Flexible operation control method for ultra-supercritical thermal power unit based on self-starting and stopping

Abstract

The application provides a flexible operation control method for ultra-supercritical thermal power generating unit based on self-starting and stopping, relates to the technical field of thermal power generation, and comprises the following steps: acquiring thermal system parameters of the thermal power generating unit in a starting and stopping process, and establishing a thermal system parameter coupling model; setting a load change rate of each stage by using a multi-stage load change trajectory; generating a load control instruction according to the multi-stage load change trajectory and a dynamic compensation algorithm based on proportional integral derivative control; acquiring peak shaving demand data from a power grid dispatching system, and judging whether the load adjustment time meets the peak shaving demand; recalculating the load change rate of each stage according to the adjusted load change trajectory, updating the dynamic compensation algorithm parameters, and generating a new load control instruction. The application realizes real-time monitoring of key parameters, evaluates the adjustment effect through deviation analysis, realizes smooth adjustment of the load of the thermal power generating unit, and improves the peak shaving capacity of the power grid and the operation efficiency of the unit.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of thermal power generation technology, and particularly relates to a flexible operation control method for ultra-supercritical thermal power generating units based on self-starting and stopping. BACKGROUND

[0002] In the flexible operation control of supercritical thermal power generating units based on self-starting and stopping, the core technical problem faced by the gradual load control is how to accurately balance between the dynamic response characteristics of the unit and the rapid changes in load demand. During the starting and stopping process of the supercritical thermal power generating unit, the thermal system parameters such as the main steam pressure, temperature, and reheat steam temperature have high coupling and large inertia, which leads to response lag and parameter fluctuation during load adjustment. The traditional load control strategy usually adopts a step or linear change mode, which cannot fully adapt to the dynamic characteristics of the unit and easily causes overshoot or oscillation, especially during large-scale load adjustment, the stability of the system is difficult to guarantee. At the same time, due to the high coordination requirement of each subsystem during the self-starting and stopping process of the unit, the setting of the load change trajectory needs to consider the dynamic response time difference of multiple links such as the boiler, steam turbine, and feedwater system. If the control is not proper, it may cause the key parameters to deviate from the safety range, and even trigger the protection action. In addition, the rapid changes in the demand of the power grid for peak shaving require the unit to complete the load adjustment in a short time, but the complex thermal characteristics of the supercritical unit also limit its rapid response capability. How to realize the smooth transition of the load under the premise of ensuring the safety of the system has become a key difficulty in the gradual load control technology. This technical problem needs in-depth research in the aspects of multi-stage load change trajectory design, stage control target optimization, and parameter dynamic compensation, so as to realize the unity of the accuracy and stability of the load control. SUMMARY

[0003] The present application provides a flexible operation control method for ultra-supercritical thermal power generating units based on self-starting and stopping, mainly including: obtaining the thermal system parameters of the thermal power generating unit during the starting and stopping process, and establishing a thermal system parameter coupling model; identifying the inertia lag characteristics of the thermal system parameters through a time series analysis method, and determining the lag time window in the load adjustment process; setting the load change rate of each stage by using a multi-stage load change trajectory; generating a load control instruction according to the multi-stage load change trajectory and a dynamic compensation algorithm based on a proportional-integral-derivative control; obtaining the peak shaving demand data from the power grid dispatching system, judging whether the load adjustment time meets the peak shaving demand, and if not, adjusting the stage division of the load change trajectory; recalculating the load change rate of each stage according to the adjusted load change trajectory, updating the parameters of the dynamic compensation algorithm, and generating a new load control instruction.

[0004] Further, the thermal system parameters include the main steam pressure, the main steam temperature, and the reheat steam temperature.

[0005] Further, the establishing of the thermal system parameter coupling model comprises: using a linear regression method in a Scikit-learn library to establish a thermal system parameter coupling model among the main steam pressure, the main steam temperature and the reheat steam temperature; determining the main steam pressure as a dependent variable, the main steam temperature and the reheat steam temperature as independent variables, and calculating regression coefficients and error terms; if the fitting degree of the thermal system parameter coupling model is lower than a preset fitting threshold, adjusting the independent variable combination, recalculating the regression coefficients, and optimizing the thermal system parameter coupling model.

[0006] Further, the identifying of the thermal system parameter inertia lag characteristic through the time series analysis method comprises: sampling the main steam pressure, the main steam temperature and the reheat steam temperature data in the thermal system parameter coupling model to form a time series data set; using an STL decomposition method to decompose the time series data set, and using an autocorrelation function to calculate the inertia lag characteristic of the thermal system parameter according to the decomposition result. Further, the multi-stage load change trajectory comprises: obtaining historical load data, and performing data cleaning, preprocessing and feature extraction; determining a target value of load adjustment according to the extracted load change characteristic and the operation requirement of the thermal system; dividing the load adjustment stages according to the change law of the historical load data for the load adjustment target value; setting a load change rate range according to the historical load change rate in each load adjustment stage; generating the multi-stage load change trajectory according to the set load change rate and the divided load adjustment stages.

[0007] Further, the dynamic compensation algorithm based on proportional integral derivative control comprises: using a proportional integral derivative control algorithm to calculate a proportional value, an integral value and a differential value for the real-time obtained parameter fluctuation value; using a formula to calculate a dynamic compensation amount according to the proportional value, the integral value and the differential value, and adjusting the load change rate to obtain a stable parameter value; determining the completion state of the load adjustment stage according to the stable parameter value.

[0008] Further, the generating of the load control instruction comprises: obtaining the multi-stage load change trajectory from the historical data, and extracting a load value and a trajectory value; using a PID control algorithm to calculate a compensation value according to the load value and the trajectory value; using a fuzzy control algorithm to generate an instruction value according to the compensation value and the trajectory value; obtaining related values from a boiler, a steam turbine and a feedwater system, and determining whether the related values are within a preset threshold range; if the related values are out of the threshold range, recalculating the compensation value and updating the instruction value; using the fuzzy control algorithm to adjust a boiler value, a steam turbine value and a feedwater value according to the updated instruction value, and coordinating the action values.

[0009] Furthermore, the determination of whether the load adjustment time meets the peak-shaving demand includes: obtaining peak-shaving demand data from the power grid dispatching system, extracting the peak-shaving demand value and the load value, and calculating the load adjustment time value using the ARIMA model; comparing the load adjustment time value with a preset peak-shaving demand threshold, and if the load adjustment time does not meet the peak-shaving demand, adjusting the stage division of the load change trajectory.

[0010] Furthermore, the adjusting the stage division of the load change trajectory refers to adjusting the stage division of the load change trajectory using a piecewise linear regression method.

[0011] Furthermore, the generation of a new load control instruction includes, for the extracted load change trajectory value, using a piecewise linear regression method to calculate the load change rate value of each stage; if the stage load change rate value exceeds a preset threshold, updating the dynamic compensation algorithm parameter value according to a preset rule; based on the updated algorithm parameter value, recalculating the load control instruction value using a least squares method; using a fuzzy control method, combining the load control instruction value and the compensation value to generate a final load control signal.

[0012] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0013] The present invention discloses a flexible operation control method for ultra-supercritical thermal power units based on automatic start-stop. The method establishes a thermal system parameter coupling model, identifies parameter inertia hysteresis characteristics, and designs a multi-stage load change trajectory and dynamic compensation algorithm to achieve smooth load adjustment. The present invention first obtains thermal parameter data from a real-time monitoring system, analyzes the parameter hysteresis characteristics, and divides the load adjustment stages accordingly. Then, based on the peak-shaving requirements of the power grid, the load change trajectory is optimized, and a dynamic compensation algorithm is used to suppress parameter fluctuations. During the adjustment process, the present invention monitors key parameters in real time, triggers necessary protection actions, and evaluates the adjustment effect through deviation analysis. Through this method, the present invention can achieve rapid and smooth adjustment of the thermal power unit load while ensuring safety, thereby improving the peak-shaving capacity of the power grid and the operating efficiency of the unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Fig. 1 The figure is an overall flow chart of the flexible operation control method of an ultra-supercritical thermal power unit based on automatic start and stop according to the present invention.

[0015] Fig. 2 Schematic diagram of parameter modeling and hysteresis analysis of the present invention.

[0016] Fig. 3 Schematic diagram of the dynamic compensation control of the present invention.

[0017] Fig. 4 This is a flow chart of generating load control instructions of the present invention. DETAILED DESCRIPTION

[0018] For a further understanding of the present application, reference will be made to the following description taken in conjunction with the accompanying drawings and embodiments. The present application will be further described in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the application. In addition, it should be noted that only parts related to the application are shown in the drawings for ease of description.

[0019] As Figs. 1-4 The embodiment based on the self-starting and stopping ultra-supercritical thermal power unit flexible operation control method can specifically include:

[0020] In step S101, the thermal system parameters of the supercritical thermal power unit in the starting and stopping process are obtained, including the main steam pressure, the main steam temperature, and the reheat steam temperature. The data is extracted from the real-time monitoring system or historical data record, and a thermal system parameter coupling model is established by using a multiple regression analysis method.

[0021] The main steam pressure, the main steam temperature, and the reheat steam temperature data of the supercritical thermal power unit in the starting and stopping process are obtained by real-time monitoring. The thermal system parameters related to the starting and stopping process are extracted from the historical data record, including the main steam pressure, the main steam temperature, and the reheat steam temperature. The real-time monitoring data and the historical data are time-aligned to ensure that the timestamps of the data points are consistent. The merged data is cleaned using the Pandas library in Python to remove missing values and outliers, and mean filtering is applied to remove noise interference to obtain a normalized data set. For the normalized data set, a linear regression method in the Scikit-learn library is used to establish a thermal system parameter coupling model between the main steam pressure, the main steam temperature, and the reheat steam temperature. In the thermal system parameter coupling model, the main steam pressure is determined as the dependent variable, and the main steam temperature and the reheat steam temperature are determined as the independent variables. The regression coefficients and error terms are calculated. If the fitting degree of the thermal system parameter model is lower than the preset threshold of 0.95, the independent variable combination is adjusted, the regression coefficients are recalculated, and the thermal system parameter coupling model is optimized. According to the optimized thermal system parameter coupling model, the quantitative relationship between the main steam pressure, the main steam temperature, and the reheat steam temperature is analyzed, and the thermal system parameter coupling law is obtained.

[0022] For example, the monitoring and analysis of thermal system parameters during the start-up and shut-down process of a supercritical thermal power unit are crucial for the safe operation of the unit. The collected main steam pressure generally fluctuates in the range of twenty-seven to thirty megapascals, the main steam temperature is maintained at five hundred eighty to six hundred degrees Celsius, and the reheat steam temperature is in the interval of five hundred seventy to five hundred ninety degrees Celsius. These data reflect the operating state of the unit during the start-up or shut-down process and need to be compared and analyzed with historical data. The historical database records the operating parameters of the unit under different loads and in different seasons. For example, during the start-up of the unit in summer, due to the high ambient temperature, the main steam temperature rises at a rate that is fifteen percent faster than in winter, which will affect the pressure change rate. Therefore, seasonal factors need to be considered during data processing. Data time alignment is the key to ensuring the accuracy of the analysis. For example, during the start-up process of a certain unit, data is recorded every ten seconds, and if communication interruption occurs, resulting in the absence of data at certain time points, linear interpolation method needs to be used for supplementation. For abnormal values, such as data points where the main steam pressure suddenly drops below ten megapascals, they should be removed. During the data cleaning process, mean filtering can effectively remove measurement noise. Taking five-point mean filtering as an example, high-frequency fluctuations in pressure measurement can be smoothed out, but the real pressure change trend will not be excessively blurred. Normalization processing unifies parameters of different dimensions to the same scale, facilitating subsequent modeling and analysis. Parameter coupling modeling needs to consider physical laws. When the unit increases the load, the rise of the main steam pressure will cause the temperature to rise, but the temperature change has a certain hysteresis. The reheat steam temperature is greatly affected by the primary superheater and has relatively weak correlation with the main steam pressure. Therefore, when establishing the regression model, a time delay term can be introduced to improve the fitting accuracy. During the model optimization process, if the correlation coefficient between the main steam pressure and the temperature is less than zero point nine, the data quality and model structure need to be re-examined. Possible reasons include unstable boiler combustion, fluctuations in the feedwater system, etc., which will affect the coupling relationship between parameters. By introducing appropriate intermediate variables or adjusting the data window length, the explanatory power of the model can be improved. The finally established thermal system parameter coupling model can quantitatively describe the relationship between the three key parameters. For example, during the stable temperature rise stage, for every one megapascal increase in the main steam pressure, the main steam temperature rises by an average of one point five degrees, and the reheat steam temperature rises by one point two degrees. This quantitative relationship helps to optimize the start-up and shut-down process control strategy and improve the reliability of the unit operation.

[0023] In step S102, based on the thermal system parameter coupling model, the parameter inertia hysteresis characteristics are identified through time series analysis method, the parameter response time is calculated, and the hysteresis time window in the load adjustment process is determined.

[0024] The main steam pressure, main steam temperature, and reheat steam temperature data in the thermal system parameter coupling model are acquired, with a sampling frequency of 1 minute, and data of 7 consecutive days are collected to form a time series data set. The STL decomposition method is used to decompose the time series data set to extract the trend item, periodic item, and residual item. According to the decomposition result, the inertia lag characteristics of the main steam pressure, main steam temperature, and reheat steam temperature are calculated using the autocorrelation function, and the lag order of each parameter is determined. Through the lag order, the response time of the main steam pressure, main steam temperature, and reheat steam temperature is calculated to obtain the dynamic characteristics of each parameter. In combination with the historical data in the load adjustment process, the Pearson correlation coefficient is used to determine the corresponding relationship between the load change and the parameter response. If the parameter response time exceeds a preset threshold, the lag time window is adjusted, and the load adjustment strategy is optimized. According to the optimized lag time window, an LSTM model is used to generate a parameter dynamic response model in the load adjustment process.

[0025] For example, in the supercritical thermal power unit thermal system parameter coupling model, the data acquisition first needs to pay attention to the integrity and continuity of the time series. Seven days of data sampled once a minute can effectively reflect the operating characteristics of the unit under different operating conditions. For example, the data of a 600 MW unit in a week shows that the main steam pressure fluctuates in the range of 27 to 30 MPa, the corresponding main steam temperature changes between 580 and 600 degrees Celsius, and the reheat steam temperature is in the interval of 570 to 590 degrees Celsius. Through time series decomposition, the parameter change law can be better understood. For example, during the daily load change process, the trend item of the main steam pressure reflects the overall direction of load adjustment, the periodic item reflects the difference between morning and evening peaks and valleys, and the residual item contains short-term fluctuation information. The data analysis of a unit shows that during the period of rapid load change in the morning and evening, the amplitude of the periodic fluctuation of the main steam pressure can reach 2 MPa. Autocorrelation function analysis reveals the lag relationship between parameters. When the load rises, the main steam pressure usually responds within three to five minutes, while the response time of the main steam temperature is seven to ten minutes, and the lag time of the reheat steam temperature is the longest, about twelve to fifteen minutes. This lag characteristic is due to the thermal inertia of the thermal system and has an important influence on the safe operation of the unit. Parameter response characteristic analysis shows that during the rapid load increase process, for every 1 MPa increase in the main steam pressure, it takes about 4 minutes to make the main steam temperature rise by 3 degrees stably. If the pressure increasing rate is too fast, it may cause temperature response lag, affecting the thermal stress control. Therefore, the load adjustment strategy needs to be optimized according to the temperature response characteristics. Correlation analysis shows that the Pearson correlation coefficient of load change and main steam pressure reaches 0.95, the correlation coefficient with main steam temperature is 0.88, and the correlation coefficient with reheat steam temperature is 0.82. This decreasing trend reflects the chain transmission characteristics of parameter response. When modeling dynamic response, the coupling relationship between parameters needs to be considered. For example, during the summer startup process of a certain unit, for every 5 degrees increase in ambient temperature, the response time of the main steam temperature will be shortened by about 8%. This temperature sensitivity requires the introduction of a temperature correction factor in modeling. At the same time, by adjusting the time window, the description accuracy of the model to the dynamic characteristics of the unit can be improved. Practice shows that setting the time window to twice the maximum lag time can better capture the parameter change law.

[0026] Step S103, for the lag time window, design a multi-stage load change trajectory, divide the load adjustment stages according to the length of the lag time window and the load adjustment target, and set the load change rate of each stage.

[0027] The historical load data is obtained from the thermal system database, and the data is cleaned and preprocessed to remove outliers and missing values. The processed data is feature extracted, and the mean, variance, maximum, minimum, and other statistical characteristics of the load are calculated, as well as the trend and periodicity of the load change. According to the extracted load change characteristics, combined with the operation requirements of the thermal system, the target value of load adjustment is determined. For the load adjustment target, according to the change rule of historical load data, the load adjustment stage is divided, and each stage corresponds to a time window. In each load adjustment stage, according to the historical load change rate, the load change rate range of this stage is set. According to the set load change rate and the divided load adjustment stage, the multi-stage load change trajectory is generated, and the load adjustment plan is formed. The temperature, pressure, flow and other parameter data are collected from the thermal system operation database, and the data is standardized. The coupling relationship model between parameters is established using multiple linear regression method, and the model parameters are estimated by least squares method. The dynamic response characteristics of each parameter are extracted from the thermal system parameter coupling model, and the response time of the parameter to the load change is calculated. The parameter response time is compared with the preset threshold value, if the response time exceeds the threshold value, the load change rate is adjusted, and the load adjustment stage is redivided. The long short-term memory network is used to predict the parameter change trend in the load adjustment process, and the prediction result is compared with the measured data to verify the model accuracy. According to the verified prediction model, the dynamic response model of the thermal system in the load adjustment process is generated.

[0028] Exemplarily, the thermal system database usually contains massive operation data, for example, a certain 600 MW unit generates more than 100,000 parameter records per day. In the data cleaning stage, the statistical feature-based outlier detection method is adopted to mark the data deviating from the mean value by more than three standard deviations as abnormal. For the missing values, the interpolation method is adopted according to the data variation trend to complete the data, ensuring the continuity of the data. In the load feature extraction stage, the unit operation mode needs to be concerned. The typical thermal power unit presents a double-peak feature on weekdays, with the load peaks of 580 MW and 550 MW at 8 am and 6 pm respectively, and the minimum load of about 300 MW at 3 am. On weekends, it presents a single-peak feature, and the load fluctuation is relatively flat. In the load adjustment stage, the unit characteristics need to be considered. Taking a certain unit as an example, the process of increasing from 300 MW to 500 MW is divided into three stages: the load growth rate is allowed to be 8 MW per minute in the initial adjustment stage, 6 MW per minute in the intermediate transition stage, and 4 MW per minute when approaching the target value, to ensure safety and stability. The parameter coupling relationship is reflected in multiple aspects. When the load increases from 400 MW to 450 MW, the main steam pressure increases from 26 MPa to 28 MPa within 3 minutes, then the main steam temperature increases from 585 degrees to 595 degrees within 7 minutes, and finally the reheat steam temperature increases from 570 degrees to 580 degrees within 12 minutes. The long short-term memory network can effectively predict the parameter variation trend. The network input includes the load change sequence of the previous four hours, and the output is the parameter prediction value of the next hour. In practical application, the prediction error of the model for the main steam pressure is controlled within 2%, and the prediction error for the temperature parameter is controlled within 3%. Through the dynamic response model, the adjustment strategy can be optimized. For example, when the unit starts in summer, the parameter responds faster due to the higher environmental temperature, and the load growth rate can be appropriately increased. The model shows that for every 5 degrees increase in environmental temperature, the main steam temperature response time is shortened by about 10%, and this feature can be used for real-time optimization of the control strategy. The parameter coupling effect is particularly obvious in the unit peak shaving process. When the load decreases rapidly, the main steam pressure decreases rapidly, but due to the thermal inertia effect, the temperature parameter has a lag in decreasing. The measured data shows that when the load decreases by 100 MW, the pressure decreases by 3 MPa within 5 minutes, while the temperature decreases by 20 degrees within 15 minutes. This lag effect needs to be considered in the dynamic response model.

[0029] In step S104, a dynamic compensation algorithm based on proportional-integral-derivative control is adopted to calculate the compensation amount according to the current parameter fluctuation at each load adjustment stage, adjust the load change rate, and suppress the parameter fluctuation.

[0030] The proportional-integral-derivative control algorithm is used to obtain the parameter fluctuation value in real time from the sensor, calculate the proportional value, integral value and derivative value. According to the proportional value, integral value and derivative value, the dynamic compensation amount is calculated using the formula. Through the dynamic compensation amount, the load change rate is adjusted, the parameter fluctuation is suppressed, and the stable parameter value is obtained. According to the stable parameter value, the completion state of the load adjustment stage is judged, and if the stable parameter value is within the preset stable range, it is determined to be completed. In each load adjustment stage, the parameter fluctuation value is monitored, and if the fluctuation value exceeds the preset fluctuation threshold, the compensation amount is recalculated. Using a multiple linear regression model, the parameter value and load value are extracted from historical data as features to train the model and obtain the response value. Through the response value, the gradient descent method is used to optimize the parameter settings of the proportional-integral-derivative control algorithm to improve the accuracy of load adjustment.

[0031] For example, the proportional-integral-derivative control algorithm mainly implements dynamic compensation for parameter fluctuation. Taking the main steam pressure fluctuation as an example, the difference between the pressure value collected by the sensor every second and the target value is the deviation value. When the pressure fluctuates from twenty-six megapascals to twenty-eight megapascals, the proportional control term responds quickly according to the deviation value, the integral control term accumulates historical deviation to eliminate steady-state error, and the derivative control term predicts the trend of deviation change. By setting the proportional coefficient to zero point six, the integral time to one hundred and twenty seconds, and the differential time to thirty seconds, the pressure value can be stabilized within five minutes. The dynamic compensation amount directly affects the load change rate. During the process of increasing the unit load from four hundred megawatts to four hundred and fifty megawatts, if the main steam temperature fluctuates greatly, the load change rate is automatically reduced. The measured data shows that for every five degrees of temperature exceeding the target value, the load growth rate decreases by twenty percent, ensuring smooth transition of parameters. The completion state of the load adjustment stage needs to consider multiple parameters. During the period of increasing the unit load from three hundred megawatts to five hundred megawatts, the main steam pressure fluctuation range is set to plus or minus zero point five megapascals, and the temperature fluctuation range is set to plus or minus three degrees. When all parameter values remain within the preset range for ten minutes, it is determined that the current adjustment stage is completed. The parameter fluctuation threshold management adopts a dynamic adjustment mechanism. For every ten degrees increase in ambient temperature, the pressure fluctuation threshold is tightened by zero point two megapascals, and the temperature fluctuation threshold is tightened by one degree. At the same time, according to the load change rate, the compensation coefficient is adjusted accordingly. For every ten percent increase in load change rate, the compensation coefficient increases by five percent. The multiple linear regression model extracts historical operating characteristics. Selecting recent operating data of a unit, taking the load value as the independent variable and the pressure and temperature as the dependent variables, a regression equation is established. The data shows that for every one hundred megawatts of load change, the pressure changes by three point two megapascals, the main steam temperature changes by fifteen degrees, and the reheat steam temperature changes by twelve degrees. When optimizing the control parameters using the gradient descent method, the mean square error of the actual response and the expected response is first calculated. By iteratively adjusting the proportional coefficient, integral time and differential time, the error is gradually reduced. The optimized parameters shorten the pressure regulation time by twenty percent and reduce the overshoot by thirty percent, significantly improving the response performance.

[0032] In step S105, according to the multi-stage load change trajectory and the dynamic compensation algorithm, a load control instruction is generated and sent to the boiler, the steam turbine and the feedwater system to coordinate the actions of the subsystems.

[0033] The multi-stage load change trajectory is obtained from historical data, and a load value and a trajectory value are extracted. A compensation value is calculated according to the load value and the trajectory value by using a PID control algorithm. According to the compensation value and the trajectory value, an instruction value is generated by using a fuzzy control algorithm and sent to the boiler, the steam turbine and the feedwater system. Related values are obtained from the boiler, the steam turbine and the feedwater system, and it is determined whether the related values are within a preset threshold range. If the related values are outside the preset threshold range, the compensation value is recalculated and the instruction value is updated by using the PID control algorithm. According to the updated instruction value, the boiler value, the steam turbine value and the feedwater value are adjusted by using the fuzzy control algorithm to coordinate the system action values. Through the coordinated action values, the completion state of the load adjustment stage is determined according to the stability of the related values, and a stable load value is obtained.

[0034] For example, the multi-stage load change trajectory reflects the actual operation path of the unit load adjustment. By extracting the corresponding relationship between the load value and the trajectory value in the historical data, a standard trajectory curve that conforms to the characteristics of the equipment can be obtained. For example, during the process of a certain unit increasing from 300 MW to 400 MW, the standard trajectory curve shows that the equipment operates most stably when the load increases by 2 MW per minute. Based on the extracted trajectory value, the proportional-integral-derivative control algorithm calculates a dynamic compensation value. When the unit load decreases from 400 MW to 350 MW, if the actual load value is 10 MW higher than the trajectory value, the controller will generate a negative compensation value to reduce the load change rate. Conversely, if the actual value is lower than the trajectory value, a positive compensation value is generated to speed up the adjustment. The fuzzy control algorithm synthesizes the instruction value according to the compensation value and the trajectory value. Taking the boiler system as an example, when the compensation value is negative and the trajectory value is in the descending interval, the given fuel quantity instruction is correspondingly reduced. In actual operation, for every 1% reduction in fuel quantity, the main steam pressure decreases by about 0.3 MPa. Similarly, the feedwater quantity instruction of the feedwater system also changes with the compensation value, and for every 1% reduction in feedwater quantity, the drum water level decreases by about 2 mm. The threshold management of the related values uses a dynamic adjustment mechanism. For every 10 degrees increase in ambient temperature, the pressure fluctuation threshold is tightened by 0.2 MPa, and the temperature fluctuation threshold is tightened by 1 degree. During the process of increasing the unit load from 300 MW to 500 MW, the main steam pressure fluctuation range is set to ±0.5 MPa, and the temperature fluctuation range is set to ±3 degrees, as the judgment criteria for related values. In the process of coordinated control, the action values of the boiler, turbine and feedwater system need to maintain a matching relationship. When the boiler fuel quantity increases by 1%, the feedwater quantity increases by 0.8% to 1.2% accordingly, to maintain the stability of the drum water level. At the same time, the turbine governing valve opening increases by 0.6% to 0.9%, to maintain the balance of the main steam pressure. This coordinated matching of action values ensures the mutual adaptation between systems. The completion status judgment of the load adjustment stage needs to consider multiple system parameters. When all related values remain within the preset range for 10 minutes, and the load value is stable within ±0.5% of the target value, it is determined that the current adjustment stage is completed. Through dynamic compensation and coordinated control, the load adjustment process is more stable, the parameter fluctuation is reduced, and the adjustment efficiency is improved.

[0035] In step S106, the peak shaving demand data is obtained from the grid dispatching system, the load adjustment time requirement is analyzed, and it is judged whether the load adjustment time meets the peak shaving demand. If not, the load change trajectory stage division is adjusted.

[0036] Peak-shaving demand data is obtained from the power grid dispatching system, and the peak-shaving demand and load values ​​are extracted. Based on the peak-shaving demand and load values, the ARIMA model is used to calculate the load adjustment time. The load adjustment time is compared with the preset peak-shaving demand threshold to determine whether the peak-shaving demand is met. If the load adjustment time does not meet the peak-shaving demand, the segmented linear regression method is used to adjust the stage division of the load change trajectory. Based on the adjusted stage division values, the least squares method is used to recalculate the trajectory value of the load change trajectory. A fuzzy control method is used to generate a load adjustment command value based on the trajectory value and the peak-shaving demand value. The load adjustment command value is sent to the load control system via the Modbus protocol to execute the load adjustment operation.

[0037] For example, the power grid dispatching system collects load changes at multiple time points and converts peak-shaving demand data into load adjustment indicators. For example, from 6:00 AM to 10:00 PM, load values ​​are collected every half hour to form a load time series. When a peak-shaving demand occurs, the current and target load values ​​are obtained, and the load adjustment interval and adjustment step size are calculated. The ARIMA model calculates the load adjustment time based on historical data. For example, a unit receives a peak-shaving demand to reduce power from 400 MW to 300 MW. By analyzing recent load adjustment data, it is predicted that this load adjustment will take 40 minutes to complete. Comparing the predicted value with the 30-minute peak-shaving demand indicates that it does not meet the requirement, necessitating optimization of the load change trajectory. The piecewise linear regression method divides the load change trajectory into multiple stages. For example, when a unit's load decreases from 500 MW to 300 MW, the original trajectory is divided into two stages: 500 to 400 MW and 400 to 300 MW. Regression analysis reveals turning points at 450 MW and 350 MW, and the trajectory is re-divided into three stages to achieve a smoother load change. The least squares method refits the load change trajectory based on the new stage division. In the 500-450 MW range, the load decreases by 3 MW per minute. In the 450-350 MW range, the load decreases by 4 MW per minute. In the 350-300 MW range, the load decreases by 2 MW per minute. This segmented adjustment method ensures the continuity of load changes while improving regulation speed. The fuzzy control method converts the trajectory value into a load adjustment command. When the actual load value exceeds the trajectory value, the controller outputs a negative adjustment value. If the actual load is 420 MW and the trajectory value is 410 MW, a negative adjustment value of 5 MW is output. Otherwise, a positive adjustment value is output to ensure that the load follows the trajectory. The load adjustment command is sent to the control module via the Modbus protocol. The protocol message contains information such as the load command value and the rate of change. After receiving the command, the control module adjusts parameters such as the coal and water supply accordingly. For every 10 MW decrease in load, the coal supply is reduced by 1-2 percent and the water supply by 0.8-1.5 percent to maintain stable unit operation. Through coordinated control, the load can smoothly transition to a new operating state.

[0038] Step S107, according to the adjusted load change trajectory, the load change rate of each stage is recalculated, the dynamic compensation algorithm parameters are updated, and a new load control instruction is generated.

[0039] The load basic data is obtained from the power grid dispatching system, and the load value and trajectory value are extracted using a data extraction tool. For the extracted load change trajectory value, the segmented linear regression method is used to calculate the load change rate value of each stage. The specific steps include segmenting the trajectory data, performing linear regression on each segment of data, and calculating the slope of each segment as the load change rate value. If the stage load change rate value exceeds the preset threshold, the dynamic compensation algorithm parameter value is updated according to the preset rule. According to the updated algorithm parameter value, the least square method is used to recalculate the load control instruction value. The specific steps include constructing a linear equation system and solving the least square solution. Using the fuzzy control method, the final control signal is generated by combining the load control instruction value and the compensation value. The specific steps include setting the fuzzy rule, fuzzy reasoning and de-fuzzification. The control signal is transmitted to the load control module through the Modbus protocol, and the load adjustment operation is performed. The specific steps include configuring the Modbus communication parameters and sending the control signal. According to the load adjustment result, the actual load data is collected, and it is judged whether the peak shaving demand is met. The specific steps include comparing the actual load value with the target load value, and judging whether the difference is within the allowed range.

[0040] For example, the power grid dispatching system collects load data of each substation and generator set in real time, and obtains load values and trajectory values through a data extraction tool. Taking a certain thermal power plant as an example, load data is collected every five minutes, including actual output, planned output and other information. Through a special data interface, the load change trajectory is extracted for nearly one hour to form time series data. The load trajectory is analyzed by piecewise linear regression method. For example, during the process of a certain unit from 500 MW to 300 MW, the slope change of adjacent data points is calculated, and a significant slope mutation is found at 450 MW and 350 MW, dividing the entire trajectory into three segments. The change rate of each segment is calculated by linear regression, the first segment is reduced by 2.5 MW per minute, the second segment is reduced by 3.5 MW per minute, and the third segment is reduced by 2 MW per minute. The dynamic compensation algorithm updates parameters based on preset rules. When the second segment load change rate exceeds the threshold of 3 MW per minute, the compensation mechanism is triggered. By increasing the time interval of this segment, the change rate is reduced to 2.8 MW per minute, ensuring the stability of load adjustment. The least square method recalculates the control command. According to the updated change rate, a linear equation set containing time and load value is constructed. The new load trajectory points are obtained by solving the equation to generate load command values every five minutes. For example, in the interval of 450 MW to 350 MW, the target load values every five minutes are 435, 420, 405, 390, 375 and 365 MW, respectively. The fuzzy controller converts the load command into a control signal. Set the fuzzy rules: when the actual load is higher than the command value and the difference increases, output a larger negative adjustment amount. For example, when the actual load is 420 MW and the command value is 405 MW, the difference is 15 MW and is increasing, output an adjustment signal of -8 MW. The control signal is sent to the load control module through the Modbus protocol. Configure communication parameters such as baud rate, data bits, etc., and use the function code write-in method to issue the control amount. The control module adjusts the coal supply and water supply after receiving the signal, and when the load is reduced by 10 MW, the coal supply is reduced by 1% to 2%. The actual load data is continuously collected for evaluation. Compared with the target value of 300 MW, when the actual load is in the range of 295 to 305 MW and remains stable, it is determined that the peak shaving demand is met. Through reasonable trajectory planning and accurate control strategy, the whole process not only ensures the timeliness of load adjustment, but also ensures the safe and stable operation of the unit.

[0041] In step S108, during the load adjustment process, key parameters are monitored in real time through sensors and data acquisition, and it is judged whether the parameters deviate from the safe range. If it deviates, a protection action is triggered to stop the load adjustment.

[0042] Key parameters are obtained by sensors and recorded in real time. According to the preset safety range, it is judged whether the parameter value exceeds the upper limit or is lower than the lower limit. The key parameters are thermal system parameters. If the parameter value deviates from the safety range, a protection action is triggered, and a stop instruction is generated. According to the stop instruction, the load adjustment process is interrupted, and the current adjustment state is recorded. The recorded adjustment state and parameter value are input into the Pandas library of Python for data cleaning and preprocessing. The isolated forest algorithm in the Scikit-learn library is used to analyze the preprocessed data, and the characteristics of the parameter deviation reason are extracted. According to the extracted characteristics, the linear regression model is used to update the safety range preset value, and the judgment condition is optimized. The updated safety range and historical parameter data are input into the ARIMA model in the Scikit-learn library to predict the parameter change trend. According to the prediction result, a new load adjustment strategy is generated by using the rule engine for the subsequent load adjustment process.

[0043] For example, during power plant operation, distributed control modules collect various key parameters, including turbine speed, boiler steam pressure, and feedwater flow, through distributed control modules. Sensors record parameter values ​​every second, forming continuous time series data. For a 600-megawatt unit, for example, the safe range for boiler main steam pressure is set between 16.8 and 17.6 MPa. When the pressure exceeds 17.6 MPa, it is determined to be an overpressure condition, triggering protection logic to automatically reduce the coal feed. At this point, the load adjustment process is paused, and the load value and related parameter status at the time of the pause are recorded. When cleaning the collected parameter data, obvious outliers, such as zero or out-of-range values ​​caused by sensor failure, are first removed. The data is then normalized to bring parameters of different dimensions to the same scale. For a particular unit, data for parameters such as steam pressure and feedwater flow over a single day is preprocessed to remove abnormal records with pressure values ​​below 10 MPa. The remaining data is then normalized. Using the isolation forest algorithm to analyze data anomalies, potential causes of parameter deviations can be identified. For example, during a unit's load reduction process, the main steam pressure fluctuated frequently. Analysis revealed this was related to a delayed response in the feedwater system. Adjustments to the feedwater pump speed control parameters were necessary to improve response speed. Furthermore, linear regression analysis based on historical data indicated that when the unit load was less than 300 megawatts, the upper safety pressure limit should be adjusted to 17.4 MPa to increase sensitivity. Time series forecasting can assess parameter trends. For example, using the steam pressure data for a particular unit over the past four hours, the forecasting model indicated a slow upward trend, predicting that the pressure would approach the upper limit in two hours. Based on this forecast, the rule engine generated a new adjustment strategy: adjusting the load reduction rate from 3 megawatts per minute to 2.5 megawatts per minute and appropriately increasing the feedwater flow rate to ensure the pressure remained within a safe range. When boiler water level deviations occurred, relevant operating data was recorded. For example, in the case of a unit experiencing high water levels, parameters such as the water level, feedwater flow rate, and steam flow rate at the time of the incident were recorded. After cleaning, the data revealed an abnormal opening of the feedwater regulating valve. Analysis revealed mechanical backlash in the valve actuator. Analysis using the isolation forest algorithm confirmed this as the primary cause of the high water level. Based on this information, the water supply control strategy was updated. Before a significant load reduction, the water supply valve opening was adjusted to a smaller position to prevent water level overshoot. Simultaneously, the prediction model indicated that if the current water supply strategy was maintained, the water level would continue to rise, necessitating further reductions in water supply. This dynamic optimization approach ensured safe unit operation while improving the stability of the regulation process.

[0044] Step S109 , based on the key parameters monitored in real time, the load adjustment effect is evaluated using deviation analysis and trend analysis methods to determine whether the load adjustment achieves the smooth transition goal. If not, the load change trajectory and dynamic compensation algorithm are optimized.

[0045] The data stream of the key parameter is acquired in real time by the sensor, the difference between the parameter value and the preset target value is calculated by the least square method, and the quantization result of the deviation degree is obtained. According to the deviation degree, the time series data of the parameter value is processed by the moving average method, and the characteristic values of the change direction and the change speed are determined. The change speed and the deviation degree are compared by using the preset threshold value, if the change speed exceeds the threshold value or the deviation degree exceeds the range, the amplitude of the load change is adjusted by the PID control algorithm, and the optimized adjustment trajectory is obtained. The control instruction of the load change is updated by the optimized adjustment trajectory, the periodic characteristics of the adjustment trajectory are extracted by using the Fourier transform, and the realization degree of the smooth transition is judged. According to the realization degree of the smooth transition, the historical data of the key parameter is grouped by using the K-means clustering algorithm, and the change direction characteristics corresponding to each group of parameter values are determined. By matching the change direction characteristics with the adjustment trajectory, the intermediate state of the load change is supplemented by using the linear interpolation method, and the continuous load adjustment sequence is obtained. According to the continuous load adjustment sequence, the future trend of the parameter value is predicted by using the ARIMA model, the preset target value of the next period is determined, and the dynamic compensation parameter is updated.

[0046] For example, a 600 MW unit in a power plant uses the least square method to monitor the difference between the main steam pressure value of the boiler and the target value of 17 MPa during operation. The pressure sensor samples once per second, and the deviation between the pressure value and the target value is calculated to determine the degree of pressure deviation. Based on the pressure data within ten minutes, a five-point moving average method is used to process the data to obtain a pressure change rate of 0.03 MPa per minute and an upward change direction. When the pressure change rate exceeds the preset threshold of 0.05 MPa per minute, the optimization control is started. The coal supply is adjusted through closed-loop proportional-integral-derivative control to reduce the pressure rise rate. According to actual operation experience, the proportional coefficient is set to 0.8, the integral time is set to 60 seconds, and the derivative time is set to 10 seconds, so that the pressure value gradually returns to the target value. Fourier transform analysis is performed on the pressure change trajectory during the adjustment process to extract the main periodic component of 300 seconds, indicating that the response is smooth. A 300 MW unit in a power plant performs cluster analysis on parameters such as turbine speed and power during peak shaving operation. Based on nearly four hours of operation data, a three-class clustering method is used to divide the operation into high-load, medium-load, and low-load conditions. The speed is 3,000 rpm and the power is greater than 270 MW in the high-load condition, and the speed fluctuates between 2,999 and 3,011 rpm in the medium-load condition. The intermediate state during load change is calculated through linear interpolation to achieve smooth transition of the speed. A gas turbine combined cycle unit uses a difference integrated moving average model to predict the boiler feedwater flow. Inputting nearly eight hours of feedwater flow data, the model shows that the feedwater demand will increase by 5% in the next two hours. Accordingly, the feedwater pump speed is adjusted in advance to avoid water level fluctuations. At the same time, combined with the load change trend, the compensation coefficient of the feedwater flow and the load is dynamically updated to improve the response accuracy. When the load of a circulating fluidized bed boiler unit is rapidly adjusted, the coal supply strategy is optimized by analyzing the bed temperature change characteristics. The least square method is used to calculate the deviation of the bed temperature from the target value of 850°C, and the change trend after moving average processing is used to determine the coal supply adjustment rate. When the bed temperature decreases at a rate of more than 1°C per minute, the coal supply is automatically increased. Fourier analysis shows that the bed temperature has a 10-minute fluctuation period, and the temperature fluctuation is suppressed by optimizing the coal supply delay parameter. Cluster analysis shows that the bed temperature and coal supply have different correlations in different load intervals, and the control parameters are adjusted adaptively to achieve dynamic stability of the bed temperature.

[0047] The above examples are only one of the preferred embodiments of the present application and should not be used to limit the protection scope of the present application, but any modification or embellishment made without substantial meaning within the main design idea and spirit of the present application, which still solves the same technical problem as the present application, should be included in the protection scope of the present application.

Claims

1. A flexible operation control method for ultra-supercritical thermal power units based on automatic start-stop, characterized in that: include: Obtain the thermal system parameters of the thermal power unit during the start-up and shutdown process, and establish a thermal system parameter coupling model; Identify the inertia lag characteristics of thermal system parameters through time series analysis and determine the lag time window during load adjustment; Using multi-stage load change trajectory, set the load change rate of each stage; A dynamic compensation algorithm based on proportional-integral-differential control is adopted to generate a load control instruction according to the multi-stage load change trajectory and the dynamic compensation algorithm; Obtain peak-shaving demand data from the power grid dispatching system to determine whether the load adjustment time meets the peak-shaving demand. If not, adjust the load change trajectory stage division; According to the adjusted load change trajectory, the load change rate of each stage is recalculated, the dynamic compensation algorithm parameters are updated, and new load control instructions are generated; The establishing of the thermal system parameter coupling model includes: Use the linear regression method in the Scikit-learn library to establish a thermodynamic system parameter coupling model between main steam pressure, main steam temperature, and reheat steam temperature; Determine the main steam pressure as the dependent variable, the main steam temperature and the reheat steam temperature as the independent variables, and calculate the regression coefficient and error term; If the fitting degree of the thermal system parameter coupling model is lower than the preset fitting threshold, the independent variable combination is adjusted, the regression coefficient is recalculated, and the thermal system parameter coupling model is optimized; The multi-stage load change trajectory includes: Obtain historical load data and perform data cleaning, preprocessing, and feature extraction; Determine the target value of load adjustment based on the extracted load variation characteristics and the operation requirements of the thermal system; For the load adjustment target value, the load adjustment stages are divided according to the changing pattern of historical load data; In each load adjustment stage, the load change rate range is set according to the historical load change rate; According to the set load change rate and the divided load adjustment stages, a multi-stage load change trajectory is generated.

2. The method according to claim 1, wherein The method of identifying the inertia hysteresis characteristics of thermal system parameters by using a time series analysis method includes: The main steam pressure, main steam temperature and reheat steam temperature data in the thermal system parameter coupling model are sampled to form a time series data set; The time series data set is decomposed using the STL decomposition method. Based on the decomposition results, the inertia lag characteristics of the thermal system parameters are calculated using the autocorrelation function.

3. The method according to claim 1, wherein The dynamic compensation algorithm based on proportional integral differential control includes: The proportional-integral-differential control algorithm is used to calculate the proportional value, integral value and differential value of the parameter fluctuation value obtained in real time; According to the proportional value, integral value and differential value, the dynamic compensation amount is calculated using the formula, and the load change rate is adjusted to obtain the stable parameter value; Based on the stable parameter value, the completion status of the load adjustment phase is determined.

4. The method according to claim 1, wherein The generating of the load control instruction comprises: Obtain multi-stage load change trajectories from historical data and extract load values ​​and trajectory values; Adopt PID control algorithm to calculate compensation value according to load value and trajectory value; According to the compensation value and trajectory value, the fuzzy control algorithm is used to generate the command value; Obtain thermal system parameters from boilers, turbines, and feedwater systems, and determine whether the thermal system parameters are within a preset threshold range; If the thermal system parameters exceed the threshold range, the compensation value is recalculated and the command value is updated; According to the updated command value, the fuzzy control algorithm is used to adjust the boiler value, steam turbine value and feed water value, and coordinate the action value. The coordinated action value refers to the unified control command value calculated after integrating the thermal system parameters and real-time deviations of the boiler, steam turbine and feed water system during the load adjustment process.

5. The method according to claim 1, wherein The determining whether the load adjustment time meets the peak load regulation requirement includes: Obtain peak-shaving demand data from the power grid dispatching system, extract peak-shaving demand values ​​and load values, and use the ARIMA model to calculate the load adjustment time value; The load adjustment time value is compared with the preset peak-shaving demand threshold. If the load adjustment time does not meet the peak-shaving demand, the stage division of the load change trajectory is adjusted.

6. The method according to claim 1, wherein The stage division of adjusting the load change trajectory refers to adjusting the stage division of the load change trajectory using a piecewise linear regression method.

7. The method according to claim 1, wherein The generating of a new load control instruction includes: Based on the extracted load change trajectory values, the load change rate values ​​at each stage are calculated using the piecewise linear regression method; If the stage load change rate value exceeds the preset threshold, the dynamic compensation algorithm parameter value is updated; According to the updated algorithm parameter values, the load control instruction value is recalculated using the least square method; The fuzzy control method is used to combine the load control command value and the compensation value to generate the final load control signal.

Images