Coal quality feedforward dust control system and method

Abstract

The present application belongs to the technical field of intelligent dust removal, and discloses a coal quality feedforward dust removal control system and method. The method comprises: collecting data and performing integrity analysis and data cleaning to obtain coal quality feedforward data X, working condition data Y and equipment data; through dynamic Pearson correlation analysis, a strong correlation subset M and a key electric field identifier F are obtained; coupling analysis is performed on X to obtain a coupling index T; theoretical dust accumulation is obtained according to T and Y analysis; dimensionality reduction is performed on the strong correlation subset, combined with the theoretical dust accumulation, to obtain a strong correlation feature subset, which is then normalized and input into an LSTM model together with F to obtain control parameters; online rolling optimization is performed based on the MOPSO algorithm to obtain optimized control parameters; and the system is controlled according to the optimized control parameters for collaborative execution. The present application effectively solves the regulation lag problem of the traditional dust removal system relying on real-time emission feedback, and breaks through the technical bottleneck that energy consumption and emission are difficult to balance under single-objective optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent dust removal, and more particularly to a coal quality feedforward dust removal control system and method. BACKGROUND

[0002] Dust generated by coal-fired power plants is one of the atmospheric pollutants. With the increasingly stringent environmental protection requirements, it is necessary to control dust emission more accurately. However, real-time changes in factors such as unit load, coal quality, and furnace air volume will change the dust concentration, particle size, and flue gas flow rate in the flue gas. The traditional electrostatic precipitator control method is single and often cannot track these changes for real-time and accurate control. To meet environmental protection requirements, more electric fields are usually invested, and the working voltage of the dust remover cannot be adjusted in real time online according to the operating conditions of the power plant, and is usually kept at high parameters, resulting in poor economic efficiency of the electric dust removal and prominent energy consumption problems.

[0003] Chinese patent application No. CN111570093A discloses an electric dust removal energy-saving control method and system based on boiler coal quantity and air quantity, which comprises: obtaining the current total coal quantity, current total air quantity, and dust content at the outlet of the electric dust remover; calculating the target value of the electric field secondary current according to a formula, and adjusting the current electric field secondary current value of the electric dust remover to the target value of the electric field secondary current. The invention takes the total air quantity of the boiler as a feedforward signal, which can increase the electric dust removal parameters in advance when the load increases, and delay the reduction of the electric dust removal parameters during load reduction, thereby improving the dynamic control effect of the dust at the outlet of the electric dust remover during dynamic changes in unit load.

[0004] Although the above method can meet most scenarios, research and actual application of the above method and prior art have found that the above method and prior art at least have the following defects:

[0005] The above method only takes the total coal quantity and air quantity as control inputs. This method implicitly assumes that the total coal quantity is completely positively correlated with the coal quality characteristics, but in practice, the ash content of different coal types may differ by several times, and the total coal quantity cannot reflect the direct impact of ash content on the ash deposition rate. The above method only adjusts the secondary current linearly through the total coal quantity and air quantity, which is essentially a single-target feedback control based on emission deviation, which cannot balance the contradiction between energy consumption and emission under high-ash conditions.

[0006] In view of this, the present application provides a coal quality feedforward dust removal control system and method to solve the above problems. SUMMARY

[0007] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the present application provides the following technical solutions: a coal quality feedforward dust removal control method, comprising the following steps:

[0008] Collect data and perform integrity analysis and data cleaning to obtain coal quality feedforward data X, working condition data Y and equipment data Z;

[0009] Obtain a strong correlation subset M and a key electric field identifier F through dynamic Pearson correlation analysis;

[0010] Perform coupling analysis on the coal quality feedforward data X to obtain a coupling index T;

[0011] Obtain a theoretical soot accumulation amount according to the coupling index T and the working condition data Y; perform PCA dimension reduction on the strong correlation subset, and combine the theoretical soot accumulation amount to obtain a strong correlation feature subset G;

[0012] Normalize the strong correlation feature subset G, and input the normalized strong correlation feature subset G and the key electric field identifier F into an LSTM model to obtain control parameters, including high-voltage rectification parameters and low-voltage vibration parameters;

[0013] Perform online rolling optimization on the control parameters based on a MOPSO algorithm to obtain optimized control parameters;

[0014] Control the system according to the optimized control parameters to perform collaborative execution.

[0015] Further, the method for obtaining the coupling index comprises:

[0016] Perform coupling analysis on the ash content and the moisture content in the coal quality feedforward data, and combine a preset moisture content threshold to analyze and obtain the coupling index.

[0017] Further, the optimized control parameters include optimized high-voltage rectification parameters and optimized low-voltage vibration parameters, and the method for the control system to perform collaborative execution comprises:

[0018] Statistically obtain a non-key electric field set and a neighboring field set of the electric field i, extract the vibration state of the electric field j at time t based on the optimized low-voltage vibration parameters, including being vibrated and not being vibrated;

[0019] Detect the number of neighboring vibration electric fields of the electric field i, take the base value of the optimized high-voltage rectification parameter as a reference, analyze and obtain a dynamic flow reduction coefficient, calculate a coal ash adjustment factor based on the dynamic flow reduction coefficient, and obtain the collaborative high-voltage rectification parameter value of the electric field i; the coal ash adjustment factor is calculated based on the real-time ash content; and the collaborative high-voltage rectification parameter value is taken as the final executed high-voltage rectification parameter value;

[0020] When the predicted emission reaches a preset emission limit value, cancel the additional flow reduction based on the emission constraint logic of the optimized high-voltage rectification parameter;

[0021] When the spark rate is higher than a preset spark rate constraint, halve the dynamic flow reduction coefficient through the device protection mechanism for linked optimization of the low-voltage vibration parameter.

[0022] Further, the method for obtaining the optimized high-voltage rectification parameters and the optimized low-voltage vibration parameters comprises:

[0023] defining a target function, the target function comprising three sub-targets of emission deviation minimization, total energy consumption minimization, and vibration energy consumption minimization;

[0024] defining the high-voltage rectification parameters and the low-voltage vibration parameters as particles;

[0025] setting a constraint relaxation function according to the ash content;

[0026] randomly generating N particles within a change range L of the high-voltage rectification parameters and the low-voltage vibration parameters output by the LSTM model, wherein L is a value limiting the change range of the particles;

[0027] adding k random disturbances to the ash content working conditions with the ash content greater than the ash content threshold, wherein k is a random disturbance value added to the ash content working conditions with the ash content greater than the ash content threshold;

[0028] standardizing the target function value to obtain a standard target, calculating the fitness of each particle according to the standard target, and performing non-dominated sorting to obtain non-dominated solutions;

[0029] screening the particles according to the non-dominated solutions, updating the individual optimal solution and the global optimal solution of the particles, adjusting the speed and position of the particles, and obtaining an updated non-dominated solution set;

[0030] storing the updated non-dominated solution set in an external archive and regularly updating the external archive;

[0031] judging whether a preset stop condition is met;

[0032] extracting the non-dominated solution set in the external archive, and screening to obtain optimized particles, i.e., the optimized high-voltage rectification parameters and the optimized low-voltage vibration parameters.

[0033] Further, the method for calculating the fitness of each particle according to the standard target comprises:

[0034] sequentially sorting the non-dominated solutions in ascending order according to each standard target to obtain a sorted solution set;

[0035] setting the crowding distance of the boundary solutions of the sorted solution set to infinity for each standard target in the sorted solution set;

[0036] for each standard target and the rth solution, updating the crowding distance according to the standard target of the r+1th solution, the standard target of the r-1th solution, and the maximum and minimum values of the standard target;

[0037] taking the crowding distance of the particle as the fitness of the particle.

[0038] Further, the method for periodically updating the external archive comprises:

[0039] adding the non-dominated solution set to the external archive, and maintaining the size of the external archive according to a preset specific strategy;

[0040] randomly selecting a global optimal solution from the external archive, and updating the particle speed and position using the global optimal solution;

[0041] checking whether a preset stop condition is met, and if yes, outputting the non-dominated solution set in the external archive; otherwise, returning to recalculate the fitness of each particle.

[0042] Further, the method for obtaining the strongly correlated subset M and the key electric field identifier F comprises:

[0043] calculating the Pearson correlation coefficients of each parameter in X, Y and Z with the emission dust concentration, screening the parameters with the Pearson correlation coefficients greater than a preset first correlation threshold R1 to obtain a first correlation subset; when the ash content is higher than an ash content threshold, screening the parameters with the Pearson correlation coefficients greater than a preset second correlation threshold R2 to obtain a second correlation subset, and combining the first correlation subset and the second correlation subset to obtain the strongly correlated subset M; wherein R1>R2;

[0044] calculating the Pearson correlation coefficients of the working condition data and the equipment data of each electric field with the emission dust concentration, and screening according to a preset key threshold to obtain the key electric field identifier F.

[0045] Further, the method for obtaining the theoretical dust accumulation amount comprises:

[0046] obtaining the theoretical flue gas amount according to X and air composition, wherein the air composition includes carbon content and nitrogen content; calculating the actual flue gas amount according to the air leakage coefficient; and analyzing the coupling index, the theoretical flue gas amount, the actual flue gas amount and the boiler load to obtain the theoretical dust accumulation amount.

[0047] Further, the data cleaning comprises missing value processing, abnormal value processing and flow data correction.

[0048] Further, the method for integrity analysis comprises:

[0049] defining the physical meaning and theoretical boundary of the parameter according to the equipment operation principle or industry standard; and defining the parameter variation interval;

[0050] collecting data in a preset period, calculating the proportion of valid data, and determining that the data is complete when the proportion of valid data is greater than a preset proportion, otherwise, determining that the data is incomplete;

[0051] for the incomplete data, resampling through parameter perturbation;

[0052] Through frequency analysis, data with data change lower than the change threshold is screened out for data resampling;

[0053] Repeat sampling until the proportion of valid data is greater than the preset proportion.

[0054] The coal quality feedforward dust removal control system implements the coal quality feedforward dust removal control method, comprising:

[0055] The data acquisition module: collects data and performs integrity analysis and data cleaning to obtain coal quality feedforward data X, working condition data Y and equipment data Z;

[0056] The first analysis module: through dynamic Pearson correlation analysis, a strong correlation subset M and a key electric field identifier F are obtained;

[0057] The second analysis module: coupling analysis is performed on the coal quality feedforward data X to obtain a coupling index T;

[0058] The feature extraction module: according to the coupling index T and the working condition data Y, the theoretical dust accumulation amount is analyzed and obtained; PCA dimension reduction is performed on the strong correlation subset, and the strong correlation feature subset G is obtained in combination with the theoretical dust accumulation amount;

[0059] The intelligent analysis module: the strong correlation feature subset G is normalized and input into the LSTM model together with the key electric field identifier F to obtain control parameters, including high-voltage rectification parameters and low-voltage vibration parameters;

[0060] The online optimization module: based on the MOPSO algorithm, the control parameters are optimized online and rolled to obtain optimized control parameters;

[0061] The collaborative optimization module: according to the optimized control parameters, the system is controlled to perform collaborative execution.

[0062] The technical effects and advantages of the coal quality feedforward dust removal control system and method of the present application are as follows:

[0063] The coal quality feedforward dust removal control method constructs a feedforward control link of coal quality characteristics-ash prediction-intelligent regulation, deeply couples coal quality parameters such as ash content and moisture content with electric dust removal operation parameters, uses dynamic Pearson correlation analysis to screen key parameters, quantifies coal quality coordination influence by coupling indicators, optimizes the feature system by PCA dimension reduction, and finally forms an intelligent control system covering prediction, optimization and execution, effectively solving the regulation lag problem of traditional dust removal systems relying on real-time emission feedback, breaking through the technical bottleneck that energy consumption and emission cannot be considered under single objective optimization, avoiding over-regulation oscillation of control parameters under high-ash conditions, realizing intelligent upgrading from experience-driven to data-algorithm-driven, and enabling the electric dust remover to not only predict ash accumulation trends in advance and dynamically generate optimal control parameters according to coal quality fluctuations under variable load conditions such as deep peak regulation, but also accurately balance high-voltage energy consumption and vibration energy consumption through sub-field cooperative execution mechanism, significantly reducing power consumption while ensuring emission compliance, and improving the adaptability and economy of the system under complex conditions. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 It is a flowchart of the coal quality feedforward dust removal control method of the present application.

[0065] Figure 2 It is a flowchart of the integrity analysis method of the present application.

[0066] Figure 3 It is a schematic diagram of the change of the LSTM model training loss value of the present application.

[0067] Figure 4 It is a schematic diagram of the structure of the coal quality feedforward dust removal control system of the present application. DETAILED DESCRIPTION

[0068] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0069] Embodiment 1:

[0070] Please refer to Figure 1 The coal quality feedforward dust removal control method provided in the embodiment includes the following steps:

[0071] Collect data and perform integrity analysis and data cleaning to obtain coal quality feedforward data X, working condition data Y and equipment data Z;

[0072] The coal quality feedforward data X includes ash content, moisture content, and sulfur content;

[0073] The working condition data Y include the concentration of the discharge port DUST, the opacity of the E / P outlet, the O2 content of the discharge port, the DUST concentration of the E / P outlet, the boiler load, the total air volume, the total coal volume, the EP inlet flue gas temperature, the dust removal active power, the unit load, the coal combustion volume, the coal ash content, the ash-slag ratio, and the related data of the air compressor system, the ash conveying system, and each conveying unit; and the working condition data Y further include air composition, which is used for calculating the flue gas volume later.

[0074] The equipment data Z include the equipment display secondary voltage, the equipment display secondary current, the equipment display primary voltage, the equipment display primary current, the equipment secondary voltage setting, the equipment secondary current setting, the equipment pulse width, the equipment pulse period, the equipment operation mode, the equipment flashover frequency setting, the number of high-energy pulses of the equipment, and the number of low-energy pulses of the equipment.

[0075] Referring to Figure 2 , the method for integrity analysis includes:

[0076] According to the equipment operation principle or the industry standard, the physical meaning and the theoretical boundary of the parameter are defined, and the data change interval is determined;

[0077] The data in the preset period is collected, the data distribution is quantified through frequency analysis, the effective data proportion is calculated, that is, the ratio of the data meeting the data change interval determined according to the equipment operation principle or the industry standard, the physical meaning of the parameter, and the theoretical boundary to the data collected in the preset period, and the qualitative integrity degree is combined with the proportion: when the effective data proportion is greater than the preset proportion, the data is judged to be complete, otherwise it is judged to be incomplete;

[0078] If the integrity is greater than 70%, the data is judged to be complete, and the process is ended;

[0079] If the integrity is less than or equal to 70%, it is determined that the “equipment operation parameter change integrity is insufficient” or the “working condition change integrity is insufficient”:

[0080] For “insufficient equipment operation parameter change integrity”, a missing parameter disturbance scheme is developed for experiment, which can be performed through parameter disturbance resampling, frequency analysis screening of low change data, and repeated operation until the effective data proportion is greater than the preset proportion;

[0081] For “insufficient working condition change integrity”, more historical data is collected synchronously for supplement;

[0082] The steps of quantifying the data distribution through frequency analysis, calculating the effective data proportion, and combining the proportion with the qualitative integrity degree are repeated until the integrity is greater than 70%, and the data integrity verification is completed.

[0083] For incomplete data, resample by parameter perturbation;

[0084] Through frequency analysis, data with data change below the change threshold are screened out for data resampling;

[0085] Repeat sampling until the proportion of valid data is greater than the preset proportion.

[0086] Data cleaning usually locates abnormalities through 3σ, empty judgment, etc., and corrects data through missing value processing, abnormal value processing, and fluctuation data correction.

[0087] In the missing value processing process, in the original collected data, especially in the device data extraction process, it is found that there are missing values. In order to ensure the effectiveness of the modeling data, these missing values are discarded;

[0088] In the abnormal value processing process, part of the data exceeds the threshold range of the index, instantaneous drift, negative number and other abnormal situations, but cannot be simply discarded to affect the continuity of time series data. The abnormality is eliminated through interpolation completion and other methods.

[0089] In the fluctuation data correction process, total coal quantity, total air quantity and other important indicators for dust prediction, but the data quality is generally not high, the numerical value fluctuates obviously and there are spikes. In order to ensure that the dust prediction is not affected by the numerical fluctuation, it is necessary to carry out smoothing processing.

[0090] Collecting data and performing integrity analysis and data cleaning to obtain coal quality feedforward data X, working condition data Y and device data Z can lay a core foundation for solving the problem of high energy consumption and deep peak regulation energy waste caused by manual or open loop control of electric dust collector through systematic data management; By determining the data change interval and quantifying the distribution, the integrity analysis of key parameters such as coal quality ash content and flue gas quantity is ensured to cover the whole working condition scene, avoiding the failure of the model under variable load and other scenes due to data loss. Data cleaning improves the reliability of parameters such as coal quality ash content and electric field current by rejecting abnormal values and correcting fluctuating data, so that the subsequent dust prediction model based on LSTM can accurately capture the correlation characteristics of coal quality-working condition-device parameters, and then realize intelligent parameter optimization of high-voltage rectification and low-voltage vibration through MOPSO multi-objective optimization algorithm. Finally, under the condition of strict environmental protection standard, not only the emission standard is ensured through closed loop control, but also the operation parameters are dynamically adjusted according to the load change during deep peak regulation, avoiding energy waste caused by high parameter operation, realizing energy saving and efficient operation of dust removal system.

[0091] Through dynamic Pearson correlation analysis, a strong correlation subset M and a key electric field identifier F are obtained.

[0092] The method for obtaining the strong correlation subset M and the key electric field identifier F includes:

[0093] Calculate the Pearson correlation coefficients of each parameter in X, Y, and Z with the emission dust concentration, screen parameters with Pearson correlation coefficients greater than a preset first correlation threshold R1, and obtain a first correlation subset; when the ash content is higher than an ash content threshold, screen parameters with Pearson correlation coefficients greater than a preset second correlation threshold R2, and obtain a second correlation subset, and combine the first correlation subset and the second correlation subset to obtain a strong correlation subset M; wherein R1>R2.

[0094] Calculate the Pearson correlation coefficients of the working condition data and the equipment data of each electric field (a total of 24) with the emission dust concentration, and screen according to a preset key threshold to obtain a key electric field identifier F.

[0095] The strong correlation subset M and the key electric field identifier F obtained through dynamic Pearson correlation analysis can accurately identify the parameter combination and the core electric field that significantly affect the emission dust concentration based on data driving, thereby providing key support for solving the high energy consumption problem caused by manual or open-loop control of the electric precipitator. The above method sets different correlation thresholds R1 and R2 (R1>R2), screens strong correlation parameters to form a first correlation subset under normal working conditions with a strict threshold R1, ensures that the control model focuses on core influencing factors, screens parameters to form a second correlation subset under complex working conditions with an ash content higher than an ash content threshold with a relatively loose correlation threshold R2, avoids missing important influencing factors due to changes in parameter correlation characteristics caused by high ash content, and combines the first correlation subset and the second correlation subset to obtain a strong correlation subset M that can comprehensively cover key parameters under different working conditions. At the same time, the key electric field identifier F screened by the correlation analysis of the electric field data and the emission concentration can clearly and efficiently determine the core action area of dust removal, and the two can jointly build an input feature system for the subsequent LSTM prediction model and provide a parameter optimization space for MOPSO multi-objective optimization, so that the electric precipitator system can dynamically adjust the high-voltage rectification and low-voltage vibration parameters based on data correlation characteristics under variable load working conditions such as deep peak regulation, avoid energy waste caused by high parameter operation, and realize the coordinated control of emission compliance and energy saving and consumption reduction.

[0096] Coupling analysis is performed on the coal quality feedforward data X to obtain a coupling index T;

[0097] The method for obtaining the coupling index T includes:

[0098] The coupling analysis on the ash content and the moisture content in the coal quality feedforward data can be performed by using typical correlation analysis, grey correlation analysis, or a specific coupling model (such as a bivariate coupling model) in multivariate statistical analysis to analyze the coupling relationship between the ash content and the moisture content. Taking the bivariate coupling model as an example, let the ash content be variable A and the moisture content be variable W, and a coupling function is constructed. , and are coefficients, obtained by historical data regression analysis, and then the coupling relationship is obtained. In combination with a preset moisture content threshold, the preset moisture content threshold is the provision of the reference coal industry standard for the moisture content of different coal types, and in combination with actual production experience and experimental data, the changes in the combustion characteristics, dust removal efficiency and other indicators of coal under different moisture contents are analyzed to determine a reasonable threshold range; after obtaining the coupling relationship of ash and moisture, the coupling index T is calculated in combination with the preset moisture content threshold. If the moisture content W is higher than the moisture content threshold , the coupling function can be adjusted by weighting, wherein the weighting coefficient is a weight factor greater than 1, which can be obtained by optimization based on a natural heuristic optimization algorithm to highlight the influence of excessive moisture content on the coupling index T; if the moisture content is lower than the moisture content threshold , the original coupling function calculation result is kept to obtain the coupling index T.

[0099] The coupling analysis of the ash and moisture in the coal quality feedforward data and the coupling index T obtained in combination with the preset moisture content threshold can quantify the influence of the synergistic effect of ash and moisture on the dust characteristics, and provide a key coal quality characteristic representation for solving the problem of energy waste caused by high-parameter operation of the electric dust collector. The above method captures the coupling effect of ash and moisture under different threshold conditions (such as the influence of increased dust adhesion on dust removal efficiency when the moisture exceeds the threshold), converts the coal quality characteristics into a calculable coupling index T, provides an input feature that can reflect the comprehensive influence of coal quality for the subsequent LSTM dust prediction model, enables the model to more accurately predict the dust emission concentration under different coal quality conditions, and at the same time provides a coal quality constraint condition for the MOPSO multi-objective optimization algorithm, so that it can dynamically adjust the high-voltage rectification parameter and the low-voltage vibration parameter according to the coupling coefficient T in the optimization process, avoid over-adjustment oscillation or energy waste caused by coal quality fluctuations, and thus realize dust removal parameter self-adaptive optimization based on coal quality characteristics in deep peak regulation and other variable load conditions, which not only ensures environmental protection emission standards, but also effectively reduces the operating energy consumption of the electric dust collector.

[0100] The theoretical dust accumulation amount is analyzed and obtained according to the coupling coefficient T and the working condition data Y; the strongly correlated subset is subjected to PCA dimension reduction, and the strongly correlated feature subset G is obtained in combination with the theoretical dust accumulation amount;

[0101] The method for obtaining the theoretical dust accumulation amount includes:

[0102] The theoretical flue gas amount is obtained according to the coal quality feedforward data X and the air composition statistics, and the air composition includes carbon content and nitrogen content; the actual flue gas amount is calculated and obtained according to the air leakage coefficient (generally 0.1-0.15); the theoretical dust accumulation amount is obtained by analyzing the coupling index, the theoretical flue gas amount, the actual flue gas amount and the boiler load.

[0103] The theoretical dust accumulation amount is obtained according to the coupling index T and the working condition data Y analysis, and the strong correlation subset M is subjected to PCA dimension reduction to obtain the strong correlation feature subset G, which can provide key technical support for solving the high energy consumption problem of the electric dust collector by constructing a precise dust accumulation quantification model and an efficient feature system. The coupling index T is obtained by coupling analysis of the ash content and moisture content in the coal quality feedforward data X, which quantifies the influence of coal quality characteristics on dust accumulation, and the actual flue gas volume and boiler load calculated based on the air composition (carbon content, nitrogen content) and the leakage coefficient in the working condition data Y, to form a theoretical dust accumulation amount calculation model reflecting the real working condition, which provides a direct basis for optimization of the vibration parameters of each electric field. Meanwhile, PCA dimension reduction is used to compress the features of the strong correlation subset M, to eliminate redundant parameters and retain the core features strongly correlated with the dust accumulation amount, to form a feature subset suitable for dust control, so that the subsequent LSTM model can more efficiently learn the mapping relationship between the dust accumulation amount and the control parameters, and the MOPSO algorithm is used to realize dynamic optimization of the vibration period, high-voltage current and other parameters, to avoid excessive vibration or high parameter operation caused by calculation deviation of the dust accumulation amount, so as to accurately match the dust removal demand and energy consumption input under variable load working conditions such as deep peak regulation, and realize the dual goals of emission standard reaching and energy saving and consumption reduction.

[0104] The strong correlation feature subset G is normalized and input into the LSTM model together with the key electric field identifier F to obtain the control parameters, including high-voltage rectification parameters (devices with an alternating current rated voltage > 1000 V or a direct current rated voltage > 1500 V can be classified as high-voltage) and low-voltage vibration parameters (devices with an alternating current rated voltage ≤ 1000 V or a direct current rated voltage ≤ 1500 V can be classified as low-voltage); wherein the model parameters of the LSTM model are obtained by optimization of the MOPSD algorithm;

[0105] The training method of the LSTM model includes:

[0106] S groups of training data are collected in advance, including the normalized strong correlation feature subset G, the key electric field identifier F, and the control parameters.

[0107] The training data are used as the input of the LSTM model, and the control parameters are used as the output of the LSTM model, with the minimum error between the output control parameters and the actual control parameters as the target, the network parameters of the LSTM model are optimized by the MOPSD algorithm to obtain the network parameters corresponding to the minimum error between the output control parameters of the LSTM model and the actual control parameters, and the LSTM model constructed by the corresponding network parameters is used as the trained LSTM model. During the training process of the LSTM model, the learning rate is reduced in an exponential decay manner, and the loss function is optimized by the small batch stochastic gradient descent method, and the change of the training loss value of the LSTM model can refer to Figure 3The model starts to decline sharply at 600 epochs, and then the loss value fluctuates in the range of 0.06-0.10. The five models with the highest accuracy in the validation set are selected, and the test set is tested to obtain the best model. The loss function adopts the binary cross-entropy function.

[0108] The MOPSO algorithm is used for online rolling optimization of the control parameters to obtain optimized control parameters; the optimized control parameters include optimized high-voltage rectification parameters and optimized low-voltage vibration parameters.

[0109] The method for obtaining the optimized control parameters includes:

[0110] A target function is defined, and the target function includes three sub-targets of minimum emission deviation, minimum total energy consumption, and minimum vibration energy consumption;

[0111] The high-voltage rectification parameters and the low-voltage vibration parameters are defined as particles.

[0112] A constraint relaxation function is set according to the ash content; the constraint relaxation function is used to linearly relax the emission limit value when the ash content is greater than an ash content threshold value, so as to avoid over-regulation oscillation in high-ash working conditions; for example:

[0113]

[0114] wherein, is a dynamic emission limit value; is a basic emission limit value; is the ash content; is a linear relaxation rate, and the range is 0.1-0.5, which is obtained by combining ash-emission-energy multi-objective experiments with field debugging; is an ash content threshold value, which is generally the critical value of ash content fluctuation causing over-regulation oscillation in historical working conditions, or is determined by statistically determining the maximum value of the ash content in a stable running interval; when , the basic emission limit value is maintained for a normal working condition, and when , the emission limit value is linearly relaxed for a high-ash working condition.

[0115] The high-voltage rectification parameters and the low-voltage vibration parameters output by the LSTM model are taken as the center points, and N particles are randomly generated within the change range of L, wherein L is a limited particle change range value.

[0116] Random disturbance k is added to the high-ash working condition with the ash content greater than the ash content threshold value, wherein k is a random disturbance value added to the ash content greater than the ash content threshold value.

[0117] The target function value is standardized to obtain a standard target, the fitness of each particle is calculated according to the standard target, and non-dominated sorting is performed to obtain non-dominated solutions.

[0118] The non-dominated solutions are screened according to the non-dominated solutions, the individual optimal solution and the global optimal solution of the particle are updated, the speed and position of the particle are adjusted according to the particle swarm algorithm, and the updated non-dominated solution set is obtained; the position corresponds to the value of the control parameter, and the speed corresponds to the search direction of the control parameter adjustment;

[0119] The updated non-dominated solution set is stored in the external archive, and the external archive is updated regularly;

[0120] It is judged whether the preset stop condition is met, such as reaching the maximum iteration number;

[0121] The non-dominated solution set in the external archive is extracted, and the optimized particle, i.e. the optimized high-voltage rectification parameter and the optimized low-voltage vibration parameter, is screened.

[0122] The method for calculating the fitness of each particle according to the standard target includes:

[0123] The non-dominated solutions are sorted in ascending order according to each standard target, and the sorted solution set is obtained;

[0124] For each standard target in the sorted solution set, the crowding distance of the boundary solution of the sorted solution set is set to infinity;

[0125] For each standard target and the rth solution, the crowding distance is updated according to the standard target of the r+1th solution, the standard target of the r-1th solution, and the maximum value and the minimum value of the standard target;

[0126] The crowding distance of the particle is used as the fitness of the particle.

[0127] The method for regularly updating the external archive includes:

[0128] The non-dominated solution set is added to the external archive, and the size of the external archive is maintained according to a preset specific strategy;

[0129] A global optimal solution is randomly selected from the external archive, and the particle speed and position are updated using the global optimal solution;

[0130] It is checked whether the preset stop condition is met, if the preset stop condition is met, the non-dominated solution set in the external archive is output; otherwise, the fitness of each particle is recalculated.

[0131] The MOPSO algorithm is used for online rolling optimization of control parameters. By constructing a multi-objective optimization framework and a dynamic parameter adjustment mechanism, the high energy consumption problem caused by manual or open-loop control of the electric precipitator can be effectively solved. The above method defines the high-voltage rectification parameters and the low-voltage vibration parameters as particles, minimizes the emission deviation, minimizes the total energy consumption, and minimizes the vibration energy consumption as the objective function, avoids over-regulation oscillation in high-ash conditions by using a constraint relaxation function, and generates an initial particle swarm based on the LSTM model output. After standardizing the target function value, the fitness based on the crowding distance is calculated to realize efficient screening of non-dominated solutions and maintenance of solution set diversity. At the same time, the historical optimal solution is stored in the external archive and updated regularly to guide the particle swarm to continuously search for better parameter combinations in the solution space. The final output of the optimized control parameters can dynamically balance the dual goals of emission compliance and energy consumption reduction in deep peak shaving and other variable load conditions, avoiding energy waste caused by high parameter operation, and responding to uncertain factors such as coal quality fluctuations through a rolling optimization mechanism to improve the stability and adaptability of the system, and realize intelligent energy-saving control of the electric precipitation process.

[0132] The control system performs collaborative execution according to the optimized high-voltage rectification parameters and the optimized low-voltage vibration parameters.

[0133] The method for collaborative execution of the control system includes:

[0134] The non-critical electric field set and the adjacent field set of electric field i are counted, and the vibration state of electric field j at time t is extracted based on the optimized low-voltage vibration parameters, including being in vibration and not being in vibration.

[0135] The number of adjacent vibration fields of electric field i is detected, and the dynamic flow reduction coefficient is obtained based on the base value of the optimized high-voltage rectification parameters. According to the dynamic flow reduction coefficient, the ash adjustment factor calculated in real time is combined to obtain the collaborative high-voltage rectification parameter value of electric field i. The ash adjustment factor is obtained according to the real-time ash calculation. The collaborative high-voltage rectification parameter value is used as the final executed high-voltage rectification parameter value.

[0136] When the predicted emission reaches the preset emission limit value, the emission constraint logic based on the optimized high-voltage rectification parameters cancels the additional flow reduction.

[0137] When the spark rate is higher than the preset spark rate constraint, the device protection mechanism for linked optimization of the low-voltage vibration parameters reduces the dynamic flow reduction coefficient by half.

[0138] According to the collaborative execution of the optimized high-voltage rectification parameter and the optimized low-voltage shaking parameter control system, through dynamic identification of non-key electric field and adjacent field shaking state, calculation of dynamic flow reduction coefficient and real-time correction of high-voltage rectification parameter combined with ash adjustment factor, precise energy saving of the electric dust removal system can be realized under the premise of ensuring emission standard. The above method first counts the non-key electric field and adjacent field shaking state, determines the dynamic flow reduction coefficient according to the number of adjacent field shaking electric field, and then optimizes the high-voltage rectification parameter combined with the ash adjustment factor calculated by real-time ash content, so that the non-key electric field can reduce additional flow without affecting the overall dust removal efficiency when the adjacent field shakes. At the same time, through the flow reduction cancellation mechanism triggered by the prediction of emission limit value and the flow reduction coefficient halving strategy of spark rate exceeding limit, a double safety guarantee is constructed to avoid over-regulation shock and equipment abnormalities under high-ash conditions. Finally, the collaborative reduction of high-voltage energy consumption and shaking energy consumption of the electric dust remover under variable load conditions such as deep peak regulation is realized. Not only does it solve the problem of energy waste caused by high-parameter operation under open-loop control, but also improves the adaptability and economy of the dust removal system under complex conditions through intelligent regulation and control of the dynamic coupling of ash characteristics and shaking state.

[0139] Embodiment 2:

[0140] Referring to Figure 4 The coal quality feedforward dust removal control system provided in the embodiment includes:

[0141] Data acquisition module: collect data and perform integrity analysis and data cleaning to obtain coal quality feedforward data X, working condition data Y and equipment data Z;

[0142] First analysis module: obtain strong correlation subset M and key electric field identifier F through dynamic Pearson correlation analysis;

[0143] Second analysis module: perform coupling analysis on X to obtain coupling index T;

[0144] Feature extraction module: analyze the theoretical ash accumulation amount according to T and Y; perform PCA dimension reduction on the strong correlation subset, and obtain strong correlation feature subset G combined with the theoretical ash accumulation amount;

[0145] Intelligent analysis module: normalize G and input it to the LSTM model together with F to obtain control parameters, including high-voltage rectification parameters and low-voltage shaking parameters;

[0146] Online optimization module: perform online rolling optimization of the control parameters based on the MOPSO algorithm to obtain optimized control parameters;

[0147] Collaborative optimization module: control the system according to the optimized control parameters to perform collaborative execution.

[0148] The embodiment realizes the non-disturbance switching of the system and the coal conveying DCS system through the development of the new intelligent system access, switching and other DCS logic configuration, including: the system sends the heartbeat wave every sampling period, outputs FALSE when not running normally, the system is disconnected, and the DCS picture alarm prompt is carried out; the system on-off button is added to the DCS picture for the operation of the operator. According to the actual needs, the total on-off button and the electric field on-off button can be added. The system on requires the electric field to be in the automatic operation state, and the system off returns to the manual state.

[0149] The above merely describes the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0150] Finally: the above merely describes the preferred embodiments of the present application, and is not used for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A coal quality feedforward dust control method, characterized in that, Includes the following steps: Collect data and perform integrity analysis and data cleaning to obtain coal quality feedforward data X, operating condition data Y, and equipment data Z; Through dynamic Pearson correlation analysis, a strongly correlated subset M and key electric field identifiers F were obtained; Methods for obtaining the strongly correlated subset M and the key electric field identifier F include: Calculate the Pearson correlation coefficients between each parameter in X, Y, and Z and the emission dust concentration. Select parameters whose Pearson correlation coefficients are greater than a preset first correlation threshold R1 to obtain the first correlation subset. When the ash content is higher than the ash content threshold, select parameters whose Pearson correlation coefficients are greater than a preset second correlation threshold R2 to obtain the second correlation subset. Summarize the first and second correlation subsets to obtain the strongly correlated subset M. Wherein, R1 > R2. Calculate the Pearson correlation coefficient between the operating data and equipment data of each electric field and the concentration of emitted dust, and filter them according to the preset key threshold to obtain the key electric field identifier F; Coupled analysis was performed on the coal quality feedforward data X to obtain the coupling index T: Methods for obtaining the coupling index include: A coupled analysis of ash and moisture content in coal quality feedforward data was performed, and the coupling index was obtained by combining the preset moisture content threshold. The theoretical ash accumulation amount is obtained by analyzing the coupling index T and the working condition data Y; PCA dimensionality reduction is performed on the strongly correlated subset, and combined with the theoretical ash accumulation amount, the strongly correlated feature subset G is obtained. The strongly correlated feature subset G is normalized and input into the LSTM model together with the key electric field identifier F to obtain control parameters, including high voltage rectification parameters and low voltage rapping parameters. The control parameters are optimized by online rolling optimization based on the MOPSO algorithm. The system coordinates execution based on optimized control parameters.

2. The coal quality feedforward dust control method according to claim 1, characterized in that, Optimizing control parameters includes optimizing high-voltage rectification parameters and low-voltage rapping parameters. Methods for coordinated execution of these parameters by the control system include: The set of non-critical electric fields and the set of neighboring fields of electric field i are statistically analyzed. Based on the optimized low-voltage rapping parameters, the rapping state of electric field j at time t is extracted, including rapping in progress and not rapping. The number of field rapping electric fields of electric field i is detected. Based on the baseline value for optimizing high-voltage rectification parameters, the dynamic current reduction coefficient is obtained. Based on the dynamic current reduction coefficient and the ash adjustment factor calculated in real time, the coordinated high-voltage rectification parameter value of electric field i is calculated. The ash adjustment factor is obtained based on real-time ash calculation. The coordinated high-voltage rectification parameter value is used as the final high-voltage rectification parameter value. When the predicted emissions reach the preset emission limit, the additional current reduction is canceled based on the emission constraint logic that optimizes the high-voltage rectifier parameters. When the spark rate exceeds the preset spark rate constraint, the equipment protection mechanism that optimizes the low-pressure rapping parameters will halve the dynamic current reduction coefficient.

3. The coal quality feedforward dust control method according to claim 2, characterized in that, Methods for obtaining optimized high-voltage rectification parameters and optimized low-voltage rapping parameters include: Define an objective function, which includes three sub-objectives: minimizing emission deviation, minimizing total energy consumption, and minimizing rapping energy consumption. The high-voltage rectification parameters and low-voltage rapping parameters are defined as particles; Set the constraint relaxation function based on the ash content; Using the high-voltage rectification parameters and low-voltage rapping parameters output by the LSTM model as the center point, N particles are randomly generated within the range of L%, where L is the value that limits the range of particle variation. For ash conditions where the ash content is greater than the ash content threshold, add a random disturbance of k%, where k is the random disturbance value added for ash conditions where the ash content is greater than the ash content threshold. The objective function value is standardized to obtain a standard objective. The fitness of each particle is calculated based on the standard objective, and non-dominated sorting is performed to obtain a non-dominated solution. The particles are filtered based on the non-dominated solutions, the individual optimal solutions and global optimal solutions of the particles are updated, the velocity and position of the particles are adjusted, and an updated set of non-dominated solutions is obtained. Store the updated set of non-dominated solutions in an external archive and update the external archive periodically; Determine whether the preset stopping conditions are met; Extract the non-dominated solution set from the external archive, and filter to obtain the optimized particles, namely the optimized high-voltage rectification parameters and the optimized low-voltage rapping parameters.

4. The coal quality feedforward dust control method according to claim 3, characterized in that, Methods for calculating the fitness of each particle based on standard objectives include: The non-dominated solutions are sorted in ascending order according to each standard objective to obtain the sorted solution set. For each standard objective in the sorted solution set, the crowding distance of the boundary solutions of the sorted solution set is set to infinity; For each standard objective and the r-th solution, update the congestion distance based on the standard objective of the (r+1)-th solution, the standard objective of the (r-1)-th solution, and the maximum and minimum values ​​of the standard objectives; The crowding distance of a particle is used as its fitness.

5. The coal quality feedforward dust control method according to claim 3, characterized in that, Methods for regularly updating external archives include: Add the non-dominated solution set to the external archive, and maintain the size of the external archive according to a preset specific strategy; Randomly select a global optimum from the external archive and update the particle velocity and position using the global optimum; Check if the preset stopping condition is met. If it is, output the set of non-dominated solutions in the external archive; otherwise, return to recalculate the fitness of each particle.

6. The coal quality feedforward dust control method according to claim 1, characterized in that, Methods for obtaining the theoretical dust accumulation amount include: The theoretical flue gas volume is obtained based on X and air composition statistics, where air composition includes carbon and nitrogen. The actual flue gas volume is then calculated based on the air leakage coefficient. The theoretical ash accumulation is obtained by analyzing the coupling index, theoretical flue gas volume, actual flue gas volume, and boiler load.

7. The coal quality feedforward dust control method according to claim 6, characterized in that, Data cleaning includes handling missing values, handling outliers, and correcting flowing data.

8. The coal quality feedforward dust control method according to claim 1, characterized in that, The methods for integrity analysis include: Based on the equipment operating principle or industry standards, define the physical meaning and theoretical boundaries of the parameters; define the parameter variation range; Collect data within a preset time period, calculate the percentage of valid data, and determine the data as complete if the percentage of valid data is greater than the preset percentage; otherwise, determine the data as incomplete. For incomplete data, resampling is performed by perturbating parameters; By using frequency analysis, data with changes below the change threshold are selected for data resampling; The percentage of repeated sampling until the valid data exceeds the preset percentage.

9. A coal quality feedforward dust removal control system, implementing the coal quality feedforward dust removal control method according to any one of claims 1-8, characterized in that, include: Data acquisition module: Collects data and performs integrity analysis and data cleaning to obtain coal quality feedforward data X, operating condition data Y, and equipment data Z; First analysis module: Through dynamic Pearson correlation analysis, obtain the strongly correlated subset M and the key electric field identifier F; The second analysis module performs coupling analysis on the coal quality feedforward data X to obtain the coupling index T. Feature extraction module: Analyzes the coupling index T and operating condition data Y to obtain the theoretical ash accumulation amount; performs PCA dimensionality reduction on the strongly correlated subset, and combines it with the theoretical ash accumulation amount to obtain the strongly correlated feature subset G; Intelligent analysis module: Normalizes the strongly correlated feature subset G and inputs it together with the key electric field identifier F into the LSTM model to obtain control parameters, including high voltage rectification parameters and low voltage rapping parameters; Online optimization module: Based on the MOPSO algorithm, the control parameters are optimized online by rolling optimization. Collaborative optimization module: Controls the system to perform collaborative execution based on optimized control parameters.

Images