CN-121979158-A - Production control method and system for unit equipment of secondary reheating unit

CN121979158ACN 121979158 ACN121979158 ACN 121979158ACN-121979158-A

Abstract

The invention discloses a unit equipment production control method and a unit equipment production control system for a secondary reheating unit, and relates to the technical field of production control. The method comprises the steps of obtaining load demand information, determining a target value of steam temperature according to the load demand information, obtaining state data of a secondary reheating unit in a preset time period, obtaining control actions through pre-trained reinforcement learning intelligent agents according to the state data and the target value of the steam temperature, evaluating the control actions according to a pre-trained dynamic model to obtain an evaluation result, and executing the control actions if the evaluation result is qualified. By introducing the reinforcement learning agent and the dynamic model, the historical trend in the state data can be used for predicting the hysteresis influence in advance and adjusting the control strategy in time, so that the deviation caused by hysteresis can be better dealt with. The problems of overshoot and untimely adjustment caused by hysteresis are reduced.

Inventors

QU LIQIU
ZHANG JIAN
KE YAN
XIN HAITAO
TANG BIN
ZHU ZILI

Assignees

国家能源集团宿迁发电有限公司
国家能源集团江苏电力有限公司
国能江苏电力工程技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260209

Claims (10)

1. A unit equipment production control method for a secondary reheating unit, the method comprising: acquiring load demand information, and determining a target value of steam temperature according to the load demand information, wherein the steam temperature comprises a main steam temperature, a primary reheating temperature and a secondary reheating temperature; Acquiring state data of the secondary reheating unit in a preset time period; preprocessing the state data and the target value of the steam temperature to obtain a feature matrix; Taking the feature matrix as input of a pre-trained reinforcement learning agent to obtain a control action; Evaluating the control action according to the pre-trained dynamic model to obtain an evaluation result; and if the evaluation result is qualified, executing the control action.
2. The unit equipment production control method for a double reheat unit according to claim 1, wherein the state data includes a main steam temperature, a single reheat temperature, a double reheat temperature, a superheater desuperheat water flow rate, a single reheat desuperheat water flow rate, a double reheat desuperheat water flow rate; the preprocessing of the state data and the target value of the steam temperature to obtain a feature matrix comprises the following steps: arranging target variables in time sequence to obtain a time sequence vector, wherein the target variables are any one variable of the state data or the target value of the steam temperature; Normalizing the time sequence vector to obtain a feature vector; and splicing the plurality of feature vectors to obtain a feature matrix.
3. The unit equipment production control method for a secondary reheat unit of claim 1, wherein the dynamic model is a modified LSTM model, the modified LSTM model comprising a convolutional layer, a long and short term memory network layer, an attention layer, and an output layer, wherein: the convolution layer is used for receiving the characteristic matrix as input and outputting a first characteristic quantity; The long-short-time memory network layer is used for carrying out secondary extraction on the first characteristic quantity to obtain a second characteristic quantity; The attention layer is used for enhancing the second characteristic quantity to obtain a third characteristic quantity; The output layer is used for outputting a prediction result according to the third characteristic quantity, and the prediction result comprises a main steam temperature prediction value, a primary reheating temperature prediction value and a secondary reheating temperature prediction value.
4. The unit equipment production control method for a secondary reheating unit according to claim 3, wherein the evaluating the control action according to the pre-trained dynamic model, the evaluating result comprising: performing recursive prediction on the state of the secondary reheating unit according to the reinforcement learning intelligent agent, the dynamic model and the control action until the preset prediction times are reached, so as to obtain hysteresis change data; and if the maximum value and the minimum value in the hysteresis change data meet a preset safety range, judging that the control action is qualified.
5. The unit equipment production control method for a secondary reheating unit according to claim 1, further comprising the step of constructing a time-series state gray level co-occurrence matrix before preprocessing the state data and the target value of the steam temperature, specifically: dividing the state data and the target value of the steam temperature into a plurality of discrete gray levels according to the respective physical change range and the safety threshold of the secondary reheating unit; And defining the distance and the direction of the time sequence state gray level co-occurrence matrix, wherein the distance corresponds to the time step difference and is used for adapting to the hysteresis characteristic of the steam temperature control of the secondary reheating unit, and the direction only keeps the forward time sequence pointing to the current time from the past time.
6. The unit equipment production control method for a secondary reheating unit according to claim 5, wherein the step of constructing a time-series state gray level co-occurrence matrix further comprises time-series data processing and gray level conversion, specifically: Acquiring state data of the preset time period and a target value of the steam temperature, wherein the state data and the target value of the steam temperature are arranged into a plurality of groups of time sequence in time sequence, each group of time sequence corresponds to a target variable, and the target variable is any one variable of the state data or the target value of the steam temperature; Each group of time sequence is mapped into a corresponding gray level sequence, so that each data point in the time sequence corresponds to a discrete gray level, and a gray level sequence set corresponding to a target variable one by one is formed.
7. The unit equipment production control method for a secondary reheating unit according to claim 6, wherein after the step of constructing the time-series state gray level co-occurrence matrix, the method further comprises a step of extracting features from the time-series state gray level co-occurrence matrix, specifically: Extracting the correlation characteristics of the time sequence state gray level co-occurrence matrix, and reflecting the linear correlation degree of two target variables on time sequence so as to identify the variables critical to the steam temperature control; extracting contrast characteristics of the time sequence state gray level co-occurrence matrix, wherein the contrast characteristics are used for reflecting gray level differences of the same target variable at different time steps so as to judge fluctuation intensity of the corresponding target variable; extracting energy characteristics of the time sequence state gray level co-occurrence matrix, wherein the energy characteristics are used for reflecting the uniformity of time sequence change of a target variable so as to judge the stable state of the variable under the current working condition; And extracting entropy features of the time sequence state gray level co-occurrence matrix, wherein the entropy features are used for reflecting the chaos of time sequence association among target variables so as to judge the stability of association among the target variables, and the entropy features are used for triggering a subsequent matrix dynamic updating mechanism.
8. The unit equipment production control method for a secondary reheating unit according to claim 7, further comprising a step of dynamically updating the time-series state gray level co-occurrence matrix, specifically: Monitoring whether the load demand information changes or not or whether the gray level sequence of any target variable changes across a safety interval or not in real time; if any change condition is monitored, calculating lag time steps of control actions and steam temperature changes under different loads through historical data, determining the time step difference optimal value of the time sequence state gray level co-occurrence matrix again, and reconstructing the time sequence state gray level co-occurrence matrix based on the new time step difference; And comparing the correlation characteristic difference of the reconstructed time sequence state gray level co-occurrence matrix and the time sequence state gray level co-occurrence matrix before updating, and if the correlation characteristic difference exceeds a preset threshold value, synchronously updating the characteristic items corresponding to the correlation characteristics in the characteristic matrix of the reinforcement learning intelligent agent.
9. The method for controlling production of a plant for a secondary reheating plant according to claim 7, wherein the step of obtaining a control action by using the feature matrix as an input of a pre-trained reinforcement learning agent comprises: After receiving the new feature matrix, the reinforcement learning intelligent agent determines the time sequence association strength and time delay relation between different state variables and steam temperature by analyzing the correlation features, and adjusts the execution time of the control action; Judging the fluctuation intensity of the steam temperature or the state variable by analyzing the contrast characteristic in the new characteristic matrix, and outputting a control action according to the fluctuation intensity, the energy characteristic and the entropy characteristic; And analyzing entropy features in the new feature matrix, and if the association between entropy feature display variables is unstable, combining features corresponding to the dynamically updated time sequence state gray level co-occurrence matrix to optimize the adjustment logic of the control action.
10. Unit plant production control system for a secondary reheating unit for implementing a unit plant production control method for a secondary reheating unit according to any of claims 1-9, characterized by comprising: The target confirmation module is used for acquiring load demand information and determining a target value of the steam temperature according to the load demand information; The data acquisition module is used for acquiring state data of the secondary reheating unit in a preset time period; The data processing module is used for preprocessing the state data and the target value of the steam temperature to obtain a feature matrix; The action strategy module is used for taking the feature matrix as the input of the pre-trained reinforcement learning intelligent agent to obtain a control action; The action evaluation module is used for evaluating the control action according to the pre-trained dynamic model to obtain an evaluation result; and the execution module is used for executing the control action if the evaluation result is qualified.

Description

Production control method and system for unit equipment of secondary reheating unit Technical Field The invention relates to the technical field of production control, in particular to a unit equipment production control method and system for a secondary reheating unit. Background The secondary reheating unit is high-efficiency power generation equipment in a modern power system, and the secondary reheating unit heats and expands steam for multiple times to realize gradual utilization of energy so as to improve heat efficiency. In the operation process of the secondary reheating unit, the steam temperature directly influences the thermal efficiency, the operation stability and the safety of the unit. Too high steam temperature can lead to the decrease of the material strength of the parts, accelerate the thermal fatigue and affect the service life of the parts, while too low steam temperature can reduce the thermal efficiency of the unit and cause energy waste. Therefore, the main task of steam temperature control is to control the temperature within a reasonable range on the premise of ensuring safety, so that the unit components are not damaged, and the thermal efficiency of the steam temperature control can be fully exerted. Currently, steam temperature control of a secondary reheat unit relies primarily on conventional PID controllers. However, the response speed and accuracy of the conventional PID controller are often not ideal when dealing with a nonlinear, large hysteresis system such as a double reheat unit. When the load or external condition of the unit is suddenly changed, the traditional PID control is easy to generate larger overshoot, so that the temperature fluctuation is large, and the safety and stability of the unit are affected. Disclosure of Invention The invention aims to solve the problem that larger overshoot is easy to generate in the background art, and provides a unit equipment production control method and a unit equipment production control system for a secondary reheating unit. In a first aspect of the implementation of the present invention, a unit equipment production control method for a secondary reheating unit is provided, and the method includes: acquiring load demand information, and determining a target value of steam temperature according to the load demand information, wherein the steam temperature comprises a main steam temperature, a primary reheating temperature and a secondary reheating temperature; Acquiring state data of the secondary reheating unit in a preset time period; preprocessing the state data and the target value of the steam temperature to obtain a feature matrix; Taking the feature matrix as input of a pre-trained reinforcement learning agent to obtain a control action; Evaluating the control action according to the pre-trained dynamic model to obtain an evaluation result; and if the evaluation result is qualified, executing the control action. Optionally, the state data includes a main steam temperature, a primary reheating temperature, a secondary reheating temperature, a superheater desuperheating water flow, a primary reheating desuperheating water flow, a secondary reheating desuperheating water flow; the preprocessing of the state data and the target value of the steam temperature to obtain a feature matrix comprises the following steps: arranging target variables in time sequence to obtain a time sequence vector, wherein the target variables are any one variable of the state data or the target value of the steam temperature; Normalizing the time sequence vector to obtain a feature vector; and splicing the plurality of feature vectors to obtain a feature matrix. Optionally, the dynamic model is an improved LSTM model, the improved LSTM model comprises a convolution layer, a long-short-time memory network layer, an attention layer and an output layer, wherein: the convolution layer is used for receiving the characteristic matrix as input and outputting a first characteristic quantity; The long-short-time memory network layer is used for carrying out secondary extraction on the first characteristic quantity to obtain a second characteristic quantity; The attention layer is used for enhancing the second characteristic quantity to obtain a third characteristic quantity; The output layer is used for outputting a prediction result according to the third characteristic quantity, and the prediction result comprises a main steam temperature prediction value, a primary reheating temperature prediction value and a secondary reheating temperature prediction value. Optionally, the training process of the reinforcement learning agent includes: according to the influence of different control actions on the steam temperature, rewards of each time step are calculated: Wherein, the Is a time stepIs a prize value for (1); Is the first The target value of the seed gas temperature,Respectively correspond to the main steam temperature a primary reheat temperature a