Deep intelligent coordination method for thermal power generating unit

Abstract

The invention provides a deep intelligent coordination method for a thermal power generating unit, which comprises the following steps of: acquiring and preprocessing data, introducing a coefficient based on coal quality analysis and dynamically adjusting in a dry state coordination optimization stage, calculating and adjusting the coal feed quantity and the feed-forward quantity of the water feed quantity according to the conditions of the unit and a power grid by combining predictive control and an intelligent feed-forward technology, and improving the load adjusting capability. In the wet-dry and dry-wet conversion steps, the water level of a separator, steam and feed water flow are continuously monitored, conversion is triggered according to set conditions, equipment such as a feed pump is coordinated to act, and stable conversion is guaranteed. During wet state coordinated optimization, the water level of the separator and the feed water flow are stabilized by using a PID or intelligent algorithm, main steam pressure and load are adjusted by using model predictive control and the like, and meanwhile, a wet state characteristic and an equipment limitation constraint instruction are considered. According to the method, the intelligent coordination level of deep peak regulation of the thermal power generating unit is comprehensively improved, and the operation efficiency, stability and flexibility of the unit are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent coordinated operation of thermal power plants, and particularly to a method for deep intelligent coordination of thermal power units; Background Art In the current energy field, thermal power units are facing many severe challenges. On the one hand, fuel prices continue to rise, and the utilization hours of power generation and the average load are constantly decreasing, which has increased the operating pressure of power generation enterprises significantly. There is an urgent need to improve the operating efficiency of the units to reduce costs and enhance market competitiveness. On the other hand, the power grid has put forward higher requirements for thermal power units, and constraints such as deep peak shaving, fast frequency modulation, and desulfurization environmental protection are becoming increasingly strict. The traditional unit operation control methods are difficult to meet these requirements.

[0002] In terms of equipment maintenance, the traditional regular inspection method has obvious disadvantages. It not only consumes a large amount of manpower and time, but also is extremely difficult to inspect in dangerous and harsh environment areas, making it difficult to comprehensively and timely discover potential problems of equipment. The fixed inspection method is prone to over-maintenance or under-maintenance of equipment due to lack of accuracy, resulting in waste of resources or safety hazards. With the development of technology, the on-demand condition-based maintenance technology has gradually matured, and it has become an inevitable trend to achieve refined equipment maintenance and transformation of maintenance management.

[0003] In addition, the situation of work safety is severe, network risks are increasing continuously, information security issues are becoming more prominent, production safety accidents occur from time to time, and the management of personnel and tools also needs to be further improved.

[0004] The self-system of thermal power units is complex, with professional separation and information fragmentation. The DCS usually cannot provide an open advanced control implementation environment, resulting in the use of an external optimization control system as a last resort, which reduces the system response speed and reliability to a certain extent. The plant-level monitoring information system (SIS) is deployed in the production management area, and many functions closely related to the production operation of the unit, such as intelligent alarm, early warning, and operation optimization, are difficult to play their effective roles. At the same time, there are many operation and maintenance management services in the power plant, and there are information island phenomena among various systems. The traditional SIS + MIS information architecture can no longer meet the urgent needs of efficient operation and maintenance and refined management. In this context, it is extremely urgent to develop a method for deep intelligent coordination of thermal power units, aiming to solve the above problems through innovative technical means, achieve the production goals of thermal power units that are safe, efficient, clean, low-carbon, and flexible, and improve the overall operation level and economic benefits of the power generation industry. Summary of the Invention

[0005] The main object of the present invention is to provide a method for deep intelligent coordination of thermal power units, so as to solve the problem that in terms of equipment maintenance, the traditional regular inspection method has obvious disadvantages, not only consuming a large amount of manpower and time, but also being extremely difficult to inspect in dangerous and harsh environment areas, making it difficult to comprehensively and timely discover potential problems of equipment.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: a deep intelligent coordination method for thermal power units, and the method includes: S1. Data acquisition and preprocessing step: Real-time collect the operation data of the thermal power unit, filter and normalize the collected data, remove outliers and noise interference, so that the data is within the range and accuracy suitable for algorithm calculation; S2. Dry-state coordination optimization step: When the unit load is in the range of 30% - 100%, introduce the fuel calorific value correction coefficient and coal-water ratio adjustment coefficient based on coal quality analysis, and dynamically adjust according to the change of coal type; Use predictive control technology to combine the current operating state of the unit and the prediction of grid load demand, calculate and dynamically adjust the feedforward quantities of coal feeding amount and water feeding amount; Adopt intelligent feedforward technology, comprehensively consider the operating state of the coal mill and the performance factors of the feed water pump, and further optimize the feedforward control to improve the load regulation ability; S3. "Wet-dry" and "dry-wet" conversion step: Continuously monitor the separator water level, steam flow rate and feed water flow rate, set threshold values and change rate judgment conditions, and trigger the one-key automatic switching of the "wet-dry" or "dry-wet" process when the conditions are met; During the switching process, coordinate the actions of equipment such as the feed water pump, turbine control valve and burner to ensure the smooth and safe conversion process and maintain the stability of the main steam pressure and unit load; S4. Wet-state coordination optimization step: When the unit load is in the range of 25% - 30%, adjust the speed of the feed water pump or the opening of the regulating valve through PID control or intelligent control algorithm to maintain the stability of the separator water level and ensure the minimum feed water flow rate; Use strategies such as model predictive control, and jointly adjust the main steam pressure and unit load by the fuel quantity and the turbine control valve. At the same time, consider the characteristics of wet-state steam-water two-phase flow and the operation limitations of equipment, and constrain and correct the control instructions; In the preferred solution, the operation data collected from the thermal power unit includes the spray water flow rate at each stage, the coal feeding amount of each layer, the feed water flow rate, the steam flow rate, the steam pressure, the load command, the separator temperature, the opening of the turbine valve, the grid frequency, and the coal quality parameter information; In the preferred solution, for the spray water flow rate at each stage, a high-precision differential pressure flow sensor is used, and its measurement principle is based on the derivative formula of Bernoulli's equation: ; Among them, is the spray water flow rate, is the discharge coefficient, is the diameter ratio, is the expansibility coefficient, is the orifice opening diameter of the throttling element, is the fluid density, is the differential pressure before and after the throttling element; the spray water flow rate data at each stage calculated by this formula can accurately reflect the spray water condition of the unit under different working conditions; For the coal feeding amount, a coal amount measuring instrument based on the principle of nuclear radiation is used, which determines the coal amount according to the absorption characteristics of coal for rays: ; wherein, is the coal feeding amount, is the initial ray intensity, is the ray intensity after penetrating the coal, is the absorption coefficient of coal for rays, is the path length of the ray passing through the coal; in this way, the coal feeding amount of each layer can be accurately measured ; Filtering process: Apply the adaptive Kalman filtering algorithm to filter the collected data; First, establish the state space model of the data: ; ; wherein, is the system state vector at time is the state transition matrix, is the input control matrix, is the input vector, is the process noise vector, is the observation vector, is the observation matrix, is the observation noise vector; Then, continuously adjust the Kalman gain according to the real-time data: ; wherein, is the predicted covariance matrix, is the observation noise covariance matrix; Finally, update the system state estimate: ; Through this adaptive Kalman filtering method, the noise interference in the data can be effectively removed, making the data smoother and more accurate, especially suitable for processing the operation data of thermal power units with dynamic change characteristics; Normalization process: Adopt the normalization method based on the standard deviation and mean of the data; For a certain data sequence , its mean is , the standard deviation is , the normalized sequence The calculation formula is: ; Outlier handling: Use an outlier detection method based on locally weighted regression scatterplot smoothing; for a given data point , perform weighted regression fitting within its local neighborhood: ; Among them, is a weight function based on the distance from data point to , for example, a Gaussian weight function can be used: ; Among them is the bandwidth parameter, which is determined by methods such as cross-validation; Calculate the residual of the data point. For a given residual threshold , if , then is considered an outlier; for the detected outliers, according to the time series characteristics of the data and the distribution of the surrounding data, methods such as linear interpolation or mean substitution based on historical data can be used for processing to ensure the integrity and reliability of the data and provide high-quality input data for the subsequent intelligent coordination algorithm; In the preferred solution, in the dry-state coordination optimization step, the calculation method of the fuel calorific value correction coefficient is: A1. Use a coal quality analysis instrument to obtain the calorific value, volatile matter, and ash content parameters of coal in real time; A2. Input the parameters in the above step A1 into a calculation model based on machine learning or empirical formula established in advance, and calculate the fuel calorific value correction coefficient. This model is trained with a large amount of historical coal quality data and corresponding unit operation data and can be continuously self-adjusted according to new coal quality data; In the preferred solution, in the dry-state coordination optimization step, the implementation method of the predictive control technology is: B1. Establish a dynamic model of the unit operation state, including a mathematical relationship model between load, steam parameters, coal feed amount, and water supply amount. This model is constructed based on historical operation data and physical mechanism analysis; B2. According to the current unit operation parameters and the predicted value of the grid load demand, use the dynamic model to predict the unit operation state in the future for a period of time; B3. Calculate the set values of the coal feed amount and the water supply amount according to the prediction results, and continuously correct the prediction model and control instructions through a feedback mechanism; In the preferred solution, the calculation steps of the fuel calorific value correction coefficient are: A101: Use an advanced on-line coal quality analysis spectrometer to obtain the calorific value of coal ( ), volatile matter ( ), ash content ( ), parameters; The spectrometer is based on the principles of infrared absorption spectroscopy and X-ray fluorescence spectroscopy. By interacting light of a specific wavelength with the coal sample, it determines the content of coal quality components according to the spectral characteristics of absorption and emission. Its measurement accuracy can reach: calorific value error within , volatile matter error within , ash content error within ; A201: Input the obtained coal quality parameters into a calculation model that combines support vector regression (SVR) and locally weighted polynomial regression (LWPR); among them, the historical coal quality data and the corresponding unit operation data set , where is the fuel calorific value correction coefficient corresponding to the th group of data; First, use SVR to perform a preliminary global trend fitting on the data, and its objective function is: ; The constraint conditions are: ; where is the weight vector, and are slack variables, is the penalty parameter, is the insensitive loss function parameter, is the kernel function that maps the input data to a high-dimensional feature space; here we choose the Gaussian kernel function: ; Then, for the new coal quality data point , use LWPR to perform further local adjustment within its local neighborhood; assume there are data points in the neighborhood, and the locally weighted polynomial regression model is: ; where is the local weight, calculated through the Gaussian weight function: ; Here is the bandwidth parameter, determined through cross-validation; through this combined method, the global generalization ability of SVR and the local adaptive ability of LWPR can be fully utilized to accurately calculate the fuel calorific value correction coefficient to adapt to the impact of different coal types on unit operation; Implementation method of predictive control technology: B101: Based on the laws of conservation of mass, energy, and momentum, as well as the physical mechanisms of thermal processes, combined with historical operation data, a dynamic model of the unit operation state is established; for the relationships between the load ( ), main steam pressure ( ), main steam temperature ( ), and coal feed rate ( ), feed water flow rate ( ), they are described by a model consisting of the following partial differential equations and algebraic equations: ; ; ; ; where is the steam density, is the steam volume, is the steam flow rate, is the coal feed rate of the th burner, is the specific enthalpy of steam, is the lower calorific value of coal, is the enthalpy value of steam, is the heat released by combustion, and are functional relationships determined according to physical properties and experience; by discretizing these equations, they are transformed into an algebraic equation set suitable for computer calculation to construct a dynamic model of the unit operation state; B201: Let the current time be , given the current unit operation parameter vector and the predicted value of the grid load demand , where, , represents the predicted values for the next time instants; using the established dynamic model, the operation state vector of the unit for a period of time in the future is predicted through iterative calculation; The fourth-order Runge - Kutta method is used for numerical solution. For the state equation in the model, where is the control input vector, including the coal feed rate and the feed water flow rate, and its iterative formula is: ; where: ; ; ; ; Here is the time step; by this method, based on the current state and control input, the future operating states of the unit such as load and steam parameters can be predicted, providing a basis for setting the coal feeding amount and water feeding amount; B301: According to the predicted operating state of the unit, aiming to minimize the load deviation and steam parameter fluctuations, calculate the set values of the coal feeding amount and water feeding amount, and establish the objective function: ; where , , are the weight coefficients at different times, and are the reference values of the main steam pressure and main steam temperature; by solving this optimization problem, the set values of the coal feeding amount and water feeding amount and ; At the same time, by comparing the actually measured operating parameters of the unit with the predicted values, calculate the deviation vector ; using this deviation vector, the recursive least squares (RLS) method is used to online correct the parameters of the dynamic model; the update formula of RLS is: ; ; ; where is the estimated value vector of the model parameters, is the gain vector, is the covariance matrix, is the forgetting factor, is the regression vector containing input and output data, is the measured output vector; through this feedback correction mechanism, the accuracy of the dynamic model and the performance of the predictive control can be continuously improved, realizing the effective control of the unit operating state; In the preferred solution, in the "wet - dry" and "dry - wet" conversion steps, the setting method of the threshold and change rate judgment conditions is as follows: For the "wet - dry" conversion, when the separator water level is lower than the set value A and the ratio of the steam flow rate to the feed water flow rate is greater than the set value B, and at the same time the change rate of the steam flow rate is greater than the set value C1, the conversion is triggered; For the "dry - wet" conversion, when the separator water level is higher than the set value D and the ratio of the steam flow rate to the feed water flow rate is less than the set value E, and at the same time the change rate of the steam flow rate is less than the set value F, the conversion is triggered; Among them, the set values A, B, C1, D, E, and F are determined according to the design parameters, operation experience, and safety requirements of the unit, and can be adjusted according to the actual operation conditions of the unit; In the preferred solution, the monitoring and judgment condition setting steps in the "wet-dry" and "dry-wet" conversion steps are as follows: For the separator water level monitoring, a high-precision capacitive liquid level sensor is used, and its measurement principle is based on the non-linear relationship between the capacitance value and the liquid level height:

[0007] Among them, is the measured capacitance value, is the initial capacitance value, which is used to correspond to the capacitance at zero liquid level, and are coefficients determined through calibration experiments, is the separator water level height; by measuring the capacitance value in real time and using the above formula to solve for the water level height inversely, the high precision and reliability of water level monitoring are ensured; The steam flow rate and the feed water flow rate are measured by high-precision vortex flow meters, and their measurement principle is based on the relationship between the frequency generated by the Karman vortex street phenomenon and the flow rate: ; Among them, is the vortex street frequency, St is the Strouhal number, which is a constant for a specific flow meter structure and fluid properties, is the fluid flow velocity, is the characteristic dimension of the flow meter; by measuring the vortex street frequency and combining parameters such as the cross-sectional area of the pipeline, the steam flow rate and the feed water flow rate can be calculated; Set the "wet-dry" conversion condition: when the separator water level is lower than the set value , and the ratio of the steam flow rate to the feed water flow rate , and at the same time the change rate of the steam flow rate is less than 1, trigger the "wet-dry" conversion; the set value ensures that the conversion is started when the separator water level is low, and the ratio and the change rate less than 1 are set to comprehensively judge whether the operation state of the unit is suitable for the "wet-dry" conversion and prevent mis-triggering; Set the "dry-wet" conversion condition: when the separator water level is higher than the set value , and the ratio of the steam flow rate to the feed water flow rate , and at the same time the change rate of the steam flow rate When this condition is met, the "dry - wet" conversion is triggered; similarly, and and are set based on the unit characteristics and operation experience to ensure that the conversion occurs under appropriate operating conditions and guarantee the safe and stable operation of the unit; Conversion process control: When the "wet - dry" conversion is triggered, a method combining model predictive control (MPC) and fuzzy logic control is adopted to coordinate the actions of the feed water pump, turbine control valve, and burner equipment; First, a dynamic model of the unit during the "wet - dry" conversion is established, considering mass and energy conservation as well as the dynamic characteristics of the equipment. For the dynamic change model of the main steam pressure : ; wherein, is the coal feeding rate, is the feed water flow rate, is the opening of the turbine control valve, and and and etc. are coefficients determined through system identification methods; this model describes the dynamic relationship between the main steam pressure and each control variable during the conversion process; Then, MPC is used to predict the operating state of the unit over a period of time in the future, and a control sequence is formulated based on the prediction results with the objective of minimizing the main steam pressure fluctuation and maintaining the unit load stable: ; wherein, and are the predicted main steam pressure and unit load, and are the reference values, is the weight coefficient, is the prediction horizon; the optimal set value sequence of control variables such as the coal feeding rate, feed water flow rate, and opening of the turbine control valve is obtained by solving this optimization problem; Meanwhile, fuzzy logic control is used to handle some non - linear factors and uncertainties that are difficult to accurately model; for example, fuzzy sets and fuzzy rules are defined to fine - tune the control variables according to factors such as the steam flow rate change rate and the separator water level change rate; For the adjustment of the opening of the turbine control valve, the following fuzzy rules are as follows: If the steam flow rate change rate is "positive large" and the separator water level change rate is "negative small", then the opening of the turbine control valve "increases moderately"; Here, "positive large", "negative small", and "moderate increase" are all fuzzy language variables. The fuzzy control output is converted into an actual control quantity through fuzzy inference and defuzzification methods, and combined with the control quantity obtained by MPC to jointly control the device action, ensuring a smooth and safe conversion process; When triggering the "dry - wet" conversion, the control method is similar, but the dynamic model and control strategy need to be adjusted according to the physical process of the "dry - wet" conversion and the characteristics of the unit; For the control of feed - water flow, during the "dry - wet" conversion process, it is necessary to quickly increase the feed - water flow to maintain the separator water level, and its dynamic model can be expressed as: ; Among them, 、 、 are coefficients; similarly, using the method of combining MPC and fuzzy logic control, a control strategy is formulated to coordinate the actions of each device, maintain the stability of the main steam pressure and unit load, and achieve a smooth "dry - wet" conversion process.

[0008] In the preferred solution, in the wet - state coordination and optimization step, the intelligent control algorithm is fuzzy control, and its implementation process includes: defining the fuzzy subsets and corresponding membership functions of the separator water level, and the fuzzy control rules for the speed of the feed - water pump or the opening of the regulating valve; According to the real - time monitored separator water level, calculate the fuzzy control quantity of the speed of the feed - water pump or the opening of the regulating valve through fuzzy inference; Perform defuzzification processing on the fuzzy control quantity to obtain the actual control output, and achieve precise control of the separator water level; In the preferred solution, for the separator water level, the defined fuzzy subsets are {very low (VL), low (L), moderate (M), high (H), very high (VH)}; Use Gaussian - type membership functions to describe each fuzzy subset, and the membership function of the fuzzy subset is: ; Among them, is the separator water level, is the central value of the water level of the "fuzzy subset", is the corresponding standard deviation; these parameters are determined through experiments or data analysis according to the normal operating water - level range and control - precision requirements of the unit; Formulation of fuzzy control rules: Formulate fuzzy control rules based on operating experience and physical principles; Among them, if the separator water level is "very low (VL)", then the speed of the feed - water pump is "maximum (MAX)" and the opening of the regulating valve is "fully open (FULL)"; If the separator water level is "low (L)", the feed pump speed is "large (LARGE)" and the regulating valve opening is "mostly open (MOSTLY_OPEN)". And so on, a complete fuzzy control rule table is formed; these rules reflect the control strategies for the feed pump speed and the regulating valve opening under different water level conditions to achieve the goal of maintaining the stability of the separator water level. Fuzzy inference and defuzzification: When the separator water level is monitored in real time, calculate its membership degrees on each fuzzy subset according to the defined membership functions , , , , ; Then, adopt the Mamdani fuzzy inference method to perform inference according to the fuzzy control rules. The same applies to the regulating valve opening. Finally, perform defuzzification processing using the centroid method to obtain the actual control outputs of the feed pump speed and the regulating valve opening ; the defuzzification formula for the feed pump speed is: ; where represents different fuzzy output values, is the corresponding quantization value; through this fuzzy control method, the nonlinearity and uncertainty in the separator water level control can be effectively handled to achieve precise control. Main steam pressure and unit load regulation: Model predictive control MPC model establishment: Considering the wet steam-water two-phase flow characteristics and equipment operation limitations, establish the dynamic models of the main steam pressure and the unit load ; for the dynamic change of the main steam pressure, based on the mass and energy conservation and the two-phase flow pressure drop model, there is: ; where is the fuel quantity, is the steam turbine governing valve opening, is the feed water flow rate, is the specific enthalpy of steam, is the steam density, is the water density, etc.; the function here is a complex functional relationship obtained through physical analysis of the wet steam-water two-phase flow process and fitting of experimental data; similarly, the dynamic model for the unit load is: ; where such as steam temperature, is a function It is also established based on physical principles and experimental data; MPC Prediction and Control: According to the current unit operation status , use the established dynamic model to predict the change trajectories of the main steam pressure and unit load within a certain period in the future , ; The prediction method adopted can be numerical solution based on the Runge - Kutta method. For example, for the prediction equation of the main steam pressure: ; Among them, is the time step, , , , are the coefficients of the Runge - Kutta method, obtained by calculating , and here is calculated according to the above - established dynamic model; Taking minimizing the main steam pressure fluctuation and maintaining the unit load stability as the objective function, the formula is: ; Among them, and are the reference values of the main steam pressure and unit load, and are the weight coefficients at different moments, reflecting the degree of emphasis on the control accuracy at different future moments; By solving this optimization problem, the optimal control sequences of the fuel quantity and the steam turbine throttle valve opening and are obtained, ; In actual control, only take the control quantities at the current moment and to act on the unit, and then repeat the above - mentioned prediction and optimization process at the next moment to achieve closed - loop control; Constraint Handling: Considering the equipment operation limitations, constrain the control instructions; For the fuel quantity , there is a lower limit and an upper limit , which are determined according to the minimum and maximum combustion capabilities of the combustion equipment; For the steam turbine throttle valve opening , there is an opening limit , which is determined by the structure and safe operation requirements of the steam turbine; During the MPC solution process, by adding these constraint conditions to the optimization problem and using a constrained optimization algorithm for solution, ensure that the obtained control instructions are within the feasible range to guarantee the safe and stable operation of the unit.

[0009] The present invention provides a deep intelligent coordination method for thermal power units. In terms of load regulation, the unit can achieve a faster load change rate of more than 1.5% Pe / min, and has a higher load regulation accuracy, which can effectively meet the strict requirements of deep peak regulation and rapid frequency regulation of the power grid, ensure the stable operation of the unit in a complex power grid environment, and enhance the overall stability and reliability of the power system. The lower limit of its load regulation range can be significantly reduced from the original 40% to 25% or even lower, which broadens the operating flexibility of the unit, enables the thermal power unit to efficiently adapt to different power demand scenarios, and improves the utilization efficiency of power resources.

[0010] In terms of operation optimization, through intelligent coordinated control, such as the introduction of advanced means such as fuel calorific value correction and coal-water ratio adjustment in dry coordinated optimization, and full consideration of steam-water two-phase flow characteristics and equipment operation restrictions in wet coordinated optimization, the overall operation efficiency of the unit has been effectively improved. In terms of boiler combustion optimization, the average boiler efficiency can be improved by about 0.5%, coal consumption can be reduced by 1.5-2g, and flue gas NOx emissions can be reduced by about 10%-20%, achieving the important goal of energy conservation and emission reduction, reducing pollution to the environment, and promoting the development of thermal power units towards clean and low-carbon.

[0011] In the field of equipment maintenance and safety assurance, with the help of intelligent detection and diagnosis technology, the equipment status can be monitored more accurately. For example, intelligent monitoring technology can detect hidden equipment failures in advance, such as coal shortage at the coal feeder inlet, sudden change in main heat steam temperature, abnormal ammonia injection flow in the denitrification system, etc. Compared with manual monitoring, it can provide early warning 1-30 minutes or even longer in advance, thus buying precious time for equipment maintenance and troubleshooting, effectively reducing equipment failure rate, reducing the number of unplanned shutdowns, extending equipment service life, ensuring safe and stable operation of the unit, reducing the risk of production safety accidents, and improving the overall safety and reliability of thermal power units.

[0012] At the operation and maintenance management level, through an integrated intelligent coordination method, the information islands between systems are broken, efficient data sharing and deep integration of business are achieved, and the efficiency and refinement of operation and maintenance management are improved. For example, the integrated air-ground joint inspection solution in intelligent maintenance management uses advanced equipment such as drones and quadruped robots combined with intelligent algorithms to achieve comprehensive and refined inspections of high, medium and low altitudes of power stations, solving the problems of traditional inspections, improving the accuracy and timeliness of inspections, and optimizing the maintenance management process. It realizes the digitization, process and standardization of maintenance business, reduces labor costs, and improves the overall benefits of production management. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1It is the flow chart of the intelligent coordination method for deep regulation of the present invention; Detailed implementation mode Embodiment 1 As Figure 1 shown, an intelligent coordination method for deep regulation of a thermal power unit, the method includes: S1. Data acquisition and preprocessing step: Real-time collect the operation data of the thermal power unit, filter and normalize the collected data, remove outliers and noise interference, so that the data is within the range and accuracy suitable for algorithm calculation; S2. Dry-state coordination optimization step: When the unit load is in the range of 30% - 100%, introduce the fuel calorific value correction coefficient and coal-water ratio adjustment coefficient based on coal quality analysis, and dynamically adjust according to the change of coal type; Use predictive control technology to combine the current operation state of the unit and the prediction of grid load demand, calculate and dynamically adjust the feedforward quantities of coal feeding amount and water supply amount; Adopt intelligent feedforward technology, comprehensively consider the operation state of the coal mill and the performance factors of the feed water pump, and further optimize the feedforward control to improve the load regulation ability; S3. "Wet-dry" and "dry-wet" conversion step: Continuously monitor the separator water level, steam flow rate and feed water flow rate, set threshold values and change rate judgment conditions, and trigger the one-key automatic switching of the "wet-dry" or "dry-wet" process when the conditions are met; During the switching process, coordinate and control the actions of equipment such as the feed water pump, turbine control valve and burner to ensure the smooth and safe conversion process and maintain the stability of the main steam pressure and unit load; S4. Wet-state coordination optimization step: When the unit load is in the range of 25% - 30%, adjust the speed of the feed water pump or the opening of the regulating valve through PID control or intelligent control algorithm to maintain the stability of the separator water level and ensure the minimum feed water flow rate; Use strategies such as model predictive control, and jointly regulate the main steam pressure and unit load by the fuel quantity and the turbine control valve. At the same time, considering the characteristics of wet-state steam-water two-phase flow and the operation limitations of equipment, the control instructions are constrained and corrected.

[0014] Data acquisition and preprocessing: Real-time collect the operation data of the thermal power unit, such as the spray water flow rate at all levels, the coal feeding amount of each layer, the feed water flow rate, the steam flow rate, the steam pressure, the load instruction, the separator temperature, the opening of the turbine valve, the grid frequency, and the coal quality parameter information. The collected data is filtered by using the adaptive Kalman filter algorithm. Its principle is to establish a state space model of the data, continuously adjust the Kalman gain according to the real-time data, and then update the system state estimation, effectively removing the noise interference in the data, making the data smoother and more accurate, and suitable for processing the dynamic change data of the unit. At the same time, adopt the normalization method based on the standard deviation and mean of the data and the local weighted regression scatter smoothing method to process outliers to ensure the integrity and reliability of the data and provide high-quality input for the subsequent algorithm.

[0015] Dry-state coordination optimization (30% - 100% load): When the unit is within this load range, after obtaining the calorific value, volatile matter, and ash content parameters of the coal using a coal quality analysis instrument, input them into a calculation model based on machine learning or empirical formulas established in advance to calculate the fuel calorific value correction coefficient. This model is trained with a large amount of historical data and can be adaptively adjusted. Then, combined with predictive control technology, based on the current operating state of the unit and the predicted value of the grid load demand, use a dynamic model constructed based on the laws of conservation of mass, energy, and momentum and the physical mechanism of the thermal process to predict the operating state of the unit, and then calculate and dynamically adjust the feedforward quantities of the coal feeding amount and the water feeding amount. At the same time, use intelligent feedforward technology to comprehensively consider the operating state of the coal mill and the performance factors of the feed water pump to further optimize the feedforward control and improve the load regulation ability.

[0016] "Wet-dry" and "dry-wet" conversion: Continuously monitor the separator water level, steam flow rate, and feed water flow rate, and trigger the conversion when specific threshold and change rate judgment conditions are met. For example, the "wet-dry" conversion condition is that the separator water level is lower than set value A, the ratio of the steam flow rate to the feed water flow rate is greater than set value B, and the change rate of the steam flow rate is greater than set value C1; the "dry-wet" conversion condition is the opposite. During the conversion process, use a method combining model predictive control MPC and fuzzy logic control to coordinate the actions of equipment such as the feed water pump, steam turbine control valve, and burner. First, establish a dynamic model of the conversion process, use MPC to predict the operating state of the unit and formulate a control sequence, and at the same time use fuzzy logic control to handle non-linear factors to ensure a smooth and safe conversion and maintain the stability of the main steam pressure and the unit load.

[0017] Wet-state coordination optimization (25% - 30% load): Within this load range, adjust the speed of the feed water pump or the opening of the regulating valve through PID control or intelligent control algorithms (such as fuzzy control) to maintain the stability of the separator water level and ensure the minimum feed water flow rate. For fuzzy control, first define the fuzzy subsets, membership functions, and corresponding control rules of the separator water level, perform fuzzy inference calculation of the control quantity based on the real-time monitored water level, and then obtain the actual control output through defuzzification. At the same time, use strategies such as model predictive control, consider the characteristics of wet-state steam-water two-phase flow and equipment operation limitations to establish dynamic models of the main steam pressure and the unit load, and accordingly adjust the main steam pressure and the unit load, and constrain and correct the control instructions.

[0018] Example 2 Further described in combination with Example 1, as Figure 1 shown, the operating data collected from the thermal power unit include the spray water flow rate at each stage, the coal feeding amount of each layer, the feed water flow rate, the steam flow rate, the steam pressure, the load instruction, the separator temperature, the steam turbine valve opening, the grid frequency, and the coal quality parameter information; In the preferred solution, high-precision differential pressure flow sensors are used for the water spray flow rates at all levels. Its measurement principle is based on a derivative formula of Bernoulli's equation: ; Among them, is the water spray flow rate, is the discharge coefficient, is the diameter ratio, is the expansibility coefficient, is the orifice diameter of the throttling element, is the fluid density, is the differential pressure before and after the throttling element; the water spray flow rate data at all levels calculated by this formula can accurately reflect the water spray situation of the unit under different working conditions; For the coal feeding amount, a coal amount measuring instrument based on the nuclear radiation principle is used, which determines the coal amount according to the absorption characteristics of coal for rays: ; Among them, is the coal feeding amount, is the initial ray intensity, is the ray intensity after penetrating the coal, is the absorption coefficient of coal for rays, is the path length of the ray passing through the coal; in this way, the coal feeding amount of each layer can be accurately measured ; Filtering process: The adaptive Kalman filtering algorithm is applied to filter the collected data; First, establish the state space model of the data: ; ; Among them, is the system state vector at time is the state transition matrix, is the input control matrix, is the input vector, is the process noise vector, is the observation vector, is the observation matrix, is the observation noise vector; Then, continuously adjust the Kalman gain according to the real-time data: ; Among them, is the predicted covariance matrix, is the observation noise covariance matrix; Finally, update the system state estimate: ; Through this adaptive Kalman filtering method, noise interference in the data can be effectively removed, making the data smoother and more accurate, especially suitable for processing the operating data of thermal power units with dynamic change characteristics; Normalization processing: Adopt a normalization method based on the standard deviation and mean of the data; for a certain data sequence , its mean is , and the standard deviation is . The formula for the normalized sequence is: ; Outlier processing: Use an outlier detection method based on the locally weighted regression scatter plot smoothing method; for a given data point , perform weighted regression fitting within its local neighborhood: ; Among them, is a weight function based on the distance from the data point to . For example, a Gaussian weight function can be adopted: ; Among them is the bandwidth parameter, which is determined by methods such as cross-validation; Calculate the residual of the data point. For a given residual threshold , if , then is considered an outlier; for the detected outliers, according to the time series characteristics of the data and the distribution of the surrounding data, methods such as linear interpolation or mean substitution based on historical data can be used for processing to ensure the integrity and reliability of the data and provide high-quality input data for the subsequent intelligent coordination algorithm.

[0019] It further illustrates that high-precision differential pressure flow sensors are used for the spray water flow at all levels in the data acquisition part. It is measured based on the derivative formula of Bernoulli's equation and can accurately reflect the spray water conditions of the unit under different operating conditions; the coal feeding amount is measured by a measuring instrument based on the principle of nuclear radiation, which can accurately measure the coal feeding amount of each layer. In the data processing link, the establishment process of the state space model of the adaptive Kalman filtering algorithm is elaborated in detail, including the functions of the state transition matrix, input control matrix, etc., and how to update the system state estimation by calculating the Kalman gain, highlighting its advantages in dealing with the dynamic change characteristics of the operating data of thermal power units, and emphasizing again the importance of data processing for the subsequent intelligent coordination algorithm to ensure that the data is within the appropriate calculation range and accuracy.

[0020] Example 3 Further illustrate in combination with Example 1, asFigure 1 As shown, in the dry state coordination and optimization step, the calculation method of the fuel calorific value correction coefficient is as follows: A1. Use a coal quality analysis instrument to obtain the calorific value, volatile matter, and ash parameters of coal in real time; A2. Input the parameters in step A1 into a calculation model based on machine learning or empirical formula established in advance, and calculate the fuel calorific value correction coefficient. This model is trained with a large amount of historical coal quality data and corresponding unit operation data, and can continuously adaptively adjust according to new coal quality data; In the preferred solution, in the dry state coordination and optimization step, the implementation method of the predictive control technology is as follows: B1. Establish a dynamic model of the unit operation state, including the mathematical relationship model between load, steam parameters, coal feed amount, and water supply amount. This model is constructed based on historical operation data and physical mechanism analysis; B2. According to the current unit operation parameters and the predicted value of the power grid load demand, use the dynamic model to predict the unit operation state in the future for a period of time; B3. Calculate the set values of the coal feed amount and water supply amount according to the prediction results, and continuously correct the prediction model and control instructions through a feedback mechanism; In the preferred solution, the calculation steps of the fuel calorific value correction coefficient are as follows: A101: Use an advanced on-line coal quality analysis spectrometer to obtain the calorific value of coal ( ), volatile matter ( ), and ash ( ) parameters in real time; This spectrometer is based on the principles of infrared absorption spectroscopy and X-ray fluorescence spectroscopy. Through the interaction of light with a specific wavelength and the coal sample, the content of coal quality components is determined according to the absorption and emission spectral characteristics; Its measurement accuracy can reach: the calorific value error is within , the volatile matter error is within , and the ash error is within ; A201: Input the obtained coal quality parameters into a calculation model combined with support vector regression (SVR) and locally weighted polynomial regression (LWPR); Among them, the historical coal quality data and the corresponding unit operation data set , where is the fuel calorific value correction coefficient corresponding to the th group of data; First, use SVR to perform a preliminary global trend fitting on the data, and its objective function is: ; The constraint condition is: ; where is the weight vector, and are slack variables, is the penalty parameter, is the insensitive loss function parameter, is the kernel function that maps the input data to a high-dimensional feature space; here we choose the Gaussian kernel function: ; Then, for the new coal quality data point , LWPR is used for further local adjustment within its local neighborhood; assuming there are data points in the neighborhood, the locally weighted polynomial regression model is: ; where is the local weight, calculated by the Gaussian weight function: ; here is the bandwidth parameter, determined by cross-validation; through this combined approach, the global generalization ability of SVR and the local adaptability of LWPR can be fully utilized to accurately calculate the fuel calorific value correction coefficient to adapt to the impact of different coal types on the unit operation; Implementation method of predictive control technology: B101: Based on the laws of conservation of mass, energy, and momentum and the physical mechanism of the thermal process, combined with historical operation data, a dynamic model of the unit operation state is established; for the relationship between the load ( ), main steam pressure ( ), main steam temperature ( ) and the coal feeding amount ( ), feed water amount ( ), it is described by the following model composed of partial differential equations and algebraic equations: ; ; ; ; where is the steam density, is the steam volume, is the steam flow rate, is the coal feeding amount of the th burner, is the specific enthalpy of the steam, is the lower calorific value of the coal, is the enthalpy value of the steam, is the heat released by combustion, and is a functional relationship determined according to physical properties and experience; by discretizing these equations, they are transformed into an algebraic equation set suitable for computer calculation, and a dynamic model of the unit operation state is constructed; B201: Let the current time be , and the current unit operation parameter vector and the predicted value of the grid load demand are known, where represents the predicted value for the next time instants; using the established dynamic model, the operation state vector of the unit for a period of time in the future is predicted through iterative calculation; The fourth-order Runge - Kutta method is used for numerical solution. For the state equation in the model, where is the control input vector, including the coal feeding amount and the water feeding amount, and its iterative formula is: ; where: ; ; ; ; Here is the time step; through this method, the operation states such as the future unit load and steam parameters can be predicted based on the current state and control input, providing a basis for setting the coal feeding amount and the water feeding amount; B301: According to the predicted unit operation state, with the goal of minimizing the load deviation and steam parameter fluctuation, calculate the set values of the coal feeding amount and the water feeding amount, and establish the objective function: ; where , , are the weight coefficients at different times, and are the reference values of the main steam pressure and the main steam temperature; by solving this optimization problem, the set values and of the coal feeding amount and the water feeding amount are obtained; Meanwhile, by comparing the actually measured unit operation parameters with the predicted values, the deviation vector is calculated; using this deviation vector, the recursive least squares method RLS is used to online correct the parameters of the dynamic model; the update formula of RLS is: ; ; ; wherein is the estimated value vector of the model parameters, is the gain vector, is the covariance matrix, is the forgetting factor, is the regression vector containing input and output data, is the measured output vector; through this feedback correction mechanism, the accuracy of the dynamic model and the performance of predictive control can be continuously improved, and the effective control of the unit operation state can be realized.

[0021] In the dry-state coordination optimization step, the calculation method of the fuel calorific value correction coefficient is described in more detail. An advanced on-line coal quality analysis spectrometer is used to obtain coal quality parameters, which is based on the principles of infrared absorption spectroscopy and X-ray fluorescence spectroscopy and has high measurement accuracy. In the calculation model, first, support vector regression (SVR) is used to perform global trend fitting. Through its objective function and constraint conditions (involving weight vector, slack variable, penalty parameter, etc.) and the selected Gaussian kernel function, the historical data is preliminarily processed, and then local weighted polynomial regression (LWPR) is used to further adjust in the local neighborhood of the new coal quality data points, giving full play to the global generalization ability of SVR and the local adaptability ability of LWPR. In terms of predictive control technology, the composition forms of partial differential equations and algebraic equations for constructing the dynamic model of the unit operation state based on physical mechanisms and historical data are described in detail, such as the relationship model between load, steam parameters, coal feed rate, and feed water flow rate, and the iterative formula for numerically solving the state equation in the model using the fourth-order Runge-Kutta method, as well as the objective function for calculating the set values of coal feed rate and feed water flow rate according to the prediction results and the update formula for online correction of the dynamic model parameters using the recursive least squares (RLS) method, comprehensively showing the technical details and implementation process of dry-state coordination optimization.

[0022] Embodiment 4 Further illustrated in combination with Embodiment 1, as Figure 1 shown, in the "wet-dry" and "dry-wet" conversion steps, the setting method of the threshold and change rate judgment conditions is as follows: For the "wet-dry" conversion, when the separator water level is lower than the set value A, the ratio of steam flow to feed water flow is greater than the set value B, and the change rate of steam flow is greater than the set value C1, the conversion is triggered; For the "dry-wet" conversion, when the separator water level is higher than the set value D, the ratio of steam flow to feed water flow is less than the set value E, and the change rate of steam flow is less than the set value F, the conversion is triggered; Among them, the set values A, B, C1, D, E, and F are determined according to the design parameters, operation experience, and safety requirements of the unit, and can be adjusted according to the actual operation conditions of the unit; The following are more detailed steps of S3: The monitoring and judgment condition setting steps in the "wet - dry" and "dry - wet" conversion steps are as follows: For the separator water level monitoring, a high - precision capacitive liquid level sensor is used, and its measurement principle is based on the non - linear relationship between the capacitance value and the liquid level height:

[0023] Among them, is the measured capacitance value, is the initial capacitance value, used to correspond to the capacitance at zero liquid level, and are coefficients determined through calibration experiments, is the separator water level height; by measuring the capacitance value in real - time and using the above formula to solve for the water level height inversely, the high precision and reliability of water level monitoring are ensured; The steam flow rate and the feed water flow rate are measured using a high - precision vortex - street flowmeter, and its measurement principle is based on the relationship between the frequency generated by the Karman vortex street phenomenon and the flow rate: ; Among them, is the vortex - street frequency, St is the Strouhal number, which is a constant for a specific flowmeter structure and fluid properties, is the fluid flow velocity, is the characteristic dimension of the flowmeter; by measuring the vortex - street frequency and combining parameters such as the cross - sectional area of the pipeline, the steam flow rate and the feed water flow rate can be calculated; Set the "wet - dry" conversion condition: When the separator water level is lower than the set value , and the ratio of the steam flow rate to the feed water flow rate , and at the same time the change rate of the steam flow rate < 1, trigger the "wet - dry" conversion; the set value ensures that the conversion is started when the separator water level is low, and the ratio and the change rate < 1 are set to comprehensively judge whether the operation state of the unit is suitable for the "wet - dry" conversion and prevent false triggering; Set the "dry - wet" conversion condition: When the separator water level is higher than the set value , and the ratio of the steam flow rate to the feed water flow rate , and at the same time the change rate of the steam flow rate When this occurs, the "dry - wet" conversion is triggered; similarly, , , The settings are based on the unit characteristics and operating experience to ensure that the conversion is carried out under appropriate working conditions and to guarantee the safe and stable operation of the unit; Conversion process control: When the "wet - dry" conversion is triggered, a method combining model predictive control (MPC) and fuzzy logic control is adopted to coordinate the actions of the feed water pump, turbine control valve, and burner equipment; First, a dynamic model of the unit during the "wet - dry" conversion is established, considering mass and energy conservation as well as the dynamic characteristics of the equipment. For the dynamic change model of the main steam pressure : ; Among them, is the coal feeding rate, is the feed water flow rate, is the opening of the turbine control valve, , , , etc. are coefficients determined through system identification methods; this model describes the dynamic relationship between the main steam pressure and each control variable during the conversion process; Then, MPC is used to predict the operating state of the unit in the future for a period of time, and a control sequence is formulated according to the prediction results with the goal of minimizing the main steam pressure fluctuation and maintaining the unit load stable: ; Among them, and are the predicted main steam pressure and unit load, and are the reference values, is the weight coefficient, is the prediction horizon; the optimal set value sequence of control variables such as the coal feeding rate, feed water flow rate, and turbine control valve opening is obtained by solving this optimization problem; At the same time, fuzzy logic control is used to handle some non - linear factors and uncertainties that are difficult to accurately model; for example, fuzzy sets and fuzzy rules are defined to fine - tune the control variables according to factors such as the steam flow rate change rate and the separator water level change rate; For the adjustment of the turbine control valve opening, there are the following fuzzy rules: If the steam flow rate change rate is "positive large" and the separator water level change rate is "negative small", then the turbine control valve opening "increases moderately"; Here, "positive large", "negative small", and "moderate increase" are all fuzzy language variables. Through fuzzy inference and defuzzification methods, the fuzzy control output is converted into an actual control quantity, which is combined with the control quantity obtained by MPC to jointly control the device actions, ensuring a smooth and safe conversion process; When triggering the "dry - wet" conversion, the control method is similar, but the dynamic model and control strategy need to be adjusted according to the physical process of the "dry - wet" conversion and the characteristics of the unit; For the control of feed water flow, during the "dry - wet" conversion process, the feed water flow needs to be rapidly increased to maintain the separator water level, and its dynamic model can be expressed as: ; Among them, 、 、 are coefficients; similarly, by using the method of combining MPC and fuzzy logic control, a control strategy is formulated to coordinate the actions of each device, maintain the stability of the main steam pressure and unit load, and achieve a smooth "dry - wet" conversion process.

[0024] In the "wet - dry" and "dry - wet" conversion steps, the principles of the sensors used for monitoring and judgment condition setting are introduced in detail. The separator water level is monitored by a high - precision capacitive liquid level sensor. Based on the non - linear relationship between the capacitance value and the liquid level height, the coefficient is determined through a calibration experiment to inversely solve the water level height, ensuring the monitoring accuracy; the steam flow and feed water flow are measured by vortex flow meters, and the flow rate is calculated based on the relationship between the frequency generated by the Karman vortex street phenomenon and the flow rate in combination with the pipeline parameters. At the same time, specific setting value examples and bases for the "wet - dry" and "dry - wet" conversion conditions are clarified, such as the specific requirements for the separator water level being lower than a certain value, the ratio of steam flow to feed water flow, and the change rate of steam flow during the "wet - dry" conversion, and how these settings comprehensively judge the unit operation status to prevent mis - triggering. In the conversion process control, the method of combining MPC and fuzzy logic control is emphasized again. The establishment of the dynamic change model of the main steam pressure and the process of MPC predictive control and fuzzy rule - based adjustment of device actions are described in detail, and it is pointed out that the control method during the "dry - wet" conversion is similar but the dynamic model and control strategy need to be adjusted according to its physical process and unit characteristics, such as the change of the dynamic model of feed water flow control and the corresponding control method adjustment.

[0025] Example 5 Further illustrated in combination with Example 1, as Figure 1 shown, in the wet - state coordination and optimization step, the intelligent control algorithm is fuzzy control, and its implementation process includes: defining the fuzzy subsets of the separator water level and the corresponding membership functions, as well as the fuzzy control rules for the rotational speed of the feed water pump or the opening degree of the regulating valve; Based on the real-time monitored water level of the separator, a fuzzy control quantity of the feed pump speed or the regulating valve opening is calculated through fuzzy inference; The fuzzy control quantity is defuzzified to obtain the actual control output, realizing the precise control of the separator water level; In the preferred solution, for the separator water level, the fuzzy subsets are defined as {very low (VL), low (L), moderate (M), high (H), very high (VH)}; A Gaussian membership function is used to describe each fuzzy subset, and the membership function of the fuzzy subset is: ; where, is the separator water level, is the central value of the "fuzzy subset" water level, is the corresponding standard deviation; these parameters are determined through experiments or data analysis according to the normal operating water level range and control accuracy requirements of the unit; Formulation of fuzzy control rules: Fuzzy control rules are formulated based on operating experience and physical principles; Among them, if the separator water level is "very low (VL)", the feed pump speed is "maximum (MAX)" and the regulating valve opening is "fully open (FULL)"; If the separator water level is "low (L)", the feed pump speed is "relatively large (LARGE)" and the regulating valve opening is "mostly open (MOSTLY_OPEN)"; And so on, forming a complete fuzzy control rule table; these rules reflect the control strategies for the feed pump speed and the regulating valve opening under different water level conditions to achieve the goal of maintaining the stability of the separator water level; Fuzzy inference and defuzzification: When the real-time monitored separator water level is obtained, the membership degrees on each fuzzy subset are calculated according to the defined membership function , , , , ; Then, the Mamdani fuzzy inference method is adopted to perform inference according to the fuzzy control rules; The same applies to the regulating valve opening; Finally, the centroid method is used for defuzzification to obtain the actual control outputs of the feed pump speed and the regulating valve opening ; the defuzzification formula for the feed pump speed is: ; where, represents different fuzzy output values, is the corresponding quantization value; through this fuzzy control method, the nonlinearity and uncertainty in the separator water level control can be effectively processed to achieve precise control; Main steam pressure and unit load regulation: Model Predictive Control (MPC) model establishment: Considering the wet steam-water two-phase flow characteristics and equipment operation limitations, establish the dynamic model of the main steam pressure and unit load For the dynamic change of the main steam pressure, based on the mass and energy conservation and the pressure drop model of two-phase flow, there is: ; where, is the fuel quantity, is the opening of the steam turbine control valve, is the feed water flow rate, is the specific enthalpy of steam, is the steam density, is the density of water, etc.; the function here is a complex functional relationship obtained through physical analysis of the wet steam-water two-phase flow process and fitting of experimental data; similarly, for the dynamic model of the unit load is: ; where, is the steam temperature, etc., and the function is also established based on physical principles and experimental data; MPC prediction and control: According to the current unit operation status , use the established dynamic model to predict the change trajectories of the main steam pressure and unit load in the future for a period of time , ; The prediction method adopted can be numerical solution based on the Runge-Kutta method. For example, for the prediction equation of the main steam pressure: ; where, is the time step, , , , are the coefficients of the Runge-Kutta method, obtained by calculating , and here is calculated according to the above established dynamic model; Taking minimizing the main steam pressure fluctuation and maintaining the unit load stability as the objective function, the formula is: ; where, and are the reference values of the main steam pressure and unit load, and is the weight coefficient at different moments, reflecting the degree of emphasis on the control accuracy at different future moments; by solving this optimization problem, the optimal control sequences of the fuel quantity and the opening of the steam turbine control valve are obtained. and , ; in actual control, only the control quantities at the current moment and act on the unit, and then the above prediction and optimization process is repeated at the next moment to achieve closed-loop control; Constraint handling: Considering the operation limitations of the equipment, the control instructions are constrained; For the fuel quantity , there is a lower limit and an upper limit , which are determined according to the minimum and maximum combustion capacities of the combustion equipment; for the opening of the steam turbine control valve , there is an opening limit , which is determined by the structure and safe operation requirements of the steam turbine; during the MPC solution process, by adding these constraint conditions to the optimization problem and using a constrained optimization algorithm for solution, it is ensured that the obtained control instructions are within the feasible range to guarantee the safe and stable operation of the unit. In the wet state coordination optimization step, the application of the fuzzy control algorithm in the separator water level control is deeply explained. The fuzzy subsets of the separator water level {very low (VL), low (L), medium (M), high (H), very high (VH)} and the adopted Gaussian membership function are defined in detail, and the method of determining the function parameters according to the unit operation water level range and control accuracy requirements is illustrated. An example of the fuzzy control rules formulated based on operation experience and physical principles is shown, such as the control strategies for the feed pump speed and the regulating valve opening corresponding to different water levels. In the fuzzy inference and defuzzification process, the method of using the Mamdani fuzzy inference method to calculate the membership degree based on the monitored water level and infer the control quantity, and the formula and principle of defuzzifying the control quantity using the centroid method to obtain the actual control output are introduced, reflecting the advantages of the fuzzy control in dealing with the nonlinearity and uncertainty of the separator water level. In terms of the main steam pressure and unit load regulation, the process and considerations of establishing a dynamic model based on mass, energy conservation, and two-phase flow pressure drop models, etc., and the equations and solution process of using the Runge-Kutta method for MPC prediction, as well as the method of setting the objective function to minimize the main steam pressure fluctuation and maintain the unit load stability and solving the control sequence are described in detail. At the same time, the importance of constraint handling in ensuring the safe operation of the equipment is emphasized, that is, setting the upper and lower limits for the fuel quantity and the opening of the steam turbine control valve according to the equipment capacity and safety requirements and handling the constraint conditions in the MPC solution.

[0026] The above embodiments are only the preferred technical solutions of the present invention and should not be regarded as limitations on the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including the equivalent replacement solutions of the technical features in the technical solutions recorded in the claims; that is, the equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A deep intelligent coordination method for thermal power units, characterized in that: The method includes: S1. Data acquisition and preprocessing step: Real-time collect the operation data of the thermal power unit, filter and normalize the collected data, remove outliers and noise interference, so that the data is within the range and accuracy suitable for algorithm calculation; S2. Dry-state coordination optimization step: When the unit load is in the range of 30% - 100%, introduce the fuel calorific value correction coefficient and coal-water ratio adjustment coefficient based on coal quality analysis, and dynamically adjust according to the change of coal type; Use predictive control technology to combine the current operation state of the unit and the prediction of grid load demand, calculate and dynamically adjust the feedforward amounts of coal feeding and water feeding; Adopt intelligent feedforward technology, comprehensively consider the operation state of the coal mill and the performance factors of the feed water pump, further optimize the feedforward control, and improve the load regulation ability; S3. "Wet-dry" and "dry-wet" conversion steps: Continuously monitor the separator water level, steam flow rate and feed water flow rate, set threshold values and change rate judgment conditions, and trigger the one-key automatic switching of the "wet-dry" or "dry-wet" process when the conditions are met; During the switching process, coordinate the actions of equipment such as the feed water pump, turbine control valve and burner to ensure the smooth and safe conversion process and maintain the stability of the main steam pressure and unit load; S4. Wet-state coordination optimization step: When the unit load is in the range of 25% - 30%, adjust the rotational speed of the feed water pump or the opening of the regulating valve through PID control or intelligent control algorithm to maintain the stability of the separator water level and ensure the minimum feed water flow rate; Use strategies such as model predictive control, and jointly adjust the main steam pressure and unit load by the fuel quantity and the turbine control valve. At the same time, considering the characteristics of wet-state steam-water two-phase flow and the operation limitations of equipment, the control instructions are constrained and corrected.

2. The deep intelligent coordination method for a thermal power unit according to claim 1, wherein: The operation data collected from the thermal power unit includes the spray water flow rate at all levels, the coal feeding amount at each layer, the feed water flow rate, the steam flow rate, the steam pressure, the load command, the separator temperature, the turbine valve opening, the grid frequency, and the coal quality parameter information.

3. The deep intelligent coordination method for a thermal power unit according to claim 2, characterized in that: For the spray water flow rate at all levels, a high-precision differential pressure flow sensor is used, and its measurement principle is based on the derivative formula of Bernoulli's equation: ； Among them, is the water spray flow rate, is the discharge coefficient, is the diameter ratio, is the expansibility coefficient, is the orifice diameter of the throttling element, is the fluid density, is the differential pressure before and after the throttling element; the water spray flow rate data at each level calculated by this formula can accurately reflect the water spray situation of the unit under different working conditions; For the coal feeding amount, a coal quantity measuring instrument based on the principle of nuclear radiation is used, which determines the coal quantity according to the absorption characteristics of coal for rays: ； Among them, is the coal feeding rate, is the initial ray intensity, is the ray intensity after penetrating the coal, is the absorption coefficient of the coal for the ray, is the path length of the ray passing through the coal; in this way, the coal feeding rate of each layer can be accurately measured ; Filtering process: Apply the adaptive Kalman filtering algorithm to filter the collected data; First, establish the state space model of the data: ； ； Among them, is the system state vector at a certain moment, is the state transition matrix, is the input control matrix, is the input vector, is the process noise vector, is the observation vector, is the observation matrix, is the observation noise vector; Then, continuously adjust the Kalman gain according to the real-time data : ； Among them, is the predicted covariance matrix, is the observation noise covariance matrix; Finally, update the system state estimation: ； Through this adaptive Kalman filtering method, the noise interference in the data can be effectively removed, making the data smoother and more accurate, especially suitable for processing the operation data of thermal power units with dynamic change characteristics; Normalization: A normalization method based on the standard deviation and mean of the data is adopted; for a certain data sequence , its mean is , the standard deviation is , and the formula for calculating the normalized sequence is: ; Outlier handling: Use an outlier detection method based on locally weighted regression scatterplot smoothing; for a given data point , perform weighted regression fitting within its local neighborhood: ； Among them, is a weight function based on the distance from the data point to For example, a Gaussian weight function can be adopted: ； wherein is a bandwidth parameter determined by methods such as cross-validation; Calculate the residuals of data points For a given residual threshold If Then it is considered that is an outlier; for the detected outliers, according to the time series characteristics of the data and the distribution of the surrounding data, methods such as linear interpolation or mean substitution based on historical data can be used for processing to ensure the integrity and reliability of the data and provide high-quality input data for subsequent intelligent coordination algorithms.

4. The deep intelligent coordination method for a thermal power unit according to claim 1, characterized in that: In the dry-state coordination optimization step, the calculation method of the fuel calorific value correction coefficient is: A1. Use a coal quality analysis instrument to obtain the calorific value, volatile matter, and ash content parameters of coal in real time; A2. Input the parameters in the above step A1 into a calculation model established in advance based on machine learning or empirical formula, calculate the fuel calorific value correction coefficient, which is trained by a large amount of historical coal quality data and the corresponding unit operation data, and can continuously adaptively adjust according to new coal quality data.

5. The deep intelligent coordination method for a thermal power unit according to claim 4, wherein: In the dry-state coordination and optimization step, the implementation method of the predictive control technology is as follows: B1. Establish a dynamic model of the unit operation state, including the mathematical relationship model between load, steam parameters, coal feeding amount, and water feeding amount. This model is constructed based on historical operation data and physical mechanism analysis. B2. According to the current unit operation parameters and the predicted value of the grid load demand, use the dynamic model to predict the unit operation state in the future for a period of time. B3. Calculate the set values of the coal feeding amount and the water feeding amount according to the prediction results, and continuously correct the prediction model and control instructions through a feedback mechanism.

6. The deep intelligent coordination method for a thermal power unit according to claim 5, characterized in that: The calculation steps of the fuel calorific value correction coefficient are as follows: A101: Use an advanced on-line coal quality analyzer spectrometer to obtain the calorific value ( ), volatile matter ( ), and ash content ( ) parameters of coal in real time; this spectrometer is based on the principles of infrared absorption spectroscopy and X-ray fluorescence spectroscopy, and determines the content of coal quality components by the interaction of light with specific wavelengths with the coal sample and according to the spectral characteristics of absorption and emission; Its measurement accuracy can reach: calorific value error within and volatile matter error within and ash content error within ; A201: Input the obtained coal quality parameters into the calculation model that combines Support Vector Regression (SVR) and Locally Weighted Polynomial Regression (LWPR); among them, the historical coal quality data and the corresponding unit operation data set , where is the fuel calorific value correction coefficient corresponding to the th group of data; First, use SVR to perform a preliminary global trend fitting on the data, and its objective function is: ； The constraint conditions are: ； where is the weight vector, and are the slack variables, is the penalty parameter, is the insensitive loss function parameter, is the kernel function that maps the input data to a high-dimensional feature space; here we choose the Gaussian kernel function: ； Then, for the new coal quality data points , LWPR is used for further local adjustment within its local neighborhood; assuming there are data points in the neighborhood, the locally weighted polynomial regression model is:[[]] ； Among them is the local weight, which is calculated by the Gaussian weight function: ； Here is the bandwidth parameter, determined by cross-validation; in this combined way, the global generalization ability of SVR and the local adaptability of LWPR can be fully utilized to accurately calculate the fuel calorific value correction coefficient , so as to adapt to the influence of different coal types on the unit operation; Implementation method of predictive control technology: B101: Based on the laws of conservation of mass, energy, and momentum, as well as the physical mechanisms of thermal processes, and combined with historical operation data, a dynamic model of the unit operation state is established; for the relationship between load ( ), main steam pressure ( ), main steam temperature ( ), and coal feed rate ( ), feed water flow ( ), it is described by a model composed of the following partial differential equations and algebraic equations: ； ； ； ； wherein is the steam density, is the steam volume, is the steam flow rate, is the coal feeding amount of the is the specific enthalpy of steam, is the lower calorific value of coal, is the enthalpy value of steam, is the heat released by combustion, and are functional relationships determined according to physical properties and experience; by discretizing these equations, they are transformed into an algebraic equation set suitable for computer calculation, and a dynamic model of the unit operation state is constructed; B201: Let the current moment be , and the current unit operation parameter vector and the predicted value of grid load demand are known. Among them, represents the predicted value for the next moments; using the established dynamic model, the operation state vector of the unit within a certain period in the future is predicted through iterative calculation. The fourth-order Runge-Kutta method is used for numerical solution of the state equations in the model , where is the control input vector, including the coal feed rate and the water feed rate, and its iterative formula is as follows: ； Where: ； ； ； ； Here is the time step; by this method, based on the current state and control input, the future operating states of the unit such as load and steam parameters can be predicted, providing a basis for setting the coal feeding amount and water feeding amount; B301. According to the predicted unit operation state, with the goal of minimizing the load deviation and steam parameter fluctuation, calculate the set values of the coal feeding amount and the water feeding amount, and establish an objective function: ； wherein 、 、 are the weighting coefficients at different times, and are the reference values of the main steam pressure and the main steam temperature; by solving this optimization problem, the set values and of the coal feeding amount and the water feeding amount are obtained; Meanwhile, by comparing the actually measured unit operation parameters with the predicted values, the deviation vector is calculated. ; Using this deviation vector, the parameters of the dynamic model are corrected online by the recursive least squares method (RLS); The update formula of RLS is as follows: ； ； ； wherein is the vector of estimated values of model parameters, is the gain vector, is the covariance matrix, is the forgetting factor, is the regression vector containing input and output data, is the measured output vector; through this feedback correction mechanism, the accuracy of the dynamic model and the performance of predictive control can be continuously improved, and the effective control of the unit operation state can be realized.

7. The deep intelligent coordination method for a thermal power unit according to claim 1, characterized in that: In the "wet-dry" and "dry-wet" conversion steps, the setting method of the threshold and change rate judgment conditions is as follows: For the "wet-dry" conversion, when the separator water level is lower than the set value A, the ratio of steam flow to water supply flow is greater than the set value B, and at the same time, the change rate of steam flow is greater than the set value C1, the conversion is triggered. For the "dry-wet" conversion, when the separator water level is higher than the set value D, the ratio of steam flow to water supply flow is less than the set value E, and at the same time, the change rate of steam flow is less than the set value F, the conversion is triggered. Among them, the set values A, B, C1, D, E, and F are determined according to the design parameters, operation experience, and safety requirements of the unit, and can be adjusted according to the actual operation conditions of the unit.

8. The deep intelligent coordination method for a thermal power unit according to claim 7, characterized in that: " The monitoring and judgment condition setting steps in the "wet-dry" and "dry-wet" conversion steps are as follows: For the monitoring of the separator water level, a high-precision capacitive liquid level sensor is used, and its measurement principle is based on the non-linear relationship between the capacitance value and the liquid level height: Among them, is the measured capacitance value, is the initial capacitance value, which is used to correspond to the capacitance at zero liquid level, and are coefficients determined through calibration experiments, is the water level height of the separator; by measuring the capacitance value in real time and using the above formula to solve for the water level height inversely, the high precision and reliability of water level monitoring are ensured; The steam flow and water supply flow are measured by high-precision vortex flow meters, and their measurement principles are based on the relationship between the frequency generated by the Karman vortex street phenomenon and the flow rate: ； Among them, is the vortex street frequency, St is the Strouhal number, which is a constant for a specific flowmeter structure and fluid properties, is the fluid flow velocity, is the characteristic dimension of the flowmeter; by measuring the vortex street frequency and combining parameters such as the cross-sectional area of the pipeline, the steam flow rate and the feed water flow rate can be calculated; Set the "wet - dry" conversion condition: When the water level in the separator is lower than the set value , and the ratio of steam flow rate to feed water flow rate , and at the same time the change rate of steam flow rate is greater than 1, trigger the "wet - dry" conversion; the set value here ensures that the conversion is started when the water level in the separator is low, and the ratio and the change rate being greater than 1 are set to comprehensively judge whether the unit operation state is suitable for the "wet - dry" conversion and prevent mis - triggering; Set the "dry - wet" conversion condition: When the water level in the separator is higher than the set value , and the ratio of steam flow rate to feed water flow rate , meanwhile the change rate of steam flow rate is [value], trigger the "dry - wet" conversion; Similarly, , , are set based on the unit characteristics and operation experience to ensure conversion under appropriate working conditions and guarantee the safe and stable operation of the unit; Conversion process control: When the "wet-dry" conversion is triggered, a method combining model predictive control MPC and fuzzy logic control is used to coordinate the actions of the feed water pump, turbine control valve, and burner equipment. First, establish a dynamic model of the unit during the "wet - dry" conversion process, considering mass, energy conservation, and equipment dynamic characteristics, for the dynamic change model of the main steam pressure : ； Among them, is the coal feeding amount, is the feed water flow rate, is the opening of the steam turbine control valve, , , , etc. are coefficients determined by the system identification method; this model describes the dynamic relationship between the main steam pressure and each control variable during the conversion process; Then, use MPC to predict the unit operation state in the future for a period of time, and formulate a control sequence according to the prediction results, with the goal of minimizing the main steam pressure fluctuation and maintaining the unit load stable as the objective function: ； Among them, and are the predicted main steam pressure and unit load, and are the reference values, is the weight coefficient, is the prediction time domain; the optimal set value sequences of control variables such as coal feeding amount, feed water flow rate, and turbine governing valve opening are obtained by solving this optimization problem; At the same time, use fuzzy logic control to handle some non-linear factors and uncertainties that are difficult to accurately model; for example, define fuzzy sets and fuzzy rules to fine-tune the control variables according to factors such as the steam flow change rate and the separator water level change rate; For the adjustment of the turbine control valve opening, there are the following fuzzy rules: If the steam flow change rate is "positive large" and the separator water level change rate is "negative small", then the turbine control valve opening "increases moderately"; Here, "positive large", "negative small", and "moderate increase" are all fuzzy language variables. Through fuzzy inference and defuzzification methods, the fuzzy control output is converted into an actual control quantity, which is combined with the control quantity obtained by MPC to jointly control the device action, ensuring a stable and safe conversion process; When triggering the "dry - wet" conversion, the control method is similar, but the dynamic model and control strategy need to be adjusted according to the physical process of the "dry - wet" conversion and the unit characteristics; For the control of the feed - water flow rate, during the "dry - wet" conversion process, it is necessary to rapidly increase the feed - water flow rate to maintain the separator water level, and its dynamic model can be expressed as: ； Among them, , , are coefficients; similarly, a control strategy is formulated by combining MPC and fuzzy logic control to coordinate the actions of each device, maintain the stability of the main steam pressure and the unit load, and achieve a smooth "dry-wet" conversion process.

9. The deep intelligent coordination method for a thermal power unit according to claim 1, characterized in that: In the wet - state coordination and optimization step, the intelligent control algorithm is fuzzy control, and its implementation process includes: defining the fuzzy subsets of the separator water level and the corresponding membership functions, as well as the fuzzy control rules for the feed - pump speed or the regulating - valve opening; According to the real - time monitored separator water level, the fuzzy control quantity of the feed - pump speed or the regulating - valve opening is calculated through fuzzy inference; The fuzzy control quantity is defuzzified to obtain the actual control output, achieving the precise control of the separator water level.

10. The method for deep intelligent coordination of a thermal power unit according to claim 9, wherein: For the separator water level, the defined fuzzy subsets are {very low (VL), low (L), moderate (M), high (H), very high (VH)}; The Gaussian - type membership function is used to describe each fuzzy subset, and the membership function of the fuzzy subset is: ； Among them, is the separator water level, is the central value of the "fuzzy subset" water level, is the corresponding standard deviation; these parameters are determined through experiments or data analysis according to the normal operating water level range and control accuracy requirements of the unit; Fuzzy control rule formulation: Fuzzy control rules are formulated based on operation experience and physical principles; Among them, if the separator water level is "very low (VL)", then the feed - pump speed is "maximum (MAX)" and the regulating - valve opening is "fully open (FULL)"; If the separator water level is "low (L)", then the feed - pump speed is "large (LARGE)" and the regulating - valve opening is "mostly open (MOSTLY_OPEN)"; And so on, forming a complete fuzzy control rule table; These rules reflect the control strategies for the feed - pump speed and the regulating - valve opening under different water - level conditions to achieve the goal of maintaining the stability of the separator water level; Fuzzy inference and defuzzification: When the water level in the separator is monitored in real time After that, calculate its membership degrees on each fuzzy subset according to the defined membership function , , , , ; Then, the Mamdani fuzzy - inference method is used to perform inference according to the fuzzy control rules; The same applies to the regulating - valve opening; Finally, the centroid method is used for clarification processing to obtain the rotational speed of the feed water pump and the actual control output of the opening degree of the regulating valve ; the clarification formula for the rotational speed of the feed water pump is as follows: ； Among them, represents different fuzzy output values, is the corresponding quantization value; Through this fuzzy control method, the nonlinearity and uncertainty in the separator water level control can be effectively handled to achieve precise control; Main steam pressure and unit load regulation: Model predictive control MPC model establishment: Considering the characteristics of wet steam-water two-phase flow and equipment operation limitations, establish the dynamic model of main steam pressure and unit load For the dynamic change of main steam pressure, based on the mass and energy conservation and the pressure drop model of two-phase flow, we have: ； Among them, is the fuel quantity, is the opening of the steam turbine control valve, is the feed water flow rate, is the specific enthalpy of steam, is the steam density, is the density of water, etc.; the function here is a complex functional relationship obtained through physical analysis of the wet steam-water two-phase flow process and fitting of experimental data; similarly, the dynamic model for the unit load is as follows: ； Among them, is the steam temperature, etc., and the function is also established based on physical principles and experimental data; MPC Prediction and Control: Based on the current unit operating status , use the established dynamic model to predict the change trajectories of the main steam pressure and unit load over a period of time in the future , ; The prediction method adopted can be a numerical solution based on the Runge-Kutta method. For example, for the prediction equation of the main steam pressure: ； Among them, is the time step, , , , are the coefficients of the Runge-Kutta method, obtained by calculating , and the here is calculated according to the dynamic model established above; Taking the minimization of the main - steam pressure fluctuation and the maintenance of the unit load stability as the objective function, the formula is: ； Among them, and are the reference values of the main steam pressure and the unit load, and are the weight coefficients at different times, reflecting the degree of emphasis on the control accuracy at different future times; by solving this optimization problem, the optimal control sequences of the fuel quantity and the turbine valve opening are obtained and , ; in actual control, only the control quantities at the current time and act on the unit, and then the above prediction and optimization process is repeated at the next time to achieve closed-loop control; Constraint handling: Considering the operation limitations of the equipment, the control instructions are constrained; For the fuel quantity , there is a lower limit and an upper limit , which are determined according to the minimum and maximum combustion capacities of the combustion equipment; for the opening of the steam turbine control valve , there are opening limits , which are determined by the structure and safe operation requirements of the steam turbine; during the MPC solution process, by adding these constraint conditions to the optimization problem and using a constrained optimization algorithm for solution, it is ensured that the obtained control instructions are within the feasible range, guaranteeing the safe and stable operation of the unit.

Images