Intelligent optimization control method for cooling tower of thermal power generating unit based on environment temperature and load

Abstract

The invention discloses a thermal power generating unit cooling tower intelligent optimization control method based on environment temperature and load, and the method comprises the following steps: building a thermal power generating unit cooling tower circulating water temperature PID control system, and presetting the proportional gain, integral time and differential time of the PID control system according to historical empirical values; establishing a neural network model, training the neural network model by using historical data of environment temperature, historical data of load of a thermal power generating unit, historical data of air inlet rate of a cooling tower and historical data of water temperature of cooling circulating water, and then predicting proportional gain, integral time and differential time of a PID control system by using the neural network model according to real-time working conditions; and establishing a genetic algorithm model, inputting a prediction result of the neural network model into the genetic algorithm model for optimization, and obtaining optimal preset parameter values of proportional gain, integral time and differential time of the PID control system. The defects in the prior art can be overcome, and the control precision of the cooling tower is improved.

Description

Technical Field

[0001] This invention relates to the field of optimized operation technology for cooling towers of thermal power units, and in particular to an intelligent optimization control method for cooling towers of thermal power units based on ambient temperature and load. Background Technology

[0002] Cooling towers are devices used in thermal power plants to cool circulating cooling water. They work by exchanging heat between the cooling water and the outside air, thus lowering the water temperature. The airflow to the cooling tower needs to be dynamically adjusted based on parameters such as ambient temperature and the load of the thermal power unit. Currently, PID closed-loop control systems are widely used. However, the control parameters of a PID closed-loop control system need to be tuned according to changes in operating conditions. Existing PID control parameter tuning methods, such as those using neural networks, lack real-time performance and cannot meet the high-precision control requirements of cooling towers. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide an intelligent optimization control method for cooling towers of thermal power units based on ambient temperature and load, which can overcome the shortcomings of the prior art and improve the control accuracy of cooling towers.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows.

[0005] A method for intelligent optimization control of cooling towers in thermal power units based on ambient temperature and load includes the following steps: A PID control system for the circulating water temperature of the cooling tower in a thermal power unit was established. The proportional gain, integral time, and derivative time of the PID control system were preset based on historical experience values. A neural network model was established and trained using historical data of ambient temperature, thermal power unit load, cooling tower air intake, and cooling circulating water temperature. The neural network model was then used to predict the proportional gain, integral time, and derivative time of the PID control system based on real-time operating conditions. A genetic algorithm model was established, and the prediction results of the neural network model were input into the genetic algorithm model for optimization to obtain the optimal preset parameter values ​​for the proportional gain, integral time, and derivative time of the PID control system.

[0006] Preferably, the neural network model includes an input layer, a hidden layer, an attention mechanism layer, and an output layer arranged sequentially. The number of neurons in the input layer is equal to the number of categories of the input data, and the number of neurons in the output layer is dynamically adjustable. The real-time ambient temperature, thermal power unit load, cooling tower air intake, and target cooling circulating water temperature are input into the input layer and standardized. The hidden layer performs nonlinear mapping and feature extraction on the standardized data. Each neuron in the hidden layer outputs a set of predicted control parameters for the PID control system. The attention mechanism layer classifies the prediction results based on the similarity of each prediction result. The output layer activates a number of neurons equal to the number of classification results. Each neuron in the output layer receives one type of prediction result and uses an activation function to obtain the final output result.

[0007] As a preferred approach, the attention mechanism layer performs weighted processing on each result data in each type of prediction result.

[0008] Preferably, the genetic algorithm model treats each prediction result output by the output layer as an individual, forming an initial population. The fitness of each individual is calculated, and a fitness threshold for the selection operation is set. Individuals with fitness below the fitness threshold are deleted. If the number of deleted individuals is less than 5% of the total number of individuals, the remaining individuals are sorted in ascending fitness order, and deletion begins from the beginning of the queue until the number of deleted individuals reaches 5% of the total number of individuals. Proportional gain data, integral time data, or derivative time data are randomly extracted from the remaining individuals for crossover operations to generate new individuals. The number of new individuals is equal to the number of individuals deleted in this iteration. Random data perturbation is performed on the individuals that have undergone crossover operations to achieve individual mutation operations.

[0009] As a preferred option, the individual fitness F = k1Os + k2Ts + k3Es, where Os is the system overshoot, Ts is the system settling time, Es is the system steady-state error, and k1 to k3 are weighting coefficients.

[0010] As a preferred method, a randomly generated mutation vector is used to perform mutation operations on the individual. The mutation vector includes the perturbation direction and the perturbation magnitude. After each mutation, the quality of the mutation result is evaluated. Based on the evaluation result, the priority coefficient of the mutation vector is determined. If the evaluation result is good, the priority coefficient is increased; if the evaluation result is bad, the priority coefficient is decreased.

[0011] Preferably, during the mutation operation, the mutated individual is compared with other individuals. If the mutated individual is identical to other individuals, the mutation operation is performed again. The beneficial effects of adopting the above technical solution are as follows: This invention improves the existing method of tuning PID parameters using neural networks. Instead of directly outputting the tuning result, it uses a specially designed genetic algorithm model to optimize the tuning result, thereby improving the accuracy of parameter tuning while ensuring the speed of parameter tuning. Attached Figure Description

[0012] Figure 1This is a schematic diagram of the present invention. Detailed Implementation

[0013] This invention addresses the problem that existing neural network models rely too heavily on prior training for PID parameter tuning, leading to inaccurate actual prediction results, and makes targeted improvements.

[0014] Reference Figure 1 The overall solution is as follows: A PID control system for the circulating water temperature of the cooling tower in a thermal power unit is established. The proportional gain, integral time, and derivative time of the PID control system are preset based on historical experience values. A neural network model is established, trained using historical data of ambient temperature, thermal power unit load, cooling tower airflow, and cooling circulating water temperature. The neural network model is then used to predict the proportional gain, integral time, and derivative time of the PID control system based on real-time operating conditions. A genetic algorithm model is established, and the prediction results of the neural network model are input into the genetic algorithm model for optimization to obtain the optimal preset parameter values ​​for the proportional gain, integral time, and derivative time of the PID control system.

[0015] In this approach, we add an attention mechanism layer to the traditional neural network model, making the neural network model consist of an input layer, a hidden layer, an attention mechanism layer, and an output layer arranged sequentially. The number of neurons in the input layer is equal to the number of categories of the input data. In addition, we set the number of neurons in the output layer to be dynamically adjustable to match the subsequent genetic algorithm optimization process.

[0016] The prediction process of the neural network model is as follows: real-time ambient temperature, thermal power unit load, cooling tower air intake, and target cooling circulating water temperature are input into the input layer and standardized. The hidden layer performs nonlinear mapping and feature extraction on the standardized data. Each neuron in the hidden layer outputs a set of predicted control parameters for the PID control system. The attention mechanism layer classifies the prediction results based on their similarity and weights the data in each category. The output layer activates a number of neurons equal to the number of classification results. Each neuron in the output layer receives one category of prediction results and uses an activation function to obtain the final output result. The neural network model of this invention outputs a set of predicted results after classification analysis, rather than the final result. This effectively reduces the accuracy requirements for training the neural network model and improves the robustness of the overall algorithm.

[0017] The genetic algorithm model treats each prediction result output by the output layer as an individual, forming an initial population, and calculates the fitness of each individual. Individual fitness F = k1Os + k2Ts + k3Es, where Os is the system overshoot, Ts is the system settling time, Es is the system steady-state error, and k1 to k3 are weight coefficients. A fitness threshold is set for the selection operation. Individuals with fitness below the threshold are deleted. If the number of deleted individuals is less than 5% of the total number of individuals, the remaining individuals are sorted in ascending order of fitness, and deletion begins from the beginning of the queue until the number of deleted individuals reaches 5% of the total number of individuals. From the remaining individuals, proportional gain data, integral time data, or derivative time data are randomly extracted and crossover operations are performed to generate new individuals. The number of new individuals is equal to the number of individuals deleted in this iteration. Random perturbation is then performed on the individuals after the crossover operation to achieve the mutation operation. Specifically, a randomly generated mutation vector is used to mutate individuals. This vector includes the direction and magnitude of the perturbation. After each mutation, the quality of the result is evaluated, and a priority coefficient is assigned to the mutation vector based on the evaluation result. A good evaluation increases the priority coefficient, while a bad evaluation decreases it. During mutation, the mutated individual is compared with other individuals. If the mutated individual is identical to other individuals, the mutation operation is repeated. This invention introduces a priority coefficient during the mutation process, enabling automatic optimization of the mutation direction and effectively improving the optimization efficiency of the genetic algorithm.

[0018] This invention proposes a PID parameter tuning method that combines neural networks and genetic algorithms. By improving the neural network and genetic algorithm, it makes them applicable to the tuning process of the PID control system of cooling towers, thereby improving the control accuracy of cooling towers.

[0019] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this invention, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0020] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent optimization control of cooling towers in thermal power units based on ambient temperature and load, characterized in that... Includes the following steps: A PID control system for the circulating water temperature of the cooling tower in a thermal power unit was established. The proportional gain, integral time, and derivative time of the PID control system were preset based on historical experience values. A neural network model was established and trained using historical data of ambient temperature, thermal power unit load, cooling tower air intake, and cooling circulating water temperature. The neural network model was then used to predict the proportional gain, integral time, and derivative time of the PID control system based on real-time operating conditions. A genetic algorithm model was established, and the prediction results of the neural network model were input into the genetic algorithm model for optimization to obtain the optimal preset parameter values ​​for the proportional gain, integral time, and derivative time of the PID control system.

2. The intelligent optimization control method for cooling towers of thermal power units based on ambient temperature and load according to claim 1, characterized in that: The neural network model includes an input layer, a hidden layer, an attention mechanism layer, and an output layer arranged sequentially. The number of neurons in the input layer is equal to the number of categories of the input data, and the number of neurons in the output layer is dynamically adjustable. The real-time ambient temperature, thermal power unit load, cooling tower air intake, and target cooling circulating water temperature are input into the input layer and standardized. The hidden layer performs nonlinear mapping and feature extraction on the standardized data. Each neuron in the hidden layer outputs a set of predicted control parameters for the PID control system. The attention mechanism layer classifies the prediction results based on the similarity of each prediction result. The output layer activates a number of neurons equal to the number of classification results. Each neuron in the output layer receives one type of prediction result and uses an activation function to obtain the final output result.

3. The intelligent optimization control method for cooling towers of thermal power units based on ambient temperature and load according to claim 2, characterized in that: The attention mechanism layer performs weighted processing on each result data in each type of prediction result.

4. The intelligent optimization control method for cooling towers of thermal power units based on ambient temperature and load according to claim 3, characterized in that: The genetic algorithm model treats each prediction result output by the output layer as an individual, forming an initial population. It calculates the fitness of each individual, sets a fitness threshold for the selection operation, and deletes individuals with fitness values ​​below the threshold. If the number of deleted individuals is less than 5% of the total number of individuals, the remaining individuals are sorted in ascending fitness order, and deletion begins from the beginning of the queue until the number of deleted individuals reaches 5% of the total number of individuals. From the remaining individuals, proportional gain data, integral time data, or derivative time data are randomly extracted for crossover operations to generate new individuals. The number of new individuals is equal to the number of individuals deleted in this iteration. Random perturbation is then performed on the individuals that have undergone crossover operations to achieve individual mutation.

5. The intelligent optimization control method for cooling towers of thermal power units based on ambient temperature and load according to claim 4, characterized in that: Individual fitness F = k1Os + k2Ts + k3Es, where Os is the system overshoot, Ts is the system settling time, Es is the system steady-state error, and k1 to k3 are weighting coefficients.

6. The intelligent optimization control method for cooling towers of thermal power units based on ambient temperature and load according to claim 5, characterized in that: Individuals are mutated using randomly generated mutation vectors, which include the direction and magnitude of the perturbation. After each mutation, the quality of the mutation result is evaluated, and a priority coefficient is assigned to the mutation vector based on the evaluation result. If the evaluation result is good, the priority coefficient is increased; if the evaluation result is bad, the priority coefficient is decreased.

7. The intelligent optimization control method for cooling towers of thermal power units based on ambient temperature and load according to claim 6, characterized in that: When performing a mutation operation, the mutated individual is compared with other individuals. If the mutated individual is the same as other individuals, the mutation operation is performed again.

Images