Burner air speed control method based on intelligent bionic optimization

Abstract

The invention relates to the technical field of combustor control, and discloses a combustor air speed control method based on intelligent bionic optimization, which comprises the following steps of: acquiring multi-source heterogeneous operation data and modeling an optimization problem; dynamic alignment and stability enhancement of multi-source heterogeneous operation data are realized; optimization of improved differential evolution control parameters based on working condition partition; on-line switching and anti-integral saturation compensation of control parameters based on working condition identification; adaptively updating a working condition partition boundary based on incremental operation data; and real-time wind speed control based on working condition partition and parameter online fusion is carried out. The method has the beneficial effects that the hysteresis response of the oxygen content and the thermal inertia characteristic of the hearth temperature are converted into prior constraint conditions in time sequence alignment, the time shifting range is limited, and the amplitude, trend and smoothness combined cost function is constructed; therefore, the multi-source heterogeneous signal not only meets time synchronization, but also conforms to the combustion mechanism of first air regulation, then oxygen generation and then temperature rise in the alignment process.

Description

Technical Field

[0001] This invention relates to the field of burner control technology, specifically to a burner wind speed control method based on intelligent biomimetic optimization. Background Technology

[0002] In industrial combustion systems, the quality of air velocity control directly affects combustion efficiency, pollutant emissions, furnace pressure stability, and equipment operational safety, making it a core component of combustion optimization control. With increasing demands for energy conservation, emission reduction, and operational stability in industrial production, modern burners often require frequent adjustments across a wide load range and must cope with various time-varying factors such as fuel composition fluctuations, environmental changes, and equipment aging. As the execution link in airflow regulation, the air velocity control system's task is to ensure that the actual air velocity quickly, stably, and smoothly tracks the setpoint under different operating conditions, thereby guaranteeing a reasonable air-fuel ratio and maintaining combustion within a high-efficiency, low-emission range.

[0003] However, the following problems still exist in the existing technology: 1. When processing multi-source heterogeneous sensor data, existing technologies typically align or interpolate signals with different sampling frequencies and response speeds to the same time point by simply aligning them according to timestamps. This ignores physical time sequence differences such as oxygen content response lag and furnace temperature thermal inertia. As a result, the high-frequency changes in wind speed and the low-frequency response of oxygen content are mistakenly regarded as synchronous causality during modeling, which weakens the ability of subsequent control optimization to characterize the real combustion mechanism.

[0004] 2. Existing control methods mainly rely on absolute measurements such as wind speed, oxygen content, and temperature to determine operating conditions. They lack quantitative characterization methods for the degree to which the combustion state deviates from the stable trajectory. When there are fluctuations in fuel calorific value or local combustion instability, it is difficult to effectively distinguish between normal operating condition adjustments and abnormal combustion states by simply relying on changes in absolute values, which reduces the sensitivity of operating condition identification to abnormal situations.

[0005] 3. Existing burner wind speed control mostly adopts a single fixed PID parameter or a simple partition hard switching strategy. A single parameter can only achieve good results under certain typical operating conditions, while its performance degrades under other operating conditions. The hard switching strategy is prone to step changes in proportional, integral, and derivative parameters when crossing partition boundaries, causing sudden action of the actuator and wind speed fluctuations, which cannot adapt to the continuous change requirements within a wide operating range.

[0006] 4. Existing operating condition zoning methods are usually based on offline clustering of historical data and used in a fixed manner. They do not have the ability to adaptively update with equipment wear, changes in fuel composition and environmental conditions. After long-term operation, the boundaries of historical zoning gradually deviate from the current actual operating area, causing online operating condition identification to incorrectly map the current operating condition to an inapplicable zoning, making the originally optimized control parameters unable to achieve the expected results. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a burner wind speed control method based on intelligent biomimetic optimization, thus resolving the problems existing in the background technology.

[0008] To achieve the above objectives, the present invention employs the following technical solution: The intelligent biomimetic optimization-based burner wind speed control method includes the following steps: S1. Collect multi-source heterogeneous operating data during the burner operation process, preprocess the collected data, and construct a multi-objective optimization problem model based on the preprocessed data; S2. Perform dynamic time alignment processing on multi-source heterogeneous operating data to construct auxiliary features that can characterize the degree of combustion deviation from the steady state, and obtain an enhanced operating condition feature vector for operating condition identification and control parameter optimization. S3. Divide the operating conditions into zones based on the historical enhanced operating condition feature vectors, optimize the PID controller parameters in each operating condition zone, and finally obtain the optimal parameter set for each operating condition zone. S4. Real-time identification of the current burner operating conditions, calling the corresponding control parameters from the optimal parameter set based on the identification results, and executing a smooth transition strategy and anti-integral saturation compensation processing during parameter switching; S5. By maintaining the sliding window data pool, the center and width of the operating condition partition are updated in small steps periodically using the latest operating data, so that the operating condition partition can continuously reflect the current system characteristics. S6. By integrating the current operating condition zone information with the online updated control parameters and combining real-time feedback deviation, the final wind speed control command is generated and the actuator is driven to achieve real-time closed-loop control of the burner wind speed.

[0009] Furthermore, optimization issues include minimizing wind speed tracking error, suppressing wind speed fluctuations during the control process, and reducing frequent operation of damper actuators, while also adapting to the shift in operating conditions caused by load changes, fuel characteristic fluctuations, and changes in environmental conditions during burner operation.

[0010] Furthermore, S2 specifically includes: S21. Construct the original operating condition feature vector, and organize the wind speed setpoint, instantaneous wind speed, oxygen content, furnace temperature and flue gas pressure into an original operating condition feature vector indexed by the wind speed sampling time. S22. Perform dynamic time warping with combustion physical constraints. Through dynamic time warping with combustion physical constraints, align low-frequency sensor data to the high-frequency wind speed time axis to obtain an aligned working condition feature vector that has a causal relationship with wind speed changes in physical time sequence. S23. Construct combustion stability indicator factors. Based on the aligned operating condition feature vector, further extract combustion stability indicator factors to characterize the degree to which the current wind-oxygen relationship deviates from the historical stable state. S24. The combustion stability indicator factor is concatenated into the aligned operating condition feature vector to form the enhanced operating condition feature vector.

[0011] Furthermore, S3 specifically refers to: S31. Perform working condition partitioning and define optimization variables. Perform fuzzy clustering based on the key working condition quantities in the enhanced working condition feature vector to form multiple working condition partitions, and define independent PID parameters for each working condition partition. S32. Construct an objective function for optimizing control parameters, incorporating speed, stability, and smoothness into the objective function simultaneously; S33. An improved differential evolution algorithm is used for parameter optimization. Two mechanisms, "current best individual guidance" and "adaptive mutation factor based on population fitness dispersion", are introduced into the standard differential evolution framework to improve the balance between global exploration and local convergence.

[0012] Furthermore, S4 specifically refers to: S41. Calculate the membership degree of the current working condition to each working condition partition, identify which working condition partition the current working condition is near based on the current enhanced working condition feature vector, and provide continuous weights for subsequent parameter fusion. S42. Perform fuzzy weighted fusion on the optimal PID parameters of each operating condition partition, and perform weighted fusion on the optimal PID parameters of each operating condition partition according to the normalized membership degree to obtain the PID parameters actually used at the current control moment. S43. Execute discrete PID control with anti-integral saturation compensation, and add a compensation term based on saturation feedback to the discrete PID control law.

[0013] Furthermore, S5 specifically refers to: S51. Construct a sliding window data pool and define update trigger conditions. Establish a sliding window data pool of fixed length and decide whether to trigger partition updates based on the number of samples and the degree of deterioration of the control effect. S52. The weighted fuzzy C-means is used to adaptively update the boundary of the working condition partition. Based on the standard fuzzy C-means clustering, time decay weight and control effect weight are introduced to incrementally fine-tune the partition center and partition width. S53. Load the updated operating condition partition parameters to decouple partition boundary updates from online control.

[0014] Furthermore, S31 specifically refers to: S311, from the enhanced working condition feature vector Extracting operating points used for operating condition partitioning Operating point It consists of the wind speed setpoint and the aligned oxygen content; S312. After normalizing all operating points, the fuzzy C-means clustering algorithm is used to divide them into... One working condition zone; S313, Divide each working condition into zones Define independent PID parameter vectors ; S314. Establish a partition center for each operating condition partition for subsequent online operating condition identification. and partition width .

[0015] Furthermore, S32 specifically refers to: S321, Divide each working condition into zones. Using historical running samples or closed-loop simulation samples within this partition, the candidate PID parameter vectors are... Conduct an evaluation of the control effectiveness; S322. Calculate three types of evaluation indicators based on the control results. The first type of evaluation indicator is the normalized average absolute error, which is used to characterize the tracking accuracy. The second type of evaluation indicator is the standard deviation of the wind speed tracking error, which is used to characterize the control stability. The third type of evaluation indicator is the average absolute value of the change in the actuator command between adjacent control cycles, which is used to characterize the smoothness of the action. S323. Sum the three types of evaluation indicators according to their weights to form the working condition zoning. The corresponding optimization objective function.

[0016] Furthermore, S33 specifically refers to: S331, Divide each working condition into zones. Initialize the population separately, and denote the population size as . , representing the number of individual candidate parameters for a single operating condition partition; S332. Perform the objective function evaluation of step S32 on each individual in the initial population to obtain the initial fitness set, and calculate the standard deviation of the initial population fitness. ; S333, Entering the... During generation iteration, first calculate the standard deviation of the current population fitness based on the current generation fitness set. And then according to Adaptive calculation of variation factor ; S334. For each target individual in the current generation, generate a mutated individual; S335. Perform a binomial crossover on the target individual and the variant individual to generate experimental individuals. The crossover probability is denoted as... ; S336. Perform boundary repair on the test individuals; S337. Calculate the objective function values ​​for the experimental individuals and the target individuals respectively, and use a greedy selection mechanism to retain the better individuals for the next generation; S338. When the maximum number of iterations is reached, or the improvement of the optimal objective function is less than the threshold for several consecutive iterations, the parameter optimization for this operating condition partition ends, and the optimal PID parameter vector is output. .

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention proposes a dynamic time warping method with combustion physics constraints, which transforms the hysteresis response of oxygen content and the thermal inertia characteristics of furnace temperature into a priori constraints in time alignment. By limiting the time shift range and constructing a joint cost function of amplitude, trend and smoothness, the multi-source heterogeneous signals can satisfy both time synchronization and conform to the combustion mechanism of "adjusting air first, then oxygen, and then raising temperature" during the alignment process, thus solving the fundamental problem of traditional time alignment methods destroying causal relationships.

[0018] 2. This invention constructs a combustion stability indicator factor, which calculates the current wind-oxygen ratio by aligning the wind speed and oxygen content, and quantifies the deviation with the statistical mean and standard deviation within the sliding window. This transforms the absolute value measurement into a characterization of the degree of deviation from the stable trajectory, enabling the operating condition feature vector to have the ability to sensitively identify abnormal states such as low load, fuel fluctuations, and local combustion instability.

[0019] 3. This invention adopts a control parameter tuning strategy that combines operating condition partitioning and fuzzy weighted fusion. First, the operating domain is divided into multiple local operating condition partitions through fuzzy C-means clustering, and the PID parameters are optimized independently for each partition. During online operation, the optimal parameters of each partition are normalized and weighted by continuous Gaussian membership, so that the control parameters achieve a smooth transition rather than a step switch during continuous changes in operating conditions, fundamentally avoiding control disturbances at the partition boundaries.

[0020] 4. This invention constructs an adaptive update mechanism for operating condition partitions based on a dual weighting of time decay and control effect. It maintains recently running samples through a sliding window data pool and assigns higher weights to samples with more recent time and worse control effect in fuzzy clustering update, so that the partition center and width can continuously drift with changes in equipment wear and fuel characteristics. At the same time, the update process is decoupled from online control and parameter mutations are prevented through smooth replacement, thus achieving long-term adaptability of the operating condition identification boundary. Attached Figure Description

[0021] Figure 1 This is a flowchart of the present invention; Figure 2This is a comparative analysis of the performance of the present invention and the conventional proportional-integral-derivative control method in terms of mean absolute error index; Figure 3 This is a comparative analysis of the present invention and the conventional proportional-integral-derivative control method in terms of error standard deviation. Figure 4 This is a comparative analysis of the present invention and the conventional proportional-integral-derivative control method in terms of the actuator's motion amplitude index; Figure 5 This is a comparative analysis of the smoothness of proportional parameters changing with operating conditions between the present invention and conventional partition switching technology; Figure 6 This is a diagram illustrating the effect of the adaptive update mechanism for operating condition partitions in this invention. Detailed Implementation

[0022] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0023] like Figure 1 As shown, this invention proposes a burner wind speed control method based on intelligent biomimetic optimization, the main contents of which are as follows: S1. Multi-source heterogeneous operation data acquisition and optimization problem modeling: Collect multi-source heterogeneous operation data during the burner operation process, preprocess the collected data, and construct a multi-objective optimization problem model based on the preprocessed data; Multi-source heterogeneous operating data is collected from the burner field control system and sensor network. This data covers key variables that reflect the dynamic characteristics of the burner, including wind speed setpoint, instantaneous wind speed measurement, flue gas oxygen content, furnace temperature, and flue gas pressure.

[0024] Among them, wind speed measurements are usually collected at a higher sampling frequency to capture rapid dynamic changes, while process parameters such as oxygen content, furnace temperature and flue gas pressure have different sampling frequencies, response lags and noise interference due to different sensor characteristics and physical response mechanisms.

[0025] The collected data needs to undergo basic preprocessing, including median filtering and low-pass filtering of the wind speed signal to remove pulse interference and high-frequency noise, and moving average or first-order smoothing of the oxygen content, furnace temperature and flue gas pressure sequences to eliminate occasional fluctuations.

[0026] Based on this, the present invention defines the burner wind speed control problem as a dynamic optimization problem with multiple operating conditions and constraints. Its core is to optimize controller parameters under different combustion conditions so that the actual wind speed can quickly, stably, and smoothly track the wind speed setpoint. This optimization problem needs to consider multiple objectives simultaneously: minimizing wind speed tracking error, suppressing wind speed fluctuations during the control process, and reducing frequent actions of the damper actuator. It also needs to adapt to the shifts in operating conditions caused by changes in load, fuel characteristics, and environmental conditions during burner operation.

[0027] S2. Dynamic alignment and stability enhancement of multi-source heterogeneous operating data: Dynamic time alignment processing is performed on multi-source heterogeneous operating data to construct auxiliary features that can characterize the degree of combustion deviation from the steady state, and obtain an enhanced operating condition feature vector for operating condition identification and control parameter optimization. Burner wind speed control relies on multiple signals, including wind speed, wind speed setpoint, oxygen content, furnace temperature, and flue gas pressure. These signals differ significantly in sampling frequency, response lag, and noise characteristics. If the original values ​​at the same moment are directly spliced ​​together for modeling, the high-frequency changes in wind speed, the low-frequency response of oxygen content, and the inertial lag of furnace temperature are easily mistakenly regarded as synchronous changes, thereby weakening the ability of subsequent control parameter optimization to characterize the real combustion mechanism.

[0028] This invention uses the wind speed sampling time as a unified time reference to perform time-series processing on oxygen content, furnace temperature, and flue gas pressure. Furthermore, it constructs auxiliary features that characterize the degree to which combustion deviates from a steady state, resulting in an enhanced operating condition feature vector for operating condition identification and control parameter optimization. The specific steps are as follows: S21. Construct the original operating condition feature vector. The wind speed setpoint, instantaneous wind speed, oxygen content, furnace temperature, and flue gas pressure are uniformly organized into an original operating condition feature vector indexed by the wind speed sampling time. The specific steps are as follows: 1) Collect the wind speed setpoint sequence, wind speed measurement sequence, oxygen content sequence, furnace temperature sequence, and flue gas pressure sequence during burner operation. Among them, the wind speed measurement sequence is preferably collected at a higher sampling frequency, such as 50Hz; the oxygen content sequence is preferably collected at a lower sampling frequency, such as 1Hz; the furnace temperature sequence can be collected by thermocouples, which usually have thermal inertia; and the flue gas pressure sequence can be collected by a pressure transmitter.

[0029] 2) Perform basic preprocessing on the data of each channel. Specifically, the wind speed measurement value is preferably first subjected to median filtering to remove pulse interference, and then subjected to low-pass filtering to suppress high-frequency noise caused by furnace pressure pulsation; the oxygen content sequence, furnace temperature sequence and flue gas pressure sequence are preferably removed by moving average or first-order smoothing to remove occasional fluctuations.

[0030] Meanwhile, to avoid the impact of dimensional differences on subsequent regularization path search, a normalized copy in the range of 0 to 1 is constructed for each channel within the path search, while the output features still retain the actual engineering units.

[0031] 3) Using the sampling time of the wind speed measurement as a unified time reference, define the first... Each wind speed sampling time is And construct the original working condition feature vector. , represented as ; in, Indicates the first The original working condition feature vector at each sampling time; express The wind speed setpoint at any given time, in m / s, represents the control target; express The wind speed measurement value at any given time, in m / s, represents the actual wind speed state; Indicates and The nearest original oxygen content record, in % . Indicates and The nearest original furnace temperature record, in °C; Indicates and The nearest original flue gas pressure record, in Pa.

[0032] It should be noted that, , and This only represents the original records organized by time index; at this point, precise time alignment conforming to the combustion mechanism has not yet been completed.

[0033] S22. Perform dynamic time warping with combustion physics constraints. The effects of burner wind speed changes on oxygen content and furnace temperature do not occur simultaneously. Instead, they usually affect oxygen content first, and then affect furnace temperature with a lag. Simply aligning them to the same moment would disrupt the causal relationship.

[0034] This invention uses dynamic time warping with combustion physics constraints to align low-frequency sensor data to a high-frequency wind speed time axis, obtaining an aligned operating condition feature vector that has a causal relationship with wind speed changes in physical time sequence. The specific steps are as follows: 1) Using wind speed time axis As a reference time axis, local candidate time shift ranges are established for the oxygen content sequence and the furnace temperature sequence, respectively. Specifically, the following definitions are made. for The time shift of oxygen content at a given moment, defined as follows: for The time shift of the furnace temperature at any given moment; Based on the burner response characteristics, restrictions lie in Within the scope, restrictions lie in Within the range, among which, This represents the maximum allowable response lag time for oxygen content, which can be taken as 2 seconds. This indicates the maximum allowable response lag time for furnace temperature, which can be set to 5 seconds. The reason for this setting is that oxygen content responds faster than wind speed, while furnace temperature responds slower than wind speed. Prior constraints can prevent reverse pairings that do not conform to the combustion mechanism from occurring in regular paths.

[0035] 2) At each wind speed sampling time Continuous candidate sampling points were established within the candidate time shift ranges for oxygen content and furnace temperature, respectively. For the oxygen content and furnace temperature values ​​corresponding to the candidate time, linear interpolation was used to obtain them from the original low-frequency sequence.

[0036] In practice, linear interpolation is implemented by finding the two nearest original sampling points before and after the target time, and then weighting the two sampling values ​​proportionally according to the relative position of the target time between these two sampling points to obtain the interpolation result for continuous time. After this processing, the low-frequency sensor data can be mapped to any continuous time consistent with the wind speed time axis.

[0037] 3) Based on the idea of ​​dynamic time warping, path search is performed on the candidate time shift paths. Specifically, a local cost matrix is ​​established between the wind speed reference time axis and the oxygen content candidate time axis, and another local cost matrix is ​​established between the wind speed reference time axis and the furnace temperature candidate time axis. Each cost value can be composed of three parts: first, the amplitude matching cost between the current candidate value and the local change in the current wind speed; second, the trend matching cost between the slope of the current candidate sequence and the trend of the local change in wind speed; and third, the smoothing penalty introduced when the time shift changes too quickly in adjacent moments.

[0038] Based on this, path search not only focuses on the similarity of single point values, but also prioritizes retaining alignment results that are continuous in time and reasonable in trend.

[0039] 4) Use dynamic programming to search for the regular path with the minimum total cost from the starting point to the ending point, and only allow the path to move in the positive time direction. At the same time, limit the time shift between adjacent moments to no more than a preset step size, thereby ensuring that the regular path satisfies monotonicity and continuity, and avoids suddenly jumping to a candidate point far away from the current physical state at a certain moment.

[0040] 5) Based on the optimal regularization path, determine each Corresponding oxygen content time shift and furnace temperature time shift And construct the aligned working condition feature vector. , represented as ; in, Indicates the first Alignment feature vectors at each sampling time; This represents the aligned oxygen content value, in percentages (%). This indicates the aligned furnace temperature, in °C. This represents the aligned flue gas pressure value, in Pa. Flue gas pressure responds quickly to changes in wind speed; in a convenient implementation method, it can be directly aligned to the nearest neighbor or through linear interpolation. No separate lag search is performed.

[0041] It should be noted that this invention directly incorporates the timing pattern of "adjusting air first, then adding oxygen, and then raising temperature" in the combustion process into the search constraints of dynamic time warping, so that the warping results are not only aligned in time, but also reasonable in mechanism, thereby improving the ability of subsequent operating condition identification and control parameter optimization to depict the real combustion dynamics.

[0042] S23, Constructing combustion stability indicator factors Using only aligned wind speed, oxygen content, furnace temperature, and flue gas pressure, although the main operating condition information has been retained, the sensitivity to abnormal conditions such as low load, fuel calorific value fluctuations, and local combustion instability is still insufficient.

[0043] This invention further extracts combustion stability indicator factors based on the aligned operating condition feature vector, which are used to characterize the degree to which the current wind-oxygen relationship deviates from the historical stable state. The specific steps are as follows: 1) At each sampling time Based on the aligned wind speed values and aligned oxygen content values Calculate the current wind oxygen ratio .

[0044] In one implementation, It can be taken as the wind speed value divided by the oxygen content value. In order to avoid the ratio being abnormally amplified due to extremely low oxygen content, a positive lower limit can be set for the oxygen content, such as 0.1%.

[0045] 2) Establish a length of [length] before the current sampling time. A sliding window is used to statistically analyze the moving average and standard deviation of the wind-oxygen ratio within the window. The moving average characterizes the recent stable wind-oxygen relationship, while the standard deviation characterizes the natural fluctuation range of the recent wind-oxygen relationship. This indicates the window length, which can be set according to the control cycle and the rate of change of operating conditions, for example, taking 20 to 200 sampling points.

[0046] 3) Calculate the combustion stability indicator factor based on the deviation between the current air-oxygen ratio and the sliding window statistical results. , represented as ; in, Indicates the first The combustion stability indicator factor at each sampling time; the larger the value, the more obvious the current wind-oxygen relationship deviates from the recent stable operating condition. Indicates the first The average value of the wind-oxygen ratio within the sliding window corresponding to each sampling time; The standard deviation of the air-oxygen ratio within the same sliding window; This represents the smallest positive number that can be divided by zero. .

[0047] 4) Limit or compress the combustion stability indicator factor to avoid excessive impact of extreme anomalies on subsequent operating condition zones.

[0048] In one implementation, it can be... Limit the range to 0 to 10, or use logarithmic compression to reduce the impact of the maximum value.

[0049] It should be noted that simple absolute measurements are difficult to reflect whether the current state deviates from the stable combustion trajectory. Combustion stability indicator factors compare the instantaneous air-oxygen relationship with recent statistical patterns, which can significantly enhance the ability of operating conditions to identify abnormal combustion states.

[0050] S24. Forming an enhanced operating condition feature vector. To unify the aligned multi-source operating condition information and combustion stability information for subsequent control parameter tuning, this invention concatenates the combustion stability indicator factor into the aligned operating condition feature vector to form an enhanced operating condition feature vector. The specific steps are as follows: 1) Align the feature vectors of the working conditions Combustion stability indicator factor By concatenating them in a fixed order, we obtain the enhanced working condition feature vector. , represented as ; in, Indicates the first The enhanced operating condition feature vector at each sampling time includes the wind speed setpoint, actual wind speed, aligned oxygen content, aligned furnace temperature, aligned flue gas pressure, and combustion stability indicator factor.

[0051] 2) The enhanced working condition feature vectors are written into the sample library in chronological order for subsequent working condition partitioning, control parameter optimization and online working condition identification. In order to ensure that the training and online identification use the same feature order, the order of each component of the enhanced working condition feature vector remains unchanged throughout the process.

[0052] S3. Improved differential evolution control parameter optimization based on operating condition partitioning: The operating condition partitioning is divided according to the historical enhanced operating condition feature vector, and the PID controller parameters are optimized in each operating condition partition. Finally, the optimal parameter set corresponding to each operating condition partition is obtained. The burner wind speed control object has characteristics such as nonlinearity, hysteresis, time-varying and multi-condition switching. Using a single fixed PID parameter often only achieves good results under a certain local condition. When the load or fuel conditions change, problems such as slow response, oscillation or frequent actuator operation are likely to occur.

[0053] This invention first divides the operating conditions into zones based on historical enhanced operating condition feature vectors, then optimizes the PID controller parameters within each operating condition zone, and finally obtains the optimal parameter set corresponding to each operating condition zone. This allows the controller parameters to no longer be a compromise solution for all operating conditions, but rather a locally optimal solution for each operating condition zone. The specific steps are as follows: S31. Perform operating condition partitioning and define optimization variables. To ensure that different control parameters correspond to different combustion conditions, this invention performs fuzzy clustering based on key operating condition quantities in the enhanced operating condition feature vector to form multiple operating condition partitions, and defines independent PID parameters for each operating condition partition. The specific steps are as follows: 1) From the enhanced working condition feature vector Extracting operating points used for operating condition partitioning Operating point Composed of wind speed setpoint and aligned oxygen content, expressed as ; in, Indicates the first The operating points of each sample are used to describe the current control target and the degree of oxygen enrichment in combustion. The reason for selecting these two quantities as the basis for partitioning is that the wind speed setpoint directly determines the target operating range, and the oxygen content directly reflects the matching relationship between air volume and fuel. Together, they determine the controller's requirements for response speed and stability.

[0054] 2) After normalizing all operating points, the fuzzy C-means clustering algorithm was used to divide them into... There are several operating condition zones, among which... This indicates the number of operating condition zones, which can be determined based on the burner's design operating point, operating range, and historical data distribution; for example, it can be 3 to 8.

[0055] In practical implementation, the fuzzy C-means clustering algorithm is used to divide the data into... Each operating condition zone, specifically, should be initialized first. The system identifies cluster centers and calculates the membership degree of each sample to each partition based on the distance from the sample to each cluster center. The cluster centers are then updated repeatedly based on the membership degree until the change in cluster centers between two adjacent iterations is less than a threshold.

[0056] 3) Divide each working condition into zones. Define independent PID parameter vectors , represented as ; in, Indicates the first The vector of control parameters to be optimized for each operating condition zone; Indicates the first Proportional parameters for each working condition zone; Indicates the first The integral time constant for each operating condition zone; Indicates the first The differential time constant of each operating condition zone.

[0057] 4) Establish a partition center for each operating condition partition for subsequent online operating condition identification. and partition width The larger the partition width, the wider the range of operating conditions covered by that partition. Indicates the first Cluster centers for each working condition zone; Indicates the first The sample scatter width of each working condition partition can be calculated from the root mean square distance of the samples in that partition to the cluster center.

[0058] S32. Construct the objective function for optimizing control parameters. Burner wind speed control not only requires small tracking error, but also a smooth control process and that the damper movement is not too violent.

[0059] This invention incorporates speed, stability, and smoothness into the objective function simultaneously, avoiding control oscillations caused by solely pursuing the minimum of a single error. The specific steps are as follows: 1) Divide each working condition into zones Using historical running samples or closed-loop simulation samples within this partition, the candidate PID parameter vectors are... Conduct an evaluation of the control effectiveness.

[0060] In practical implementation, the samples in this partition can be input into the burner wind speed control model or the historical playback model in chronological order to obtain the wind speed tracking error sequence and the actuator command sequence.

[0061] 2) Calculate three types of evaluation indicators based on the control results. Specifically, the first type of evaluation indicator is the normalized average absolute error, which is used to characterize the tracking accuracy; the second type of evaluation indicator is the standard deviation of the wind speed tracking error, which is used to characterize the control stability; and the third type of evaluation indicator is the average absolute value of the change in the actuator command between adjacent control cycles, which is used to characterize the smoothness of the action.

[0062] 3) The three types of evaluation indicators are weighted and summed to form the working condition zones. Corresponding optimization objective function , represented as ; in, Indicates the first Each operating condition zone in the candidate PID parameter vector The overall evaluation value is as follows: the smaller the value, the better the control performance. This represents the normalized mean absolute error. This represents the standard deviation of the wind speed tracking error; This represents the average absolute value of the changes in the execution mechanism's instructions; The weighting coefficients representing tracking accuracy. Weighting coefficients representing stability The weighting coefficient representing smoothness can be set according to actual control requirements; for example, it can be set to... , , .

[0063] In practical implementation, the normalized mean absolute error It is preferable to use the average value of the ratio of the absolute value of each sample error to the corresponding wind speed setpoint to reduce the impact of the inconsistency of error dimensions between different setpoint ranges; the change in actuator command is preferably calculated using the absolute value of the difference between damper opening commands in adjacent control cycles to punish excessively frequent or large adjustment actions.

[0064] It should be noted that if the burner wind speed control only pursues rapid error convergence, it often leads to frequent damper operation or even induces secondary oscillations. By incorporating both smoothness and stability into the objective function, control parameters more suitable for industrial operation can be obtained.

[0065] S33. Parameter optimization is performed using an improved differential evolution algorithm. Conventional differential evolution algorithms suffer from insufficient early exploration and a tendency to get trapped in local optima in multimodal nonconvex optimization problems.

[0066] This invention introduces two mechanisms into the standard differential evolution framework: "current best individual guidance" and "adaptive mutation factor based on population fitness dispersion," to improve the balance between global exploration and local convergence. The specific steps are as follows: 1) Divide each working condition into zones Initialize the population separately, and denote the population size as . , which represents the number of individual candidate parameters for a single operating condition partition, for example, it can be 20 to 60; Each individual is a PID parameter vector, and the initial values ​​of each individual are randomly generated within a preset boundary range, for example, The value can be taken within the empirical proportional gain range. The value can be taken within the allowable range of the integration time constant. It can take values ​​within the allowable range of differential time constants.

[0067] 2) Perform the objective function evaluation of step S32 on each individual in the initial population to obtain the initial fitness set, and calculate the standard deviation of the initial population fitness. , which represents the degree of dispersion of the fitness of the 0th generation population, and is used to reflect the initial population diversity.

[0068] 3) Enter the During generation iteration, first calculate the standard deviation of the current population fitness based on the current generation fitness set. And then according to Adaptive calculation of variation factor , represented as ; in, Indicates the first The adaptive variation factor of the generation; This indicates the lower limit of the variation factor, which can be between 0.2 and 0.4. This represents the upper limit of the variation factor, which can range from 0.8 to 1.0; Indicates the first Standard deviation of population fitness.

[0069] Based on this, when the current generation population has large differences, the variation factor is large, which is conducive to expanding the search range; when the current generation population gradually converges, the variation factor decreases, which is conducive to local fine search.

[0070] 4) For each target individual in the current generation, generate a mutated individual. Specifically, the mutation method adopts a combination of "current best individual + difference vector + random perturbation vector". This is because the current best individual can guide the convergence, and the random perturbation term can suppress the population from prematurely gathering in local regions. This is expressed as: ; in, Indicates the first The first working condition zone is in the The generation targeting the first Mutant individuals generated from a target individual; Indicates the first The individual with the best fitness in the generation; This refers to the first individual randomly selected from the current population. This indicates the second individual randomly selected from the current population. This indicates the third individual randomly selected from the current population. This refers to the fourth individual randomly selected from the current population. The four individuals are several distinct individuals randomly selected from the current population. This represents the random disturbance coefficient, which can be between 0.05 and 0.3.

[0071] 5) Perform a binomial crossover on the target individual and the variant individual to generate experimental individuals. The crossover probability is denoted as . It can be fixed at 0.9.

[0072] In the specific implementation, a random number between 0 and 1 is generated for each dimension parameter. When the random number is no greater than 1, the random number is determined. When the experimental individual is selected, this dimension is taken from the variant individual; otherwise, it is taken from the target individual. At the same time, at least one dimension must be randomly assigned from the variant individuals to ensure that there are indeed differences between the experimental individuals and the target individuals.

[0073] 6) Perform boundary repair on the test individuals. Specifically, when a certain dimension parameter exceeds the preset parameter range, the dimension can be pulled back to the allowable range by truncation, mirroring, or re-random sampling, thereby avoiding invalid parameters from entering the fitness evaluation stage.

[0074] 7) Calculate the objective function values ​​of the experimental individuals and the target individuals respectively, and use a greedy selection mechanism to retain the better individuals for the next generation. That is, when the objective function value of the experimental individual is smaller, the experimental individual replaces the target individual; otherwise, the original target individual is retained.

[0075] 8) When the maximum number of iterations is reached, or when the improvement of the optimal objective function is less than the threshold for several consecutive iterations, the parameter optimization for this operating condition partition ends, and the optimal PID parameter vector is output. , indicating the first The optimal PID parameter vector obtained after optimization for each operating condition partition corresponds to the proportional, integral, and derivative parameters, respectively. , indicating the first The optimal PID parameter vector for each operating condition zone; in, This represents the optimal scaling parameter. This represents the optimal integration time constant. This represents the optimal differential time constant.

[0076] It should be noted that this invention no longer uses a fixed mutation factor, but automatically adjusts the search intensity according to the degree of dispersion of the population fitness. Furthermore, it utilizes both optimal individual guidance and random difference perturbation in the mutation operation, thereby taking into account both the early global search and the later local refinement, and reducing the probability of the control parameter tuning getting trapped in a local optimum.

[0077] S4. Online switching of control parameters and anti-integral saturation compensation based on operating condition identification: Real-time identification of the current operating condition of the burner; calling the corresponding control parameters from the optimal parameter set according to the identification results; and executing a smooth transition strategy and anti-integral saturation compensation processing during parameter switching.

[0078] When the burner is running online, the operating conditions will continuously drift with changes in load, fuel and environment. If only fixed parameters of a certain operating condition zone are used, or if hard switching is made directly at the zone boundary, it is easy to cause sudden changes in control parameters and oscillations in the actuator.

[0079] This invention calculates the membership degree of the current operating condition to each operating condition zone online, smoothly fuses the optimal PID parameters of each operating condition zone, and adds anti-integral saturation compensation at the controller output to ensure parameter continuity and actuator safety under wide operating conditions. The specific steps are as follows: S41. Calculate the membership degree of the current operating condition to each operating condition zone. This invention identifies which operating condition zone the current operating condition is located in based on the current enhanced operating condition feature vector, and provides continuous weights for subsequent parameter fusion. The specific steps are as follows: 1) At the current control moment Obtain the feature vector of the current enhanced working condition. Extract the current operating point from it. , represents the operating point at the current control moment, and, Composed of the current wind speed setpoint and the current aligned oxygen content, it is represented as ; in, This represents the enhanced operating condition feature vector at the current control moment.

[0080] 2) Based on the zoning centers of each working condition obtained in step S301 and partition width The similarity between the current operating point and each operating condition partition is calculated using a Gaussian membership function, expressed as follows: ; in, Indicates the current operating point For the first Membership degree of each working condition zone; Indicates the current operating point and the first... Euclidean distance between the centers of each working condition zone; The larger the value, the wider the coverage of the adjacent operating conditions.

[0081] 3) Normalize all membership degrees to obtain normalized membership degrees. ; In practice, the membership degree of the current working point to a certain working condition partition can be divided by the sum of the membership degrees of all working condition partitions. When the sum of all membership degrees is too small, a very small positive number can be added to prevent division by zero.

[0082] It should be noted that the operating point usually does not fall exactly at the center of a certain operating zone. Using continuous membership rather than hard classification can provide continuous weights for the smooth transition of subsequent PID parameters.

[0083] S42. Perform fuzzy weighted fusion of the optimal PID parameters for each operating condition zone. To avoid sudden changes caused by hard parameter switching, this invention performs weighted fusion of the optimal PID parameters for each operating condition zone according to normalized membership degrees to obtain the PID parameters actually used at the current control moment. The specific steps are as follows: 1) Read the optimal PID parameter vector corresponding to each working condition partition. This includes the optimal scaling parameter. Optimal integration time constant and optimal differential time constant .

[0084] 2) Based on normalized membership degree The weighted summation of the optimal parameters for each operating condition zone yields the actual PID parameters at the current control moment, expressed as: , , ; in, This indicates the proportion of parameters actually in use at the current control moment. This represents the integral time constant actually used at the current control moment. This represents the differential time constant actually used at the current control moment.

[0085] 3) The fused current PID parameters are sent to the online controller. After this processing, when the current operating point is close to the center of a certain operating zone, the parameter weight of that operating zone will dominate. When the current operating point is between two or more operating zones, the control parameters will smoothly transition between these operating zones, thereby suppressing control oscillations near the zone boundaries.

[0086] S403, Perform discrete PID control with anti-integral saturation compensation. The burner damper actuator has a physical limit. If the controller output exceeds the limit for a long time and the integral term continues to accumulate, it will form obvious integral saturation, resulting in slow controller recovery when the error reverses.

[0087] This invention adds a compensation term based on saturation feedback to the discrete PID control law. The specific steps are as follows: 1) Define the lower limit of the opening degree of the damper actuator as: The maximum opening degree is ,in, 0% is acceptable. 100% is acceptable; The controller's original output at the previous moment Execute the limit and obtain the actual execution output from the previous moment. , represented as ; in, Indicates time Actual execution instructions after bandwidth limiting This represents the function that takes the minimum value. This represents the function that takes the maximum value.

[0088] It should be noted that nested maximum and minimum value functions are used to modify the controller's raw output. The limitation, firstly, Ensure that the output is not lower than the lower limit ,Then, Ensure that the result of the previous step does not exceed the upper limit. ,final, It is restricted to The actual instructions executed within the interval.

[0089] 2) Calculate the wind speed tracking error at the current control moment. This is the difference between the current wind speed setpoint and the current wind speed measurement.

[0090] 3) Calculate the current control increment using an incremental discrete PID control law. And an anti-integral saturation compensation term is added to it, expressed as: ; in, This represents the control increment at the current control moment; This indicates the wind speed tracking error at the current control moment. This indicates the wind speed tracking error in the previous control cycle; This indicates the wind speed tracking error in the first two control cycles; Indicates the control cycle; This represents the anti-integral saturation compensation coefficient, which can take values ​​between 0 and 1, for example, 0.1 to 0.5; This represents the output difference before and after the previous time limit. This difference is 0 when saturation has not occurred, and is a correction amount opposite to the direction of over-limit when saturation occurs.

[0091] 4) Update the current controller's original output based on the actual output executed at the previous moment and the current control increment. Then, the amplitude limiting is performed again to obtain the actual output at the current moment. And send the value to the damper actuator.

[0092] 5) Buffer the error, control increment, raw output and actual execution output at the current moment for continued use in the next control cycle.

[0093] It should be noted that this invention combines the PID parameters after partition smoothing and fusion with anti-integral saturation compensation, so that the controller can still maintain good continuity and recoverability when operating conditions change and actuator limit exists at the same time, reducing the hysteresis and oscillation after damper saturation.

[0094] S5. The working condition partition boundary is adaptively updated based on incremental operating data. By maintaining a sliding window data pool, the center and width of the working condition partition are updated in small steps periodically using the latest operating data, so that the working condition partition can continuously reflect the current system characteristics.

[0095] After long-term operation, wear and tear, changes in fuel composition, changes in environmental conditions, and sensor drift can cause the boundaries of historical operating condition zones to gradually deviate from the current actual operating area. If the operating condition zone boundaries are not updated, even if each operating condition zone already has better parameters, the online identification stage may incorrectly map the current operating condition to an unsuitable operating condition zone.

[0096] This invention maintains a sliding window data pool and periodically updates the center and width of the operating condition partitions in small steps using the latest operational data, enabling the operating condition partitions to continuously reflect the current system characteristics. The specific steps are as follows: S51. Construct a sliding window data pool and define update trigger conditions. To correct the operating condition partition boundaries using recent operational data, this invention establishes a fixed-length sliding window data pool and determines whether to trigger partition updates based on the number of samples and the degree of deterioration in control performance. The specific steps are as follows: 1) Maintain a fixed-length sliding window data pool. Used to store the most recent There are 10 valid running samples, among which, This represents the sliding window data pool; This indicates the data pool capacity, which can be set according to the device's operating rhythm; for example, it can take 500 to 5000 samples.

[0097] 2) For each sample entering the sliding window data pool, at least the operating point should be recorded. and corresponding control effect evaluation indicators ,in, Indicates the sample time Corresponding operating points; This indicates the control effectiveness evaluation index corresponding to this sample.

[0098] 3) Calculate control performance evaluation indicators based on wind speed tracking error and actuator action. , represented as ; in, This represents the motion weighting coefficient, used to balance the impact of tracking error and actuator motion on the control effect, and can be taken from 0.1 to 1.0; The larger the value, the worse the control performance of the sample under the corresponding operating condition. Indicates the sample time The corresponding control increment; Indicates the sample time The corresponding wind speed tracking error.

[0099] 4) Set the trigger conditions for the work condition partition update. Specifically, either of the following two types of trigger conditions can be used: First, the number of new samples in the sliding window data pool has reached the update threshold. Second, the average value of the control effect evaluation index within the sliding window data pool. continuous The statistical period was worse than the historical average. 1.2 times; in, This indicates the threshold of new samples required to trigger an update; This represents the average value of the control effectiveness evaluation index within the current window; Indicators representing the historical long-term average control effectiveness evaluation indicators; This indicates the number of consecutive degradation determinations.

[0100] S52. Adaptively update the working condition partition boundaries using weighted fuzzy C-means. To make the boundary of the operating condition partition more focused on recent operating status and areas of weak control, this invention introduces time decay weights and control effect weights on the basis of standard fuzzy C-means clustering, and performs incremental fine-tuning on the partition center and partition width. The specific steps are as follows: 1) When any of the triggering conditions in step S501 is met, retrieve the data from the sliding window data pool. The system reads all operating points and control performance evaluation indicators, and calculates the comprehensive weight for each sample. This gives higher weight to samples that are more recent in time and have poorer control effects during partition updates, as shown below: ; in, Indicates sample The overall weight; This represents the time decay coefficient, with values ​​between 0 and 1, for example, 0.95 to 0.999; Indicates the current update time; Indicates sample The sampling time; Indicates sample Evaluation indicators of control effectiveness; This represents the average value of the control effect evaluation index within the current sliding window data pool; This represents the standard deviation of the control performance evaluation index within the current sliding window data pool; This represents the weighting coefficient for the control effect, used to adjust the emphasis on samples with poor control. It represents the smallest positive number that is not divisible by zero.

[0101] 2) Using the current operating condition partition center as the initial center for weighted fuzzy C-means clustering, weighted clustering is performed on the operating condition points in the sliding window data pool. Specifically, the fuzzy membership degree is first calculated based on the distance of the sample to each partition center, and then combined with the comprehensive weight. Update the center of each partition, and then repeat the iteration until the change in the center of each partition is less than the threshold. In this way, while preserving the continuity of the original partition structure, the partition boundaries are guided to adjust to the areas with weak control in the near future.

[0102] 3) Update the first Partition center of each working condition zone , represented as ; in, Indicates the updated number Each working condition zone center; Indicates sample For the first Fuzzy membership degree of each working condition partition; This represents the fuzziness index, with 2 being the preferred value. Indicates sample The corresponding operating point.

[0103] 4) Recalculate the partition number based on the updated partition center. The partition width of each working condition zone Specifically, the weighted root mean square distance between the partition sample and the center of the updated partition can be used as the partition width to characterize the actual coverage of the updated partition.

[0104] 5) To avoid excessive disturbance to online recognition caused by a single update, a smooth replacement is performed on the updated partition center and partition width. For example, the old parameters and the new parameters are linearly fused at a fixed ratio before being written into the online recognition module, or a maximum allowable update step size is set and the excess is truncated, thereby preventing drastic fluctuations in partition boundaries due to short-term abnormal data.

[0105] It should be noted that this invention does not simply redo the clustering with the latest data. Instead, it uses a "time decay + control effect weighting" approach to focus updates on recent operating conditions with poor control quality. This allows the operating condition partition boundaries to better align with the system's time-varying characteristics while avoiding excessive damage to the long-term effective partition structure.

[0106] S53, Load the updated operating condition partition parameters To ensure that real-time control is not interrupted by the partition update process, this invention decouples partition boundary updates from online control. The specific steps are as follows: 1) In a non-real-time update thread, complete the update of the working condition partition center and partition width in step S502 to obtain the updated set of working condition partition centers. and the updated set of partition widths .

[0107] 2) Perform a validity check on the updated results. The validity check should include at least the following: Check whether the number of samples in each working condition partition is higher than the minimum sample threshold; whether the width of each partition is greater than the minimum allowed width; and whether the updated partition center is still within the preset working condition range.

[0108] 3) After the update result passes the validity check, at the end of the current control cycle, the partition center in the online identification module is set. and partition width Replace with and This ensures that the membership calculation in step S401 uses the latest working condition partition boundary.

[0109] 4) After the replacement is completed, continue to use steps S402 and S403 to perform online parameter fusion and wind speed control. This allows the operating condition zone boundary to adaptively track the dynamic changes of the burner during long-term operation without interrupting real-time control.

[0110] S6. Real-time wind speed control based on working condition zoning and online parameter fusion: It integrates the current working condition zoning information with the online updated control parameters, combines real-time feedback deviation, generates the final wind speed control command and drives the actuator to achieve real-time closed-loop control of burner wind speed. In practical applications, this invention performs operating condition identification, parameter fusion, anti-integral saturation compensation, and output limiting sequentially within each control cycle according to a fixed control cycle, thereby forming a closed-loop real-time wind speed control.

[0111] First, at the current control moment, the controller acquires the latest wind speed setpoint, instantaneous wind speed measurement, and time-aligned oxygen content, furnace temperature, and flue gas pressure to construct the current enhanced operating condition feature vector. Then, using the center and width of each working condition partition obtained by clustering historical data in advance, the Gaussian membership degree of the current working condition point to each working condition partition is calculated and normalized to obtain a set of continuous weights that reflect the similarity between the current working condition and each partition. Then, the system reads the optimal PID parameters corresponding to each operating condition partition, including proportional parameters, integral time constant and derivative time constant, and uses normalized membership degree as weight to perform weighted fusion of the PID parameters of each partition to generate the PID parameters actually used in the current control cycle. Then, after obtaining the fused PID parameters, the controller calculates the tracking error based on the current wind speed setpoint and the actual wind speed measurement, and uses an incremental discrete PID control law to calculate the control increment. When calculating the control increment, the controller introduces an anti-integral saturation compensation term. This term is based on the difference between the controller's original output and the actual limited output at the previous moment. When integral saturation occurs, it generates a reverse correction amount to suppress the excessive accumulation of the integral term. Finally, the controller adds the actual execution output of the previous moment to the current control increment to obtain the original control output of the current moment, and performs amplitude limiting processing to ensure that the output value is within the allowable opening range of the damper actuator. The limited final command is then sent to the damper actuator to drive the damper to adjust the wind speed.

[0112] The entire control process is executed cyclically in each cycle, enabling the controller to achieve a smooth transition of PID parameters when the burner operating conditions change continuously, and effectively suppressing the control lag and oscillation caused by actuator saturation, thereby ensuring the speed, stability and safety of wind speed control.

[0113] In one embodiment, such as Figures 2-4As shown, a comparative analysis of control performance under different operating conditions was conducted to verify the comprehensive control performance of the wind speed control method based on intelligent biomimetic optimization proposed in this invention under different combustion conditions. The experiment selected three typical operating conditions: low load, medium load, and high load. A bar chart was used to compare the performance of this technology with the conventional proportional-integral-derivative (PI-DE) control method in three key indicators: mean absolute error, standard deviation of error, and actuator amplitude. Conventional PI-DE control refers to a classic control method that uses a fixed set of proportional, integral, and derivative parameters. These parameters are usually tuned empirically under a typical operating condition and remain unchanged under all operating conditions. This is currently the most widely used technical solution in industrial burner control. The horizontal axis of the bar chart represents three different combustion conditions, and the vertical axes represent "mean absolute error (m / s)," "standard deviation of error (m / s)," and "amplitude of action (percentage)." The mean absolute error reflects the overall accuracy of wind speed tracking, the standard deviation of the error reflects the stability of the control process, and the amplitude of the action measures the intensity of the action of the damper actuator during the adjustment process. The smaller the amplitude of the action, the smoother the control process and the less wear on the actuator. Figures 2-4 Two different filled textures are used in the bar chart to represent conventional proportional-integral-derivative (PI-DE) control and this technology, respectively, arranged side-by-side for easy comparison. The data distribution in the bar chart clearly shows that this technology significantly outperforms conventional PI-DE control in all three indicators across the three operating conditions. Specifically, the mean absolute error of this technology is lower than that of the conventional method in all operating conditions, indicating that this technology can more accurately track the setpoint to the actual wind speed; the standard deviation of this technology is also significantly smaller, meaning that the wind speed tracking process is more stable and less volatile; in terms of the actuator's action amplitude, this technology also shows a significant advantage, indicating that its control strategy is smoother and the damper adjustment frequency and amplitude are lower. Experimental results demonstrate the advantages of this invention through operating condition partitioning and parameter fusion, optimizing control parameters separately for different operating conditions, avoiding the performance degradation caused by traditional single fixed parameters taking a compromise solution for all operating conditions, and achieving a comprehensive improvement in speed, stability, and smoothness.

[0114] In one embodiment, such as Figure 5As shown, a comparative analysis of the smoothness of proportional parameters changing with operating conditions is conducted to verify whether the parameter fuzzy weighted fusion mechanism based on operating condition membership in this invention can effectively suppress abrupt changes in control parameters when operating conditions change continuously. The experiment simulates the dynamic process of the wind speed setpoint continuously increasing from 5 meters per second to 20 meters per second, comparing the proportional parameter change curves under the parameter smoothing transition strategy of this technology with those under the traditional hard switching strategy. The traditional hard switching strategy refers to directly selecting the optimal proportional parameter corresponding to the nearest partition as the current control parameter based on the distance between the current operating point and the center of each operating partition. This strategy can cause a step jump in control parameters when the operating condition crosses the partition boundary. The horizontal axis of the line graph represents the "wind speed setpoint (m / s)," reflecting the continuous change process of the burner's target wind speed; the vertical axis represents the "proportional parameter (dimensionless)," a key parameter in the proportional-integral-derivative controller that determines the current error response intensity, and its value directly affects the speed and stability of control. Figure 5 The three points at 7.5 m / s, 12.5 m / s, and 17.5 m / s correspond to the center positions of the three operating condition zones. The blue dashed line represents the proportional parameter change trajectory under the hard switching strategy, and the orange solid line represents the change trajectory under the fuzzy weighted fusion strategy of this technology. As can be seen from the blue line graph, the proportional parameter under the hard switching strategy exhibits a clear step-like jump when crossing zone boundaries, especially near the centers of the three zones, where the parameter value suddenly jumps from the value of the previous zone to the value of the next zone. This jump can easily trigger sudden action of the actuator in actual control, causing wind speed fluctuations. In contrast, the proportional parameter curve under the fuzzy weighted fusion strategy of this technology exhibits a smooth transition shape, with the parameter value changing continuously when crossing zone boundaries without any step points. The curve retains the dominant characteristics of the optimal parameter in the zone near the zone center, while achieving a gradual transition of parameters between zones. Experimental results show that this technology, through Gaussian membership functions and normalized weighted fusion, effectively avoids the control disturbances caused by parameter hard switching, enabling the controller to achieve smooth parameter evolution when operating conditions change continuously.

[0115] In one embodiment, such as Figure 6 As shown, the effect of the working condition partition adaptive update mechanism of the present invention is illustrated in the form of a scatter plot and a center offset arrow. Figure 6 In Figure (a), the distribution of operating points formed by historical operating data and their initial cluster centers are shown. The horizontal axis represents the wind speed setpoint in meters per second, and the vertical axis represents the aligned oxygen content in percentage. Different colors represent the three initial operating condition partitions of the fuzzy clustering, and the red crosses are the initial cluster centers of each partition. Figure 6(b) shows the distribution of new operating point data collected after a period of operation, overlaid with the updated partition centers obtained through weighted fuzzy C-means clustering. Green squares represent the updated centers, red crosses represent the original center positions, and gray arrows indicate the direction and magnitude of the shift in each partition center due to operating condition drift. Figure 6 As can be seen from (b) in the figure, due to the influence of factors such as equipment wear and changes in fuel characteristics during long-term operation, the operating point of the low-load area moves slightly to the upper right, while the operating point of the high-load area shifts to the lower right. The updated partition center can follow this drift trend, making the partition boundary more consistent with the current operating state. Figure 6 In (a) of this paper, the original partition boundaries are based on historical data. If they are not updated for a long time, the current operating condition will be incorrectly mapped to an inapplicable partition during subsequent online identification. This invention combines a sliding window data pool with time decay and control effect weights to ensure that partition boundary updates can both preserve the long-term effective structure and reflect recent changes in weak control areas, thereby ensuring that operating condition identification and parameter fusion are always based on accurate operating condition partitions.

Claims

1. A burner wind speed control method based on intelligent biomimetic optimization, characterized in that, Includes the following steps: S1. Collect multi-source heterogeneous operating data during the burner operation process, preprocess the collected data, and construct a multi-objective optimization problem model based on the preprocessed data; S2. Perform dynamic time alignment processing on the multi-source heterogeneous operating data to construct auxiliary features that can characterize the degree of combustion deviation from the steady state, and obtain an enhanced operating condition feature vector for operating condition identification and control parameter optimization. S3. Divide the operating conditions into zones based on the historical enhanced operating condition feature vectors, optimize the PID controller parameters in each operating condition zone, and finally obtain the optimal parameter set for each operating condition zone. S4. Real-time identification of the current operating conditions of the burner, calling the corresponding control parameters from the optimal parameter set according to the identification results, and performing a smooth transition strategy and anti-integral saturation compensation processing during parameter switching. S5. By maintaining the sliding window data pool, the center and width of the operating condition partition are updated in small steps periodically using the latest operating data, so that the operating condition partition can continuously reflect the current system characteristics. S6. By integrating the current operating condition zone information with the online updated control parameters and combining real-time feedback deviation, the final wind speed control command is generated and the actuator is driven to achieve real-time closed-loop control of the burner wind speed.

2. The burner wind speed control method based on intelligent bionic optimization according to claim 1, characterized in that, The optimization problem includes minimizing wind speed tracking error, suppressing wind speed fluctuations during the control process, and reducing the frequent operation of the damper actuator. It also needs to adapt to the shift in operating conditions caused by load changes, fuel characteristic fluctuations, and changes in environmental conditions during burner operation.

3. The burner wind speed control method based on intelligent bionic optimization according to claim 1, characterized in that, S2 specifically includes: S21. Construct the original operating condition feature vector, and organize the wind speed setpoint, instantaneous wind speed, oxygen content, furnace temperature and flue gas pressure into an original operating condition feature vector indexed by the wind speed sampling time. S22. Perform dynamic time warping with combustion physical constraints. Through dynamic time warping with combustion physical constraints, align low-frequency sensor data to the high-frequency wind speed time axis to obtain an aligned working condition feature vector that has a causal relationship with wind speed changes in physical time sequence. S23. Construct combustion stability indicator factors. Based on the aligned operating condition feature vector, further extract combustion stability indicator factors to characterize the degree to which the current wind-oxygen relationship deviates from the historical stable state. S24. The combustion stability indicator factor is concatenated into the aligned operating condition feature vector to form the enhanced operating condition feature vector.

4. The burner wind speed control method based on intelligent bionic optimization according to claim 1, characterized in that, S3 specifically refers to: S31. Perform working condition partitioning and define optimization variables. Perform fuzzy clustering based on the key working condition quantities in the enhanced working condition feature vector to form multiple working condition partitions, and define independent PID parameters for each working condition partition. S32. Construct an objective function for optimizing control parameters, incorporating speed, stability, and smoothness into the objective function simultaneously; S33. An improved differential evolution algorithm is used for parameter optimization. Two mechanisms, "current best individual guidance" and "adaptive mutation factor based on population fitness dispersion", are introduced into the standard differential evolution framework to improve the balance between global exploration and local convergence.

5. The burner wind speed control method based on intelligent bionic optimization according to claim 1, characterized in that, S4 specifically refers to: S41. Calculate the membership degree of the current working condition to each working condition partition, identify which working condition partition the current working condition is near based on the current enhanced working condition feature vector, and provide continuous weights for subsequent parameter fusion. S42. Perform fuzzy weighted fusion on the optimal PID parameters of each operating condition partition, and perform weighted fusion on the optimal PID parameters of each operating condition partition according to the normalized membership degree to obtain the PID parameters actually used at the current control moment. S43. Execute discrete PID control with anti-integral saturation compensation, and add a compensation term based on saturation feedback to the discrete PID control law.

6. The burner wind speed control method based on intelligent bionic optimization according to claim 1, characterized in that, S5 specifically refers to: S51. Construct a sliding window data pool and define update trigger conditions. Establish a sliding window data pool of fixed length and decide whether to trigger partition updates based on the number of samples and the degree of deterioration of the control effect. S52. The weighted fuzzy C-means is used to adaptively update the boundary of the working condition partition. Based on the standard fuzzy C-means clustering, time decay weight and control effect weight are introduced to incrementally fine-tune the partition center and partition width. S53. Load the updated operating condition partition parameters to decouple partition boundary updates from online control.

7. The burner wind speed control method based on intelligent bionic optimization according to claim 4, characterized in that, S31 specifically refers to: S311, from the enhanced working condition feature vector Extracting operating points used for operating condition partitioning Operating point It consists of the wind speed setpoint and the aligned oxygen content; S312. After normalizing all operating points, the fuzzy C-means clustering algorithm is used to divide them into... One working condition zone; S313, Divide each working condition into zones Define independent PID parameter vectors ; S314. Establish a partition center for each operating condition partition for subsequent online operating condition identification. and partition width .

8. The burner wind speed control method based on intelligent bionic optimization according to claim 4, characterized in that, S32 specifically refers to: S321, Divide each working condition into zones. Using historical running samples or closed-loop simulation samples within this partition, the candidate PID parameter vectors are... Conduct an evaluation of the control effectiveness; S322. Calculate three types of evaluation indicators based on the control results. The first type of evaluation indicator is the normalized average absolute error, which is used to characterize the tracking accuracy. The second type of evaluation indicator is the standard deviation of the wind speed tracking error, which is used to characterize the control stability. The third type of evaluation indicator is the average absolute value of the change in the actuator command between adjacent control cycles, which is used to characterize the smoothness of the action. S323. Sum the three types of evaluation indicators according to their weights to form the working condition zoning. The corresponding optimization objective function.

9. The burner wind speed control method based on intelligent bionic optimization according to claim 4, characterized in that, S33 specifically refers to: S331, Divide each working condition into zones Initialize the population separately, and denote the population size as . , representing the number of individual candidate parameters for a single operating condition partition; S332. Perform the objective function evaluation of step S32 on each individual in the initial population to obtain the initial fitness set, and calculate the standard deviation of the initial population fitness. ; S333, Entering the... During generation iteration, first calculate the standard deviation of the current population fitness based on the current generation fitness set. And then according to Adaptive calculation of variation factor ; S334. For each target individual in the current generation, generate a mutated individual; S335. Perform a binomial crossover on the target individual and the variant individual to generate experimental individuals. The crossover probability is denoted as... ; S336. Perform boundary repair on the test individuals; S337. Calculate the objective function values ​​for the experimental individuals and the target individuals respectively, and use a greedy selection mechanism to retain the better individuals for the next generation; S338. When the maximum number of iterations is reached, or the improvement of the optimal objective function is less than the threshold for several consecutive iterations, the parameter optimization for this operating condition partition ends, and the optimal PID parameter vector is output. .

Images