiency than the original operating conditions under different loads. The optimized thermal efficiency ranges from 94.28% to 94.41%, and the efficiency improvement ratio is about 0.2% compared with that before optimization.

[0175] The method of the present invention (combustion optimization - multi - model control) is compared and analyzed with model control and multi - model control. The comparison results of the control effects under variable load conditions, the control effects under stable load conditions, the probability density distribution comparison of the outlet NOx concentration under variable load conditions, and the probability density distribution comparison of the outlet NOx concentration under stable load conditions are as Figures 7 - 10 shown. According to Figure 7 , Figure 8 shown, the method proposed by the present invention has a smaller fluctuation range, a faster variable - load control response speed under variable - load conditions, and the outlet NOx concentration curve is closer to the set 35mg / m 3 emission set value, and its fluctuation range is ±8.10mg / m 3 ; under stable - load conditions, while ensuring the relative stability of the outlet NOx concentration of denitration, it can quickly and accurately callback when the model deviates from the target value, and the fluctuation range of the outlet NOx concentration is ±5.80mg / m 3 . According to Figure 9 , under variable - load conditions, the mean value of the controlled variable of the method proposed by the present invention is the same as the set value and has a lower variance; while for the multi - model control and single - model control methods, there are certain deviations between the mean value and the set value. According to Figure 10 , under stable - load conditions, the method proposed by the present invention has equally excellent control characteristics (accurate set - value tracking and small variance), which are better than the other two control methods, greatly improving the overall control effect of the denitration device.

[0176] The present invention has been described in detail above in combination with embodiments, but the content described is only the specific implementation manner of the present invention, and it cannot be understood as a limitation of the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements made according to the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.

Claims

1. A collaborative optimization control method for boiler combustion-denitration process, characterized in that it includes a combustion optimization control module based on CO monitoring, a total reductant control module based on parameter prediction including air volume, and a sectional injection control module based on a sectional injection volume distribution table; The combustion optimization control module based on CO monitoring is used to monitor the change of CO concentration and regulate the air volume; The total reductant control module based on parameter prediction including air volume is used to monitor the key parameters affecting the outlet NOx concentration, predict the total amount of reductant required for the denitration device, control the frequency of the reductant main pipe pump, and achieve the control goal of minimizing the reductant consumption under the condition that the outlet NOx concentration meets the standard; The sectional injection control module based on the sectional injection volume distribution table controls the opening degree of each injection valve in the flue in real time according to the sectional injection volume distribution table established according to the NOx flow velocity and concentration distribution in the flue under different working conditions; The control method includes the following steps: (1) Based on historical data, use the long short-term memory neural network algorithm to establish a neural network model between combustion efficiency and CO concentration; (2) Based on historical data analysis, on-site tests, and particle swarm parameter optimization method, obtain the prediction models of coal quantity Coal, total reductant injection amount L, air volume X and outlet NOx concentration under different working conditions of the boiler, and the prediction model of air volume X and CO concentration; G Nox = f(Coal, L, X) G co = f(X) Wherein, G Nox is the predicted model of the NOx concentration at the outlet, which is related to the coal quantity, the total amount of reductant injected, and the air volume. G co is the predicted model of the CO concentration, which is related to the air volume; (3) According to the distribution of the sectional injection ports of the denitration device, divide the flue cross-section into multiple sections, and through on-site tests, determine the distribution of NOx concentration and flue gas flow velocity in the flue gas at each section under multiple working conditions, and analyze to obtain the sectional injection volume distribution table under multiple working conditions; (4) The combustion optimization control module optimizes the combustion efficiency in the furnace by controlling the air volume based on the real-time CO concentration, and takes the air volume control instruction as a feed-forward prediction and inputs it into the total reductant control module; (5) According to the real-time working condition data, the intelligent algorithm server judges the real-time boiler working condition through the fuzzy rule base, determines the multi-condition prediction model to be adopted and the weight of each prediction model, and calculates and adjusts the frequency of the reductant main pipe pump or the opening degree of the total ammonia water regulating valve to control the total reductant injection amount; (6) According to the real-time working condition data, select the corresponding sectional reductant injection volume distribution table, finally determine the real-time opening degree of each valve, and adjust the opening degree of each sectional reductant injection valve in real time; The system constructed based on the above method includes a boiler, a denitration device and its auxiliary facilities, on-line monitoring equipment, an intelligent algorithm server and a control device. The intelligent algorithm server realizes real-time communication with the on-line monitoring equipment through the object linking and embedding process control service. The control device is connected to the intelligent algorithm server and implements control according to the instructions of the intelligent algorithm server; The monitoring indicators of the on-line monitoring equipment include flue gas outlet temperature, NOx concentration, CO concentration, primary air volume, secondary air volume, flue gas oxygen content, coal quantity, reductant main pipe flow pump frequency, reductant main pipe flow, and opening degree of each area injection valve; the on-line monitoring equipment has the functions of input and output of historical data and real-time data, and realizes information intercommunication and real-time feedback with the intelligent algorithm server; The reducing agent includes ammonia water, urea solution and liquid ammonia with different concentrations; The denitration device includes SCR and SNCR; The control objects of the denitration device include the opening degree of the reducing agent injection valve in each zone and the frequency of the reducing agent main pipe pump.

2. The coordinated optimization control method for the boiler combustion-denitration process according to claim 1, characterized in that Step (1) specifically includes the following steps: Step A1: Collect the CO concentration under different load conditions of the boiler, calculate the combustion efficiency by the inverse balance method, preprocess the data, input it into the long short-term memory neural network model for training, and establish the long short-term memory neural network model between the combustion efficiency and the CO concentration; Step A2: According to the long short-term memory neural network model between the obtained combustion efficiency and the CO concentration, determine the CO concentration range when the combustion efficiency is higher than the average value, and set the above CO concentration range as the control target.

3. The coordinated optimization control method for the boiler combustion-denitration process according to claim 2, characterized in that: In step A1, the formula for calculating the combustion efficiency is as follows: where η is the combustion efficiency, Q 1 is the output heat, Q r is the input heat, Q 2 is the heat in the flue gas, Q 3 is the heat loss due to some combustible gas carried away by the flue gas, Q 4 is the heat loss due to pulverized coal mixed with fly ash and not participating in combustion, Q 5 is the heat from the furnace wall and pipes to the surrounding environment, Q 6 is the heat carried away by the ash slag and cooling; In step A1, the method for preprocessing the data includes four steps: outlier removal, CEMS purge period removal, data smoothing and normalization; Outlier removal is carried out by data cleaning, deleting the outliers and replacing them with the average value of adjacent data; CEMS purge period removal is carried out by data cleaning, clearing the data in the half hour after the CEMS purge signal is connected to avoid the influence of invalid data on model training; Data smoothing adopts the two-dimensional Gaussian filtering method, and the specific formula is as follows: In the formula, G(x,y) is the value after Gaussian filtering, x,y are the data to be filtered, and σ is a constant that determines the width of the Gaussian filtering function; Normalization processing uses the min-max normalization method to normalize the preprocessed data; The method for establishing the long short-term memory neural network model between the combustion efficiency and the CO concentration includes constructing a validation set and a training set and training the neural network model with the training set. The specific process is as follows: Randomly divide the processed data into a validation set and a training set, use the training set to train the long short-term memory neural network model. Each time, input the training set data segment into the long short-term memory layer of the long short-term memory neural network model, the output data is processed by the fully connected layer and then output to the attention layer, and then processed by another fully connected layer to obtain the boiler combustion efficiency value calculated according to the CO concentration; Subsequently, compare this value with the actual boiler combustion efficiency value, backpropagate the error, and correct the parameters of each fully connected layer. Repeat this process to obtain the required long short-term memory neural network model.

4. The coordinated optimization control method for the boiler combustion-denitration process according to claim 3, characterized in that Step (2) specifically includes the following steps: Step B1: Analyze the historical operation data of the unit, perform clustering analysis based on the boiler load status and the load rising and falling rates, determine the characteristics of the load rising, flat, and falling conditions of the boiler at high, slightly high, medium, slightly low, and low loads, and use them for the next step of experiments; among them, the load status is high load when it is 90 - 100%, slightly high load when it is 70 - 90%, medium load when it is 50 - 70%, slightly low load when it is 40 - 50%, and low load when it is <40%. Step B2: According to the boiler condition characteristics determined in Step B1, conduct experiments with the reductant injection amount unchanged and the reductant injection amount step - changing within each condition range, and obtain the experimental data of the outlet NOx concentration corresponding to parameters including air volume, coal amount, and reductant injection amount under different conditions, as well as the experimental data of the CO concentration corresponding to the air volume. Step B3: Use the particle swarm algorithm to optimize the prediction model parameters including the proportional coefficient, lag coefficient, and inertia coefficient, and obtain the prediction models between the air volume, coal amount, reductant injection amount parameters that best conform to the actual system characteristics and the outlet NOx concentration under different conditions, as well as the prediction model between the air volume and the CO concentration; the above - mentioned prediction models are all represented by a first - order inertial system model with pure lag: where K pi represents the proportional link, T pi represents the response link, T di represents the pure dead-time link, U represents the model input, i.e., the influencing parameters including the air volume, coal volume, and reductant injection volume, Y represents the model output, i.e., the influenced parameters including the outlet NOx concentration and CO concentration, and G represents the system transfer function.

5. The collaborative optimization control method for the boiler combustion - denitration process according to claim 4, characterized in that: In Step B2, the injection amount amplitude of the reductant injection amount step - changing experiment is 5 L / h to 50 L / h; The optimization objective of the particle swarm optimization algorithm adopted in step B3 is to obtain the model parameters with the highest goodness of fit R 2 with the experimental results, which specifically includes the following steps: Step B31: Input the influence parameter data obtained from the experiment into the model. Based on the model of the parameter to be optimized, obtain the step response curve of the corresponding influenced parameter. Compare this curve with the actual influenced parameter curve obtained from the experiment, and calculate the goodness of fit R 2 , and use this value as the fitness of the particle swarm optimization algorithm to participate in the optimization process; Step B32: Based on the characteristics of the particle swarm algorithm, optimize by continuously updating the position and velocity of the particles until the convergence condition is reached, and obtain the goodness of fit R 2 The maximum model parameter, thereby obtaining the prediction model that best conforms to the characteristics of the current working condition.

6. The collaborative optimization control method for the boiler combustion - denitration process according to claim 4, characterized in that: Step (3) specifically includes the following steps: Step C1: Divide the flue gas cross - section into N regions, each region is numbered i, i ∈ [1, N], and the number of regions is equal to the number of injection valves; Step C2: According to the boiler condition characteristics determined in Step B1, adopt the method of on - site experiments within each condition range, measure and analyze the NOx concentration and flow rate in each partition of the flue gas cross - section determined in Step C1, and determine the distribution characteristics of the NOx concentration and flow rate in each partition under different conditions; Step C3: According to the distribution characteristics of the NOx concentration and flow rate in each partition under the above - mentioned different conditions, determine the proportion of the actual required reductant in the total reductant amount in each region, and establish a distribution table of the partition injection amounts under each condition; The values in the distribution table of the partition injection amounts are as follows: Where: K i is the value in the distribution table of the partition injection amount, C i is the NOx concentration in the i-th partition, V i is the flue gas flow velocity in the i-th partition.

7. The collaborative optimization control method for the boiler combustion - denitration process according to claim 6, characterized in that: Step (4) specifically includes the following steps: Step D1: Collect the CO concentration in the boiler, and determine whether the current boiler efficiency reaches the optimum according to the CO concentration control target area determined in Step A2; Step D2: If there is room for improvement in the current boiler efficiency, then through the intelligent algorithm server, based on the CO concentration prediction model, use the NSGA - Ⅱ algorithm to optimize the control of the air volume; The optimization objective function of the boiler combustion efficiency is defined as follows: s.t X min <X<X max Where X is the air volume, c co is the CO concentration, X max and X min are the upper and lower limits of the air volume parameter respectively, is the boiler efficiency optimization coefficient, is a function of the outlet NOx concentration varying with the air volume; At the same time, add the following constraints to the above optimization problem: where X t+1 is the air volume parameter optimized by the NSGA-II algorithm, and X t is the air volume parameter before optimization.

8. The collaborative optimization control method for the boiler combustion - denitration process according to claim 7, characterized in that: In step (5), it specifically includes the following steps: Step E1: According to the operating condition characteristics of the boiler determined in step B1, determine the rules of the fuzzy rule base. The input parameters of the fuzzy rule base are the boiler load state and the load rising / falling rate. The boiler load state is divided into α fuzzy subsets, and the load rising / falling rate is divided into β fuzzy subsets. The output is the current operating condition of the boiler, and there are α×β kinds of operating conditions in total; Step E2: If the operating condition output after defuzzification is high, medium, or low load, only use the prediction model corresponding to the operating condition; if the operating condition output after defuzzification is slightly high or slightly low, use the prediction models under its adjacent operating conditions at the same time, and perform weighted summation on their calculation results according to their corresponding membership degrees to obtain the final result; Step E3: Obtain the air volume, coal volume, outlet NOx concentration, and the measured value of the total injection amount of the denitration device in real time. Determine the prediction model and weight required for calculation under the current working condition according to the above fuzzy rules. Use the prediction models corresponding to the air volume and coal volume for feed-forward prediction, use the outlet NOx concentration as the feedback correction parameter, and calculate the next total injection amount L in real time through model predictive control with feed-forward all , and adjust the frequency of the reductant main pipe pump.

9. According to the boiler combustion-denitration process collaborative optimization control method described in claim 8, it is characterized in that: Step (6) specifically includes the following steps: Step F1: According to the current operating condition determined in step E3, the control strategy selects the corresponding partition injection amount distribution table for the next calculation; Step F2: Combine the total amount of reducing agent injection L obtained in Step E3 all , and calculate the opening degree corresponding to each valve in combination with the injection amount distribution table for each zone. The amount of reducing agent required in each zone is L i The formula is as follows: L i = K i L all where L all is the total amount of reductant injected, in L i is the amount of reductant required in each zone.

Images