A boiler steam staging optimization control method and system

Abstract

The application discloses a boiler steam grading optimization control method and system, relates to the technical field of optimization control, and comprises the following steps: collecting boiler multi-source data, preprocessing the data to obtain normalized data, generating measurement values through amplitude limiting and smoothing, calling a prediction model to calculate a confidence degree and generate a confidence state vector, determining a boiler operation level and a target weight vector, defining control variables and generating a candidate scheme set, rolling prediction with the confidence state vector as an initial value, calculating four types of evaluation values, aggregating a comprehensive evaluation value according to the target weight, determining a leading scheme and a secondary leading scheme, combining a steam quality deviation, external disturbance and mismatch degree to obtain a strategy discrimination value, executing switching tracking or anti-disturbance strategy iterative optimization, screening a degraded scheme, combining the leading scheme and a median sequence for flexible correction and boundary projection, and outputting a target control value sequence to be converted into a control amount and then being issued.

Description

Technical Field

[0001] This invention relates to the field of optimization control technology, and in particular to a boiler steam staged optimization control method and system. Background Technology

[0002] As the core equipment for converting the chemical energy of fuel into thermal energy in thermal power units, the quality of steam parameters in large coal-fired power plant boilers directly determines the safety and economy of steam turbine generator units. With the continuous increase in the penetration rate of new energy sources in my country's power system, deep peak shaving and rapid load changes have become the norm for thermal power units. This places much higher technical requirements on the dynamic response capability, multi-condition adaptability, and robust stability of boiler steam control systems than on traditional base load operation. The boiler steam control system is essentially a complex thermal engineering object with strong nonlinearity, large inertia, strong coupling of multiple variables, and pure time delay characteristics. Its controlled variables include multiple key parameters such as main steam pressure, main steam temperature, main steam flow rate, steam drum water level, furnace negative pressure, and flue gas oxygen content, while the manipulated variables involve multiple execution channels such as coal feed rate, primary air volume, secondary air volume, induced draft volume, feedwater flow rate, desuperheating water flow rate, and main steam regulating valve opening.

[0003] However, existing technologies lack a dynamic fusion mechanism between the reliability of measured values ​​and model predictions, resulting in distorted state estimations when sensors fail or operating conditions change drastically. Furthermore, there is a lack of flexible correction and reconstruction mechanisms for deteriorating candidate solutions during the iteration process. When population diversity declines or encounters strong disturbances, it is difficult to quickly restore effective search directions, thus limiting the overall optimization effect of steam quality and safety indicators. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a boiler steam staged optimization control method and system, which solves the problem that the existing technology has not established a dynamic fusion mechanism between the reliability of measured values ​​and model predictions, and it is difficult to quickly recover the effective search direction, resulting in limited comprehensive optimization effect of steam quality and safety indicators.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a boiler steam staged optimization control method, which includes, Multi-source boiler operation data is collected and preprocessed to obtain normalized data. After obtaining the changes of each measured variable, amplitude limiting and recursive smoothing are performed to generate measured values ​​for prediction. The confidence level is calculated and a confidence state vector is generated. Based on the confidence state vector, the current boiler operation level is determined and a target weight vector corresponding to the operation level is constructed. Based on the measured variables, control variables are defined, and after the first candidate scheme is formed by shifting and padding with zeros, a disturbance is applied to generate a set of candidate schemes. With the credible state vector as the initial state, the control variables are used as external inputs and the prediction model is called to perform rolling predictions to obtain the boiler prediction state at each prediction step. The steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordination control evaluation value of each candidate scheme are calculated. After performing grouping and aggregation based on the target weight vector, the comprehensive evaluation value is obtained by combining the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value and coordination control evaluation value, and the dominant and secondary dominant schemes are determined. The steam quality deviation degree, external disturbance intensity and mismatch degree are calculated and weighted summed to obtain the strategy discrimination value. Based on the strategy discrimination value, the dominant tracking or anti-disturbance exploration update strategy is switched to iteratively optimize the candidate schemes to obtain a new set of candidate schemes. The degree of degradation is determined based on the new candidate scheme set. After screening the degradation schemes based on the degree of degradation, a median control sequence is constructed. The degradation schemes are then flexibly modified and range-corrected in conjunction with the dominant scheme. Finally, the target control value sequence is output and converted into control variables for distribution.

[0007] As a preferred embodiment of the boiler steam staged optimization control method of the present invention, the step of collecting multi-source boiler operating data, preprocessing it to obtain normalized data, and acquiring the changes in each measured variable are as follows: By uniformly collecting multi-source data from various measurement points, performing time alignment, missing value imputation, and normalization, we obtain the normalized data of the previous control time and the normalized data of the current time. The change in the normalized value at the current time relative to the normalized value at the previous control time is obtained by subtracting the normalized value at the previous control time from the normalized value at the current time. .

[0008] As a preferred embodiment of the boiler steam staged optimization control method of the present invention, the steps of performing amplitude limiting and recursive smoothing, generating measured values ​​for prediction, and calculating the confidence level to perform confidence state vector generation are as follows: Establish a database of maximum permissible rates of change, and multiply the maximum rate of change of each normalized value by the data sampling period to obtain the maximum permissible change of each normalized value; Based on the maximum change and the change Adjust the normalized values ​​of each measurement variable to obtain the amplitude limit value, perform recursive smoothing on the amplitude limit value, and generate the recursive smoothed measurement value at the current time. The pre-trained prediction model is invoked to predict the current value based on the normalized value of the previous control time. The predicted value is then subtracted from the recursively smoothed measurement value at the current time to obtain the prediction bias of each measurement variable. Based on the prediction bias, the exponential decay function of the Laplace kernel is used to calculate the credibility of each measured variable. The results are then compared with a preset credibility threshold to generate the final valid state value and concatenate them into a credibility state vector.

[0009] As a preferred embodiment of the boiler steam staged optimization control method of the present invention, the step of determining the current boiler operating level based on the reliable state vector and constructing the target weight vector corresponding to the operating level refers to calculating the boiler load ratio and the preset low load level threshold by extracting the final effective state value of the main steam flow based on the reliable state vector. Medium load level threshold High load level threshold The comparison is performed to generate the current boiler operating level, and then the corresponding target weight vector is constructed using the analytic hierarchy process.

[0010] As a preferred embodiment of the boiler steam staged optimization control method of the present invention, the defined control variables refer to the coal feed rate, primary air volume, secondary air volume, induced draft volume, feed water volume, desuperheating water flow rate, and main steam regulating valve opening as control variables.

[0011] As a preferred embodiment of the boiler steam staged optimization control method of the present invention, the calculation of the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordinated control evaluation value of each candidate scheme is as follows: Obtain the main steam pressure deviation, main steam temperature deviation, and main steam flow deviation, perform weighted aggregation over the entire rolling prediction length, and generate a steam quality evaluation value. Calculate the predicted steam drum water level, predicted furnace negative pressure, and predicted flue gas oxygen content for each prediction step. Weighted summation is performed for the temperature rise rate that exceeds the limit to generate the equipment safety evaluation value. The cumulative action of the coal feeder, air damper, induced draft damper, water supply valve, desuperheating water valve, and main steam regulating valve over the entire rolling prediction length is calculated separately to generate the actuator action evaluation value. The coordination deviations of wind and coal, steam and water, and cooling are calculated and weighted to obtain the coordination control evaluation value.

[0012] As a preferred embodiment of the boiler steam staged optimization control method of the present invention, the step of performing group aggregation based on the target weight vector, and then combining the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordinated control evaluation value to obtain a comprehensive evaluation value and determine the dominant scheme and the secondary dominant scheme, as detailed below: Based on the target weight vector, the evaluation dimensions are grouped and aggregated according to their physical attributes. After generating the weights of each evaluation value, the weighted sum of each evaluation value is obtained to obtain the comprehensive evaluation value of the candidate solution. After sorting all candidate schemes according to their comprehensive evaluation values ​​from smallest to largest, a sorted list is obtained. The candidate scheme with the smallest comprehensive evaluation value in the sorted list is determined as the dominant scheme for the current control period, and the candidate scheme with the second largest comprehensive evaluation value is determined as the secondary dominant scheme.

[0013] As a preferred embodiment of the boiler steam staged optimization control method of the present invention, the following steps are taken: the calculated steam quality deviation degree, external disturbance intensity, and mismatch degree are weighted and summed to obtain a strategy discrimination value. Based on the strategy discrimination value, the dominant tracking or anti-disturbance exploration update strategy is switched to iteratively optimize the candidate schemes to obtain a new set of candidate schemes, as follows: The arithmetic mean of the steam quality evaluation values ​​for the prediction steps within the rolling prediction length is used as the degree of steam quality deviation. Calculate the intensity of external disturbance based on changes in boiler load demand; The degree of mismatch is calculated based on each final valid state value and predicted value in the credible state vector; The strategy discrimination value is obtained by weighted summation of steam quality deviation, external disturbance intensity, and mismatch degree. The policy discrimination value is compared with the preset policy switching threshold. If the policy discrimination value is less than or equal to the preset policy switching threshold, the dominant tracking update policy is used to update the candidate solution. If the policy discrimination value is greater than the preset policy switching threshold, the anti-disturbance exploration update policy is used, and all updated candidate solutions are combined into a new candidate solution set.

[0014] As a preferred embodiment of the boiler steam staged optimization control method of the present invention, the following steps are taken: First, the degree of degradation is determined based on a new set of candidate solutions. Then, after screening deteriorating solutions using the degree of degradation, a median control sequence is constructed. Next, the deteriorating solutions are flexibly and range-wise corrected in conjunction with the dominant solution. Finally, the target control value sequence is output and converted into control quantities for distribution. Specifically, the steps are as follows: The minimum comprehensive evaluation value is selected from the updated candidate solutions as the optimal evaluation value. The comprehensive evaluation values ​​of each updated candidate solution are then normalized to obtain the degree of degradation. Embed a consecutive no-improvement counter for each updated candidate solution, and record the number of consecutive no-improvements for each updated candidate solution through the consecutive no-improvement counter; Based on the number of consecutive periods without improvement and the degree of degradation, it is jointly determined whether the updated candidate solution is a degraded solution; Based on the median control sequence of the dominant scheme and all updated candidate schemes, the degraded scheme is flexibly modified to obtain the reconstructed candidate scheme; Based on the candidate reconstruction scheme, check whether the control variables meet the allowable range. If they exceed the allowable range of the control variables, directly correct them to the corresponding boundary values. After the maximum number of iterations is reached, the target control value sequence of the control variables is output. The target values ​​in the target control value sequence are converted into control quantities and then sent to the corresponding actuator control loops via wireless transmission technology to achieve optimized control.

[0015] Secondly, the present invention provides a boiler steam staged optimization control system, comprising: The vector generation module is used for preprocessing, obtaining the change amount, performing amplitude limiting and recursive smoothing, generating measured values ​​for prediction, calculating the confidence level to perform confidence state vector generation, determining the current boiler operating level based on the confidence state vector, and constructing the target weight vector corresponding to the operating level. The evaluation value calculation module is used to define control variables, and after moving forward and padding with zeros to form the first candidate scheme, it applies perturbation to generate a set of candidate schemes. With the credible state vector as the initial state, the control variables are used as external inputs and the prediction model is called to perform rolling prediction to obtain the boiler prediction state at each prediction step. The module also calculates the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordinated control evaluation value for each candidate scheme. The scheme update module is used to perform grouping and aggregation based on the target weight vector, and then combine the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value and coordination control evaluation value to obtain a comprehensive evaluation value and determine the dominant scheme and the secondary dominant scheme. It calculates the steam quality deviation degree, external disturbance intensity and mismatch degree and performs weighted summation to obtain the strategy discrimination value. Based on the strategy discrimination value, it switches the dominant tracking or anti-disturbance exploration update strategy to iteratively optimize the candidate schemes and obtain a new set of candidate schemes. The control quantity distribution module is used to determine the degree of degradation based on the new candidate scheme set, filter the degradation schemes using the degree of degradation, construct the median control sequence, combine the dominant scheme to perform flexible correction and range correction on the degradation schemes, and output the target control value sequence to convert it into control quantities for distribution.

[0016] The beneficial effects of this invention are as follows: By constructing a database of maximum permissible rate of change and a recursive smoothing mechanism, combined with a credibility fusion strategy based on the Laplace kernel, measurement noise and abnormal interference are effectively suppressed. Furthermore, through a load grading mechanism and dynamic weight vector, an adaptive balance between safety, economy, and stability objectives under different operating conditions is achieved. Secondly, by dynamically switching the dominant tracking and anti-disturbance exploration update strategies through policy discriminant values, and by implementing flexible correction and boundary projection reconstruction based on the dominant-median sequence for deteriorated schemes, the global optimization ability and population diversity of the optimization algorithm under strong disturbances are significantly enhanced. Therefore, this invention significantly improves the dynamic quality and steady-state accuracy of the boiler steam control system in complex operating environments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the boiler steam staged optimization control method in Example 1.

[0019] Figure 2 This is a structural diagram of the boiler steam staged optimization control system in Example 1.

[0020] Figure 3 This is a flowchart of obtaining the new candidate scheme set in Example 1. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Example 1, referring to Figures 1-3 This is the first embodiment of the present invention, which provides a boiler steam staged optimization control method, including the following steps: S1. Collect multi-source boiler operation data, preprocess it to obtain normalized data, and obtain the changes of each measurement variable. Then, perform amplitude limiting and recursive smoothing to generate measurement values ​​for prediction. Calculate the confidence level and generate a confidence state vector. Based on the confidence state vector, determine the current boiler operation level and construct the target weight vector corresponding to the operation level.

[0025] S1.1. By uniformly collecting multi-source data from various measuring points, time alignment, missing value imputation, and normalization are performed to obtain the normalized data of the previous control time and the normalized data of the current time.

[0026] Specifically, by uniformly collecting multi-source data from various measuring points, performing time alignment, missing value imputation, and normalization, normalized data from the previous control time is obtained. During the current control cycle, data acquisition is performed for time alignment, missing value imputation, and normalization to obtain the normalized data for the current moment.

[0027] This method involves collecting multi-source data at various measuring points using various acquisition devices. For example, pressure or differential pressure transmitters are installed at the superheater outlet main steam pipe and the upper flue gas passage of the furnace to collect the main steam pressure and furnace negative pressure; thermocouples or resistance thermometers are inserted into the central flow area of ​​the superheater outlet pipe to collect the main steam temperature; velocity flow meters are installed at the straight sections of the corresponding pipes or ducts to collect the main steam flow, feedwater flow, desuperheating water flow, primary air volume, secondary air volume, and induced draft volume; a differential pressure level meter is used to collect the steam drum water level; a zirconia oxygen analyzer is used at the tail flue to collect the flue gas oxygen content; and a weighing meter is used to collect the coal feed rate.

[0028] S1.2. Subtract the normalized value from the previous control time from the current normalized value to obtain the change in the normalized value at the current time relative to the normalized value at the previous control time. .

[0029] S1.3 Establish a database of maximum permissible rates of change. Multiply the maximum rate of change of each normalized value by the data sampling period to obtain the maximum permissible change of each normalized value.

[0030] It should be noted that the maximum permissible rate of change database is pre-established based on the actual response capabilities of each component of the boiler and stores the maximum permissible rate of change for each measured variable per unit time. For example, for the main steam pressure, the normal operating pressure range can be set to 14MPa to 17MPa, and the maximum permissible rate of change for the main steam pressure can be set to 0.03MPa per second based on the pressure vessel safety operation requirements and the turbine inlet pressure change limits. For the main steam temperature, the rated main steam temperature can be set to 540℃, the normal adjustment range can be set to 520℃ to 560℃, and the maximum permissible rate of change for the main steam temperature can be set to 1.5℃ per second based on the superheater metal thermal stress limits and the desuperheating water adjustment capabilities. For the main steam flow rate, the rated steam flow rate can be set to 300 tons / hour, and the maximum allowable rate of change of the main steam flow rate can be set to 1.2 tons per second, based on the load regulation capacity of the boiler combustion system. For the steam drum water level variable, the normal steam drum water level control range can be set to within 150 mm above and below the reference water level, and the maximum allowable rate of change of the steam drum water level can be set to 8 mm per second, based on the steam drum water storage space and the regulation speed of the feedwater system. For the furnace negative pressure variable, the normal operating furnace negative pressure can be set to -50 Pa to -1 Pa, based on the example. Within a 50Pa range, based on the furnace sealing structure and the induced draft fan's adjustment capacity, the maximum allowable rate of change of furnace negative pressure is set to 20Pa per second. For the variable flue gas oxygen content, the combustion control target oxygen content can be set to 3.5% to 5%, and based on combustion stability requirements, the maximum allowable rate of change of flue gas oxygen content is set to 0.15 percentage points per second. For the variable coal feed rate, the maximum boiler coal feed rate can be set to 45 tons / hour, and based on the coal feeder's conveying capacity and the combustion chamber's thermal inertia, the maximum allowable rate of change of coal feed rate is set to 0.25 tons per second. For primary air volume variables, the rated primary air volume can be set to 90,000 cubic meters per hour, and the maximum allowable rate of change of primary air volume can be set to 800 cubic meters per second, based on the pulverizing system and fan response capabilities. For secondary air volume variables, the rated secondary air volume can be set to 120,000 cubic meters per hour, and the maximum allowable rate of change of secondary air volume can be set to 1,200 cubic meters per second, based on combustion air distribution stability requirements. For induced draft volume variables, the rated induced draft volume can be set to 160,000 cubic meters per hour, and the maximum allowable rate of change of secondary air volume can be set to 160,000 cubic meters per hour, based on the induced draft fan adjustment performance. To ensure stable furnace pressure, the maximum allowable rate of change of induced draft volume is set to 1500 cubic meters per second. For feedwater flow rate, the rated feedwater flow rate can be set to 320 tons per hour, and the maximum allowable rate of change of feedwater flow rate is set to 1.5 tons per second based on the feedwater pump's adjustment capacity and the steam-water circulation stability requirements. For desuperheating water flow rate, the rated desuperheating water flow rate can be set to 20 tons per hour, and the maximum allowable rate of change of desuperheating water flow rate is set to 0.15 tons per second based on the spray valve's response speed and the superheater's temperature adjustment requirements.

[0031] S1.4, Based on the maximum change and the change The normalized values ​​of each measurement variable are adjusted to obtain the amplitude limit value. The amplitude limit value is then recursively smoothed to generate the recursively smoothed measurement value at the current time.

[0032] The amplitude limit value is obtained by the following expression: ; In the formula, Indicates the first The measured variables at the current time The amplitude limit, Indicates the first The measured variables at the previous control time The normalized value, Indicates the first The maximum change of a measured variable This represents the change in the normalized value at the current time relative to the normalized value at the previous control time. Indicates the first The measured variables at the current time The normalized value, Indicates otherwise; The threshold value is recursively smoothed using the following expression: ; In the formula, Indicates the first The measured variables at the current time The measured value after recursion and smoothing. Indicates the first The recursive smoothing coefficient of each measured variable, Indicates the first The measured variables at the previous control time The measured value after recursive smoothing.

[0033] It should be noted that the recursive smoothing coefficient Separate settings can be configured for data that changes rapidly and data that changes slowly. For example, for variables that change rapidly and require a quick response, a larger recursive smoothing coefficient can be set to increase the proportion of the current sampled data in the smoothing result. For instance, the recursive smoothing coefficient for the main steam pressure variable can be set to 0.65, and the recursive smoothing coefficient for the main steam flow variable can be set to 0.70. The recursive smoothing coefficient for the coal feed rate variable is set to 0.75; the recursive smoothing coefficients for the primary air volume and secondary air volume variables are set to 0.70. For variables that change slowly and are easily affected by transient disturbances, a smaller recursive smoothing coefficient is set to enhance the role of historical data. For example, the recursive smoothing coefficient for the steam drum water level variable is set to 0.45; the recursive smoothing coefficient for the furnace negative pressure variable is set to 0.50; the recursive smoothing coefficient for the flue gas oxygen content variable is set to 0.40; the recursive smoothing coefficient for the feedwater flow rate variable is set to 0.55; and the recursive smoothing coefficient for the desuperheating water flow rate variable is set to 0.60.

[0034] S1.5. Call the pre-trained prediction model, predict the current value based on the normalized value of the previous control time, and subtract the predicted value from the recursively smoothed measurement value at the current time to obtain the prediction deviation of each measurement variable.

[0035] It should be noted that the pre-trained prediction model can be set as a long short-term memory network, including an input layer, hidden layers, and an output layer. The model training uses the mean squared error between the predicted and actual values ​​as the loss function, and uses the Adam optimizer to update the model parameters iteratively. When the loss function value no longer decreases, the iteration stops, and the trained prediction model is output. The training data can be collected from historical operating data under stable boiler operation and different load change conditions, and the input and output samples are divided according to the control cycle. For example, the training sample uses data from 60 consecutive control cycles as a time window, and twelve normalized measurement variables are collected in each control cycle. Therefore, each training sample contains 60 sets of historical state data.

[0036] S1.6 Based on the prediction bias, the exponential decay function of the Laplace kernel is used to calculate the credibility of each measured variable. The results are then compared with the preset credibility threshold to generate the final valid state value and concatenate them into a credibility state vector.

[0037] The confidence level of each measured variable is calculated using the following expression: ; In the formula, Indicates the first The measured variables at the current time The credibility of This represents the natural exponential function. Represents absolute value. Indicates the first The measured variables at the current time Prediction bias, Indicates the first The standard deviation of the prediction bias of each measured variable (which can be set by statistically analyzing the standard deviation of the prediction bias sequence at historical control times). The credibility level is compared with a preset credibility threshold. If the credibility level is greater than or equal to the preset credibility threshold, the current measurement value is considered credible, and the recursively smoothed measurement value is directly used as the final valid state value. Otherwise, the current measurement value is considered to have some anomaly, and the credibility level is used as a weight to fuse the smoothed value and the predicted value to obtain the final valid state value of the measurement variable. The final valid state values ​​of all measured variables are concatenated to generate the reliable state vector at the current moment.

[0038] It should be noted that the confidence threshold can be set based on the statistical results of historical prediction deviations; however, in this invention, the confidence threshold can be set to 0.85, under which approximately 95% or more of the measurement data under normal operating conditions can be directly used as the recursive smoothed value, while the measurement value and the predicted value can be fused in a timely manner under conditions of sudden load changes, sensor abnormalities, or model mismatch. Secondly, when using confidence level as a weight to fuse smoothed and predicted values, the measured value is directly multiplied by the confidence level, while the predicted value needs to be multiplied by the confidence level. After generating the weights required for the predicted values, multiply them by the predicted values. Finally, add the results of the two multiplications together to obtain the final effective state.

[0039] Furthermore, based on the reliable state vector, the final effective state value of the main steam flow is extracted, and the ratio between the final effective state value and the rated main steam flow of the boiler is used as the boiler load ratio at the current moment. The current boiler load ratio is sequentially compared with the preset low load level threshold. Medium load level threshold High load level threshold The comparison is performed to generate the current boiler operating level.

[0040] The expression for generating the current boiler operating level is: ; In the formula, Indicates the current moment Boiler operating level, Indicates a low load level. Indicates medium load level. Indicates a high load level. Indicates the peak load level. This indicates the load percentage corresponding to the boiler's minimum stable combustion load. Indicates the current moment The boiler load ratio, This indicates the load ratio corresponding to the boiler's maximum continuous evaporation capacity (BMCR).

[0041] It should be noted that the load ratio corresponding to the boiler's maximum continuous evaporation capacity (BMCR) can be set to the boiler's rated value. Furthermore, when the boiler is operating at low load, the combustion system needs to maintain sufficient fuel input and air supply. When the load is below 45% of the rated load, combustion stability, flue gas oxygen content, and furnace temperature changes significantly affect the control effect. Therefore, the low load threshold is exemplarily set... The value is set to 0.45; secondly, boilers are typically in their economic operating range between 45% and 70% load, where combustion efficiency is high and equipment thermal inertia is stable. Therefore, the threshold value for medium load levels is appropriate. The present invention can exemplarily take 0.7; furthermore, when the boiler load exceeds 90%, the rate of change of steam pressure, steam temperature, and thermal stress on the heating surface increases, therefore the high load level threshold... It can be set to 0.90 for example; Since boiler load may fluctuate continuously around a certain boundary value during actual operation, directly switching operating levels based on interval boundaries would lead to frequent switching of operating levels, causing continuous changes in the prediction model and evaluation weights, thereby reducing control stability. Therefore, this invention also provides a hysteresis mechanism for control, specifically: setting the boundary value between two adjacent operating levels as... The corresponding hysteresis width is When a boiler switches from a lower operating level to a higher operating level, the following conditions should be met: ; When a boiler switches from a higher operating level to a lower operating level, the following conditions should be met: ; Regarding hysteresis width The present invention can set a hysteresis width of 0.03 between low load level and medium load level, a hysteresis width of 0.03 between medium load level and high load level, and a hysteresis width of 0.02 between high load level and peak load level, thereby avoiding frequent switching of operating level caused by fluctuation of boiler load ratio near level boundary.

[0042] Furthermore, based on the current boiler operating level, a corresponding target weight vector is set. The target weight vector includes combustion stability weight, flue gas oxygen content control weight, unit steam fuel consumption control weight, main steam pressure control weight, main steam temperature control weight, furnace negative pressure control weight, steam drum water level control weight, main steam temperature rise rate control weight, and actuator operation smoothness weight.

[0043] It should be noted that: setting the corresponding target weight vector requires selecting historical operating data under typical steady-state conditions for each operating level. Based on the normalized root mean square values ​​of the nine optimization targets (combustion stability, flue gas oxygen content, unit steam fuel consumption, main steam pressure deviation, main steam temperature deviation, furnace negative pressure, steam drum water level, main steam temperature rise rate, and actuator smoothness), the relative importance ranking of each target under the operating level is determined using the standard analytic hierarchy process (AHP) or entropy weight method, and the target weight vector is generated accordingly. This invention preferably uses the analytic hierarchy process (AHP). For the judgment matrix in the analytic hierarchy process (AHP), boiler operation experts can construct a 9th-order positive and negative judgment matrix by comparing the pairwise importance of the nine optimization targets according to the Saaty 1~9 scaling method.

[0044] S2. Define control variables based on measured variables, and after moving forward and padding with zeros to form the first candidate scheme, apply perturbation to generate a set of candidate schemes. With the credible state vector as the initial state, use the control variables as external input and call the prediction model to perform rolling prediction to obtain the boiler prediction state at each prediction step. Calculate the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordinated control evaluation value for each candidate scheme.

[0045] Specifically, based on the reliable state vector of the current boiler operating level, the coal feed rate, primary air volume, secondary air volume, induced draft volume, water feed rate, and desuperheating water flow rate are extracted. The opening setpoint of the main steam regulating valve at the current control moment is read from the control instruction unit to obtain the opening of the main steam regulating valve. The extracted coal feed rate, primary air volume, secondary air volume, induced draft volume, feed water volume, desuperheating water flow rate and the opening of the main steam regulating valve are defined as control variables. The control variables of the previous control moment are extracted, then moved forward in time sequence, and zero increment is added at the end to form the first candidate scheme. Based on the first candidate scheme, disturbances within a preset range are applied to the coal feed rate, primary air volume, secondary air volume, induced draft volume, water feed rate, desuperheating water flow rate, and main steam regulating valve opening to form the remaining candidate schemes. All candidate schemes are combined to obtain the final candidate scheme set.

[0046] The disturbance is generated using a relative disturbance method to determine the disturbance values ​​for each control variable, specifically: Let the allowable disturbance ratio of the corresponding control variable be... The current control variable is allowed to operate within a certain range. , and They represent the first Upper and lower limits of each control variable; The corresponding disturbance value of the control variable is calculated using the following expression: ; In the formula, Indicates the first The disturbance values ​​of each control variable.

[0047] It should be noted that: perturbation ratio The perturbation ratios are configured according to the response capabilities, control precision requirements, and operating levels of different actuators. For example, the coal feed rate and air volume directly affect the furnace heat load and combustion state, requiring a certain search range; therefore, a relatively large perturbation ratio is set. The water feed rate and main steam regulating valve directly affect the stability of the steam-water system; therefore, a smaller perturbation ratio is set. Thus, the exemplary values ​​for the perturbation ratios of each control variable are as follows: coal feed rate perturbation ratio 0.05, primary air volume perturbation ratio 0.04, secondary air volume perturbation ratio 0.05, induced draft volume perturbation ratio 0.04, water feed rate perturbation ratio 0.03, desuperheating water flow rate perturbation ratio 0.05, and main steam regulating valve opening perturbation ratio 0.02.

[0048] Furthermore, for each candidate scheme in the final candidate scheme set, the credible state vector of the current boiler operating level is used as the initial state, and the sequence of control variables formed by the control variables in the candidate scheme is used as the external input. Then, the pre-trained prediction model (i.e. the pre-trained prediction model set with a long short-term memory network in the previous step) is called to perform rolling prediction to obtain the boiler prediction state under each prediction step.

[0049] The boiler prediction status includes prediction of main steam pressure, main steam temperature, main steam flow rate, steam drum water level, flue gas oxygen content, furnace negative pressure, coal feed rate, primary air volume, secondary air volume, induced draft volume, feedwater flow rate, and desuperheating water flow rate.

[0050] S2.1 Obtain the main steam pressure deviation, main steam temperature deviation, and main steam flow deviation, perform weighted aggregation over the entire rolling prediction length, and generate steam quality evaluation values.

[0051] Specifically, for each prediction step, the predicted main steam pressure is subtracted from the preset main steam pressure reference value to obtain the main steam pressure deviation; The main steam temperature deviation is obtained by subtracting the preset main steam temperature reference value from the predicted main steam temperature. The deviation of the main steam flow rate is obtained by subtracting the preset steam demand flow rate from the predicted main steam flow rate. Based on the main steam pressure deviation, main steam temperature deviation, and main steam flow rate deviation, calculate the steam quality evaluation value over the entire rolling prediction length.

[0052] The expression for calculating the steam quality evaluation value over the entire rolling prediction length is as follows: ; In the formula, Indicates the first Steam quality evaluation values ​​for each candidate scheme Indicates the rolling prediction length. Indicates the first Each candidate solution in the prediction step Main steam pressure deviation, This indicates the reference value for the main steam pressure. Indicates the first Each candidate solution in the prediction step Main steam temperature deviation, This indicates the reference value for the main steam temperature. Indicates the first Each candidate solution in the prediction step Main steam flow deviation, This indicates the reference value for the main steam flow rate. This indicates the weight of the main steam pressure term. This indicates the weight of the main steam temperature term. This indicates the weight of the main steam flow rate item.

[0053] It should be noted that all reference values ​​remain constant throughout the entire boiler operation. The rated operating parameters of the boiler can be used to determine the values ​​of each baseline. Secondly, in the boiler operation safety requirements, the main steam pressure directly affects the turbine operation safety, so it is set with the highest weight; the main steam temperature affects the equipment thermal stress and steam quality, so it is set with the second weight; the main steam flow mainly reflects the load tracking capability, so it is set with a relatively low weight. As for the weight setting, the constraint usually needs to ensure that the sum of the weights is 1. Therefore, the weight of the main steam pressure is set to 0.40, the weight of the main steam temperature is set to 0.35, and the weight of the main steam flow is set to 0.25.

[0054] S2.2 Calculate the predicted steam drum water level, predicted furnace negative pressure, and predicted flue gas oxygen content for each prediction step. The temperature rise rate is weighted and summed for the limits to generate the equipment safety evaluation value.

[0055] Specifically, after obtaining the allowable upper and lower limits of the steam drum water level, the allowable upper and lower limits of the furnace negative pressure, and the allowable upper and lower limits of the flue gas oxygen content based on the boiler design manual, the excess steam drum water level, the excess furnace negative pressure, and the excess flue gas oxygen content are calculated by combining the predicted steam drum water level, the predicted furnace negative pressure, and the predicted flue gas oxygen content. Based on the predicted main steam temperature, the predicted temperature rise rate between two adjacent prediction steps is calculated, and the temperature rise rate exceeding the limit is obtained. The various exceeding limits are combined to form the equipment safety evaluation value.

[0056] The expression for calculating the excess quantity is as follows: ; In the formula, Indicates the first Among the candidate solutions, the first one is... The parameters in the prediction step The quantity exceeding the limit (the first one here) The parameter refers to any one of the following: steam drum water level, furnace negative pressure, and flue gas oxygen content. This indicates the operation of retrieving the maximum value. Indicates the first Among the candidate solutions, the first one is... The parameters in the prediction step The predicted value, Indicates the first The maximum allowed limit for each parameter. Indicates the first The lower limit of each parameter; After calculating the predicted temperature rise rate between two adjacent prediction steps, the temperature rise rate limit is obtained, expressed as: ; ; In the formula, Indicates the first Each candidate solution in the prediction step With prediction step Predicted temperature rise rate between Indicates the first Each candidate solution in the prediction step The predicted main steam temperature is as follows. Indicates the first Each candidate solution in the prediction step The predicted main steam temperature is as follows. Indicates the control cycle. Indicates the first Each candidate solution in the prediction step The more limited the rate of temperature rise, the better. This indicates the maximum allowable rate of temperature rise (which can be set using the rated parameters in the boiler design manual). The various limits are combined to form the equipment safety evaluation value, expressed as: ; In the formula, Indicates the first Equipment safety evaluation values ​​for each candidate solution. This indicates that the water level in the steam drum exceeds the limit. This indicates that the negative pressure in the furnace exceeds the limit. This indicates that the higher the oxygen content in the flue gas, the greater the limit. This indicates the weight of the steam drum water level exceeding the limit. This indicates the weight of exceeding the limit for negative pressure in the furnace. This indicates the weight of the oxygen content in the flue gas as it exceeds the limit. This indicates the weight of the rate of temperature rise exceeding the limit.

[0057] It should be noted that: Excessive steam drum water level may cause water carryover in the steam, while excessively low steam drum water level may disrupt water circulation and cause water shortage on the heating surfaces. Therefore, exceeding the steam drum water level limit has a direct impact on the safety of the boiler steam-water system and is thus given a high weight. Insufficient furnace negative pressure or a shift to positive pressure may cause flue gas and flames to spray out of the furnace. Excessively high absolute furnace negative pressure may increase air leakage and affect combustion stability, while also increasing the load on the induced draft system. Therefore, its weight is second only to the steam drum water level. A rapid rise in main steam temperature will increase the thermal stress on the superheater, main steam pipelines, and related metal components of the turbine, and may shorten the service life of the heating surfaces. However, exceeding the temperature rise rate limit usually has a certain... The weights of the boiler's oxygen content can be lower than those corresponding to the steam drum water level and furnace negative pressure, given the fixed thermal inertia and settling time. Too low an oxygen content in the flue gas may cause incomplete combustion, while too high an oxygen content will increase exhaust heat loss and reduce boiler efficiency. Exceeding these limits mainly reflects combustion completeness and economy, and can indirectly characterize the stable state of combustion; therefore, their weights can be set to the minimum. Considering the constraint that the sum of the weights is 1, the weights corresponding to the steam drum water level exceeding the limit can be set to 0.35, the furnace negative pressure exceeding the limit to 0.30, the main steam temperature rise rate exceeding the limit to 0.20, and the flue gas oxygen content exceeding the limit to 0.15.

[0058] S2.3. Calculate the cumulative action of the coal feeder, air supply damper, induced draft damper, water supply valve, desuperheating water valve, and main steam regulating valve over the entire rolling prediction length, and generate the actuator action evaluation value.

[0059] The expression for generating the actuator motion evaluation value is as follows: ; In the formula, Indicates the first The evaluation value of the implementing agency actions of each candidate scheme. Indicates the length of the control cycle. Indicates the total number of implementing agencies. Indicates the first Each implementing agency in the prediction step The corresponding change value of the control quantity.

[0060] It should be noted that the change in control quantity is calculated by subtracting the control value of the previous prediction step from the control setpoint of the subsequent prediction step, and taking the absolute value of the difference. This absolute value is used as the action quantity of the actuator in the corresponding prediction step. The use of an absolute value is to ensure that both opening and closing actions are included in the cumulative action quantity, avoiding the cancellation of actions in opposite directions during the accumulation process.

[0061] S2.4 Calculate the wind-coal coordination deviation, steam-water coordination deviation, and desuperheating coordination deviation, and perform weighted aggregation to obtain the coordination control evaluation value.

[0062] Specifically, the air-coal coordination deviation is as follows: the predicted primary air volume and the predicted secondary air volume are added together and then divided by the predicted coal feed to obtain the predicted air-coal ratio. The predicted air-coal ratio is then subtracted from the preset target air-coal ratio and then divided by the target air-coal ratio. The absolute value of the result is then used to obtain the air-coal coordination deviation. It should be noted that the target air-to-coal ratio can be predetermined based on the theoretical air volume, excess air coefficient, primary and secondary air configuration ratio, and historical stable operating data of the boiler's designed coal type. In practice, data from at least 30 minutes of continuous stable operation of the boiler at the corresponding load level can be selected. After excluding periods of start-up / shutdown, soot blowing, pulverizer switching, and sensor failures, the ratio of total air supply to coal feed at each sampling time can be calculated, and the average value of these ratios can be taken as the initial target air-to-coal ratio.

[0063] The steam-water coordination deviation is specifically calculated by multiplying the preset steam-water coordination coefficient by the predicted main steam flow rate, then subtracting the multiplication result from the predicted feedwater flow rate and dividing by the rated feedwater flow rate (the boiler's nameplate technical parameters), and taking the absolute value to obtain the steam-water coordination deviation.

[0064] It should be noted that the steam-water coordination coefficient is used to represent the feedwater demand corresponding to a unit main steam flow rate. It can be determined based on the boiler's continuous blowdown volume, steam-water losses, the desuperheating water regression relationship, and the measurement aperture. When the predicted main steam flow rate and predicted feedwater flow rate use the same statistical boundary, and blowdown and steam-water losses are ignored, the steam-water coordination coefficient can be set to a value close to one. For example, the steam-water coordination coefficient can be set in the range of 1.00 to 1.08, preferably 1.07.

[0065] The desuperheating coordination deviation is specifically calculated using the predicted desuperheating water flow rate, predicted main steam temperature, and main steam temperature reference value. The expression is as follows: ; In the formula, Indicates the first Each candidate solution in the prediction step The temperature reduction coordination deviation Indicates the first Each candidate solution in the prediction step Predicted desuperheating water flow rate, This represents the temperature reduction coordination ratio coefficient. Indicates the first Each candidate solution in the prediction step Predicted main steam temperature, This indicates the reference value for the main steam temperature. This indicates the rated desuperheating water flow rate (technical parameters on the boiler's nameplate).

[0066] It should be noted that the desuperheating coordination ratio coefficient is used to convert the deviation of the main steam temperature relative to the reference value into the corresponding desuperheating water flow rate adjustment requirement, and its unit should match the desuperheating water flow rate unit and the temperature unit. For example, the desuperheating coordination ratio coefficient can be set to 0.20–0.60 tons / hour·℃, preferably 0.30–0.50 tons / hour·℃. A preferred value is 0.40 tons / hour·℃. This value is suitable for subcritical or supercritical coal-fired boilers with a rated main steam temperature of approximately 540℃ and a rated desuperheating water flow rate of approximately 20 tons / hour, and ensures that the desuperheating water adjustment amount maintains a linear correspondence with the main steam temperature deviation.

[0067] The weighted aggregation specifically involves weighting and aggregating the various coordination deviations to obtain the coordination control evaluation value, expressed as: ; In the formula, Indicates the first The coordination and control evaluation value of each candidate scheme. Indicates the first Each candidate solution in the prediction step The deviation in the coordination of wind and coal, Indicates the first Each candidate solution in the prediction step Soft drink coordination deviation, The weight representing the deviation in wind and coal coordination. The weights representing the soft drink coordination deviations, This indicates the weight of the temperature reduction coordination deviation.

[0068] It should be noted that the weights for air-fuel coordination deviation, steam-water coordination deviation, and desuperheating coordination deviation can be set to 0.40, 0.35, and 0.25 respectively, for example. The basis for these values ​​is that the air-fuel coordination deviation directly reflects the matching relationship between fuel and combustion air, and has a relatively direct impact on combustion stability, furnace heat load, and flue gas oxygen content, thus it is given a high weight. The steam-water coordination deviation affects the steam drum water level and steam-water circulation stability, and is given the second highest weight. The desuperheating coordination deviation mainly reflects the degree of matching between the desuperheating water supply and the main steam temperature deviation. Considering that the boiler body has a certain thermal inertia, it can be given a relatively low weight.

[0069] S3. After performing grouping and aggregation based on the target weight vector, combine the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value and coordination control evaluation value to obtain a comprehensive evaluation value and determine the dominant and secondary dominant schemes. Calculate the steam quality deviation degree, external disturbance intensity and mismatch degree and perform weighted summation to obtain the strategy discrimination value. Based on the strategy discrimination value, switch between dominant tracking or anti-disturbance exploration update strategy to iteratively optimize the candidate schemes and obtain a new set of candidate schemes.

[0070] S3.1 Based on the target weight vector, group and aggregate according to the physical attributes of the evaluation dimensions, generate the weight of each evaluation value, and then sum the weighted evaluation values ​​to obtain the comprehensive evaluation value of the candidate solutions.

[0071] Specifically, the main steam pressure control weight and the main steam temperature control weight are extracted from the target weight vector, and the average of the two weights is taken as the weight of the steam quality evaluation value. ; The weights of the equipment safety evaluation value are obtained by arithmetically averaging the weights of combustion stability, flue gas oxygen content control, furnace negative pressure control, steam drum water level control, and main steam temperature rise rate control. ; The weight of the smoothness of the actuator's action is used as the weight of the actuator's action evaluation value. ; The weights of the coordinated control evaluation value are obtained by arithmetically averaging the control weights for unit steam fuel consumption, main steam pressure, and main steam temperature. ; Multiply each evaluation value by its corresponding weight and sum them to obtain the comprehensive evaluation value of the candidate solution. The expression is as follows: ; In the formula, Indicates the first The comprehensive evaluation value of each candidate solution.

[0072] It should be noted that, due to the differences in the physical meaning, number of components, and original numerical range of the four evaluation values, historical data from at least thirty consecutive operating days under normal boiler conditions can be collected. The 95th percentile values ​​of the four evaluation values ​​can be used as the normalization benchmark for the corresponding evaluation values. After normalizing each evaluation value, a weighted summation is then performed.

[0073] S3.2 After sorting all candidate schemes according to their comprehensive evaluation values ​​from smallest to largest, a sorted list is obtained. The candidate scheme with the smallest comprehensive evaluation value in the sorted list is determined as the dominant scheme for the current control period, and the candidate scheme with the second largest comprehensive evaluation value is determined as the secondary dominant scheme.

[0074] It should be noted that the dominant scheme is used to characterize the control direction with the best comprehensive evaluation result in the current candidate scheme set, and its corresponding control variable sequence serves as the main reference for candidate scheme updates and final control output within the current control cycle; The secondary dominant scheme is used to characterize the effective control direction that is second best in overall evaluation results besides the dominant scheme. The secondary dominant scheme does not directly replace the output of the dominant scheme, but participates in the candidate scheme update as an auxiliary reference scheme to provide a second control direction that is different from the dominant scheme.

[0075] S3.3 Calculate the arithmetic mean of the steam quality evaluation values ​​of the prediction steps within the rolling prediction length, and use it as the degree of steam quality deviation.

[0076] It should be noted that the steam quality deviation degree is used to characterize the overall level of steam quality stability maintained by the candidate scheme throughout the entire rolling forecast range. The smaller the steam quality deviation degree, the more closely the main steam pressure, main steam temperature, and main steam flow rate of the candidate scheme remain close to the corresponding target values ​​in each forecast step throughout the entire rolling forecast length, with smaller steam quality fluctuations and better control stability. The larger the steam quality deviation degree, the more the candidate scheme deviates from the target value continuously or has large quality fluctuations during the forecasting process, with poor steam quality stability.

[0077] S3.4 Calculate the intensity of external disturbances based on changes in boiler load demand.

[0078] Specifically, based on the current steam demand and the steam demand at the previous control time, the rate of change of steam demand between two adjacent control times is calculated. Multiply the rate of change in steam demand by the preset load change influence coefficient to obtain the current disturbance intensity of the boiler, which is used as the external disturbance intensity.

[0079] The rate of change in steam demand between two adjacent control moments is calculated using the following expression: ; In the formula, This represents the rate of change in steam demand between the current steam demand and the steam demand at the previous control time. Indicates the current time Steam demand, Indicates the previous control time. Steam demand, This indicates the boiler's rated steam flow rate (obtained from the boiler nameplate).

[0080] It should be noted that the load change influence coefficient is used to convert the rate of change in steam demand into the intensity of external disturbances that can participate in the judgment of operating status and the switching of control strategies. Furthermore, the influence coefficient is a preset parameter greater than zero; for example, the load change influence coefficient is set in the range of 0.5 to 3, preferably 1.5. This ensures that the intensity of external disturbances maintains a good linear response relationship with the rate of change in steam demand.

[0081] S3.5 Calculate the degree of mismatch based on each final valid state value and predicted value in the credible state vector.

[0082] The expression for calculating the degree of mismatch is as follows: ; In the formula, Indicates the current time The degree of mismatch, Indicates the number of measured variables. Indicates the first The measured variables at the current time The final effective state value, Indicates the first Predicted values ​​of each measured variable, Indicates the first The normalized baseline value of each measurement variable.

[0083] It should be noted that the degree of mismatch is used to characterize the overall deviation level between the final effective state value in the credible state vector and the predicted value output by the state prediction model; while the normalized reference value is preferably set as the rated value, such as the rated main steam pressure, rated main steam temperature, and rated steam flow rate of the boiler, which can all be obtained from the boiler's nameplate or design manual.

[0084] S3.6. The steam quality deviation, external disturbance intensity, and mismatch degree are weighted and summed to obtain the strategy discrimination value.

[0085] The weighted summation is expressed as follows: ; In the formula, Indicates the current time The policy discriminant value, The weight representing the degree of deviation in steam quality. The weights representing the intensity of external disturbances. Weights that indicate the degree of mismatch.

[0086] It should be noted that the weights for steam quality deviation, external disturbance intensity, and model mismatch should all be non-negative. To ensure a clear and stable numerical range for the strategy discrimination value, it is preferable that the sum of the three weights equals one; and the three factors—steam quality, external load disturbance, and model mismatch—account for primary, secondary, and auxiliary discrimination roles, respectively. Among them, steam quality, as the boiler control objective, accounts for the primary proportion of the comprehensive discrimination result; external disturbance, as a factor of operating environment change, accounts for the secondary proportion; and mismatch, as an evaluation of model reliability, accounts for the auxiliary proportion. Based on this proportion, the weight of steam quality deviation can be set to 0.4, the weight of external disturbance intensity to 0.35, and the weight of mismatch to 0.25, with the sum of the three weights being one.

[0087] S3.7. Compare the policy discrimination value with the preset policy switching threshold. If the policy discrimination value is less than or equal to the preset policy switching threshold, the dominant tracking update policy is used to update the candidate scheme. If the policy discrimination value is greater than the preset policy switching threshold, the anti-disturbance exploration update policy is used, and all updated candidate schemes are combined into a new candidate scheme set.

[0088] The candidate solutions are updated using a dominant tracking update strategy, expressed as follows: ; In the formula, Indicates the updated number One candidate solution, Indicates the current time The One candidate solution, Indicates the tracking percentage of the leading scheme. Indicates the current time The dominant solution, Indicates the tracking ratio of the secondary dominant scheme. Indicates the secondary dominant option; The disturbance-resistant exploration update strategy is adopted, and the expression is: ; In the formula, Indicates the first The exploration perturbation vector of each candidate scheme.

[0089] It should be noted that both the dominant tracking update strategy and the disturbance-resistant exploration update strategy are used to iteratively update the control sequences in the current candidate scheme set; and the strategy switching threshold can be determined based on the boiler's historical stable operation data and disturbance operation data, for example, it can be set to 0.05 to 0.2, and preferably to 0.1. The tracking ratio of the leading scheme Tracking ratio of secondary dominant schemes All are preset parameters greater than or equal to zero. The dominant solution tracking ratio controls the extent to which candidate solutions move towards the dominant solution, while the secondary dominant solution tracking ratio controls the extent to which candidate solutions absorb features from the secondary dominant solution. Preferably, the dominant solution tracking ratio is greater than the secondary dominant solution tracking ratio, so that the dominant solution becomes the main update direction during the stable operation phase, and the secondary dominant solution only serves as an auxiliary update direction; for example, the dominant solution tracking ratio is set to 0.6, and the secondary dominant solution tracking ratio is set to 0.2. When using this set of values, candidate solutions mainly move towards the current optimal solution, while retaining some features of the original candidate solutions and absorbing information from the secondary optimal solution to a certain extent; Secondly, the exploration perturbation vector can be generated using random numbers. In one implementation, a uniform random number between -1 and 1 is generated for each element of the exploration perturbation vector using a random number generator, and then multiplied by the exploration step size of the corresponding control variable. The exploration step size can be determined based on the allowable range of the control variable.

[0090] S4. Determine the degree of degradation based on the new candidate scheme set, screen the degraded schemes using the degree of degradation, construct the median control sequence, combine the dominant scheme to perform flexible correction and range correction on the degraded schemes, and output the target control value sequence to convert it into control quantity for distribution.

[0091] S4.1 Select the minimum comprehensive evaluation value from the updated candidate schemes as the optimal evaluation value, normalize the comprehensive evaluation values ​​of each updated candidate scheme, and obtain the degree of degradation.

[0092] Normalization, in particular, is expressed as: ; In the formula, Indicates the updated number The degree of degradation corresponding to each candidate solution This represents the optimal evaluation value. This indicates a small positive number that prevents the denominator from being zero (e.g., ).

[0093] It should be noted that the new candidate solution set includes the updated candidate solutions and the recalculated comprehensive evaluation value; while the degree of degradation is used to characterize the relative increase in the comprehensive evaluation value of each updated candidate solution relative to the best comprehensive evaluation value in the current new candidate solution set.

[0094] S4.2. Embed a continuous no-improvement counter for each updated candidate solution, and record the number of consecutive no-improvements for each updated candidate solution through the continuous no-improvement counter.

[0095] Specifically, if the comprehensive evaluation value meets the requirements... If the updated candidate solution is considered improved, the consecutive periods of no improvement in the consecutive no-improvement counter will be reset to zero, and the overall evaluation value will satisfy the following condition. If the current updated candidate solution is not improved, then the number of consecutive unimproved times in the consecutive unimproved counter is incremented by one. Indicates the first The candidate solutions at the current moment The overall evaluation value, Indicates the first The candidate solutions were in the previous control time. The overall evaluation value.

[0096] It should be noted that a maximum count value should be set for the continuously unimproved counter to avoid integer overflow during long-term operation. The maximum count value can be set to one hundred, one thousand, or other upper limits allowed by the data type. When the counter reaches its maximum value, if there is still no improvement, it will remain at the maximum value and will not continue to increase.

[0097] S4.3. Based on the number of consecutive times without improvement and the degree of degradation, jointly determine whether the updated candidate solution is a degraded solution.

[0098] Specifically, the number of consecutive unimproved attempts and the degree of degradation are compared with the preset degradation threshold and the maximum number of consecutive unimproved attempts, respectively. If the number of consecutive unimproved attempts is greater than or equal to the maximum number of consecutive unimproved attempts or the degree of degradation is greater than the degradation threshold, the corresponding updated candidate solution is determined as the degraded solution. If neither of these conditions is met, the current candidate solution is retained and iterative updates continue.

[0099] It should be noted that the maximum number of consecutive no-improvement cycles is a positive integer and can be determined based on the allowed number of candidate iterations and the length of the control period. For example, if many iterations are performed in each control period and the interval between single updates is short, the maximum number of consecutive no-improvement cycles can be set to five to ten times; if a candidate solution is updated once between adjacent control moments, the maximum number can be set to three to six times. Preferably, the present invention sets the maximum number of consecutive no-improvement cycles to five. With this preferred value, candidate solutions are allowed to undergo a small number of no-improvement updates to avoid being immediately eliminated due to a single random disturbance, constraint correction, or evaluation noise. Secondly, the degradation threshold is a non-negative number. If the degradation threshold is set too small, candidate solutions whose comprehensive evaluation value is only slightly worse than the optimal solution may also be judged as degraded, thereby reducing the diversity of the candidate solution set. If the degradation threshold is set too large, significantly worse candidate solutions will still be retained for a long time, increasing invalid predictions and comprehensive evaluation calculations. Therefore, this invention sets the degradation threshold to 0.5. With a value of 0.5, once the comprehensive evaluation cost of a candidate solution relative to the current optimal solution reaches a preset significant level, it can be determined as a degraded solution without having to wait for the number of consecutive unimproved times to reach the maximum value.

[0100] S4.4. Based on the median control sequence of the dominant scheme and all updated candidate schemes, the degraded scheme is flexibly modified to obtain the reconstructed candidate scheme.

[0101] Specifically, the median control sequence is obtained by sorting the control variables corresponding to all updated candidate schemes element by element and taking the median value. Extract the dominant scheme and median control sequence, and combine them with the preset dominant scheme correction ratio to perform flexible correction on the deteriorated scheme, so as to obtain the reconstructed control sequence as a reconstruction candidate scheme. The flexible modification of the degradation scheme is expressed as follows: ; In the formula, Indicates by the first Reconstruction candidate solutions are generated through flexible modification of each candidate solution. Indicates the proportion of the main scheme to be revised. Indicates the median control sequence correction ratio. This represents the median control sequence corresponding to all updated candidate schemes. Indicates the first One candidate solution.

[0102] It should be noted that: flexible correction is used to reconstruct candidate schemes that have been identified as deteriorating schemes, rather than directly deleting them all or completely replacing them with the dominant scheme; and the candidate scheme range of the median control sequence preferably uses all valid candidate schemes that have been updated in this round. Secondly, when obtaining the median control sequence, if the number of valid candidate schemes is odd, the values ​​at each element position are sorted, and the value in the middle is directly selected as the median value of that element. For example, if there are five valid candidate schemes, the five values ​​at a certain element position are sorted, and the third value after sorting is selected. If the number of valid candidate schemes is even, the average of the two middle values ​​after sorting can be used as the median value of that element. For example, if there are six valid candidate schemes, the third and fourth values ​​after sorting are selected, and their average is taken. For control variables that can only be adjusted according to discrete gears, the average value should be rounded according to the minimum adjustment resolution of the actuator after obtaining the average value. It should also be noted that: the dominant scheme correction ratio represents the proportion of dominant scheme control information introduced into the reconstructed candidate schemes, and the median control sequence correction ratio represents the proportion of the median control level introduced into the candidate scheme set. Both the dominant scheme correction ratio and the median control sequence correction ratio should be values ​​not less than zero and not greater than one, and their sum should be less than or equal to one. Under this condition, the remaining proportion corresponding to the deteriorated scheme will not become negative; therefore, this invention exemplarily sets the dominant scheme correction ratio to 0.5 and the median control sequence correction ratio to 0.2. Under these exemplary values, each control element of the reconstructed candidate scheme can primarily refer to the dominant scheme, while absorbing a portion of the median control level from the candidate scheme set and retaining a small amount of information from the original deteriorated scheme.

[0103] S4.5. Based on the candidate reconstruction scheme, query whether the control variables meet the allowable range. If they exceed the allowable range of the control variables, directly correct them to the corresponding boundary values.

[0104] The expression for whether the allowed range is met is: , Indicates the first The lower limit of the allowable range for each control variable. Indicates by the first Among the candidate reconstructing schemes generated from the candidate schemes, the first one is... One control variable, Indicates the first The upper limit of the allowed range for each control variable.

[0105] It should be noted that the allowable ranges of the control variables include: the minimum and maximum allowable coal feed rates of the coal feeder; the minimum and maximum allowable airflow rates of the primary air fan; the minimum and maximum allowable airflow rates of the secondary air fan; the minimum and maximum allowable airflow rates of the induced draft fan; the minimum allowable opening or minimum allowable flow rate and corresponding maximum value of the feedwater regulating valve; the allowable adjustment range of the desuperheating water regulating valve; and the minimum and maximum allowable opening of the main steam regulating valve. Here, "minimum" refers to the lower limit, and "maximum" refers to the upper limit. The allowable ranges of each control variable can be obtained from the equipment design manual or equipment operating procedures. Secondly, the boundary projection method is used to correct to the corresponding boundary value. For example, when the control variable exceeds the allowable range, a new control value is not recalculated, nor is proportional scaling performed. Instead, the corresponding boundary value is directly assigned so that the corrected control variable is exactly at the boundary of the allowable range.

[0106] S4.6 After the maximum number of iterations, the target control value sequence of the control variables is output. The target values ​​in the target control value sequence are converted into control quantities and then sent to the corresponding actuator control loops through wireless transmission technology to achieve optimized control.

[0107] The actuator control circuit includes a coal feeding control circuit, a primary air control circuit, a secondary air control circuit, an induced draft control circuit, a water supply control circuit, a desuperheating water control circuit, and a main steam regulating valve control circuit.

[0108] It should be noted that converting target values ​​in the target control value sequence into control variables can be done through control loop calibration relationships. For example, for a coal feeding control loop, the coal feeder calibration curve or coal feeder flow characteristic table is read, and the corresponding coal feeder speed setpoint is looked up based on the target coal feed rate. If the target coal feed rate is between two calibration points, the corresponding coal feeder speed setpoint is calculated using linear interpolation. After calculation, this speed setpoint is used as the control command for the coal feeding control loop. For both primary and secondary air control loops, read the calibration relationship between fan frequency and air volume, or read the calibration relationship between damper opening and air volume. For example, if the controlled object is a variable frequency fan, calculate the corresponding fan frequency based on the target air volume; if the controlled object is damper adjustment, query the corresponding damper opening based on the target air volume. For the induced draft control loop, read the induced draft fan operating characteristic curve, determine the corresponding induced draft fan frequency or damper opening according to the target induced draft volume, and use it as the set value for the induced draft control loop; For the water supply control loop, the target water supply flow rate is converted into the corresponding control valve opening set value based on the flow characteristic curve of the water supply regulating valve or the corresponding relationship of the valve flow coefficient. For the desuperheating water control loop, read the flow calibration relationship of the desuperheating water regulating valve, determine the corresponding valve opening according to the target desuperheating water flow, and output it to the desuperheating water control loop; For the main steam regulating valve control loop, the output main steam pressure is used as the target pressure and set to the main steam pressure controller. The standard PID controller automatically obtains the valve opening control signal based on the pressure deviation.

[0109] This embodiment also provides a boiler steam staged optimization control system, including: The vector generation module is used for preprocessing, obtaining the change amount, performing amplitude limiting and recursive smoothing, generating measured values ​​for prediction, calculating the confidence level to perform confidence state vector generation, determining the current boiler operating level based on the confidence state vector, and constructing the target weight vector corresponding to the operating level. The evaluation value calculation module is used to define control variables, and after moving forward and padding with zeros to form the first candidate scheme, it applies perturbation to generate a set of candidate schemes. With the credible state vector as the initial state, the control variables are used as external inputs and the prediction model is called to perform rolling prediction to obtain the boiler prediction state at each prediction step. The module also calculates the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordinated control evaluation value for each candidate scheme. The scheme update module is used to perform grouping and aggregation based on the target weight vector, and then combine the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value and coordination control evaluation value to obtain a comprehensive evaluation value and determine the dominant scheme and the secondary dominant scheme. It calculates the steam quality deviation degree, external disturbance intensity and mismatch degree and performs weighted summation to obtain the strategy discrimination value. Based on the strategy discrimination value, it switches the dominant tracking or anti-disturbance exploration update strategy to iteratively optimize the candidate schemes and obtain a new set of candidate schemes. The control quantity distribution module is used to determine the degree of degradation based on the new candidate scheme set, filter the degradation schemes using the degree of degradation, construct the median control sequence, combine the dominant scheme to perform flexible correction and range correction on the degradation schemes, and output the target control value sequence to convert it into control quantities for distribution.

[0110] In summary, by constructing a database of maximum permissible rates of change and a recursive smoothing mechanism, combined with a credibility fusion strategy based on the Laplace kernel, measurement noise and abnormal interference are effectively suppressed. Furthermore, through a load grading mechanism and dynamic weight vector, an adaptive balance between safety, economy, and stability objectives under different operating conditions is achieved. Secondly, by dynamically switching the dominant tracking and anti-disturbance exploration update strategies through policy discriminant values, and by implementing flexible correction and boundary projection reconstruction based on the dominant-median sequence for deteriorated schemes, the global optimization ability and population diversity of the optimization algorithm under strong disturbances are significantly enhanced. Therefore, this invention significantly improves the dynamic quality and steady-state accuracy of the boiler steam control system in complex operating environments.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A boiler steam staged optimization control method, characterized in that: include, Multi-source boiler operation data is collected and preprocessed to obtain normalized data. After obtaining the changes of each measured variable, amplitude limiting and recursive smoothing are performed to generate measured values ​​for prediction. The confidence level is calculated and a confidence state vector is generated. Based on the confidence state vector, the current boiler operation level is determined and a target weight vector corresponding to the operation level is constructed. Based on the measured variables, control variables are defined, and after the first candidate scheme is formed by shifting and padding with zeros, a disturbance is applied to generate a set of candidate schemes. With the credible state vector as the initial state, the control variables are used as external inputs and the prediction model is called to perform rolling predictions to obtain the boiler prediction state at each prediction step. The steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordination control evaluation value of each candidate scheme are calculated. After performing grouping and aggregation based on the target weight vector, the comprehensive evaluation value is obtained by combining the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value and coordination control evaluation value, and the dominant and secondary dominant schemes are determined. The steam quality deviation degree, external disturbance intensity and mismatch degree are calculated and weighted summed to obtain the strategy discrimination value. Based on the strategy discrimination value, the dominant tracking or anti-disturbance exploration update strategy is switched to iteratively optimize the candidate schemes to obtain a new set of candidate schemes. The degree of degradation is determined based on the new candidate scheme set. After screening the degradation schemes based on the degree of degradation, a median control sequence is constructed. The degradation schemes are then flexibly modified and range-corrected in conjunction with the dominant scheme. Finally, the target control value sequence is output and converted into control variables for distribution.

2. The boiler steam staged optimization control method as described in claim 1, characterized in that: The collected multi-source boiler operation data is preprocessed to obtain normalized data and the changes in each measured variable are obtained, as detailed below: By uniformly collecting multi-source data from various measurement points, performing time alignment, missing value imputation, and normalization, we obtain the normalized data of the previous control time and the normalized data of the current time. The change in the normalized value at the current time relative to the normalized value at the previous control time is obtained by subtracting the normalized value at the previous control time from the normalized value at the current time. .

3. The boiler steam staged optimization control method as described in claim 2, characterized in that: The process involves performing amplitude limiting and recursive smoothing, generating measured values ​​for prediction, and calculating the confidence level to generate a confidence state vector, as detailed below: Establish a database of maximum permissible rates of change, and multiply the maximum rate of change of each normalized value by the data sampling period to obtain the maximum permissible change of each normalized value; Based on the maximum change and the change Adjust the normalized values ​​of each measurement variable to obtain the amplitude limit value, perform recursive smoothing on the amplitude limit value, and generate the recursive smoothed measurement value at the current time. The pre-trained prediction model is invoked to predict the current value based on the normalized value of the previous control time. The predicted value is then subtracted from the recursively smoothed measurement value at the current time to obtain the prediction bias of each measurement variable. Based on the prediction bias, the exponential decay function of the Laplace kernel is used to calculate the credibility of each measured variable. The results are then compared with a preset credibility threshold to generate the final valid state value and concatenate them into a credibility state vector.

4. The boiler steam staged optimization control method as described in claim 3, characterized in that: The process of determining the current boiler operating level based on a reliable state vector and constructing a target weight vector corresponding to that operating level refers to extracting the final effective state value of the main steam flow based on the reliable state vector, calculating the boiler load ratio, and comparing it with a preset low load level threshold. Medium load level threshold High load level threshold The comparison is performed to generate the current boiler operating level, and then the corresponding target weight vector is constructed using the analytic hierarchy process.

5. The boiler steam staged optimization control method as described in claim 4, characterized in that: The defined control variables refer to the coal feed rate, primary air volume, secondary air volume, induced draft volume, water supply volume, desuperheating water flow rate, and main steam regulating valve opening as control variables.

6. The boiler steam staged optimization control method as described in claim 5, characterized in that: The calculation of steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordination control evaluation value for each candidate scheme is as follows: Obtain the main steam pressure deviation, main steam temperature deviation, and main steam flow deviation, perform weighted aggregation over the entire rolling prediction length, and generate a steam quality evaluation value. Calculate the predicted steam drum water level, predicted furnace negative pressure, and predicted flue gas oxygen content for each prediction step. Weighted summation is performed for the temperature rise rate that exceeds the limit to generate the equipment safety evaluation value. The cumulative action of the coal feeder, air damper, induced draft damper, water supply valve, desuperheating water valve, and main steam regulating valve over the entire rolling prediction length is calculated separately to generate the actuator action evaluation value. The coordination deviations of wind and coal, steam and water, and cooling are calculated and weighted to obtain the coordination control evaluation value.

7. The boiler steam staged optimization control method as described in claim 6, characterized in that: After performing grouping and aggregation based on the target weight vector, a comprehensive evaluation value is obtained by combining the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordination control evaluation value, and the dominant and secondary dominant schemes are determined, as follows: Based on the target weight vector, the evaluation dimensions are grouped and aggregated according to their physical attributes. After generating the weights of each evaluation value, the weighted sum of each evaluation value is obtained to obtain the comprehensive evaluation value of the candidate solution. After sorting all candidate schemes according to their comprehensive evaluation values ​​from smallest to largest, a sorted list is obtained. The candidate scheme with the smallest comprehensive evaluation value in the sorted list is determined as the dominant scheme for the current control period, and the candidate scheme with the second largest comprehensive evaluation value is determined as the secondary dominant scheme.

8. The boiler steam staged optimization control method as described in claim 7, characterized in that: The calculated steam quality deviation, external disturbance intensity, and mismatch degree are weighted and summed to obtain a strategy discrimination value. Based on the strategy discrimination value, the dominant tracking or disturbance rejection exploration update strategy is switched to iteratively optimize the candidate schemes, resulting in a new set of candidate schemes, as follows: The arithmetic mean of the steam quality evaluation values ​​for the prediction steps within the rolling prediction length is used as the degree of steam quality deviation. Calculate the intensity of external disturbance based on changes in boiler load demand; The degree of mismatch is calculated based on each final valid state value and predicted value in the credible state vector; The strategy discrimination value is obtained by weighted summation of steam quality deviation, external disturbance intensity, and mismatch degree. The policy discrimination value is compared with the preset policy switching threshold. If the policy discrimination value is less than or equal to the preset policy switching threshold, the dominant tracking update policy is used to update the candidate solution. If the policy discrimination value is greater than the preset policy switching threshold, the anti-disturbance exploration update policy is used, and all updated candidate solutions are combined into a new candidate solution set.

9. The boiler steam staged optimization control method as described in claim 8, characterized in that: The process involves determining the degree of degradation based on a new set of candidate solutions, filtering degraded solutions using the degree of degradation, constructing a median control sequence, and then performing flexible and range adjustments on the degraded solutions in conjunction with the dominant solution. Finally, the target control value sequence is output and converted into control variables for distribution. The details are as follows: The minimum comprehensive evaluation value is selected from the updated candidate solutions as the optimal evaluation value. The comprehensive evaluation values ​​of each updated candidate solution are then normalized to obtain the degree of degradation. Embed a consecutive no-improvement counter for each updated candidate solution, and record the number of consecutive no-improvements for each updated candidate solution through the consecutive no-improvement counter; Based on the number of consecutive periods without improvement and the degree of degradation, it is jointly determined whether the updated candidate solution is a degraded solution; Based on the median control sequence of the dominant scheme and all updated candidate schemes, the degraded scheme is flexibly modified to obtain the reconstructed candidate scheme; Based on the candidate reconstruction scheme, check whether the control variables meet the allowable range. If they exceed the allowable range of the control variables, directly correct them to the corresponding boundary values. After the maximum number of iterations is reached, the target control value sequence of the control variables is output. The target values ​​in the target control value sequence are converted into control quantities and then sent to the corresponding actuator control loops via wireless transmission technology to achieve optimized control.

10. A boiler steam staged optimization control system, based on the boiler steam staged optimization control method according to any one of claims 1 to 9, characterized in that: include, The vector generation module is used for preprocessing, obtaining the change amount, performing amplitude limiting and recursive smoothing, generating measured values ​​for prediction, calculating the confidence level to perform confidence state vector generation, determining the current boiler operating level based on the confidence state vector, and constructing the target weight vector corresponding to the operating level. The evaluation value calculation module is used to define control variables, and after moving forward and padding with zeros to form the first candidate scheme, it applies perturbation to generate a set of candidate schemes. With the credible state vector as the initial state, the control variables are used as external inputs and the prediction model is called to perform rolling prediction to obtain the boiler prediction state at each prediction step. The module also calculates the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value, and coordinated control evaluation value for each candidate scheme. The scheme update module is used to perform grouping and aggregation based on the target weight vector, and then combine the steam quality evaluation value, equipment safety evaluation value, actuator action evaluation value and coordination control evaluation value to obtain a comprehensive evaluation value and determine the dominant scheme and the secondary dominant scheme. It calculates the steam quality deviation degree, external disturbance intensity and mismatch degree and performs weighted summation to obtain the strategy discrimination value. Based on the strategy discrimination value, it switches the dominant tracking or anti-disturbance exploration update strategy to iteratively optimize the candidate schemes and obtain a new set of candidate schemes. The control quantity distribution module is used to determine the degree of degradation based on the new candidate scheme set, filter the degradation schemes using the degree of degradation, construct the median control sequence, combine the dominant scheme to perform flexible correction and range correction on the degradation schemes, and output the target control value sequence to convert it into control quantities for distribution.

Images