Air pre-heater temperature control optimization method for thermal power generation

Abstract

The invention discloses an air pre-heater temperature control optimization method for thermal power generation, and belongs to the technical field of control optimization, and the method comprises the steps: constructing an air pre-heater inlet air temperature PID control system, obtaining an actual air temperature feedback value through an inlet air temperature monitoring module, and transmitting the actual air temperature feedback value to an inlet air temperature error calculation module; the inlet air temperature error calculation module receives the target temperature value and compares the target temperature with a feedback value so as to calculate and output a real-time error signal and transmit the real-time error signal to the inlet air temperature PID controller module, and the inlet air temperature PID controller module subjected to parameter optimization calculates a control quantity according to the real-time error signal; and the control quantity is output to the inlet air temperature adjusting module so as to adjust the inlet air temperature of the air pre-heater. The risk of corrosion and dust deposition of the low-temperature section of the air pre-heater can be effectively reduced, and the high use value is achieved for guaranteeing safe operation of a thermal power generation system and improving the fine operation level.

Description

An Optimization Method for Air Preheater Temperature Control in Thermal Power Generation Technical Field

[0001] This invention belongs to the technical field of control optimization, and particularly relates to an optimization method for air preheater temperature control in thermal power generation. Background Technology

[0002] The air preheater is the core heat exchange equipment in the boiler system of a thermal power plant. It is responsible for heating the combustion air using the waste heat from the flue gas. Its operating status is an important guarantee for the thermal efficiency and reliability of the generator set. During operation, the temperature of the inlet air directly affects the metal wall temperature of the low-temperature end heating surface of the air preheater. If the temperature is too low, water vapor and sulfides in the flue gas will condense on the surface of the air preheater, causing severe low-temperature corrosion and ash blockage. These problems will lead to increased resistance on the flue gas side of the equipment, reduced heat exchange efficiency, and may even cause safety hazards such as damage to core heat exchange components, increased power consumption of the induced draft fan, and unplanned shutdown of the unit. To control the inlet air temperature of the air preheater, commonly used solutions include air heater systems, hot air recirculation systems, and flue gas bypass devices. These systems maintain temperature stability by adjusting the flow rates on the steam or air side. However, due to nonlinear disturbances caused by drastic load changes, parameter drift caused by ambient temperature variations, and actuator response lag, traditional PID control strategies often rely on manual parameter tuning based on experience. When facing variable operating conditions or complex disturbances, these strategies often struggle to guarantee the accuracy of the control system, leading to problems such as large overshoot, excessively long settling times, and unstable control performance. Therefore, it is necessary to introduce a more efficient control method. Summary of the Invention

[0003] To overcome the technical problems described in the background section, this invention provides an optimization method for air preheater temperature control in thermal power generation. It introduces an improved, groundbreaking platypus optimization algorithm to optimize and tune the PID control parameters, which can effectively reduce the risk of corrosion and ash accumulation in the low-temperature section of the air preheater. This method has high application value in ensuring the safe operation of the thermal power generation system and improving the level of refined operation.

[0004] The technical solution of this invention is as follows: an optimization method for air preheater temperature control in thermal power generation, comprising the following steps: S1, constructing a PID control system for air preheater inlet air temperature, including an inlet air temperature error calculation module, an inlet air temperature PID controller module, an improved breakthrough platypus optimization algorithm module, an inlet air temperature regulation module, and an inlet air temperature monitoring module; S2, introducing an improved breakthrough platypus optimization algorithm and including at least the following improvement strategies: S21, introducing an enhanced population initialization strategy in the initialization stage, generating candidate individuals through a combination of chaotic mapping and back learning, and obtaining a formal initial population after boundary repair, fitness evaluation, and screening; S22, introducing a differential evolution parameter adaptive update strategy in the search evolution stage, setting differential variable asynchronous long factor and crossover probability for each individual, and adaptively updating according to probability triggering; S23, introducing Levy flight in the position perturbation stage. An enhanced oviparous perturbation strategy is implemented by embedding the Levy flight step generation process into the oviparous burst mechanism to construct an oviparous burst displacement oriented towards the historical optimal position offset relationship; S24, a neighborhood electric field-guided perturbation strategy is introduced in the neighborhood guidance stage, which identifies the set of superior neighbors whose cost function values ​​are better than the current individual, and constructs electric field-guided perturbations based on the cost difference and positional relationship between the superior neighbors and the current individual; S25, a stagnation detection and restart strategy is introduced in the iterative evolution stage, which updates the stagnation count by detecting whether the historical optimal solution has significantly improved, and performs random restarts on some individuals with poor fitness when the stagnation threshold is reached; S3, an improved breakthrough platypus optimization algorithm is used to optimize the parameters of the inlet air temperature PID controller to obtain the optimal PID control parameters; S4, the obtained optimal PID control parameters are set as the control parameters of the inlet air temperature PID controller module to achieve optimized control of the air preheater inlet air temperature.

[0005] Furthermore, in the air preheater inlet air temperature PID control system constructed in step S1, the actual air temperature feedback value is obtained by the inlet air temperature monitoring module and transmitted to the inlet air temperature error calculation module. The inlet air temperature error calculation module receives the target temperature value and compares the target temperature with the feedback value to calculate and output a real-time error signal, which is then transmitted to the inlet air temperature PID controller module. The improved breakthrough platypus optimization algorithm module is used to optimize the PID control parameters of the inlet air temperature PID controller module. The inlet air temperature PID controller module with optimized parameters calculates the control quantity based on the real-time error signal and outputs the control quantity to the inlet air temperature regulation module to regulate the air preheater inlet air temperature.

[0006] Furthermore, the enhanced population initialization in step S21 includes the following steps: Step S211, generating an initial chaotic variable vector. The components of the initial chaotic variable vector are taken from the sampling interval. and avoid special point sets Then, following the successive evolution of the Logistic chaotic mapping, the state variable evolution formula is: In the formula, Indicates the first The variable vector after the second chaotic evolution. Indicates the first The variable vector after the second chaotic evolution. This represents the control parameters for the Logistic mapping. This represents element-wise multiplication; Step S212: Map the chaotic variables obtained from each evolution to the search space to form the initial chaotic population. The mapping formula is: In the formula, Indicates the first The position vector of a chaotic initial individual. Represents the lower boundary vector of the search space. Represents the boundary vector in the search space. Indicates the first The variable vector obtained from the second chaotic evolution; Step S213: Perform reverse learning on the initial chaotic population to generate a reverse population. The reverse learning formula is: In the formula, Indicates the first The position vectors of the reverse-learned individuals; Step S214: Merge the initial chaotic population and the reverse-learned population into a total candidate population, and perform boundary repair on each individual in the total candidate population. The boundary repair formula is: In the formula, Indicates the first candidate species in the total candidate population. The position vector of each individual; Step S215: Calculate the cost function value of each individual in the total candidate population. When an anomaly occurs in the cost function calculation, assign the corresponding cost function value as a penalty value. Step S216: Sort the total candidate population in ascending order of cost function value, and select the population with the smallest cost function value. Each individual is used as the initial population, and the initial historical global optimal position is determined accordingly. and the corresponding historical best fitness value .

[0007] Furthermore, the differential evolution parameter adaptive update strategy in step S22 includes the following steps: Step S221, assigning an independent differential variable asynchronous long factor to each individual during the initialization phase. With cross probability Step S222: During the search evolution process, with a 10% probability, the first... The difference-variable asynchronous long factor of each individual is updated, and the update formula is: In the formula, This represents the updated difference-variable long factor. The fundamental parameter representing the differential variable asynchronous long factor is... The span parameter represents the differential variable asynchronous long factor. This represents an independent random number that follows a uniform distribution on the closed interval [0, 1]; when the trigger condition is not met, Keep the previous state unchanged; Step S223: During the search evolution process, with a probability of 10%, adjust the first... The crossover probability of each individual is updated using the following formula: In the formula, This represents the crossover probability after the i-th individual is updated. This represents an independent random number that follows a uniform distribution on the closed interval [0, 1]; when the trigger condition is not met, Keep the previous state unchanged; Step S224: In the land search mode, utilize the updated differential variable asynchronous long factor. Perform differential mutation to generate a mutation vector. The calculation formula is as follows: In the formula, Indicates the first The difference mutation vector corresponding to each individual , , This represents three distinct indices randomly selected from the population, none of which are equal to the current individual's index. The backup individual location; Step S225: Perform a crossover operation on the mutation vector to generate candidate pattern locations, and ensure that at least one dimension comes from the mutation vector through a forced dimension mechanism. The crossover formula is: In the formula, Indicates the first The individual in the first Candidate mode components on the dimension, The difference mutation vector is represented at the th Components on the dimension, Indicates the first The interval used in dimensional cross-determination Uniformly distributed independent random numbers This indicates a randomly selected mandatory inheritance dimension index. Indicates the first The individual in the first Backup location components on the dimension.

[0008] Furthermore, the Lévy flight-enhanced oviparous perturbation strategy in step S23 includes the following steps: Step S231, first initialize the Lévy oviparous displacement to a zero vector, and generate burst determination random numbers. When satisfied At that time, the oviposition burst mechanism is triggered, in which, Indicates the oviparous burst trigger threshold; Step S232: After triggering the oviparous burst mechanism, set the Levy flight stability distribution index parameter. And calculate the scale parameter in the Mantegna method. The calculation formula is: In the formula, Represents random variables standard deviation Represents the Gamma function. The parameters represent the Lévy flight stability distribution index; step S233: generate the Lévy flight step size vector based on the Mantegna method, calculated using the following formula: In the formula, Indicates the first The Lévy flight step size vector for each individual. This indicates that the expression follows a pattern with a mean of 0 and a variance of . A normally distributed random vector, Let represent a random vector that follows a standard normal distribution. This indicates element-wise absolute value operation; Step S234: Based on the Levy flight step size vector and the offset of the current individual relative to the historical best position, construct the final Levy oviparous displacement, calculated using the following formula: In the formula, Indicates the first The displacement vector of the oviparous burst of each individual. This represents the displacement amplitude parameter of the oviparous burst. Indicates the first Backup locations for each individual instance This indicates the historical global optimal position; when the oviparous burst mechanism is not triggered. Keep it as a zero vector.

[0009] Furthermore, the neighborhood electric field-guided perturbation strategy in step S24 includes the following steps: Step S241, identifying individuals in the population with cost function values ​​superior to the current individual. For all individuals, construct a set of best neighbor indexes, calculated using the following formula: In the formula, Indicates the first A set of best neighbor indexes for each individual. Indicates the first The cost function value of each backup individual. Indicates the first The cost function value of each backup individual; Step S242: If the set of best neighbor indices is not empty, calculate the distance between the current individual and each best neighbor. The calculation formula is: In the formula, Indicates the first Individual and the first Euclidean distance between each good neighbor Represents the L2 norm, This indicates a small positive quantity that prevents the denominator from being zero; Step S243: Calculate the guiding weight based on the relationship between the cost function value difference and the distance. The calculation formula is as follows: In the formula, Indicates the guiding weight. Indicates the first The cost function value of each backup individual. Show the first The cost function value of each backup individual. Indicates the first Individual and the first The Euclidean distance between each good neighbor; Step S244: Accumulate and average the guiding effects of all good neighbors to form the final electric field guided perturbation, calculated by the following formula: In the formula, Indicates the first The final electric field-guided perturbation vector of each individual, This indicates the intensity of the disturbance guided by the electric field in the neighborhood. This indicates the number of individuals in the best neighbor index set; when the best neighbor index set is empty... Keep it as a zero vector.

[0010] Furthermore, the stagnation detection and restart strategy in step S25 includes the following steps: Step S251, Initialize the stagnation counter that has not shown significant improvement continuously. And set the maximum number of consecutive stagnation algebras. Step S252: After each generation of search, extract the optimal fitness of the current formal population. The calculation formula is as follows: In the formula, This represents the optimal fitness value in the current generation of the formal population. Indicates the first The fitness value of a formal individual. Indicates the population size; Step S253: If the significant improvement condition is met, update the historical best solution and reset the stagnation counter to zero; if the significant improvement condition is not met, increment the stagnation counter. The significant improvement condition is: In the formula, This represents the historical best fitness value. This indicates a significant improvement in the judgment threshold; Step S254: When the stall counter reaches the maximum number of consecutive stall generations, restart according to the restart ratio. The number of individuals to be restarted is determined by the following formula: In the formula, Indicates the number of individuals restarted. This represents the floor function. Indicates the restart ratio; Step S255: Select the population with the worst fitness... Each individual is re-randomly initialized into the search space, and the calculation formula is: In the formula, Indicates the first A new position for each restarted individual. This represents a random vector that follows a uniform distribution on the closed interval [0, 1] for each dimension; Step S256: Recalculate the fitness of the restarted individual. If an anomaly occurs in the fitness calculation, assign the corresponding fitness value as a penalty value. And reset the stall counter to zero.

[0011] Furthermore, step S3, which utilizes the improved breakthrough platypus optimization algorithm to optimize the PID control parameters for the inlet air temperature, includes the following steps: Step S801: Perform parameter initialization and storage pre-allocation, setting the basic control parameters, adaptive update parameters, stagnation detection parameters, and convergence information storage space; Step S802: Utilize chaotic mapping and reverse learning strategies to perform enhanced population initialization, obtaining the formal initial population and historical global optimal solutions through chaotic variable evolution, search space mapping, reverse individual generation, candidate set merging, boundary repair, and fitness screening; Step S803: Enter the main iteration. The process iterates and performs dynamic weight scheduling, generating global and local search weights based on the current iteration progress, and backing up the current formal population and its fitness. Step S804 involves adaptive parameter updates for each individual, updating the differential variable asynchronous long factor and crossover probability using a probability-triggered method to provide individual-level adaptive control parameters for subsequent land mode searches. Step S805 involves executing an adaptive dual-mode search, switching between underwater and land modes based on the mode determination results. The underwater mode generates search results with local perturbations oriented towards historical optimal positions, while the land mode generates results based on differential... Candidate mode positions for mutation and crossover operations; Step S806: Execute Levy-enhanced oviparous perturbation, generating oviparous burst displacements in a probabilistic triggering manner, and constructing large-span random perturbations by combining the relative offset between the current individual and the historical best position; Step S807: Execute neighborhood electric field-guided perturbation, by identifying a set of superior neighbors that are better than the current individual, calculating the neighbor-guided weights and accumulating them to form electric field-guided perturbations, thereby enhancing the individual's ability to migrate to superior regions; Step S808: Execute integrated position update, boundary repair, and greedy selection, coordinating the dual-mode search results, Levy oviparous displacements, and neighborhood electric field perturbations. The same assembly is used to obtain new positions, and after boundary repair and fitness calculation, the formal population is updated by a strict greedy criterion; Step S809: Execute the global optimal update, stagnation detection and restart mechanism. When the historical optimal solution is detected to have not improved significantly, the stagnation count is accumulated, and after reaching the stagnation threshold, the individuals with the worst fitness are randomly restarted to restore population diversity; Step S810: Record convergence information and determine the termination condition. In each generation, the historical global optimal fitness and historical global optimal position are recorded, and after reaching the maximum number of iterations, the optimal solution, optimal fitness and corresponding convergence process information are output.

[0012] The beneficial effects of this invention due to the adoption of the above-mentioned technology are as follows.

[0013] This invention addresses the challenges of precise inlet air temperature control and the potential for acid dew point corrosion and heating surface blockage caused by the severe load fluctuations and large hysteresis characteristics of air preheaters in thermal power generation systems. It introduces an improved, groundbreaking platypus optimization algorithm to optimize and tune PID control parameters. By integrating chaotic mapping and back-learning strategies during the initialization phase, the algorithm significantly improves the distribution quality and exploration potential of the initial population within the complex parameter space. Furthermore, by utilizing a differential evolution parameter adaptive update strategy and a neighborhood electric field-guided perturbation mechanism, the algorithm enhances its parameter adjustment flexibility and orientation towards advantageous regions during the evolution process. By embedding the Lévy flight-enhanced oviposition perturbation strategy and the stagnation detection and restart mechanism, the stagnation phenomenon in the search process is effectively overcome, the algorithm's ability to escape local extrema is enhanced, and the optimal combination of PID control parameters can be found in the full range of operating conditions. This enables the air preheater inlet air temperature PID control system to exhibit shorter settling time, smaller overshoot, and extremely high steady-state accuracy when facing drastic changes in unit load or sudden changes in ambient temperature. It can effectively reduce the risk of corrosion and ash accumulation in the low-temperature section of the air preheater, and has high application value for ensuring the safe operation of the thermal power generation system and improving the level of refined operation. Attached Figure Description

[0014] Figure 1 is a schematic diagram of the process of the present invention.

[0015] Figure 2 is a schematic diagram of the process of tuning the PID control parameters for inlet air temperature using the improved breakthrough platypus optimization algorithm of the present invention.

[0016] Figure 3 is a comparison of the optimal fitness convergence curves of the breakthrough platypus optimization algorithm of this invention and the improved breakthrough platypus optimization algorithm.

[0017] Figure 4 is a comparison of the system response curves of the breakthrough platypus optimization algorithm of this invention and the improved breakthrough platypus optimization algorithm.

[0018] Figure 5 is a comparison of the PID control parameter Kp optimization curves of the breakthrough platypus optimization algorithm of the present invention and the improved breakthrough platypus optimization algorithm.

[0019] Figure 6 is a comparison of the PID control parameter Ki optimization curves of the breakthrough platypus optimization algorithm of the present invention and the improved breakthrough platypus optimization algorithm.

[0020] Figure 7 is a comparison of the PID control parameter Kd optimization curves of the breakthrough platypus optimization algorithm of the present invention and the improved breakthrough platypus optimization algorithm. Detailed Implementation

[0021] Example 1: As shown in Figure 1, this invention provides an optimization method for air preheater temperature control in thermal power generation, including the following steps: S1, constructing an air preheater inlet air temperature PID control system, including an inlet air temperature error calculation module, an inlet air temperature PID controller module, an improved breakthrough platypus optimization algorithm module, an inlet air temperature adjustment module, and an inlet air temperature monitoring module. In the air preheater inlet air temperature PID control system constructed in step S1, the inlet air temperature monitoring module obtains the actual air temperature feedback value and transmits it to the inlet air temperature error calculation module. The inlet air temperature error calculation module receives the target temperature value and compares the target temperature with the feedback value to calculate and output the actual air temperature. The real-time error signal is transmitted to the inlet air temperature PID controller module. An improved breakthrough platypus optimization algorithm module optimizes the PID control parameters of the inlet air temperature PID controller module. The optimized inlet air temperature PID controller module calculates the control quantity based on the real-time error signal and outputs this control quantity to the inlet air temperature regulation module to regulate the inlet air temperature of the air preheater. S2. An improved breakthrough platypus optimization algorithm is introduced, including at least the following improvement strategies: S21. An enhanced population initialization strategy is introduced in the initialization stage. Candidate individuals are generated through a combination of chaotic mapping and reverse learning, and formal initial individuals are obtained after boundary repair, fitness evaluation, and screening. The population is improved to enhance the distribution quality of the initial population in the search space; S22, a differential evolution parameter adaptive update strategy is introduced in the search evolution stage. By setting differential variable asynchronous length factors and crossover probabilities for each individual, and performing adaptive updates according to probability triggering, the parameter adjustment capability during the search process is improved; S23, a Lévy flight-enhanced oviparous perturbation strategy is introduced in the position perturbation stage. By embedding the Lévy flight step size generation process into the oviparous burst mechanism, an oviparous burst displacement oriented towards the historical optimal position offset relationship is constructed to enhance the ability to escape local optimal regions; S24, a neighborhood electric field-guided perturbation strategy is introduced in the neighborhood guidance stage. By identifying the set of superior neighbors whose cost function values ​​are better than the current individual, and based on... The cost difference and positional relationship between the best neighbor and the current individual are used to construct an electric field-guided perturbation to improve the population's ability to migrate to the dominant region; S25, a stagnation detection and restart strategy is introduced in the iterative evolution stage. The stagnation count is updated by detecting whether the historical best solution has been significantly improved, and random restarts are performed on some individuals with poor fitness when the stagnation threshold is reached to restore population diversity and improve search ability in the stagnation state; S3, the improved breakthrough platypus optimization algorithm is used to optimize the parameters of the inlet air temperature PID controller. In the iterative process, dynamic weight scheduling, dual-mode search, position comprehensive update, boundary repair, greedy selection, global optimal update and termination judgment are combined to obtain the optimal PID control parameters;S4. Set the optimal proportional, integral, and derivative parameters obtained through optimization as the control parameters of the inlet air temperature PID controller to achieve optimized control of the air preheater inlet air temperature.

[0022] As shown in Figure 2, step S3, which uses the improved breakthrough platypus optimization algorithm to optimize the PID control parameters for the inlet air temperature, includes the following steps: Step S801: Perform parameter initialization and storage pre-allocation, setting the basic control parameters, adaptive update parameters, stagnation detection parameters, and convergence information storage space; Step S802: Perform enhanced population initialization using chaotic mapping and reverse learning strategies, obtaining the formal initial population and historical global optimal solutions through chaotic variable evolution, search space mapping, reverse individual generation, candidate set merging, boundary repair, and fitness screening; Step S803: ... Step S804 involves initiating an iterative loop and executing dynamic weight scheduling. Global and local search weights are generated based on the current iteration progress, and the current formal population and its fitness are backed up. Step S805 involves adaptively updating parameters for each individual, updating the differential variable asynchronous long factor and crossover probability using a probability-triggered method to provide individual-level adaptive control parameters for subsequent land mode searches. Step S806 involves executing an adaptive dual-mode search, switching between underwater and land modes based on the mode determination results. The underwater mode generates search results with local perturbations oriented towards historical optimal positions, while the land mode generates results based on... Candidate mode positions for differential mutation and crossover operations; Step S806: Execute Lévy flight enhanced oviposition perturbation, generating oviposition burst displacement in a probabilistic triggering manner, and constructing a large-span random perturbation by combining the relative offset between the current individual and the historical best position; Step S807: Execute neighborhood electric field guided perturbation, by identifying a set of superior neighbors that are better than the current individual, calculating the neighbor guided weight, and accumulating it to form an electric field guided perturbation, so as to enhance the individual's ability to migrate to high-quality areas; Step S808: Execute integrated position update, boundary repair, and greedy selection, combining the dual-mode search results, Lévy oviposition displacement, and neighborhood electric field perturbation. Collaborative assembly yields new positions, and after boundary repair and fitness calculation, the formal population is updated using a strict greedy criterion. Step S809: Execute a global optimal update, stagnation detection, and restart mechanism. When no significant improvement is detected in the historical optimal solution, accumulate stagnation counts. After reaching the stagnation threshold, randomly restart the individuals with the worst fitness to restore population diversity. Step S810: Record convergence information and determine termination conditions. Record the historical global optimal fitness and historical global optimal position in each generation. After reaching the maximum number of iterations, output the optimal solution, optimal fitness, and corresponding convergence process information.

[0023] The transfer function of the controlled object in the PID control system for air preheater inlet air temperature is: In the formula, This indicates that the controlled object passes a function. Represents the physical gain of the object. Represents the object's inertial time constant. Indicates the pure time delay of the object input. This represents the Laplace operator, and after system identification, the corresponding specific transfer function is: The unity negative feedback closed-loop transfer function is: In the formula, Represents the closed-loop transfer function. This represents the transfer function of the PID controller. This indicates that the controlled object passes a function.

[0024] In the PID control system for air preheater inlet air temperature, the step amplitude is defined as: In the formula, Indicates the step amplitude. Indicates the target output. This represents the initial output of the system, where , , The values ​​were taken as 80, 95, and 15 in sequence; the simulation time series is as follows: In the formula, Represents the simulated time series. Indicates the sampling step size. Indicates the simulation termination time, where , The values ​​are taken as 1 and 800 respectively; the formula for judging the divergence of the closed-loop response is: In the formula, Indicates the response divergence threshold. Indicates the target output. This represents the initial output of the system and sets the maximum allowable absolute amplitude threshold for the closed-loop response. Normalization error is defined as: In the formula, Indicates the normalization error. Indicates time The system output, Indicates the target output. This represents the step amplitude; the absolute error index for integral time is: In the formula, Indicates the absolute error index of integration time. Represents a time variable. Indicates the normalization error. Indicates the simulation termination time; the overshoot ratio is defined as: In the formula, Indicates the overshoot ratio. This represents the maximum output value during the response process. Indicates the target output. Indicates the step amplitude; the overshoot / overlimit penalty term is: In the formula, This indicates the penalty for overshooting or exceeding limits. Indicates the overshoot ratio. This indicates the maximum allowable overshoot ratio, set to 0.05; the penalty for exceeding the settling time limit is: In the formula, This indicates the penalty for exceeding the adjustment time limit. Indicates the actual adjustment time. This indicates the maximum allowable adjustment time, while the setting... =0.02, =400; The control energy proxy item is: In the formula, Indicates the control energy proxy item, Represents the proportionality coefficient. Represents the integral coefficient. Represents the differential coefficients; in summary, the corresponding constructed fitness function is: In the formula, This represents the fitness function value. This indicates the weight of the absolute error index for integration time. This indicates the weighting of overshoot and settling time penalties. This indicates the penalty for overshooting or exceeding limits. This indicates the penalty for exceeding the adjustment time limit. This indicates the weight of the control energy proxy term. Indicates the control energy proxy item, , , The fitness function takes values ​​of 1, 100, and 0.5 respectively. When the PID parameters exceed the search boundary, violate physical parameter coupling constraints, the closed-loop response diverges, or simulation calculations show abnormalities, the fitness function takes the following values: In the formula, This represents an infinite penalty value.

[0025] The frequency domain transfer function of the PID controller for the air preheater inlet air temperature is: In the formula, This represents the transfer function of the PID controller. Indicates the controller direction symbol. This represents the proportionality coefficient. Represents the integral coefficient. Represents the differential coefficient. Let represent the Laplace operator, and 0.02 represent the filter coefficients; the frequency domain control law is: In the formula, This indicates the controller output signal. This represents the transfer function of the PID controller. The Laplace transform of the error signal is represented; after substituting it into the controller expression, the frequency domain control law is written as: In the formula, This indicates the controller output signal. Indicates the controller direction symbol. This represents the proportionality coefficient. Represents the integral coefficient. Represents the differential coefficient. This represents the Laplace transform of the error signal; and the error signal is defined as: In the formula, Indicates time The control error signal, Indicates the target output. Indicates time The system output; the time-domain control law of the PID controller for the air preheater inlet air temperature is: In the formula, Indicates time The controller output, Indicates the controller direction symbol. This represents the proportionality coefficient. This indicates the control error signal. Indicates the term of action of integration. Represents the differential component of the filter. Represents the integral variable; the filtered differential components satisfy: In the formula, 0.02 represents the filter coefficient. Represents the differential component of the filter. Represents the differential coefficient. It represents the derivative of the error signal with respect to time.

[0026] The improved breakthrough platypus optimization algorithm is used to optimize the PID control parameters for inlet air temperature. The specific steps include: Step 1, performing parameter initialization and storage pre-allocation; Step 11, setting the basic control parameters, including the dynamic weighting index. =2.0, Levi's oviparous burst trigger probability =0.15, amplitude base of oviparous burst displacement =0.1, intensity of disturbance guided by the neighboring electric field =0.5 and the switching threshold between underwater mode and land mode =0.5, Population size N=30; Step 12, Initialize the adaptive parameters of the probability-triggered adaptive update mechanism for control parameters; Set the first... The difference-variable asynchronous length factor of each individual =0.5, No. Crossover probability of individuals =0.9, population size =30; Step 13: Initialize the stall detection counter and the maximum stall threshold; Set the number of iterations that do not show significant improvement. =0, the maximum allowed number of consecutive stagnant algebras =15; Step 14: Pre-allocate storage space for convergence curve and historical optimal solution; ,and In the formula, Indicates length is The convergence curve vector, Indicates size is The historical optimal solution trajectory matrix, This represents the maximum number of iterations and has a value of 100. This represents the dimension of the problem and has a value of 3.

[0027] Step 2: Perform enhanced population initialization using chaotic mapping and reverse learning strategies; Step 21: Generate initial chaotic variable vectors. Each of its components is composed of random numbers in the interval (0, 1) that are not equal to 0.25, 0.5, or 0.75 to avoid mapping fixed points, and evolves successively according to the Logistic mapping. In the formula, Indicates the first The variable vector after the second chaotic evolution. Represents the variable vector after the next chaotic evolution, and constants. This represents the control parameters for the Logistic mapping. This represents element-wise multiplication; step 22: Map the chaotic variables obtained from each evolution to the search space to form the initial chaotic population. , In the formula, Indicates the first The position vector of a chaotic initial individual. This represents the lower boundary vector of the search space, with values ​​in the range [0.0, 0.0, 0.0]. Let represent the boundary vector of the search space and take the value [1.568, 1.254, 0.156]. Indicates the first The variable vector obtained from the second chaotic evolution. Indicates the population size and takes the value 30; Step 23: Perform reverse learning on the chaotic initial population to generate a reverse population; , In the formula, Indicates the first The position vector of a backward-learned individual. Represents the lower boundary vector of the search space. Represents the boundary vector in the search space. Indicates the first The position vectors of the initial chaotic individuals; step 24: merge the chaotic population and the reverse population into a group of size... The candidate set; In the formula, This represents the total candidate population formed by row-wise concatenation of the chaotic population and the reverse population. Represents the initial chaotic population. Represents the reverse learning population, symbol This indicates that the data is concatenated row by row; step 25: Perform boundary repair on each individual in the overall candidate population; , In the formula, Indicates the first candidate species in the total candidate population. The position vector of each individual Represents the boundary vector in the search space. Represents the lower boundary vector of the search space. This represents the element-wise minimization operator, used to ensure that the individual position does not exceed the upper boundary vector of the search space. , This represents the element-wise maximization operator, used to ensure that the individual position is not lower than the lower boundary vector of the search space. Step 26: Calculation If an anomaly occurs during the calculation of the cost function value of each candidate individual, it will be assigned a large penalty value. In the formula, Indicates the first candidate species in the total candidate population. The cost function value for each individual. This represents a large penalty value used when the cost function calculation fails; it is only used in the sorting and selection phases, and is taken as... Step 27: Sort the total candidate population in ascending order of fitness, and select the top candidates with the smallest error. Each individual is used as the initial population, and the initial historical global optimal position is determined accordingly. and the corresponding historical best fitness value .

[0028] Step 3: Enter the main iteration loop and execute dynamic weight scheduling; Step 31: In the... Calculate the global search weight; In the formula, Indicates the first Global search weight of the generation Indicates the current iteration number. Indicates the maximum number of iterations. Indicates the dynamic weight index; Step 32: Calculate the local search weight; In the formula, Indicates the first Local search weight of the generation Indicates the first The global search weights of the generation; step 33, back up the current formal population and its cost function values, and obtain and It is used for generating candidate solutions and calculating perturbations for the current generation.

[0029] Step 4, for the first Each individual performs adaptive updates of differential evolution parameters; step 41, updates the differential variable asynchronous long factor with a 10% probability. In the formula, Indicates the first The updated difference transforms into an asynchronous long factor for each individual. Indicates that it conforms to the interval The update is performed only when the random trigger condition is met, otherwise, it is executed using uniformly distributed random numbers. Keep the previous state unchanged; step 42, update the crossover probability with a 10% probability. In the formula, Indicates the first The crossover probability after updating each individual Indicates that it conforms to the interval The update is performed only if the random trigger condition is met, otherwise, a uniformly distributed random number is used. Keep the previous state unchanged.

[0030] Step 5: Perform adaptive dual-mode search and generate candidate mode positions; Step 51: Generate random numbers for mode determination. If satisfied Then, perform a local normal perturbation search in underwater mode. In the formula, Indicates the first Candidate locations generated for each individual in underwater mode Indicates the first Backup locations for each individual instance Let represent a random vector that follows a standard normal distribution. This represents the historical global optimal position, and the constant 0.5 represents the local perturbation scaling factor; step 52, if the following conditions are met... Then, the random baseline difference mutation strategy in the land mode is executed. In the formula, Indicates the first The difference mutation vector corresponding to each individual , , This represents three distinct indices randomly selected from the population, none of which are equal to the current individual's index. Backup individual locations, This represents the updated differential variable asynchronous long factor; Step 53: Perform binary crossover on the mutation vector of the land pattern to generate candidate pattern positions, and ensure that at least one dimension comes from the mutation vector through a forced dimension mechanism. In the formula, Indicates the first The individual in the first Candidate mode components on the dimension, The difference mutation vector is represented at the th Components on the dimension, Indicates the first The interval used in dimensional cross-determination Uniform random numbers, Indicates the first The crossover probability of each individual, Indicates in arrive A dimension index randomly selected from among them. Indicates the first The individual in the first Backup location components on the dimension.

[0031] Step 6: Execute the Levy flight enhanced oviposition perturbation; Step 61: First initialize the oviposition displacement to a zero vector, generating an interval... Uniformly distributed burst determination random numbers As a trigger condition for the burst, if the following conditions are met... This triggers the burst mechanism and uses the Mantegna method to generate the Levy step size; ,in, The exponential parameter representing the Lévy flight stability distribution; In the formula, Represents the random variable in the Mantegna method standard deviation Represents the Gamma function. Represents pi (π). The exponential parameter representing the Lévy flight stability distribution; In the formula, Indicates the first The Lévy flight step size vector for each individual. Indicates that it follows the mean. variance is A normally distributed random vector, Let represent a random vector that follows a standard normal distribution. The exponential parameter representing the stable distribution of Lévy flight. This indicates element-wise absolute value operation; Step 62: Based on the large span characteristic of the Levy step size and the offset of the current individual relative to the optimal position, construct the oviparous displacement; In the formula, Indicates the first The displacement vector of the oviparous burst of each individual. The amplitude base of the oviparous burst displacement. Indicates the first The Lévy flight step size vector for each individual. Indicates the first Backup locations for each individual instance This represents the historical global best position up to the current moment. This indicates element-wise multiplication; if the burst mechanism is not triggered, then... Keep it as a zero vector.

[0032] Step 7: Implement the improved neighborhood electric field guided perturbation; In the formula, Indicates the first A set of best neighbor indexes for each individual. Indicates the first The cost function value of each backup individual. Indicates the first The cost function value of the backup individual; step 72, if the set of best neighbors is not empty, then calculate the cost function value of the backup individual; The distance of each individual relative to each of its best neighbors In the formula, Indicates the first Individual and the first Euclidean distance between each good neighbor Represents the L2 norm, and They represent the first The and the first The location vector of each backup individual This indicates the smallest positive number that prevents the denominator from being zero; Step 73, calculate the... Individuals are affected by the first The guiding weight when a good neighbor plays a role. In the formula, Indicates the guiding weight. Indicates the first The cost function value of each backup individual. Indicates the first The cost function value of each backup individual. Indicates the first Individual and the first Step 74: Accumulate and average the guiding effects of all good neighbors to form the final electric field perturbation. In the formula, Indicates the first The final electric field-guided perturbation vector of each individual, This indicates the intensity of the disturbance guided by the electric field in the neighborhood. This represents the number of individuals in the set of best neighbors. Indicates the first The best neighbors to the first The guiding weight of each individual, This represents the direction vector from the current individual to the best neighbor. If the set of best neighbors is empty, then... Keep it as a zero vector.

[0033] Step 8: Perform comprehensive position update, boundary repair, and greedy selection; Step 81: After completion , and After the calculation, generate random numbers independently again. Used for comprehensive weight update branch determination, this determination is independent of the pattern determination in step 5; step 82, if satisfied ,and If the weight is 0.5, then the new positions are assembled according to the global weight. In the formula, Indicates the first The new position after the individual units are fully assembled Indicates the first Backup locations for each individual instance This indicates the global search weight of the current generation. Indicates the position of the candidate pattern generated by the dual-mode search. Indicates the first The displacement vector of the oviparous burst of each individual. Indicates the first The final electric field-guided perturbation vector for each individual; step 83, if the following conditions are met... ,and If the weight is 0.5, then the new position is assembled according to the local weight. In the formula, Indicates the first The new position after the individual units are fully assembled Indicates the first Backup locations for each individual instance This indicates the local search weight of the current generation. Indicates the position of the candidate pattern generated by the dual-mode search. Indicates the first The displacement vector of the oviparous burst of each individual. Indicates the first The final electric field-guided perturbation vector for each individual; step 84, perform reflection repair on the dimensions exceeding the upper boundary in the new location, that is, when hour, In the formula, Indicates the repaired number The individual in the first Positional components in the dimension, Indicates the first The upper boundary of the dimension, Indicates compliance with the interval Uniformly distributed independent random numbers Indicates the first The individual in the first Backup location components on the dimension; step 85, perform reflection repair on dimensions below the lower boundary in the new location, that is, when hour, In the formula, Indicates the repaired number The individual in the first Positional components in the dimension, Indicates the first The lower boundary of the dimension, Indicates compliance with the interval Uniformly distributed independent random numbers Indicates the first The individual in the first Backup location components on the dimension; Step 86: Calculate the fitness of the new location. If an anomaly occurs during fitness calculation, assign it a large penalty value. In the formula, This represents the cost function value corresponding to the new position. This indicates the large penalty value used when the cost function calculation fails; step 87: Update the population using a strictly greedy selection method. In the formula, Indicates the updated position. Indicates the new location after repair. This indicates the backup location of the i-th individual. This represents the cost function value at the new location. This indicates the first [number] in the official population before the update. The cost function value of each individual; step 88: if the new position is accepted, then update its fitness value synchronously; otherwise, retain the original fitness value.

[0034] Step 9: Perform global optimal update, stall detection, and restart; Step 91: Extract the optimal fitness in the current official population; In the formula, Indicates the current number The optimal fitness value in the formal population, Indicates the first The fitness value of a formal individual. Indicates the population size; Step 92: If the significant improvement condition is met, update the historical best and clear the stall counter; In the formula, This represents the optimal fitness value of the current generation. This represents the historical best fitness value. This represents the threshold for significant improvement; step 93: if the condition for significant improvement is not met, then let... Step 94: When the stall counter reaches the maximum allowed stall generation, a restart mechanism is triggered. In the formula, Indicates the number of individuals restarted. This represents the floor function. Indicates population size; step 95, select the worst-case scenario... Each individual is re-randomly initialized within the search space; , In the formula, Indicates the first A new position for each restarted individual. Represents the lower boundary vector of the search space. Represents the boundary vector in the search space. Let represent a random vector whose dimensions are uniformly distributed on the closed interval [0, 1]. This represents element-wise multiplication. Indicates the number of individuals restarted; step 96: Recalculate the fitness of the restarted individuals, and assign a value if an anomaly occurs in the calculation. Then the stall counter is reset to zero.

[0035] Step 10: Record convergence information and determine termination conditions; Step 101: Record the historical global optimal fitness of the current generation; In the formula, Indicates the first The convergence curve records the values ​​of the generation. Indicates the historical best fitness value; step 102: Record the historical global best position of the current generation; In the formula, Indicates the first The historical global optimal position vector is recorded. This represents the historical global optimal solution up to the current generation. Indicates the current iteration count; step 103, when the iteration count reaches the maximum iteration count. When the algorithm terminates, the optimal solution is output. , , and If the maximum number of iterations has not been reached, proceed to the next iteration.

[0036] By implementing the Breakthrough Platypus Optimization Algorithm and its improved version in a Matlab simulation environment to optimize the PID control parameters of the controlled object, the results are shown in Figures 3 to 7. As shown in Figure 3, in the comparison of optimal fitness convergence characteristics, the improved Breakthrough Platypus Optimization Algorithm (solid line) exhibits significant evolutionary pressure and global search efficiency, achieving a sharp decrease in fitness value in the early stages of the first 10 iterations, and finally converging to a final fitness value of approximately 758.30 after about 70 iterations. In contrast, the Breakthrough Platypus Optimization Algorithm (dashed line) shows a clear stepwise convergence characteristic, reflecting that the original algorithm is prone to getting trapped in local optima, exhibiting significant search stagnation, and ultimately achieving a convergence value of only 9. 89.51; This shows that the improved breakthrough platypus optimization algorithm can produce a lower fitness value, effectively proving its stronger optimization accuracy when dealing with high-dimensional nonlinear search spaces; As shown in Figure 4, in terms of settling time, the system driven by the improved breakthrough platypus optimization algorithm can enter steady state relatively faster; In terms of overshoot, the improved breakthrough platypus optimization algorithm effectively suppresses overshoot in the initial stage of response, resulting in a smoother waveform without obvious oscillations; In terms of steady-state performance, both the breakthrough platypus optimization algorithm and the improved breakthrough platypus optimization algorithm can eventually stabilize the system output at the target value of 95, but the improved breakthrough platypus optimization algorithm has a significant advantage in quality control during the transition process; As shown in Figures 5, 6, and 7, In the optimization process, the improved breakthrough platypus optimization algorithm maintained a large exploration range and eventually stabilized at around 1.5381. During the optimization process, although both the breakthrough platypus optimization algorithm and the improved breakthrough platypus optimization algorithm eventually converged to 0.0190, the improved breakthrough platypus optimization algorithm had a larger jump amplitude in the early stages of iteration, which verified its stronger walking ability in the parameter space. In the optimization process, the improved and groundbreaking platypus optimization algorithm will... The parameter value was pushed to a higher value, approximately 0.1495, which generally helps improve the response of the system's differential element, explaining why it achieves smaller overshoot and faster adjustment speed. In the later stages of iterations after the 80th iteration, the trajectories of all three sets of parameters tended to be horizontal straight lines without fluctuations, confirming that this optimized system has high numerical stability. In summary, the improved breakthrough platypus optimization algorithm is significantly superior to the breakthrough platypus optimization algorithm in terms of convergence accuracy, global optimization capability, and dynamic response quality of the closed-loop system. Although the improved algorithm slightly increases the execution time due to the introduction of a more complex search mechanism, from 19.63 seconds for the breakthrough platypus optimization algorithm to 22.57 seconds for the improved breakthrough platypus optimization algorithm, the control accuracy gain brought by the improved breakthrough platypus optimization algorithm completely offsets the time cost. The optimal control parameters finally tuned to 1.5381, 0.0190, and 0.1495 enable relatively better control quality.

Claims

1. A method for optimizing air preheater temperature control in thermal power generation, characterized in that: The process includes the following steps: S1. Constructing a PID control system for the air preheater inlet air temperature, including an inlet air temperature error calculation module, an inlet air temperature PID controller module, an improved breakthrough platypus optimization algorithm module, an inlet air temperature regulation module, and an inlet air temperature monitoring module; S2. Introducing an improved breakthrough platypus optimization algorithm, including at least the following improvement strategies: S21. In the initialization phase, an enhanced population initialization strategy is introduced, generating candidate individuals through a combination of chaotic mapping and back learning, and obtaining the formal initial population after boundary repair, fitness evaluation, and screening; S22. In the search evolution... In the first stage, a differential evolution parameter adaptive update strategy is introduced. This involves setting a differential variable asynchronous length factor and crossover probability for each individual and then adaptively updating the parameters according to a probability-triggered method. In the second stage, a Levy flight-enhanced oviparous perturbation strategy is introduced. This involves embedding the Levy flight step size generation process into the oviparous burst mechanism to construct an oviparous burst displacement oriented towards the historical optimal position offset relationship. In the third stage, a neighborhood electric field-guided perturbation strategy is introduced. This involves identifying a set of superior neighbors whose cost function values ​​are better than the current individual's and constructing an electric field-guided perturbation based on the cost difference and positional relationship between the superior neighbors and the current individual. S25. In the iterative evolution stage, a stagnation detection and restart strategy is introduced. The stagnation count is updated by detecting whether the historical best solution has been significantly improved, and random restarts are performed on some individuals with poor fitness when the stagnation threshold is reached. S3. The parameters of the inlet air temperature PID controller are optimized using the improved breakthrough platypus optimization algorithm to obtain the optimal PID control parameters. S4. The obtained optimal PID control parameters are set as the control parameters of the inlet air temperature PID controller module to achieve optimized control of the air preheater inlet air temperature.

2. The method for optimizing air preheater temperature control in thermal power generation according to claim 1, characterized in that: In the air preheater inlet air temperature PID control system constructed in step S1, the actual air temperature feedback value is obtained by the inlet air temperature monitoring module and transmitted to the inlet air temperature error calculation module. The inlet air temperature error calculation module receives the target temperature value and compares the target temperature with the feedback value to calculate and output a real-time error signal, which is then transmitted to the inlet air temperature PID controller module. The improved breakthrough platypus optimization algorithm module is used to optimize the PID control parameters of the inlet air temperature PID controller module. The inlet air temperature PID controller module with optimized parameters calculates the control quantity based on the real-time error signal and outputs the control quantity to the inlet air temperature regulation module to regulate the air preheater inlet air temperature.

3. The method for optimizing air preheater temperature control in thermal power generation according to claim 1, characterized in that: The enhanced population initialization in step S21 includes the following steps: Step S211, generating the initial chaotic variable vector. The components of the initial chaotic variable vector are taken from the sampling interval. and avoid special point sets Then, following the successive evolution of the Logistic chaotic mapping, the state variable evolution formula is: In the formula, Indicates the first The variable vector after the second chaotic evolution. Indicates the first The variable vector after the second chaotic evolution. This represents the control parameters for the Logistic mapping. This represents element-wise multiplication; Step S212: Map the chaotic variables obtained from each evolution to the search space to form the initial chaotic population. The mapping formula is: In the formula, Indicates the first The position vector of a chaotic initial individual. Represents the lower boundary vector of the search space. Represents the boundary vector in the search space. Indicates the first The variable vector obtained from the second chaotic evolution; Step S213: Perform reverse learning on the initial chaotic population to generate a reverse population. The reverse learning formula is: In the formula, Indicates the first The position vectors of the reverse-learned individuals; Step S214: Merge the initial chaotic population and the reverse-learned population into a total candidate population, and perform boundary repair on each individual in the total candidate population. The boundary repair formula is: In the formula, Indicates the first candidate species in the total candidate population. The position vector of each individual; Step S215: Calculate the cost function value of each individual in the total candidate population. When an anomaly occurs in the cost function calculation, assign the corresponding cost function value as a penalty value. Step S216: Sort the total candidate population in ascending order of cost function value, and select the population with the smallest cost function value. Each individual is used as the initial population, and the initial historical global optimal position is determined accordingly. and the corresponding historical best fitness value 。 4. The method for optimizing air preheater temperature control in thermal power generation according to claim 1, characterized in that: The differential evolution parameter adaptive update strategy in step S22 includes the following steps: Step S221, during the initialization phase, assign an independent differential variable asynchronous long factor to each individual. With cross probability Step S222: During the search evolution process, with a 10% probability, the first... The difference-variable asynchronous long factor of each individual is updated, and the update formula is: In the formula, This represents the updated difference-variable long factor. The fundamental parameter representing the differential variable asynchronous long factor is... The span parameter represents the differential variable asynchronous long factor. This represents an independent random number that follows a uniform distribution on the closed interval [0, 1]; when the trigger condition is not met, Keep the previous state unchanged; Step S223: During the search evolution process, with a probability of 10%, adjust the first... The crossover probability of each individual is updated using the following formula: In the formula, This represents the crossover probability after the i-th individual is updated. This represents an independent random number that follows a uniform distribution on the closed interval [0, 1]; when the trigger condition is not met, Keep the previous state unchanged; Step S224: In the land search mode, utilize the updated differential variable asynchronous long factor. Perform differential mutation to generate a mutation vector. The calculation formula is as follows: In the formula, Indicates the first The difference mutation vector corresponding to each individual 、 、 This represents three distinct indices randomly selected from the population, none of which are equal to the current individual's index. The backup individual location; Step S225: Perform a crossover operation on the mutation vector to generate candidate pattern locations, and ensure that at least one dimension comes from the mutation vector through a forced dimension mechanism. The crossover formula is: In the formula, Indicates the first The individual in the first Candidate mode components on the dimension, The difference mutated vector is represented in the th case. Components on the dimension, Indicates the first The interval used in dimensional cross-determination Uniformly distributed independent random numbers This indicates a randomly selected mandatory inheritance dimension index. Indicates the first The individual in the first Backup location components on the dimension.

5. The method for optimizing air preheater temperature control in thermal power generation according to claim 1, characterized in that: The Lévy flight-enhanced oviposition perturbation strategy in step S23 includes the following steps: Step S231, first initialize the Lévy oviposition displacement to a zero vector, and generate burst determination random numbers. When satisfied At that time, the oviposition burst mechanism is triggered, in which, Indicates the oviparous burst trigger threshold; Step S232: After triggering the oviparous burst mechanism, set the Levy flight stability distribution index parameter. And calculate the scale parameter in the Mantegna method. The calculation formula is: In the formula, Represents random variables standard deviation Represents the Gamma function. The parameters represent the Lévy flight stability distribution index; step S233: generate the Lévy flight step size vector based on the Mantegna method, calculated using the following formula: In the formula, Indicates the first The Lévy flight step size vector for each individual. This indicates that the expression follows a pattern with a mean of 0 and a variance of . A normally distributed random vector, Let represent a random vector that follows a standard normal distribution. This indicates element-wise absolute value operation; Step S234: Based on the Levy flight step size vector and the offset of the current individual relative to the historical best position, construct the final Levy oviparous displacement, calculated using the following formula: In the formula, Indicates the first The displacement vector of the oviparous burst of each individual. This represents the displacement amplitude parameter of the oviparous burst. Indicates the first Backup locations for each individual instance This indicates the historical global optimal position; when the oviparous burst mechanism is not triggered. Keep it as a zero vector.

6. The method for optimizing air preheater temperature control in thermal power generation according to claim 1, characterized in that: The neighborhood electric field-guided perturbation strategy in step S24 includes the following steps: Step S241, Identify individuals in the population with cost function values ​​superior to the current individual. For all individuals, construct a set of best neighbor indexes, calculated using the following formula: In the formula, Indicates the first A set of best neighbor indexes for each individual. Indicates the first The cost function value of each backup individual. Indicates the first The cost function value of each backup individual; Step S242: If the set of best neighbor indexes is not empty, calculate the distance between the current individual and each best neighbor, using the following formula: In the formula, Indicates the first Individual and the first Euclidean distance between each good neighbor Represents the L2 norm, This indicates a small positive quantity that prevents the denominator from being zero; Step S243: Calculate the guiding weight based on the relationship between the cost function value difference and the distance. The calculation formula is as follows: In the formula, Indicates the guiding weight. Indicates the first The cost function value of each backup individual. Indicates the first The cost function value of each backup individual. Indicates the first Individual and the first The Euclidean distance between each good neighbor; Step S244: Accumulate and average the guiding effects of all good neighbors to form the final electric field guided perturbation, calculated by the following formula: In the formula, Indicates the first The final electric field-guided perturbation vector of each individual, This indicates the intensity of the disturbance guided by the electric field in the neighborhood. This indicates the number of individuals in the best neighbor index set; when the best neighbor index set is empty. Keep it as a zero vector.

7. The method for optimizing air preheater temperature control in thermal power generation according to claim 1, characterized in that: The stagnation detection and restart strategy in step S25 includes the following steps: Step S251, Initialize the stagnation counter that has not shown significant improvement. And set the maximum number of consecutive stagnation algebras. Step S252: After each generation of search, extract the optimal fitness of the current formal population. The calculation formula is as follows: In the formula, This represents the optimal fitness value in the current generation of the formal population. Indicates the first The fitness value of a formal individual. Indicates the population size; Step S253: If the significant improvement condition is met, update the historical best solution and reset the stagnation counter to zero; if the significant improvement condition is not met, increment the stagnation counter. The significant improvement condition is: In the formula, This represents the historical best fitness value. This indicates a significant improvement in the judgment threshold; Step S254: When the stall counter reaches the maximum number of consecutive stall generations, restart according to the restart ratio. The number of individuals to be restarted is determined by the following formula: In the formula, Indicates the number of individuals restarted. This represents the floor function. Indicates the restart ratio; Step S255: Select the population with the worst fitness... Each individual is re-randomly initialized into the search space, and the calculation formula is: In the formula, Indicates the first A new position for each restarted individual. This represents a random vector that follows a uniform distribution on the closed interval [0, 1] for each dimension; Step S256: Recalculate the fitness of the restarted individual. If an anomaly occurs in the fitness calculation, assign the corresponding fitness value as a penalty value. And reset the stall counter to zero.

8. The method for optimizing air preheater temperature control in thermal power generation according to claim 1, characterized in that: Step S3, which utilizes the improved breakthrough platypus optimization algorithm to optimize the PID control parameters for inlet air temperature, includes the following steps: Step S801: Perform parameter initialization and storage pre-allocation, setting basic control parameters, adaptive update parameters, stagnation detection parameters, and convergence information storage space; Step S802: Utilize chaotic mapping and reverse learning strategies to perform enhanced population initialization, obtaining the formal initial population and historical global optimal solutions through chaotic variable evolution, search space mapping, reverse individual generation, candidate set merging, boundary repair, and fitness screening; Step S803: Enter the main iteration loop and... Perform dynamic weight scheduling, generating global and local search weights based on the current iteration progress, and backing up the current formal population and its fitness; Step S804: Perform adaptive parameter updates for each individual, updating the differential variable asynchronous long factor and crossover probability respectively in a probability-triggered manner, providing individual-level adaptive control parameters for subsequent land mode searches; Step S805: Perform adaptive dual-mode search, switching between underwater and land modes based on the mode determination results, wherein the underwater mode is used to generate local perturbation search results oriented towards the historical optimal position, and the land mode is used to generate results based on differential mutation and... Candidate mode positions for cross-operations; Step S806: Execute Levy-enhanced oviparous perturbation, generating oviparous burst displacements in a probabilistic triggering manner, and constructing large-span random perturbations by combining the relative offset between the current individual and the historical best position; Step S807: Execute neighborhood electric field guided perturbation, by identifying a set of superior neighbors that are better than the current individual, calculating the neighbor guided weights and accumulating them to form electric field guided perturbations, thereby enhancing the individual's ability to migrate to superior regions; Step S808: Execute integrated position update, boundary repair and greedy selection, coordinating the dual-mode search results, Levy oviparous displacements and neighborhood electric field perturbations. Assemble the new position and update the formal population according to the strict greedy criterion after boundary repair and fitness calculation; Step S809: Execute the global optimal update, stagnation detection and restart mechanism. When the historical optimal solution has not been significantly improved, accumulate the stagnation count and perform random restart on the individuals with the worst fitness after reaching the stagnation threshold to restore population diversity; Step S810: Record convergence information and determine the termination condition. Record the historical global optimal fitness and historical global optimal position in each generation, and output the optimal solution, optimal fitness and corresponding convergence process information after reaching the maximum number of iterations.

Images