Intelligent combined control method fusing high-performance PC and high-control stability DCS

Abstract

The application discloses a kind of intelligent joint control methods of fusing high-performance PC and high control stability DCS, comprising the following steps: build high-performance PC and the two-way MODBUS communication link of DCS, load section machine furnace coupling dynamic model intelligent identification, full load multi-agent active disturbance rejection control architecture construction, sub-agent active disturbance rejection controller design, full load adaptive primary frequency modulation control optimization, control instruction safety check and write-back execution, rolling window online parameter iterative optimization;The application realizes the two-way closed loop fusion of high-performance PC and DCS relying on MODBUS communication, without replacing original DCS hardware, can give system complex algorithm bearing capacity, reduce the cost of transformation and upgrading. Through load section machine furnace coupling modeling and multi-agent active disturbance rejection control, cover 20%-100% full load operation interval, improve the tracking ability and anti-interference performance of unit variable load process.

Description

Technical Field

[0001] This invention relates to the field of industrial process control and generator set operation optimization technology, specifically to an intelligent joint control method that integrates a high-performance PC and a high-control-stability DCS. Background Technology

[0002] Distributed control systems (DCS) are the core control carriers of coal-fired power generating units, possessing high reliability, strong real-time performance, and stable field execution capabilities, making them the mainstream architecture for industrial production control. However, limited by the computing performance of embedded controllers, traditional DCS can only support basic control algorithms such as PID, making it difficult to deploy complex multivariable coupled control, system identification, and active disturbance rejection algorithms. This results in large lag in the boiler-turbine coupling and limited control quality under strong disturbance scenarios.

[0003] With the increasing demand for deep peak shaving from generator sets, flexible operation at 20%-100% full load places higher demands on control accuracy and response speed. Existing DCS upgrade solutions mostly involve replacing controller hardware, which has problems such as high modification costs, long downtime, and insufficient compatibility. Some external optimization systems only use one-way data reading and open-loop guidance, which cannot achieve deep integration of closed-loop control. Communication delays and defects in data interaction mechanisms can also easily lead to control asynchrony.

[0004] Currently, control schemes that integrate DCS and external computing units generally suffer from problems such as insufficient algorithm adaptability, low full load coverage, and imperfect safety redundancy mechanisms, making it difficult to simultaneously meet the dual requirements of complex computing capacity and on-site control stability. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent joint control method that integrates a high-performance PC and a high-control-stability DCS to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent joint control method integrating a high-performance PC and a high-control-stability DCS, comprising the following steps:

[0007] Step 1: Establish a MODBUS communication link between the high-performance computing PC and the distributed control system (DCS), configure a bidirectional data mapping register, and realize the acquisition and transmission of process operation data from the DCS to the PC, as well as the write-back of control commands from the PC to the DCS.

[0008] Step 2: Divide and preprocess the acquired full-load operation data of the generator set into operating conditions, and perform intelligent identification of dynamic models for load segments in the turbine-boiler coupling process to obtain the turbine-boiler coupling dynamic model corresponding to each load segment.

[0009] Step 3: Construct a multi-agent active disturbance rejection control architecture covering the full load range of 20%-100%, divide the controlled object into sub-control agents according to the function of the controlled object, and set up a coordinating agent to achieve decoupling and matching of each sub-control agent;

[0010] Step 4: Design an active disturbance rejection controller for each sub-control agent. Estimate the total system disturbance in real time by extending the state observer. Combine the output prediction of the machine-furnace coupling dynamic model to generate the basic control quantity. Then, add the total disturbance compensation quantity to obtain the final control correction quantity.

[0011] Step 5: Based on the grid frequency deviation and the current unit load conditions, generate a primary frequency regulation dynamic compensation command to adapt to the frequency regulation response characteristics of different load segments;

[0012] Step 6: The control correction value generated by the active disturbance rejection control is superimposed and verified with the primary frequency modulation dynamic compensation command. After rate limiting and amplitude limiting, it is written back to the control execution loop of the DCS through the MODBUS communication link.

[0013] Step 7: Set a rolling time window to iteratively update the parameters of the turbine-furnace coupling dynamic model and the active disturbance rejection controller based on the latest operating data collected periodically, thereby achieving online optimization of the control strategy.

[0014] Furthermore, in step one, the MODBUS communication link uses the Ethernet TCP protocol for transmission. The process analog quantities and switch quantities of the DCS are mapped to a continuous holding register address range. The PC reads the register data according to a fixed sampling period. The control instructions generated by the PC are mapped to the writable register address range of the DCS. Each set of instructions corresponds to a double register check bit. When the data is inconsistent, the current frame is discarded and the valid instructions of the previous frame are retained.

[0015] Furthermore, in step two, the turbine-furnace coupled dynamic model adopts an incremental state-space structure, and its calculation equation is as follows:

[0016]

[0017] In the formula, The system output increment vector at time k contains the increments of four controlled variables: main steam pressure, main steam temperature, unit active power, and heating extraction steam pressure. for The control input increment vector at any given time includes four control quantity increments: fuel quantity command, total feedwater flow command, high-pressure valve opening command, and first-stage desuperheating water flow command. for The measurable disturbance increment vector at any given time includes the increments of three disturbance quantities: ambient temperature, heating steam extraction flow rate, and raw coal received basis lower heating value. The state transition matrix represents the inertial coupling characteristics of the controlled variable itself. The control gain matrix represents the strength of the effect of the control quantity on the controlled quantity. This is the disturbance gain matrix, which characterizes the degree of influence of the measurable disturbance on the controlled variable; This is the current sampling time number.

[0018] Furthermore, in step three, the sub-control agents are divided into three categories: boiler combustion agent, steam-water system agent, and turbine heating agent. The controlled variables of the boiler combustion agent are the main steam pressure and furnace negative pressure, and the control variables are the total fuel quantity command, the total air supply command, and the total induced draft command. The controlled variables of the steam-water system agent are the main steam temperature and the reheat steam temperature, and the control variables are the first-stage desuperheating water flow command, the second-stage desuperheating water flow command, and the flue gas damper opening command. The controlled variables of the turbine heating agent are the unit active power and the heating extraction steam pressure, and the control variables are the high-pressure regulating valve opening command and the extraction steam regulating valve opening command. The coordinating agent receives the status information of each sub-agent and adjusts the setpoint weights of each sub-agent with the goal of minimizing the boiler-turbine energy balance deviation.

[0019] Furthermore, in step four, the control law calculation equation for the active disturbance rejection controller is:

[0020]

[0021] In the formula, Let k be the initial control input at time k; For the controller's proportional gain parameter; The setpoint of the controlled variable at time k; The state observation value of the controlled variable output by the extended state observer at time k; The total system disturbance observation value output by the extended state observer at time k includes unmodeled system dynamics, unknown external disturbances, and coupling disturbances between subsystems; This is the control gain compensation factor, used to match the gain relationship between the control variable and the controlled variable.

[0022] Furthermore, in step five, the calculation equation for the primary frequency modulation dynamic compensation command is:

[0023]

[0024] In the formula, The value of the frequency modulation command correction at time k; This is the frequency modulation fundamental gain coefficient; Let k be the power grid frequency deviation value at time k; The rated frequency of the power grid; The rated active power of the unit; This is the load adaptive coefficient, and its value varies with the current unit load rate. Dynamic adjustment: the 20%-40% load range corresponds to the first coefficient range, and the 40%-100% load range corresponds to the second coefficient range.

[0025] Furthermore, in step two, the operating conditions are divided into four continuous intervals according to the unit load rate: 20%-40% low load interval, 40%-60% medium-low load interval, 60%-80% medium-high load interval, and 80%-100% high load interval. Each load interval is identified with an independent model parameter matrix. During the control process, the model parameters of the corresponding interval are called according to the real-time load rate. At the boundary of adjacent load intervals, a linear interpolation method is used to achieve a smooth transition.

[0026] Furthermore, in step six, the instruction write-back process is equipped with a three-level security verification mechanism: the first level is amplitude verification, which forcibly limits the control correction amount to the boundary value when it exceeds the preset upper and lower limits; the second level is rate verification, which limits the change amount of the instruction in two adjacent frames to the maximum rate when it exceeds the preset change rate; the third level is communication status verification, which stops instruction issuance and triggers the DCS to automatically switch to the native control logic when no response is received from the DCS end or data verification error is found in multiple consecutive frames.

[0027] Furthermore, in step seven, the length of the rolling time window is set to an integer multiple of the fixed sampling period. During each iteration, the latest running data within the window is substituted into the identification algorithm, and the model parameters are updated using a weighted recursive method, with the weight of the new data being higher than that of the old data. The controller parameters are optimized with the goal of minimizing the sum of squares of the control deviations, and are iteratively adjusted using the gradient descent method. The parameter update step size is dynamically adjusted according to the load fluctuation amplitude.

[0028] Furthermore, the extended state observer in step four is a third-order linear extended state observer. The inputs are the measured value of the controlled variable and the output value of the control variable, and the outputs are the observed value of the controlled variable state, the observed value of the controlled variable rate of change, and the observed value of the total system disturbance, respectively. The observer gain parameter is configured according to the characteristics of the model for each load segment and is adjusted synchronously with the switching of load intervals.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] This invention leverages MODBUS communication to achieve bidirectional closed-loop fusion of a high-performance PC and a DCS, enabling the system to handle complex algorithms without replacing existing DCS hardware, thus reducing upgrade costs. Through load-segment boiler-turbine coupling modeling and multi-agent active disturbance rejection control, it covers the 20%-100% full-load operating range, improving the unit's tracking capability and disturbance rejection performance during load changes. An extended state observer is introduced to uniformly compensate for internal and external disturbances, mitigating control lag and fluctuations caused by boiler-turbine coupling. Combined with load-adaptive primary frequency regulation optimization, it balances frequency regulation response capability and operational stability across different load segments. Multi-level safety checks and disturbance-free switching mechanisms ensure operational reliability under joint control modes. The overall solution achieves a complementary advantage of complex computing power and high control stability, providing a feasible control architecture for flexible full-load operation of coal-fired power generating units. Attached Figure Description

[0031] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figure 1 This invention provides an intelligent joint control method that integrates a high-performance PC and a high-control-stability DCS, comprising the following steps:

[0034] Step 1: Establish a high-performance bidirectional MODBUS communication link between the PC and DCS.

[0035] This step constructs a joint control physical architecture consisting of complex upper-layer computation and stable lower-layer execution. A high-performance PC serves as an external computing unit, undertaking computationally intensive tasks such as model identification, multivariable control algorithm calculations, and parameter optimization iterations. The DCS, as the field control execution unit, handles tasks with high real-time and reliability requirements, including field signal acquisition, lower-level loop control, equipment driving, and safety interlocking. The two systems achieve bidirectional data interaction via the MODBUS communication protocol, forming a complete closed-loop control link.

[0036] The communication link uses industrial Ethernet as the physical medium and the MODBUS-TCP protocol for data transmission, adapting to the electromagnetic environment and long-distance transmission requirements of industrial sites. The communication architecture adopts a master-slave mode, with a high-performance PC acting as the master station initiating data requests and issuing commands, and the DCS acting as the slave station responding to requests and returning data. The DCS side pre-allocates a contiguous holding register address space in the controller memory, dividing it into two independent address ranges: a read-only data area and a writable data area. These two address ranges do not overlap, avoiding read / write operation conflicts.

[0037] The read-only data area is used to map all process operation data acquired by the DCS, covering two main categories: analog signals and digital signals. Analog signals are divided into four categories according to the system: boiler side, turbine side, heating side, and grid side. Boiler side parameters include total fuel quantity, total forced draft, total induced draft, drum water level, main steam pressure, main steam temperature, reheat steam temperature, furnace negative pressure, flue gas temperature in each section, feedwater main flow rate, first-stage desuperheating water flow rate, and second-stage desuperheating water flow rate. Turbine side parameters include high-pressure regulating valve opening, intermediate-pressure regulating valve opening, unit active power, unit reactive power, turbine speed, regulating stage pressure, extraction steam pressure in each section, extraction steam temperature in each section, and exhaust steam temperature. Heating side parameters include heating extraction steam pressure, heating extraction steam temperature, total heating extraction steam flow rate, and heating return water temperature. Grid side parameters include real-time grid frequency and bus voltage. Digital signals include the operating status, fault status, interlocking action status, and valve limit signals of major auxiliary equipment.

[0038] The data mapping rules are as follows: each analog quantity corresponds to two consecutive holding registers, stored in floating-point format, with the numerical range corresponding one-to-one with the physical quantity range; each digital quantity corresponds to one bit in the register, arranged sequentially in byte order, and a single register can carry 16 digital quantity signals. After reading the registers, the PC parses the data according to the pre-agreed mapping rules and restores it to the corresponding physical quantity value and status signal.

[0039] The writable data area is used to map control commands and configuration parameters issued by the PC, including correction commands for each control loop, controller parameter settings, mode switching commands, etc. Each group of control commands corresponds to three consecutive registers, where the first two registers store the floating-point value of the command, and the third register stores the checksum. The checksum is generated by summing the values ​​of the first two registers, that is, summing the binary values ​​of the first two registers byte by byte, and taking the lower eight bits as the checksum. When the PC issues a command, it writes the command value and the checksum simultaneously; after the DCS receives the command, it first calculates the sum and checksum of the currently received value according to the same rules, compares it with the received checksum, if they match, it is determined to be a valid command and the control value is updated; if they do not match, it is determined to be a communication error, the current frame is discarded and the valid command of the previous frame remains unchanged, to avoid command jumps caused by communication interference.

[0040] The communication sampling period can be configured according to control requirements. Data acquisition and command issuance are executed synchronously within the same period to ensure the consistency of control timing. On the PC, each communication cycle first performs a batch read operation on the read-only data area to acquire all running data. Immediately after acquisition, the control algorithm is started to generate control correction commands for each loop. After calculation, a batch write operation on the writable data area is immediately performed to send the commands to the DCS. A complete "read-calculate-write" closed loop is completed within a single cycle, minimizing the impact of communication latency on control performance.

[0041] The communication link is equipped with a heartbeat detection mechanism. The PC sends a heartbeat command every fixed number of communication cycles, and the command is written to a dedicated heartbeat register. After receiving the heartbeat command, the DCS returns a heartbeat response code in the corresponding response register. If the PC does not receive a correct heartbeat response for several consecutive cycles, it is determined that the communication has been interrupted. The PC immediately stops issuing all control commands, and the DCS automatically switches to the native PID control mode through its built-in timeout judgment logic. The switching process uses a gradual change in command weight to achieve a smooth transition, ensuring the safe operation of the unit.

[0042] Step 2: Intelligent Identification of Dynamic Model of Coupling Between Boiler and Turbine in Different Load Sections

[0043] This step establishes a mathematical model that accurately reflects the dynamic characteristics of the boiler-turbine coupling, providing a model foundation for subsequent control algorithms. The boiler-turbine coupling process of a coal-fired power generating unit is characterized by large inertia, large lag, strong nonlinearity, and strong coupling of multiple variables. The dynamic characteristics differ significantly under different load conditions. Therefore, a load segment identification method is adopted to improve the model's adaptability to all operating conditions.

[0044] First, the collected raw operating data is preprocessed to eliminate the impact of data noise and outliers on the identification accuracy. The preprocessing process consists of three steps: outlier removal, filtering and noise reduction, and data normalization.

[0045] Outlier removal employs a dual-discrimination method: the first level is physical range discrimination, where upper and lower limits of the physical range that conform to the characteristics of the equipment are set for each operating parameter, and data exceeding the range are directly identified as outliers and removed; the second level is statistical discrimination, which uses the 3σ criterion, sets a fixed-length sliding window, and calculates the mean and standard deviation of the same parameter data within the window. Data that deviates from the mean by more than three times the standard deviation is identified as outliers, and is replaced by linear interpolation of adjacent normal data within the window to eliminate abnormal data caused by sensor jumps and signal interference.

[0046] The filtering and denoising employs a moving average filtering algorithm. A fixed-length sliding window is set, and the arithmetic mean of the data within the window is used as the filtered output for the current moment. This suppresses high-frequency noise interference from the field sensors while preserving the dynamic trend of the data. The window length is configured according to the inertial characteristics of the parameters; a longer window is used for parameters with large inertia, and a shorter window is used for rapidly changing parameters.

[0047] Data normalization employs the extreme value normalization method. For each parameter, the original value is linearly mapped to the [0,1] interval based on its upper and lower limits of range. This eliminates the influence of the difference in the dimensions of different parameters on the identification process and improves the convergence speed and numerical stability of the identification algorithm.

[0048] After data preprocessing, operating conditions are divided based on the ratio of active power to rated power (i.e., load rate). The full load operating range of 20%-100% is divided into four consecutive load intervals: 20%-40% low load interval, 40%-60% medium-low load interval, 60%-80% medium-high load interval, and 80%-100% high load interval. A hysteresis judgment mechanism is used for operating condition division. Different interval switching thresholds are used for load rate increases and decreases, with a fixed threshold difference to avoid frequent switching of model parameters when the load fluctuates slightly near the interval boundaries.

[0049] For each load range, a dynamic model identification of the turbine-boiler coupling was performed separately. The model adopts an incremental state-space structure, which can intuitively reflect the dynamic coupling relationship between the control variable, disturbance variable, and controlled variable, and facilitates the design and implementation of subsequent control algorithms. The calculation equations of the model are as follows:

[0050]

[0051] In the formula, The system output increment vector at time k is a four-dimensional column vector. The four elements are, in order, the main steam pressure increment, the main steam temperature increment, the unit active power increment, and the heating extraction steam pressure increment, covering the core controlled parameters of the boiler-turbine coupling process. for The control input increment vector at any given time is a four-dimensional column vector. The four elements are, in order, the total fuel quantity command increment, the total feedwater flow command increment, the high-pressure valve opening command increment, and the first-stage desuperheating water flow command increment, which correspond to the core control means that affect the controlled parameters. for The measurable disturbance increment vector at any given time is a three-dimensional column vector, with the three elements being the ambient temperature increment, the heating steam flow rate increment, and the raw coal received basis lower heating value increment, corresponding to the main external disturbance factors that can be detected in real time during the production process. It is a 4×4 order state transition matrix, where each element represents the strength of the inertial coupling between the corresponding controlled variables; It is a 4×4 order control gain matrix, where each element represents the gain and dynamic characteristics of the corresponding control quantity on the corresponding controlled quantity. It is a 4×3 order perturbation gain matrix, where each element represents the degree of influence of the corresponding perturbation on the corresponding controlled variable; This is the sequence number of the current sampling time, and the time interval between adjacent sequence numbers is equal to the communication sampling period.

[0052] Model identification employs a recursive least squares algorithm. For each load interval, historical operating data within that interval is extracted, converted into incremental form, and then input into the identification algorithm to iteratively solve the matrix. , , The parameter values ​​are set. The identification process is configured with convergence criteria; iteration stops when the parameter change over multiple consecutive iterations is less than a preset threshold, and the final model parameters for that load range are output.

[0053] The model parameters of adjacent load intervals are smoothly transitioned using linear interpolation. When the unit load rate is in the boundary region of two adjacent intervals, the model parameters of the two intervals are weighted according to the relative position of the current load rate within the interval to obtain the actual model parameters under the current operating condition. The weights are proportional to the distance from the load rate to the interval boundary, thus avoiding control fluctuations caused by sudden parameter changes.

[0054] Step 3: Construction of a Full-Load Multi-Agent Disturbance Rejection Control Architecture

[0055] This step decomposes the overall control system into multiple independent sub-control agents, each responsible for the control task of a specific controlled object. Simultaneously, by coordinating the agents to handle the coupling relationships between subsystems, the design complexity of the controller is simplified, while the system's scalability and maintainability are improved. The sub-control agents are divided into three categories based on the functional attributes of the controlled object: boiler combustion agent, steam-water system agent, and steam turbine heating agent.

[0056] The control objective of the boiler combustion intelligent system is to maintain stable main steam pressure and furnace negative pressure, ensuring energy supply and demand balance and a reasonable air-fuel ratio during boiler combustion. Its inputs include main steam pressure setpoint, furnace negative pressure setpoint, measured main steam pressure, measured furnace negative pressure, total fuel quantity feedback, total forced draft feedback, and total induced draft feedback. Outputs include total fuel quantity correction commands, total forced draft correction commands, and total induced draft correction commands. The boiler combustion intelligent system primarily addresses combustion-side disturbances caused by coal quality fluctuations and load changes by rapidly adjusting the fuel and air ratio to stabilize the main steam pressure and provide a stable steam source for the turbine side.

[0057] The control objective of the steam-water system intelligent agent is to maintain stable main steam and reheat steam temperatures, ensuring that steam parameters meet the turbine operating requirements. Its inputs include main steam temperature setpoint, reheat steam temperature setpoint, measured main steam temperature, measured reheat steam temperature, first-stage desuperheating water flow feedback, second-stage desuperheating water flow feedback, and flue gas damper opening feedback. Its outputs are first-stage desuperheating water flow correction commands, second-stage desuperheating water flow correction commands, and flue gas damper opening correction commands. The steam-water system intelligent agent primarily addresses steam temperature deviations caused by load changes, desuperheating water disturbances, and flue gas heat exchange fluctuations. Through coordinated adjustment of the desuperheating water and flue gas dampers, it controls the steam temperature within the allowable range.

[0058] The control objective of the steam turbine heating intelligent agent is to maintain the stability of the unit's active power and heating extraction steam pressure, meeting the grid load command and heating load demand. Its inputs include the unit's active power setpoint, heating extraction steam pressure setpoint, measured unit active power, measured heating extraction steam pressure, high-pressure control valve opening feedback value, and extraction steam regulating valve opening feedback value; the outputs are high-pressure control valve opening correction commands and extraction steam regulating valve opening correction commands. The steam turbine heating intelligent agent primarily addresses the power and extraction steam pressure deviations caused by changes in grid load commands and fluctuations in heating user loads by adjusting the control valve and extraction steam valve openings to quickly track load demand.

[0059] The coordinating agent sits atop the three sub-control agents, responsible for coordinating energy balance and decoupling subsystems on the boiler and turbine sides. The coordinating agent collects real-time data on the controlled variable deviations and control outputs of each sub-agent, and calculates the boiler energy supply and demand deviations. When deviations exceed preset ranges, it dynamically adjusts the setpoint tracking rate and control weights of each sub-agent. For example, during load increases, the coordinating agent appropriately increases the control weight of the boiler combustion agent to accelerate fuel increase and pre-store steam energy, preventing a significant drop in main steam pressure. During load decreases, it appropriately increases the control weight of the turbine heating agent to control the valve closing rate, preventing a pressure spike. The coordinating agent does not directly output control commands; it achieves coordinated control only by adjusting the internal parameters of the sub-agents, avoiding conflicts with the sub-agents' control logic.

[0060] The parameter configuration of the multi-agent architecture also adopts a load-segment adaptation approach. Each load segment corresponds to a set of agent control parameters, which are switched synchronously according to the operating conditions to ensure consistent control performance across the entire load segment. In the low-load segment, the unit has weak heat storage capacity and large inertia of the controlled object, so the parameter configuration focuses on stability and slow response; in the high-load segment, the unit has fast response speed and large adjustment margin, so the parameter configuration focuses on speed and disturbance rejection.

[0061] Step 4: Design of Sub-Agent Active Disturbance Rejection Controller

[0062] This step employs active disturbance rejection control (ADRC) as the core control algorithm for each sub-agent. It unifies the unmodeled dynamics, parameter perturbations, coupling effects between subsystems, and unknown external disturbances into a single "total disturbance," which is estimated in real time using an extended state observer and compensated for in the control input. This significantly improves the system's disturbance rejection capability and control accuracy.

[0063] Each sub-control agent is equipped with an independent third-order linear extended state observer. The observer inputs are the measured value of the controlled variable and the control output from the controller. The outputs three state observations: the first is the state observation of the controlled variable, tracking its actual value; the second is the rate of change observation of the controlled variable, tracking its trend; and the third is the total system disturbance observation, tracking the combined effect of all internal and external disturbances. The observer's gain parameter is configured according to the model characteristics of the corresponding load segment to ensure the speed and stability of the observation process and avoid excessive amplification of observation noise. The observer's update cycle is consistent with the communication sampling cycle, with one state update completed in each sampling cycle.

[0064] The control law of the active disturbance rejection controller adopts a proportional control plus aggregate disturbance feedforward compensation structure, and its calculation equation is as follows:

[0065]

[0066] In the formula, Let k be the initial control input at time k; For the controller's proportional gain parameter; The setpoint of the controlled variable at time k; The state observation value of the controlled variable output by the extended state observer at time k; The total system disturbance observation value output by the extended state observer at time k includes unmodeled system dynamics, unknown external disturbances, and coupling disturbances between subsystems; This is the control gain compensation factor, used to match the gain relationship between the control variable and the controlled variable.

[0067] The first part of the control law The proportional feedback term generates a basic control action based on the deviation between the setpoint and the observed value, pushing the controlled variable towards the setpoint; Part Two The total disturbance compensation term converts the observed total disturbance into a compensation value for the control variable, thus offsetting the impact of the disturbance on the controlled variable in advance. Through real-time compensation of the total disturbance, the active disturbance rejection controller can effectively mitigate the effects of adverse factors such as large lag in the turbine-furnace coupling and fluctuations in coal quality, improve the stability of the control process, and reduce the fluctuation amplitude of the controlled parameters.

[0068] Initial control quantity output by the controller The control quantity needs to be allocated through a control quantity allocation process, which breaks it down into specific instructions for the corresponding actuators. For example, the initial control quantity of the boiler combustion intelligent body is decomposed into fuel quantity instructions, air supply quantity instructions, and induced draft quantity instructions, which are allocated according to the preset air-fuel ratio and air-smoke balance relationship to ensure the economy and safety of the combustion process and avoid abnormal operating conditions such as oxygen-deficient combustion or positive pressure smoke injection.

[0069] The proportional gain of the controller varies depending on the load range. Control gain compensation factor Different configuration values ​​are used for the gain parameters of the observer and the control object. In the low-load range, the controlled object has high inertia and low gain; therefore, the proportional gain and observer gain are appropriately increased to improve response speed. In the high-load range, the controlled object responds quickly and is prone to fluctuations; therefore, the gain parameters are appropriately decreased to improve control stability. Parameter switching also employs a smooth interpolation transition method to avoid control shocks during changes in operating conditions.

[0070] Step 5: Full-load adaptive primary frequency regulation control optimization

[0071] Primary frequency regulation is the core function of generator units participating in grid frequency adjustment. Traditional DCS primary frequency regulation uses fixed-parameter control logic, which makes it difficult to balance the frequency regulation performance of different load segments: at low load segments, the units have weak heat storage capacity, and fixed frequency regulation parameters can easily lead to large fluctuations in steam parameters, affecting operational safety; at high load segments, the units have great frequency regulation potential, and fixed parameters cannot fully utilize the frequency regulation capability, making it difficult to meet the grid frequency regulation requirements. This step, based on the original DCS primary frequency regulation logic, generates dynamic compensation commands from the high-performance PC side, which are then superimposed on the original frequency regulation commands to optimize primary frequency regulation performance. The compensation commands are dynamically calculated based on the grid frequency deviation and the current load conditions, and their calculation equation is as follows:

[0072]

[0073] In the formula, The value of the frequency modulation command correction at time k; This is the frequency modulation fundamental gain coefficient; Let k be the power grid frequency deviation value at time k; The rated frequency of the power grid; The rated active power of the unit; This is the load adaptive coefficient, and its value varies with the current unit load rate. Dynamic adjustment: the 20%-40% load range corresponds to the first coefficient range, and the 40%-100% load range corresponds to the second coefficient range.

[0074] Load adaptive coefficient A two-stage configuration is adopted: the first coefficient range corresponds to the 20%-40% low load range, with relatively small coefficient values ​​to reduce the amplitude of frequency regulation and avoid large fluctuations in unit steam parameters during frequency regulation, ensuring the stability of low-load operation; the second coefficient range corresponds to the 40%-100% medium-high load range, with relatively large coefficient values ​​to fully utilize the unit's heat storage and regulation capabilities, improve frequency regulation response speed and contribution, and meet the grid frequency regulation requirements. At the boundary between the two ranges, the coefficients adopt a linear smooth transition to avoid abrupt changes in frequency regulation commands when the load crosses the boundary.

[0075] The frequency regulation compensation command is equipped with dead-zone and amplitude limiting protection. When the grid frequency deviation is less than the dead-zone threshold, no compensation command is generated to avoid unit fluctuations caused by frequent minor adjustments. When the compensation command exceeds the preset upper and lower limits, it is forcibly limited to the boundary values ​​to prevent excessive frequency regulation from affecting the normal operation of the unit. The rate of change of the compensation command is also limited to ensure that the frequency regulation process is stable and controllable.

[0076] The generated primary frequency regulation compensation command and the load control correction amount generated by the active disturbance rejection controller are superimposed on the corresponding channels and act together on the turbine regulating valve control loop to achieve coordinated operation of load closed-loop control and primary frequency regulation function. The two do not conflict with each other and jointly maintain the stability of unit power and grid frequency.

[0077] Step Six: Control Command Security Verification and Write-back Execution

[0078] This step is responsible for integrating and verifying the various control commands generated by the PC side, ensuring that the commands are safe and reliable before sending them to the DCS for execution. It is the core security barrier of the joint control architecture, preventing abnormal commands from threatening the safe operation of the unit.

[0079] First, command superposition and channel allocation are performed: the control correction quantities output by the boiler combustion intelligent agent, the steam-water system intelligent agent, and the turbine heating intelligent agent are superimposed with the primary frequency regulation dynamic compensation command on the corresponding channels to obtain the total correction command for each actuator. The total correction command is incrementally superimposed on the output of the original PID controller in the DCS to form the final actuator command. The advantage of this incremental correction method is that when the joint control mode is exited, the DCS side can directly restore the original control logic without adjusting any parameters, achieving a truly seamless switchover, and the switchover process will not impact the unit operation.

[0080] Subsequently, a three-level security verification mechanism is executed to verify the general correction command layer by layer, ensuring that the command meets the requirements for safe operation:

[0081] The first level is amplitude verification. For each control channel, upper and lower limit amplitude values ​​for the correction command are preset. These amplitude values ​​are determined comprehensively based on the adjustable range of the actuator and the safe operation requirements of the unit. When the calculated total correction command exceeds the upper or lower limit, the command value is forcibly limited to the corresponding boundary value to avoid excessive correction causing the actuator to be fully open or fully closed, which could lead to safety risks such as large fluctuations in steam parameters.

[0082] The second level is rate verification. For each control channel, a maximum allowable change within a unit sampling period is preset. When the change in command between two adjacent sampling periods exceeds this maximum allowable change, the command is limited according to the maximum change rate, so that the command gradually approaches the target value, avoiding mechanical shocks and fluctuations in operating parameters caused by sudden changes in commands.

[0083] The third level is communication status verification. The PC side monitors the communication status with the DCS in real time, including abnormal situations such as communication response timeout, data verification errors, and loss of heartbeat signals. When a minor communication anomaly occurs in a single frame, the issuance of commands for the current cycle is paused, the command value of the previous cycle is maintained, and communication is awaited to be restored in the next cycle. When communication anomalies occur in multiple consecutive cycles, it is determined to be a communication interruption. All control commands are immediately stopped, and a mode switching signal is sent to the DCS, triggering the DCS side to automatically switch to the native control mode. The original PID loop of the DCS undertakes the control task completely, and the PC side only retains the data acquisition function. After communication is restored, joint control is resumed according to the operator's instructions.

[0084] After security verification, the instructions are written to the corresponding writable register address on the DCS side according to the register mapping rules of the MODBUS communication protocol. Each instruction frame is accompanied by a checksum. After receiving the data, the DCS side performs data verification again, and after confirming that it is correct, it is superimposed on the original control loop.

[0085] The DCS side is equipped with dedicated control switching logic, configuring two operating states: "joint control mode" and "native control mode". Operators can manually switch modes according to operational needs, or the switching can be automatically triggered by communication abnormalities or fault signals. The switching process adopts a smooth transition mechanism, with the weight of the joint control command gradually increasing or decreasing over several cycles to avoid sudden command jumps during mode switching and ensure a smooth and uninterrupted switching process.

[0086] Step 7: Online parameter iterative optimization of the scrolling window

[0087] The characteristics of coal-fired power generating units drift slowly over time due to factors such as equipment wear and tear, and changes in the type of coal fed into the furnace. The performance of models and controllers with fixed parameters gradually declines, making it impossible to maintain optimal performance in the long term. This step utilizes an online iterative optimization mechanism to continuously update model and controller parameters, ensuring long-term stable control performance and reducing the workload of manual parameter tuning.

[0088] The online update of model parameters adopts a rolling window recursive identification method. A fixed-length rolling time window is set, with the window length being an integer multiple of the sampling period. The latest operating data is continuously stored within the window. When new data enters the window, the oldest data is removed from the window, realizing the rolling update of data. Every fixed update cycle, the preprocessed operating data within the window is converted into incremental form and substituted into the recursive identification algorithm to perform an iterative update of the matrix parameters of the turbine-furnace coupling model.

[0089] The iterative process employs a weighted strategy, with newer data receiving higher weights and older data receiving lower weights. The weights change linearly with the data's position within the window. This weighting method allows the model to track slow changes in equipment characteristics while retaining sufficient historical operating condition adaptability, avoiding parameter oscillations caused by random fluctuations in new data.

[0090] Model parameter updates are subject to a limit on the magnitude of change; the parameter change in a single iteration must not exceed a preset proportional threshold to avoid parameter mutations caused by abnormal data. Updated parameters undergo rationality verification; they only take effect if they conform to physical laws and equipment characteristics. Otherwise, the previous version of parameters is retained to ensure the model always possesses physical interpretability. Online optimization of controller parameters aims to minimize control deviation and employs gradient descent for iterative adjustment. The sum of squared deviations between the setpoint and measured values ​​of the controlled variable is selected as the optimization objective function. The gradient of the objective function with respect to the controller gain parameter is calculated at fixed intervals, and the parameters are adjusted along the negative gradient direction to gradually reduce control deviation and improve control accuracy. The parameter adjustment step size is dynamically adjusted according to load fluctuations: a smaller step size is used when the load is stable to avoid frequent parameter fluctuations; a larger step size is used when the load changes significantly to accelerate parameter convergence.

[0091] The parameter optimization process is subject to strict safety constraints. The adjustment range of all parameters must not exceed the pre-set upper and lower limits to ensure the controller remains within a stable operating range and prevent control divergence caused by parameters exceeding these limits. The optimization calculation process is executed asynchronously in the background, without affecting the instruction output of the current control cycle. Parameter updates take effect in the next control cycle, avoiding the optimization calculation from consuming the main control link's computing resources and ensuring the real-time performance and stability of the core control loop.

[0092] In addition to the boiler-turbine coupling model parameters and the active disturbance rejection controller parameters, the primary frequency regulation load adaptive coefficient is also included in the rolling optimization scope. Using the comprehensive evaluation index of frequency regulation response speed and steam parameter fluctuation amplitude as the optimization objective, the coefficient values ​​for different load ranges are dynamically fine-tuned based on historical frequency regulation operation data. While meeting the grid frequency regulation performance requirements, the impact of frequency regulation actions on the unit's main parameters is minimized, achieving a balance between frequency regulation capability and operational stability.

[0093] The rolling optimization process includes a manual intervention interface, allowing operators to pause the online optimization function at any time based on actual on-site conditions, lock the current model and controller parameters, or manually reset the parameters to their initial default values. This balances the convenience of automatic optimization with the flexibility of manual operation. All parameter change records during the optimization process are stored, including the change time, parameter values ​​before and after the change, and the operating conditions that triggered the optimization, facilitating subsequent operational analysis and fault tracing.

[0094] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent joint control method integrating high-performance PC and high-control-stability DCS, characterized by: Includes the following steps: Step 1: Establish a MODBUS communication link between the high-performance computing PC and the distributed control system (DCS), configure a bidirectional data mapping register, and realize the acquisition and transmission of process operation data from the DCS to the PC, as well as the write-back of control commands from the PC to the DCS. Step 2: Divide and preprocess the acquired full-load operation data of the generator set into operating conditions, and perform intelligent identification of dynamic models for load segments in the turbine-boiler coupling process to obtain the turbine-boiler coupling dynamic model corresponding to each load segment. Step 3: Construct a multi-agent active disturbance rejection control architecture covering the full load range of 20%-100%, divide the controlled object into sub-control agents according to the function of the controlled object, and set up a coordinating agent to achieve decoupling and matching of each sub-control agent; Step 4: Design an active disturbance rejection controller for each sub-control agent. Estimate the total system disturbance in real time by extending the state observer. Combine the output prediction of the machine-furnace coupling dynamic model to generate the basic control quantity. Then, add the total disturbance compensation quantity to obtain the final control correction quantity. Step 5: Based on the grid frequency deviation and the current unit load conditions, generate a primary frequency regulation dynamic compensation command to adapt to the frequency regulation response characteristics of different load segments; Step 6: The control correction value generated by the active disturbance rejection control is superimposed and verified with the primary frequency modulation dynamic compensation command. After rate limiting and amplitude limiting, it is written back to the control execution loop of the DCS through the MODBUS communication link. Step 7: Set a rolling time window to iteratively update the parameters of the turbine-furnace coupling dynamic model and the active disturbance rejection controller based on the latest operating data collected periodically, thereby achieving online optimization of the control strategy.

2. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: In step one, the MODBUS communication link uses the Ethernet TCP protocol for transmission. The process analog quantity and switch quantity status of the DCS are mapped to a continuous holding register address range. The PC reads the register data according to a fixed sampling period. The control instructions generated by the PC are mapped to the writable register address range of the DCS. Each set of instructions corresponds to a double register check bit. When the data is inconsistent, the current frame is discarded and the valid instructions of the previous frame are retained.

3. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: In step two, the turbine-furnace coupled dynamic model adopts an incremental state-space structure, and its calculation equation is as follows: In the formula, The system output increment vector at time k contains the increments of four controlled variables: main steam pressure, main steam temperature, unit active power, and heating extraction steam pressure. for The control input increment vector at any given time includes four control quantity increments: fuel quantity command, total feedwater flow command, high-pressure valve opening command, and first-stage desuperheating water flow command. for The measurable disturbance increment vector at any given time includes the increments of three disturbance quantities: ambient temperature, heating steam extraction flow rate, and raw coal received basis lower heating value. The state transition matrix represents the inertial coupling characteristics of the controlled variable itself. The control gain matrix represents the strength of the effect of the control quantity on the controlled quantity. This is the disturbance gain matrix, which characterizes the degree of influence of the measurable disturbance on the controlled variable; This is the current sampling time number.

4. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: In step three, the sub-control agents are divided into three categories: boiler combustion agent, steam-water system agent, and turbine heating agent. The controlled variables of the boiler combustion agent are the main steam pressure and furnace negative pressure, and the control variables are the total fuel quantity command, the total air supply command, and the total induced draft command. The controlled variables of the steam-water system agent are the main steam temperature and the reheat steam temperature, and the control variables are the first-stage desuperheating water flow command, the second-stage desuperheating water flow command, and the flue gas damper opening command. The controlled variables of the turbine heating agent are the unit active power and the heating extraction steam pressure, and the control variables are the high-pressure regulating valve opening command and the extraction steam regulating valve opening command. The coordinating agent receives the status information of each sub-agent and adjusts the setpoint weights of each sub-agent with the goal of minimizing the boiler-turbine energy balance deviation.

5. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: In step four, the control law calculation equation for the active disturbance rejection controller is: In the formula, Let k be the initial control input at time k; For the controller's proportional gain parameter; The setpoint of the controlled variable at time k; The state observation value of the controlled variable output by the extended state observer at time k; The total system disturbance observation value output by the extended state observer at time k includes unmodeled system dynamics, unknown external disturbances, and coupling disturbances between subsystems; This is the control gain compensation factor, used to match the gain relationship between the control variable and the controlled variable.

6. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: In step five, the calculation equation for the primary frequency modulation dynamic compensation command is: In the formula, The value of the frequency modulation command correction at time k; This is the frequency modulation fundamental gain coefficient; Let k be the power grid frequency deviation value at time k; The rated frequency of the power grid; The rated active power of the unit; This is the load adaptive coefficient, and its value varies with the current unit load rate. Dynamic adjustment: the 20%-40% load range corresponds to the first coefficient range, and the 40%-100% load range corresponds to the second coefficient range.

7. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: In step two, the operating conditions are divided into four continuous intervals based on the unit load rate: 20%-40% low load interval, 40%-60% medium-low load interval, 60%-80% medium-high load interval, and 80%-100% high load interval. Each load interval is identified with an independent model parameter matrix. During the control process, the model parameters of the corresponding interval are called according to the real-time load rate. At the boundary of adjacent load intervals, a linear interpolation method is used to achieve a smooth transition.

8. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: In step six, the instruction write-back process is equipped with a three-level security verification mechanism: the first level is amplitude verification, which forcibly limits the control correction amount to the boundary value when it exceeds the preset upper and lower limits; the second level is rate verification, which limits the change amount of the instruction in two adjacent frames to the maximum rate when it exceeds the preset change rate; the third level is communication status verification, which stops instruction issuance and triggers the DCS to automatically switch to the native control logic when no response is received from the DCS end or the data verification error occurs in multiple consecutive frames.

9. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: In step seven, the length of the rolling time window is set to an integer multiple of the fixed sampling period. During each iteration, the latest running data in the window is substituted into the identification algorithm, and the model parameters are updated using a weighted recursive method, with the weight of the new data being higher than that of the old data. The controller parameters are optimized with the goal of minimizing the sum of squares of the control deviations, and are iteratively adjusted using the gradient descent method. The parameter update step size is dynamically adjusted according to the load fluctuation.

10. The intelligent joint control method integrating a high-performance PC and a high-control-stability DCS according to claim 1, characterized in that: The extended state observer in step four is a third-order linear extended state observer. The inputs are the measured value of the controlled variable and the output value of the control variable. The outputs are the state observation value of the controlled variable, the rate of change observation value of the controlled variable, and the total disturbance observation value of the system, respectively. The observer gain parameters are configured according to the characteristics of each load segment model and are adjusted synchronously as the load interval switches.

Images