A thermoelectric coupling control system for a coal-fired boiler

Abstract

The present application belongs to the field of coal-fired boiler cogeneration control, and discloses a heat and electricity coupling control system for a coal-fired boiler. The present application can improve the consistency, integrity and reliability of multi-source heterogeneous operation data by collecting time sequence operation data of the coal-fired boiler, the steam turbine generator, the heat exchanger and the heat and electricity coupling device, and sequentially performing time alignment, abnormal value correction, missing value completion and feature extraction, thereby providing a reliable data basis for subsequent load prediction and control decision. The heat load and the electricity load are predicted in multiple time scales by a time convolution network based on an attention mechanism, and a fuzzy C-means clustering algorithm is combined to identify the dynamic working condition of the boiler, so that the control system can perceive the change trend of the heat and electricity load in advance, corresponding control strategies are adopted for deep regulation steady state, load increase transient state, load decrease transient state and rated steady state, and the response lag and parameter fluctuation existing in the traditional feedback control are reduced.

Description

Technical Field

[0001] This invention belongs to the field of cogeneration control for coal-fired boilers, and specifically relates to a thermoelectric coupling control system for coal-fired boilers. Background Technology

[0002] Cogeneration systems using coal-fired boilers generate steam through boiler combustion, which then drives a steam turbine generator to produce electricity. Simultaneously, heat energy is supplied to the heating network via steam extraction or heat exchange devices, enabling the cascade utilization of fuel energy. These systems are widely used in industrial production and district heating. In actual operation, there is a strong dynamic coupling relationship between the boiler combustion side, the steam turbine generator side, and the heating network supply side. Changes in operating parameters such as coal feed rate, air distribution rate, feedwater flow rate, main steam parameters, and steam extraction rate all simultaneously affect the unit's power generation load, heating capacity, combustion efficiency, and operational stability.

[0003] As the proportion of new energy sources such as wind power and photovoltaics connected to the grid continues to increase, coal-fired combined heat and power (CHP) units need to undertake more frequent peak-shaving tasks. Electricity load commands and heat user demands exhibit characteristics such as large fluctuations, rapid changes, and different time scales. Traditional CHP control typically employs fixed control logic, PID feedback control, or preset operating curves, adjusting coal feed, air volume, steam pressure, and extraction valve opening based on current parameter deviations. This type of control relies primarily on feedback correction after disturbances occur, making it difficult to anticipate future changes in heat and electricity loads. During load increases / decreases and deep peak shaving, issues such as asynchronous CHP responses, main steam pressure fluctuations, reduced combustion efficiency, and frequent adjustments to control parameters can easily arise.

[0004] Existing technologies have proposed optimization schemes for coal-fired power plant operating parameters based on operational data fusion, load forecasting, and multi-objective optimization algorithms. For example, by collecting coal quality data, real-time operating condition data, control system parameters, and environmental emission data, the multi-source data is standardized, and a time-series forecasting model is used to obtain a predicted load sequence. Then, using combustion efficiency, pollutant emissions, and response delay as optimization objectives, control parameters such as air supply volume, primary and secondary air ratio, coal blending ratio, and burner start-up and shutdown status are generated. Such schemes can improve the adaptive capability of coal-fired unit parameter adjustment to a certain extent.

[0005] However, existing operating parameter optimization technologies mostly focus on the configuration of combustion-side parameters and the balance between power generation load, energy efficiency, and emissions, while neglecting the thermoelectric coupling relationship between the boiler, turbine, and heating network. This makes it difficult to coordinate control variables such as coal feed, air distribution, water supply, and heating steam extraction according to heat and electrical load demands at different time scales. Furthermore, multi-source sensor data may suffer from inconsistent sampling periods, outliers, and missing data, affecting the accuracy of load forecasting and control decisions. Under frequent load variations, the risk of slagging on boiler heating surfaces and the accumulation of thermal stress in the turbine also change dynamically with the control process, making it difficult for existing methods to feed back the equipment health status to the control parameter optimization process in real time. Summary of the Invention

[0006] The purpose of this invention is to overcome the problem that existing thermoelectric coupling control methods for coal-fired boilers are difficult to predict changes in thermoelectric load in advance, and to provide a thermoelectric coupling control system for coal-fired boilers.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a thermoelectric coupling control method for a coal-fired boiler, comprising the following steps: Time-series operation data of coal-fired boilers, steam turbine generators, heat exchangers of heating networks and thermoelectric coupling devices are collected. Time alignment, outlier correction, missing value completion and feature extraction are performed on the time-series operation data to obtain a standardized operation dataset. Standardized operation datasets, external meteorological data, and power grid dispatch information are input into a time convolutional network based on an attention mechanism to obtain multi-timescale heat load prediction sequences and electrical load prediction sequences. Based on the furnace flame center height, flue gas temperature distribution characteristics, and historical load change rate, a fuzzy C-means clustering algorithm is used to determine the dynamic operating condition category of the boiler. Based on the standardized operation dataset, calculate the risk of slagging on the boiler heating surface and the accumulation of thermal stress in the steam turbine, obtain the equipment health status parameters, and determine the corresponding safety constraints based on the equipment health status parameters. The heat load prediction sequence, the electricity load prediction sequence, the boiler dynamic operating condition category, and the equipment health status parameters are input into a two-layer optimization model consisting of a deep deterministic strategy gradient algorithm and a sequential quadratic programming algorithm. The deep deterministic strategy gradient algorithm determines the coarse adjustment reference value and penalty coefficient of the heat-electric power ratio. The coarse adjustment reference value of the heat-electric power ratio is used as the initial solution. The boiler-turbine coupling nonlinear mechanism model is used as the equality constraint, and the safety constraint condition is used as the inequality constraint. The sequential quadratic programming algorithm is used to obtain the heat-electric power allocation result and the control variable sequence. The sequence of control variables is converted into drive commands for the actuator. Based on the drive commands and the operation feedback data of the coal-fired boiler, the coal feed rate, air distribution rate, water feed rate and the opening of the heating steam extraction valve are adjusted in a closed loop. Based on the updated risk of boiler heating surface slagging and turbine thermal stress accumulation during the closed-loop regulation process, the safety constraints are adaptively modified, and the thermoelectric power allocation results and control variable sequence are updated based on the modified safety constraints.

[0008] A further improvement of this invention lies in the following method for collecting time-series operational data from coal-fired boilers, steam turbine generators, heat exchangers in heating networks, and thermoelectric coupling devices, and performing time alignment, outlier correction, missing value completion, and feature extraction on the time-series operational data to obtain a standardized operational dataset: Collect time-series operating data of coal-fired boilers, steam turbine generators, heat exchangers in heating networks, and thermoelectric coupling devices; The combustion-side data of coal-fired boilers, the steam turbine generator-side data, and the heating network-side data are aligned with a unified timestamp, and the time-series operation data of different sampling periods are resampled to the same synchronization time base. The isolated forest algorithm is used to calculate the outlier score of data points in each time window. Data points with outlier scores exceeding the preset outlier threshold are identified as outlier data points. The outlier data points are then corrected based on the normal data before and after them. The missing data segments in the time series data were completed using cubic spline interpolation. Extract furnace temperature distribution, flue gas oxygen content, main steam parameters, reheat steam parameters, extraction steam parameters, and generator output power from the time-series operational data after outlier correction and missing value completion to form a standardized operational dataset.

[0009] A further improvement of this invention lies in inputting standardized operational datasets, external meteorological data, and power grid dispatch information into a time-convolutional network based on an attention mechanism to obtain multi-time-scale heat load prediction sequences and electrical load prediction sequences. The specific method for determining the boiler's dynamic operating condition category using a fuzzy C-means clustering algorithm, based on furnace flame center height, flue gas temperature distribution characteristics, and historical load change rates, is as follows: The standardized operational dataset from a preset historical time period, the ambient temperature, humidity and wind speed from a preset future time period, and the automatic power generation control commands of the power grid are input into a temporal convolutional network based on an attention mechanism. Temporal features of the standardized running dataset are extracted by dilated convolutional layers in a temporal convolutional network. The weights of the temporal features at each historical moment are determined by an attention mechanism, and the heat load prediction intervals and electricity load prediction intervals corresponding to multiple future prediction periods are output. Based on the furnace flame center height, flue gas temperature distribution characteristics, and historical load change rate, the fuzzy C-means clustering algorithm is used to calculate the membership degree of the current operating condition to the deep adjustment steady state, the load increase transient state, the load decrease transient state, and the rated steady state, respectively. The operating condition with the highest membership degree is determined as the boiler dynamic operating condition category.

[0010] A further improvement of this invention lies in the following method for calculating the risk of slagging on the boiler heating surface and the accumulation of thermal stress in the turbine based on a standardized operating dataset, obtaining equipment health status parameters, and determining the corresponding safety constraints based on these parameters: Extract the furnace outlet flue gas temperature, fly ash carbon content, and coal softening temperature from the standardized operating dataset. Input the furnace outlet flue gas temperature, fly ash carbon content, and coal softening temperature into the random forest regression model to obtain the slagging probability of the screen-type superheater and the high-temperature superheater. The turbine regulating stage temperature and rotor stress concentration factor are extracted from the standardized operation dataset. The thermal stress change process of the turbine rotor is determined based on the turbine regulating stage temperature and rotor stress concentration factor. The equivalent cycle number corresponding to the load change is counted using the rainflow counting method to obtain the turbine thermal stress accumulation. The equipment health status parameters are determined based on the slagging probability and the cumulative thermal stress of the steam turbine, and the corresponding safety constraints are determined based on the equipment health status parameters.

[0011] A further improvement of this invention lies in inputting the heat load prediction sequence, electrical load prediction sequence, boiler dynamic operating condition category, and equipment health status parameters into a two-layer optimization model composed of a deep deterministic strategy gradient algorithm and a sequential quadratic programming algorithm. The deep deterministic strategy gradient algorithm determines the coarse-adjustment reference value and penalty coefficient for the heat-to-electric power ratio. Using the coarse-adjustment reference value for the heat-to-electric power ratio as the initial solution, and the boiler-turbine coupling nonlinear mechanism model as the equality constraint, and the safety constraint condition as the inequality constraint, the specific method for obtaining the heat-to-electric power allocation result and the control variable sequence through the sequential quadratic programming algorithm is as follows: The heat load prediction sequence, the electrical load prediction sequence, the boiler dynamic operating condition category, and the equipment health status parameters are used as the state inputs of the deep deterministic policy gradient algorithm. The actor network of the deep deterministic policy gradient algorithm outputs the coarse adjustment reference value of the thermoelectric power ratio, the boiler combustion intensity correction coefficient, and the penalty coefficient. The coarse adjustment reference value of the thermoelectric power ratio is used as the initial solution of the sequential quadratic programming algorithm. Based on the boiler-turbine coupled nonlinear mechanism model, the equation constraints between coal feed rate, water flow rate and turbine regulating valve opening and main steam pressure, main steam temperature and power generation are established. Inequality constraints are established based on the rate of change of the metal temperature of the heated surface, the fluctuation range of the main steam pressure, the fluctuation range of the steam drum water level, the flue gas temperature at the furnace outlet, and the concentration of pollutant emissions. Using the cumulative coal consumption within a preset optimization period as the objective function, the sequential quadratic programming algorithm is used for iterative solution to obtain the thermal power allocation results, as well as the coal feed sequence, air distribution sequence, water supply flow sequence, and heating extraction steam valve opening sequence.

[0012] A further improvement of this invention lies in converting the sequence of control variables into drive commands for the actuator. The specific method for closed-loop regulation of coal feed rate, air distribution rate, water feed rate, and heating steam extraction valve opening based on the drive commands and operational feedback data from the coal-fired boiler is as follows: The coal feed rate sequence is used as a feedforward signal input to the coal feed rate feedforward-feedback controller. The feedback correction amount is determined based on the deviation between the measured value of the main steam pressure and the set value of the main steam pressure. The feedforward signal and the feedback correction amount are superimposed to generate the coal feeder drive command. The theoretical air volume is determined based on the real-time coal feed rate, and the theoretical air volume is corrected based on the oxygen content of the flue gas at the furnace outlet. The primary air volume and secondary air volume are determined by the air-coal ratio cross-limiting controller. Based on the unit load command and the turbine regulating valve opening, the main steam pressure setpoint is determined by a rolling optimization method, and the feedwater flow rate is adjusted according to the main steam pressure setpoint. Adjust the opening degree of the heating extraction steam valves according to the opening sequence of the heating extraction steam valves and the supply and return water parameters of the heating network.

[0013] A further improvement of this invention lies in the adaptive correction of safety constraints based on the updated risk of boiler heating surface slagging and turbine thermal stress accumulation during the closed-loop regulation process. The specific method for updating the thermoelectric power allocation result and control variable sequence based on the corrected safety constraints is as follows: The updated boiler heating surface slagging risk is compared with the preset slagging risk threshold. When the boiler heating surface slagging risk exceeds the preset slagging risk threshold, the safety constraints corresponding to the furnace outlet flue gas temperature and the furnace outlet flue gas temperature change rate in the dual-layer optimization model are tightened. The updated turbine thermal stress accumulation is compared with the preset thermal stress threshold. When the turbine thermal stress accumulation exceeds the preset thermal stress threshold, the safety constraints corresponding to the unit load change rate and the heat transfer surface metal temperature change rate in the dual-layer optimization model are tightened. The revised safety constraints are re-input into the sequential quadratic programming algorithm to continuously update the thermoelectric power allocation results and the sequence of control variables.

[0014] In a second aspect, the present invention provides a thermoelectric coupling control system for a coal-fired boiler, comprising: The multi-source data fusion and preprocessing module is used to collect time-series operation data of coal-fired boilers, steam turbine generators, heat exchangers of heating networks and thermoelectric coupling devices, and to perform time alignment, outlier correction, missing value completion and feature extraction on the time-series operation data to obtain a standardized operation dataset. The load forecasting and operating condition identification module is used to input standardized operating datasets, external meteorological data and power grid dispatch information into a time convolutional network based on an attention mechanism to obtain multi-timescale heat load forecasting sequences and electrical load forecasting sequences. Based on the furnace flame center height, flue gas temperature distribution characteristics and historical load change rate, the fuzzy C-means clustering algorithm is used to determine the dynamic operating condition category of the boiler. The health status assessment and adaptive correction module is used to calculate the risk of slagging on the boiler heating surface and the accumulation of thermal stress in the steam turbine based on the standardized operation dataset, obtain equipment health status parameters, and determine the corresponding safety constraints based on the equipment health status parameters. The hybrid intelligent optimization decision module is used to input the heat load prediction sequence, the electricity load prediction sequence, the boiler dynamic operating condition category, and the equipment health status parameters into a two-layer optimization model composed of a deep deterministic strategy gradient algorithm and a sequential quadratic programming algorithm. The deep deterministic strategy gradient algorithm determines the coarse adjustment reference value and penalty coefficient of the heat-electric power ratio. The coarse adjustment reference value of the heat-electric power ratio is used as the initial solution. The boiler-turbine coupling nonlinear mechanism model is used as the equality constraint, and the safety constraint condition is used as the inequality constraint. The sequential quadratic programming algorithm obtains the heat-electric power allocation result and the control variable sequence. The dynamic coordination control execution module is used to convert the sequence of control variables into drive commands for the actuators. Based on the drive commands and the operation feedback data of the coal-fired boiler, it performs closed-loop regulation of the coal feed rate, air distribution rate, water feed rate, and the opening of the heating extraction steam valve. The adaptive correction module is used to adaptively correct the safety constraints based on the boiler heating surface slagging risk and turbine thermal stress accumulation updated during the closed-loop regulation process, and update the thermoelectric power allocation results and control variable sequence based on the corrected safety constraints.

[0015] A further improvement of the present invention is that the dynamic coordination control execution module includes a coal feed rate feedforward-feedback controller, an air-coal ratio cross-limit controller, and a sliding pressure setpoint optimizer. The coal feed rate feedforward-feedback controller is used to take the coal feed rate sequence as a feedforward signal and determine the feedback correction amount based on the real-time deviation of the main steam pressure. The air-coal ratio cross-limit controller is used to determine the primary air volume and secondary air volume based on the coal feed rate and the excess air coefficient at the furnace outlet, and to limit the primary air volume and secondary air volume to a minimum. The sliding pressure setpoint optimizer is used to determine the main steam pressure setpoint in a rolling manner based on the unit load command and the turbine regulating valve opening, using a model predictive control method.

[0016] A further improvement of the present invention is that it also includes a human-computer interaction and edge computing gateway module, which is connected to a multi-source data fusion and preprocessing module, a load forecasting and operating condition identification module, a health status assessment and adaptive correction module, a hybrid intelligent optimization decision-making module and a dynamic coordination control execution module, respectively. The human-computer interaction and edge computing gateway module is equipped with an interpretable artificial intelligence component, which is used to calculate the contribution of each input feature to the output control variable in the hybrid intelligent optimization decision module, and generate corresponding decision basis data. The human-computer interaction and edge computing gateway module supports OPC UA and MQTT protocols, which are used to upload the operating parameters, equipment health status parameters and system self-test status of coal-fired boilers to the cloud production management system and receive power grid dispatch information.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention improves the consistency, completeness, and reliability of multi-source heterogeneous operating data by collecting time-series operating data from coal-fired boilers, steam turbine generators, heat exchangers in heating networks, and thermoelectric coupling devices, and sequentially performing time alignment, outlier correction, missing value completion, and feature extraction. This provides a reliable data foundation for subsequent load forecasting and control decisions. By using a time-scale prediction of thermal and electrical loads through an attention-based temporal convolutional network and combining it with a fuzzy C-means clustering algorithm to identify boiler dynamic conditions, the control system can anticipate trends in thermal and electrical load changes and implement corresponding control strategies for deep-adjustment steady state, load increase transients, load decrease transients, and rated steady state, reducing the response lag and parameter fluctuations inherent in traditional feedback control. By calculating the risk of slagging on boiler heating surfaces and the accumulation of thermal stress in the steam turbine, and transforming the equipment health status into safety constraints for a two-layer optimization model, the invention can meet thermal and electrical load demands while considering equipment safety and lifespan. By combining a deep deterministic strategy gradient algorithm with a sequential quadratic programming algorithm, the invention balances global optimization capability and local solution accuracy to obtain reasonable thermal and electrical power allocation results and control variable sequences. Furthermore, by implementing closed-loop regulation of coal feed rate, air volume, water flow rate, and heating steam extraction valve opening, and dynamically correcting safety constraints based on updated slagging risk and thermal stress accumulation, closed-loop control of prediction, optimization, execution, and correction is achieved, thereby improving the thermal-electric coordination capability, operational stability, and overall energy utilization efficiency of coal-fired boilers. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall system modules of the present invention.

[0019] Figure 2 This is a schematic diagram of the internal two-layer architecture data flow of the hybrid intelligent optimization decision module of the present invention; Figure 3This is a schematic diagram of the signal interaction between the dynamic coordination control execution module and the boiler field actuator of the present invention. Detailed Implementation

[0020] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0021] like Figure 1 As shown, the thermoelectric coupling control system of the present invention includes a multi-source data fusion and preprocessing module 100, a load prediction and operating condition identification module 200, a hybrid intelligent optimization decision-making module 300, a dynamic coordinated control execution module 400, a health status assessment and adaptive correction module 500, and a human-machine interaction and edge computing gateway module 600. The health status assessment function in the health status assessment and adaptive correction module 500 is used to calculate the slagging risk of the boiler heating surface and the thermal stress accumulation of the turbine. The adaptive correction function can be deployed as an independent adaptive correction module to correct the safety constraints in the hybrid intelligent optimization decision-making module 300 based on the updated slagging risk and thermal stress accumulation.

[0022] In this embodiment, the multi-source data fusion and preprocessing module 100 is connected to sensor groups on the coal-fired boiler, steam turbine generator, heat exchanger of the heating network, and thermoelectric coupling device. The sensor groups include thermocouples, pressure transmitters, vortex flow meters, zirconia oxygen meters, and an acoustic temperature measuring device for detecting the height of the furnace flame center. Each sensor can transmit its acquired time-series data to the multi-source data fusion and preprocessing module 100 via a 4-20mA current loop, Modbus TCP / IP protocol, or industrial Ethernet.

[0023] The time-series operational data includes at least the combustion-side data of the coal-fired boiler, the turbine generator-side data, and the heating network-side data. The combustion-side data of the coal-fired boiler includes the furnace temperature field distribution, furnace outlet flue gas temperature, wall temperatures of each heating surface, flue gas oxygen content, fly ash carbon content, coal softening temperature, coal feed rate, primary air volume, and secondary air volume. The turbine generator-side data includes main steam pressure, main steam temperature, main steam flow rate, reheat steam pressure, reheat steam temperature, turbine regulating stage temperature, turbine regulating valve opening, extraction steam pressure, extraction steam temperature, and generator set output power. The heating network-side data includes the heating network supply water temperature, return water temperature, supply water flow rate, return water flow rate, heating network pressure, and the opening of the heating extraction valve.

[0024] Because different sensors have different sampling periods—for example, the sampling period for pressure signals is 100ms, for temperature signals it's 1s, and for flow signals it's 5s—a heterogeneous timing data alignment unit is first used to configure a unified timestamp for each timing data set. Then, timing data with different sampling periods are resampled to a unified 5s synchronization time base. For data with sampling frequencies higher than the 5s time base, downsampling can be performed based on data within the corresponding time interval. For data lacking synchronization data between adjacent sampling moments, linear interpolation can be used to obtain the data values ​​at the synchronization moment.

[0025] After time alignment, the Isolation Forest algorithm is used to detect anomalies in data points within each time window. The Isolation Forest algorithm calculates anomaly scores based on the average path length of each data point across multiple randomly isolated trees. When the anomaly score of a data point exceeds a preset anomaly threshold, that data point is marked as an anomaly. In this embodiment, the preset anomaly threshold is set to 0.65. For each marked anomaly, a weighted average is calculated based on the normal data preceding and following it, and the resulting weighted average replaces the original anomaly.

[0026] For data gaps caused by sensor malfunctions, network packet loss, or communication interruptions, cubic spline interpolation is used for data completion. After time alignment, outlier correction, and missing value completion, feature extraction and standardization are performed on each parameter to generate a continuous and complete standardized runtime dataset. The standardized runtime dataset can be represented as:

[0027] In the formula, This represents a standardized running dataset; Indicates the first The running state vector at each synchronous time point; This indicates the number of synchronization time points contained in the running data.

[0028] In this embodiment, each operating state vector includes a total of 48 feature dimensions, such as furnace temperature field distribution characteristics, flue gas oxygen content, main steam parameters, reheat steam parameters, extraction steam parameters, heating network supply and return water parameters, and generator set output power.

[0029] After the standardized operation dataset is generated, it is sent to the load forecasting and operating condition identification module 200, the health status assessment and adaptive correction module 500, and the hybrid intelligent optimization decision-making module 300.

[0030] The load forecasting and operating condition identification module 200 acquires external meteorological data and power grid dispatch information through the human-machine interaction and edge computing gateway module 600. The external meteorological data includes ambient temperature, ambient humidity, and wind speed for a preset time period in the future, while the power grid dispatch information includes the automatic generation control command sequence and target load command issued by the power grid dispatch center.

[0031] The load forecasting and operating condition identification module 200 internally deploys a pre-trained temporal convolutional network based on an attention mechanism. The temporal convolutional network takes as input a standardized operating dataset within a preset historical time period and external meteorological data and power grid dispatch information within a preset future time period. Preferably, the preset historical time period is 2 hours before the current time, and the preset future time period is 4 hours after the current time.

[0032] When the synchronization time base of the standardized operating dataset is 5 seconds, the input window length of the temporal convolutional network can be set to 1440 time steps, corresponding to the historical operating data of the two hours prior to the current moment. The temporal convolutional network expands its receptive field layer by layer through multiple layers of causal convolution and dilated convolution, ensuring that the data at the prediction moment does not use information from after the prediction moment. The attention mechanism determines the weights of temporal features based on the correlation between the operating characteristics of each historical moment and the current prediction target, causing the network to focus on historical moments related to load changes. For example, when the unit is about to increase its load, the attention mechanism can increase the feature weights of the load change rate, main steam pressure change rate, and coal feed rate change rate within the 20 minutes prior to the current moment.

[0033] The temporal convolutional network outputs heat load prediction sequences and electrical load prediction sequences corresponding to multiple prediction time periods. In one specific setting, the output window length is set to 240 time steps and 480 time steps, respectively, corresponding to the load prediction results for the next 20 minutes and the next 40 minutes. In other settings, multiple prediction windows for the next 30 minutes, 1 hour, 2 hours, and 4 hours can be set according to scheduling needs, thereby forming multi-timescale prediction results for the next 30 minutes to 4 hours.

[0034] The prediction results output by the temporal convolutional network include the heat load prediction interval, the electrical load prediction interval, and the corresponding confidence level for each prediction result. In this embodiment, the load prediction interval is represented by a 90% confidence interval. For example, when the current electrical load is 200MW, the temporal convolutional network predicts that the electrical load will increase to 230MW in 15 minutes, corresponding to a prediction confidence level of 85%. The predicted heat load and electrical load sequences can be represented as follows:

[0035] and

[0036] In the formula, Indicates the current time; This indicates the number of time steps within the prediction time range.

[0037] While performing load forecasting, the load forecasting and operating condition identification module 200 identifies the dynamic operating condition of the boiler based on the furnace flame center height, flue gas temperature distribution characteristics, and historical load change rate. The furnace flame center height can be obtained by acoustic temperature measuring devices installed at different heights in the furnace, and the flue gas temperature distribution characteristics can be extracted from the temperature measuring points at the furnace outlet and each heating surface area.

[0038] The furnace flame center height, flue gas temperature distribution characteristics, and historical load change rate are used to form an operating condition feature vector. A fuzzy C-means clustering algorithm is then used to calculate the membership degree of this feature vector to four typical operating condition categories: deep-adjustment steady state, load increase transient state, load decrease transient state, and rated steady state. The fuzzy C-means clustering algorithm outputs four membership degree values. The operating condition category with the highest membership degree is determined as the current dynamic operating condition category of the boiler and sent to the hybrid intelligent optimization decision module 300.

[0039] The Health Status Assessment and Adaptive Correction Module 500 calculates the risk of slagging on the boiler heating surfaces and the accumulation of thermal stress in the turbine based on a standardized operating dataset.

[0040] To assess the risk of slagging on boiler heating surfaces, the flue gas temperature at the furnace outlet, fly ash carbon content, and coal softening temperature are first extracted from a standardized operating dataset and then input into a pre-trained random forest regression model. The random forest regression model consists of multiple regression decision trees, each outputting a corresponding predicted value for slagging risk. The outputs of each regression decision tree are then integrated to obtain the slagging probability of the screen-type superheater and the high-temperature superheater.

[0041] For the accumulation of thermal stress in steam turbines, the temperature of the regulating stage and the stress concentration factor of the rotor are extracted from the standardized operation dataset. The thermal stress change process of the steam turbine rotor is determined according to the temperature change process of the regulating stage. Then, the irregular thermal stress change process is decomposed into load rise and fall cycles of different amplitudes using the rainflow counting method. The equivalent cycle number corresponding to each load rise and fall cycle is counted, and the cumulative fatigue life loss of the steam turbine rotor is calculated based on the equivalent cycle number.

[0042] The probability of slagging on the boiler heating surface, the thermal stress of the steam turbine, the equivalent number of cycles, and the cumulative fatigue life loss together constitute the equipment health status parameters, which can be denoted as a health factor vector:

[0043] The safety constraints in the hybrid intelligent optimization decision module 300 are determined based on the equipment health status parameters. These safety constraints include at least the following: the rate of change of the heated surface metal temperature; the furnace outlet flue gas temperature; the rate of change of the furnace outlet flue gas temperature; the unit load change rate; and the turbine thermal stress.

[0044] like Figure 2 As shown, the hybrid intelligent optimization decision module 300 adopts a two-layer optimization model consisting of an upper-layer deep deterministic policy gradient agent and a lower-layer sequential quadratic programming solver.

[0045] The upper-layer deep deterministic policy gradient agent consists of an actor network and an evaluator network. The actor network receives the load forecast sequence, the current boiler dynamic condition category, and the health factor vector as state inputs, i.e., the state inputs include:

[0046] In the formula, This represents the heat load forecast sequence and the electricity load forecast sequence within the future forecast time range; Indicates the current dynamic operating condition category of the boiler; This indicates the current health status parameters of the device.

[0047] The actor network outputs a continuous action vector with a dimension of 3:

[0048] In the formula, This indicates the coarse adjustment reference value for the thermoelectric power ratio; This represents the boiler combustion intensity correction factor; This represents the penalty coefficient. The evaluator network is used to evaluate the value of the action output by the actor network and updates the network parameters of both the actor network and the evaluator network based on the reward signal obtained from environmental feedback.

[0049] The reward function for a deep deterministic policy gradient agent is:

[0050] In the formula, Indicates the reward value; This indicates the coal consumption under the corresponding control action; This indicates the amount of nitrogen oxides emitted under the corresponding control action; This indicates an additional penalty that occurs when the thermal stress of the steam turbine exceeds the safety threshold. , and These represent the weighting coefficients corresponding to coal consumption, nitrogen oxide emissions, and equipment thermal stress, respectively.

[0051] Through the aforementioned reward function, the deep deterministic policy gradient agent balances the reduction of coal consumption, the reduction of pollutant emissions, and the protection of unit equipment, and outputs a coarse-adjustment reference value for the thermoelectric power ratio that meets the current load demand and the health status of the equipment.

[0052] The lower-level sequential quadratic programming solver receives the coarse-tuned reference value of the thermoelectric power ratio output by the actor network as the initial solution. The sequential quadratic programming solver incorporates a reduced-order boiler-turbine coupled nonlinear mechanism model. This mechanism model is based on the conservation of mass and energy, with the coal feed rate as a reference. Water supply flow rate Steam turbine regulating valve opening The primary air volume, secondary air volume, and the opening degree of the heating steam extraction valve are used as inputs, with the main steam pressure as the input. Main steam temperature Power generation And heating power as output.

[0053] A boiler-turbine coupled nonlinear mechanism model is used to establish equality constraints between coal feed rate, feedwater flow rate, turbine regulating valve opening, main steam pressure, main steam temperature, and power generation. Inequality constraints are established based on equipment health status parameters, boiler operating safety boundaries, and pollutant emission limits.

[0054] In this embodiment, the inequality constraints include at least:

[0055]

[0056]

[0057]

[0058] In the formula, This represents the measured value of the main steam pressure; This indicates the main steam pressure setpoint; This represents the measured value of the steam drum water level; This indicates the setpoint for the steam drum water level; This indicates the temperature of the flue gas at the furnace outlet; This indicates the concentration of nitrogen oxide emissions.

[0059] In addition, the inequality constraints also include the rate of change of the temperature of the heated surface metal, the thermal stress of the steam turbine, the rate of change of the unit load, and the upper and lower limits of the heating parameters.

[0060] The sequential quadratic programming solver aims to minimize the cumulative coal consumption within the next 15 minutes. It iteratively solves the problem while satisfying heat load demand, electricity load demand, boiler-turbine coupling, and safety constraints, obtaining the heat and power allocation results and the optimal control variable sequence. The optimal control variable sequence includes the coal feed rate sequence, air distribution sequence, water feed rate sequence, and heating extraction steam valve opening sequence, which can be expressed as follows:

[0061] In one specific setup, the optimization timeframe is 15 minutes into the future, and the time interval between adjacent control commands in the control variable sequence is 30 seconds. Therefore, each type of control variable includes 30 control command values. The sequential quadratic programming solver can obtain a control scheme that satisfies all safety constraints within seconds.

[0062] The dynamic coordination control execution module 400 receives the control variable sequence output by the hybrid intelligent optimization decision module 300 and converts the control variable sequence into drive instructions that the underlying actuators can recognize. For example... Figure 3 As shown, the dynamic coordination control execution module 400 includes a coal feed rate feedforward-feedback controller, an air-coal ratio cross-limit controller, and a sliding pressure setpoint optimizer.

[0063] The coal feed rate feedforward-feedback controller will input the coal feed rate sequence This serves as a feedforward signal, which is then sent to the coal feeder frequency converter. Simultaneously, the measured main steam pressure is collected, and the deviation between the measured value and the setpoint is calculated. This deviation is input to the PI controller, which generates a feedback correction. The feedforward signal and the feedback correction are superimposed to form the coal feeder drive command, which the coal feeder frequency converter uses to adjust the coal feed rate.

[0064] The air-coal ratio cross-limit controller calculates the theoretical air volume based on the real-time coal feed rate, performs closed-loop correction on the theoretical air volume based on the oxygen content of the flue gas at the furnace outlet, and determines the primary and secondary air volumes based on the trend of coal feed rate changes. The air-coal ratio cross-limit controller sets minimum air volume safety limits for the primary and secondary air volumes to ensure that the primary air volume is never lower than the lower limit required to prevent backfire.

[0065] For example, when the coal feed rate rapidly increases from 30t / h to 35t / h, the air-coal ratio cross-limiting controller first increases the secondary air volume, and then increases the primary air volume after a delay of 2-3 seconds. This ensures complete combustion of the pulverized coal airflow while preventing the burners from being extinguished due to a sudden increase in primary air volume. When the coal feed rate decreases, the air-coal ratio cross-limiting controller first reduces the primary air volume, and then reduces the secondary air volume, to avoid excessive fuel or oxygen fluctuations in the furnace.

[0066] The sliding pressure setpoint optimizer uses a model predictive control method to continuously calculate the main steam pressure setpoint based on the unit load command and the turbine regulating valve opening. When the unit is under partial load, the turbine regulating valve opening is typically maintained at 70%–85%. The sliding pressure setpoint optimizer determines the main steam pressure setpoint based on the turbine regulating valve opening, the current main steam pressure, and the predicted electrical and thermal load demands, replacing the fixed sliding pressure operating curve and reducing turbine throttling losses.

[0067] The dynamic coordination control execution module 400 adjusts the feedwater flow rate according to the main steam pressure setpoint, and adjusts the opening of the heating extraction steam valve according to the opening sequence of the heating extraction steam valve and the supply and return water parameters of the heating network. When the heating network supply water temperature is lower than the setpoint or the heat load forecast result increases, the opening of the heating extraction steam valve is increased; when the heating network supply water temperature is higher than the setpoint or the predicted heat load decreases, the opening of the heating extraction steam valve is decreased.

[0068] During the closed-loop regulation process, the health status assessment and adaptive correction module 500 continuously acquires updated operational feedback data and recalculates the risk of slagging on the boiler heating surface and the accumulation of thermal stress in the turbine.

[0069] In one specific embodiment, the detected flue gas temperature at the furnace outlet is 950°C, the softening temperature of the current coal type is approximately 1100°C, and the carbon content of the fly ash is approximately 3%. Inputting the above data into a random forest regression model yields a slagging probability of 75% for the high-temperature superheater. Since the slagging probability exceeds the preset slagging risk threshold of 60%, the adaptive correction module sends a safety constraint correction request to the hybrid intelligent optimization decision module 300.

[0070] Upon receiving a correction request, the hybrid intelligent optimization decision module 300 tightens the furnace outlet flue gas temperature constraint in the sequential quadratic programming solver and adds a safety constraint that the rate of change of the furnace outlet flue gas temperature does not exceed +5℃ / min. When resolving the problem, the sequential quadratic programming solver prioritizes increasing the opening of the upper burnout damper or adjusting the upper secondary air volume to reduce the flue gas temperature at the furnace flame center and in the high-temperature heating surface area. This adjustment process may reduce combustion efficiency by approximately 0.2%, but it reduces the risk of slagging in the high-temperature superheater, avoiding furnace shutdown and slagging removal due to severe slagging.

[0071] When the risk of slagging exceeds a preset threshold, the frequency of soot blowing can be increased according to actual operational needs to remove ash and slagging from the surfaces of the screen-type superheater and high-temperature superheater. The amount of increase in the soot blowing frequency can be determined based on the difference between the slagging probability and the preset slagging risk threshold.

[0072] In another specific embodiment, the turbine rotor low-cycle fatigue life loss tracking unit uses the rainflow counting method to record that the unit has experienced three load increase / decrease cycles with a load change range of 50MW on that day, corresponding to a cumulative consumption of 0.05% of the design fatigue life. When the cumulative fatigue life loss approaches the preset warning value, the adaptive correction module tightens the unit load change rate constraint and the heated surface metal temperature change rate constraint.

[0073] When the load change rate command issued by the dispatch center exceeds 15 MW / min, the system refuses to execute the portion exceeding the safety constraint range and sends feedback information about the unit's limited ramp-up capability to the dispatch center through the human-machine interaction and edge computing gateway module 600, requesting a reduction in the load change rate. In other embodiments, when the accumulated fatigue life loss approaches the design life limit, the original unit load change rate constraint can be tightened by 30%.

[0074] The revised safety constraints are re-input into the sequential quadratic programming solver. Based on the revised constraints, the sequential quadratic programming solver recalculates the thermal power allocation results, as well as the coal feed sequence, air distribution sequence, water supply flow sequence, and heating extraction steam valve opening sequence, thereby forming an adaptive safety closed loop of interaction between equipment health status and control decisions.

[0075] The Human-Computer Interaction and Edge Computing Gateway Module 600 provides a web-based visual monitoring interface. This interface displays the heat load forecast curve, electrical load forecast curve, current boiler dynamic operating condition category, boiler heating surface slagging probability, turbine thermal stress accumulation, thermal and electrical power distribution results, and the current and predicted values ​​of each control variable.

[0076] The human-computer interaction and edge computing gateway module 600 is equipped with an interpretable artificial intelligence component. This component can call the SHAP library to calculate the contribution of each input feature in the hybrid intelligent optimization decision-making module 300 to the Shapley value of the output control variable, and generate decision basis text and a visual heatmap based on the contribution.

[0077] For example, in a specific coal feed rate optimization decision, the load forecast contributes 40% to the coal feed rate decision, the risk of boiler heating surface slagging contributes 25%, and the remaining contribution is determined by factors such as the current operating condition, main steam pressure deviation, heating network load, and turbine thermal stress. The visual monitoring interface overlays the corresponding thermograms on the load trend graph and the boiler 3D model, enabling operators to identify the main factors affecting the current control decision.

[0078] When the system predicts that the future load will decrease, the interpretable artificial intelligence component generates corresponding decision-making basis data to explain why the system reduces the upper secondary air volume or adjusts the coal feed, so that operators can understand the decision-making process of the hybrid intelligent optimization decision module 300.

[0079] The Human-Machine Interaction and Edge Computing Gateway Module 600 supports OPC UA and MQTT protocols. It can interact with the distributed control system of coal-fired boilers, the digital electro-hydraulic control system of steam turbines, and the heating network control system via OPC UA protocol. It can also upload more than 20 key operating parameters, equipment health status parameters, key control commands, and system self-test reports to the enterprise private cloud platform via MQTT protocol through the 5G network.

[0080] The enterprise private cloud platform can store historical system operation data and perform statistical analysis on coal consumption, heat supply, power generation, pollutant emissions, and equipment health status during different operating cycles. The human-machine interaction and edge computing gateway module 600 is also used to receive remote target load commands from the cloud dispatching system or the power grid dispatching center, realizing data communication between the local control system and the cloud production management system.

[0081] The overall working process of this invention is as follows: First, the multi-source data fusion and preprocessing module 100 collects field sensor data from the coal-fired boiler, steam turbine generator, heat exchanger of the heating network, and thermoelectric coupling device. It performs time alignment, outlier correction, missing value completion, and feature extraction on the field sensor data to obtain a standardized operating dataset. The load forecasting and operating condition identification module 200 predicts future heat and electrical loads based on the standardized operating dataset, external meteorological data, and power grid dispatch information, and identifies the current dynamic operating condition of the boiler. The health status assessment and adaptive correction module 500 calculates the risk of slagging on the boiler heating surface and the accumulation of thermal stress in the steam turbine to obtain equipment health status. The system calculates the state parameters and determines the safety constraints; the hybrid intelligent optimization decision module 300 provides a coarse adjustment reference value for the thermoelectric power ratio through the upper-level deep deterministic strategy gradient intelligent agent, and then calculates the control variable sequence that meets the safety constraints and economic operation requirements through the lower-level sequential quadratic programming solver; the dynamic coordination control execution module 400 converts the control variable sequence into the corresponding drive commands for the coal feeding, air distribution, water supply and heating steam extraction valves, and performs closed-loop adjustment based on the operation feedback data; the adaptive correction module dynamically corrects the safety constraints based on the updated slagging risk and turbine thermal stress accumulation, and continuously updates the thermoelectric power allocation results and control variable sequence.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A thermoelectric coupling control method for a coal-fired boiler, characterized in that, Includes the following steps: Time-series operation data of coal-fired boilers, steam turbine generators, heat exchangers of heating networks and thermoelectric coupling devices are collected. Time alignment, outlier correction, missing value completion and feature extraction are performed on the time-series operation data to obtain a standardized operation dataset. Standardized operation datasets, external meteorological data, and power grid dispatch information are input into a time convolutional network based on an attention mechanism to obtain multi-timescale heat load prediction sequences and electrical load prediction sequences. Based on the furnace flame center height, flue gas temperature distribution characteristics, and historical load change rate, a fuzzy C-means clustering algorithm is used to determine the dynamic operating condition category of the boiler. Based on the standardized operation dataset, calculate the risk of slagging on the boiler heating surface and the accumulation of thermal stress in the steam turbine, obtain the equipment health status parameters, and determine the corresponding safety constraints based on the equipment health status parameters. The heat load prediction sequence, the electricity load prediction sequence, the boiler dynamic operating condition category, and the equipment health status parameters are input into a two-layer optimization model consisting of a deep deterministic strategy gradient algorithm and a sequential quadratic programming algorithm. The deep deterministic strategy gradient algorithm determines the coarse adjustment reference value and penalty coefficient of the heat-electric power ratio. The coarse adjustment reference value of the heat-electric power ratio is used as the initial solution. The boiler-turbine coupling nonlinear mechanism model is used as the equality constraint, and the safety constraint condition is used as the inequality constraint. The sequential quadratic programming algorithm is used to obtain the heat-electric power allocation result and the control variable sequence. The sequence of control variables is converted into drive commands for the actuator. Based on the drive commands and the operation feedback data of the coal-fired boiler, the coal feed rate, air distribution rate, water feed rate and the opening of the heating steam extraction valve are adjusted in a closed loop. Based on the updated risk of boiler heating surface slagging and turbine thermal stress accumulation during the closed-loop regulation process, the safety constraints are adaptively modified, and the thermoelectric power allocation results and control variable sequence are updated based on the modified safety constraints.

2. The thermoelectric coupling control method for a coal-fired boiler according to claim 1, characterized in that, The specific method for collecting time-series operational data from coal-fired boilers, steam turbine generators, heat exchangers in heating networks, and thermoelectric coupling devices, and performing time alignment, outlier correction, missing value completion, and feature extraction on the time-series operational data to obtain a standardized operational dataset is as follows: Collect time-series operating data of coal-fired boilers, steam turbine generators, heat exchangers in heating networks, and thermoelectric coupling devices; The combustion-side data of coal-fired boilers, the steam turbine generator-side data, and the heating network-side data are aligned with a unified timestamp, and the time-series operation data of different sampling periods are resampled to the same synchronization time base. The isolated forest algorithm is used to calculate the outlier score of data points in each time window. Data points with outlier scores exceeding the preset outlier threshold are identified as outlier data points. The outlier data points are then corrected based on the normal data before and after them. The missing data segments in the time series data were completed using cubic spline interpolation. Extract furnace temperature distribution, flue gas oxygen content, main steam parameters, reheat steam parameters, extraction steam parameters, and generator output power from the time-series operational data after outlier correction and missing value completion to form a standardized operational dataset.

3. The thermoelectric coupling control method for a coal-fired boiler according to claim 1, characterized in that, Standardized operational datasets, external meteorological data, and power grid dispatch information are input into a time-convolutional network based on an attention mechanism to obtain multi-timescale heat load prediction sequences and electrical load prediction sequences. The specific method for determining the boiler's dynamic operating condition category using a fuzzy C-means clustering algorithm, based on furnace flame center height, flue gas temperature distribution characteristics, and historical load change rates, is as follows: The standardized operational dataset from a preset historical time period, the ambient temperature, humidity and wind speed from a preset future time period, and the automatic power generation control commands of the power grid are input into a temporal convolutional network based on an attention mechanism. Temporal features of the standardized running dataset are extracted by dilated convolutional layers in a temporal convolutional network. The weights of the temporal features at each historical moment are determined by an attention mechanism, and the heat load prediction intervals and electricity load prediction intervals corresponding to multiple future prediction periods are output. Based on the furnace flame center height, flue gas temperature distribution characteristics, and historical load change rate, the fuzzy C-means clustering algorithm is used to calculate the membership degree of the current operating condition to the deep adjustment steady state, the load increase transient state, the load decrease transient state, and the rated steady state, respectively. The operating condition with the highest membership degree is determined as the boiler dynamic operating condition category.

4. The thermoelectric coupling control method for a coal-fired boiler according to claim 1, characterized in that, The risk of slagging on the boiler heating surface and the accumulation of thermal stress in the steam turbine are calculated based on the standardized operation dataset to obtain equipment health status parameters. The specific method for determining the corresponding safety constraints based on the equipment health status parameters is as follows: Extract the furnace outlet flue gas temperature, fly ash carbon content, and coal softening temperature from the standardized operating dataset. Input the furnace outlet flue gas temperature, fly ash carbon content, and coal softening temperature into the random forest regression model to obtain the slagging probability of the screen-type superheater and the high-temperature superheater. The turbine regulating stage temperature and rotor stress concentration factor are extracted from the standardized operation dataset. The thermal stress change process of the turbine rotor is determined based on the turbine regulating stage temperature and rotor stress concentration factor. The equivalent cycle number corresponding to the load change is counted using the rainflow counting method to obtain the turbine thermal stress accumulation. The equipment health status parameters are determined based on the slagging probability and the cumulative thermal stress of the steam turbine, and the corresponding safety constraints are determined based on the equipment health status parameters.

5. The thermoelectric coupling control method for a coal-fired boiler according to claim 1, characterized in that, The heat load forecast sequence, electricity load forecast sequence, boiler dynamic operating condition category, and equipment health status parameters are input into a two-layer optimization model consisting of a deep deterministic strategy gradient algorithm and a sequential quadratic programming algorithm. The deep deterministic strategy gradient algorithm determines the coarse adjustment reference value and penalty coefficient for the heat-to-electric power ratio. Using the coarse adjustment reference value for the heat-to-electric power ratio as the initial solution, the boiler-turbine coupling nonlinear mechanism model is used as the equality constraint, and the safety constraint condition is used as the inequality constraint. The specific method for obtaining the heat-to-electric power allocation result and the control variable sequence through the sequential quadratic programming algorithm is as follows: The heat load prediction sequence, the electrical load prediction sequence, the boiler dynamic operating condition category, and the equipment health status parameters are used as the state inputs of the deep deterministic policy gradient algorithm. The actor network of the deep deterministic policy gradient algorithm outputs the coarse adjustment reference value of the thermoelectric power ratio, the boiler combustion intensity correction coefficient, and the penalty coefficient. The coarse adjustment reference value of the thermoelectric power ratio is used as the initial solution of the sequential quadratic programming algorithm. Based on the boiler-turbine coupled nonlinear mechanism model, the equation constraints between coal feed rate, water flow rate and turbine regulating valve opening and main steam pressure, main steam temperature and power generation are established. Inequality constraints are established based on the rate of change of the metal temperature of the heated surface, the fluctuation range of the main steam pressure, the fluctuation range of the steam drum water level, the flue gas temperature at the furnace outlet, and the concentration of pollutant emissions. Using the cumulative coal consumption within a preset optimization period as the objective function, the sequential quadratic programming algorithm is used for iterative solution to obtain the thermal power allocation results, as well as the coal feed sequence, air distribution sequence, water supply flow sequence, and heating extraction steam valve opening sequence.

6. The thermoelectric coupling control method for a coal-fired boiler according to claim 1, characterized in that, The specific method for converting the control variable sequence into drive commands for the actuator, and then using these drive commands and operational feedback data from the coal-fired boiler to perform closed-loop regulation of the coal feed rate, air distribution rate, water feed rate, and heating steam extraction valve opening is as follows: The coal feed rate sequence is used as a feedforward signal input to the coal feed rate feedforward-feedback controller. The feedback correction amount is determined based on the deviation between the measured value of the main steam pressure and the set value of the main steam pressure. The feedforward signal and the feedback correction amount are superimposed to generate the coal feeder drive command. The theoretical air volume is determined based on the real-time coal feed rate, and the theoretical air volume is corrected based on the oxygen content of the flue gas at the furnace outlet. The primary air volume and secondary air volume are determined by the air-coal ratio cross-limiting controller. Based on the unit load command and the turbine regulating valve opening, the main steam pressure setpoint is determined by a rolling optimization method, and the feedwater flow rate is adjusted according to the main steam pressure setpoint. Adjust the opening degree of the heating extraction steam valves according to the opening sequence of the heating extraction steam valves and the supply and return water parameters of the heating network.

7. The thermoelectric coupling control method for a coal-fired boiler according to claim 1, characterized in that, Based on the updated boiler heating surface slagging risk and turbine thermal stress accumulation during the closed-loop regulation process, the safety constraints are adaptively corrected. The specific method for updating the thermoelectric power allocation results and control variable sequence based on the corrected safety constraints is as follows: The updated boiler heating surface slagging risk is compared with the preset slagging risk threshold. When the boiler heating surface slagging risk exceeds the preset slagging risk threshold, the safety constraints corresponding to the furnace outlet flue gas temperature and the furnace outlet flue gas temperature change rate in the dual-layer optimization model are tightened. The updated turbine thermal stress accumulation is compared with the preset thermal stress threshold. When the turbine thermal stress accumulation exceeds the preset thermal stress threshold, the safety constraints corresponding to the unit load change rate and the heat transfer surface metal temperature change rate in the dual-layer optimization model are tightened. The revised safety constraints are re-input into the sequential quadratic programming algorithm to continuously update the thermoelectric power allocation results and the sequence of control variables.

8. A thermoelectric coupling control system for a coal-fired boiler, characterized in that, include: The multi-source data fusion and preprocessing module is used to collect time-series operation data of coal-fired boilers, steam turbine generators, heat exchangers of heating networks and thermoelectric coupling devices, and to perform time alignment, outlier correction, missing value completion and feature extraction on the time-series operation data to obtain a standardized operation dataset. The load forecasting and operating condition identification module is used to input standardized operating datasets, external meteorological data and power grid dispatch information into a time convolutional network based on an attention mechanism to obtain multi-timescale heat load forecasting sequences and electrical load forecasting sequences. Based on the furnace flame center height, flue gas temperature distribution characteristics and historical load change rate, the fuzzy C-means clustering algorithm is used to determine the dynamic operating condition category of the boiler. The health status assessment and adaptive correction module is used to calculate the risk of slagging on the boiler heating surface and the accumulation of thermal stress in the steam turbine based on the standardized operation dataset, obtain equipment health status parameters, and determine the corresponding safety constraints based on the equipment health status parameters. The hybrid intelligent optimization decision module is used to input the heat load prediction sequence, the electricity load prediction sequence, the boiler dynamic operating condition category, and the equipment health status parameters into a two-layer optimization model composed of a deep deterministic strategy gradient algorithm and a sequential quadratic programming algorithm. The deep deterministic strategy gradient algorithm determines the coarse adjustment reference value and penalty coefficient of the heat-electric power ratio. The coarse adjustment reference value of the heat-electric power ratio is used as the initial solution. The boiler-turbine coupling nonlinear mechanism model is used as the equality constraint, and the safety constraint condition is used as the inequality constraint. The sequential quadratic programming algorithm obtains the heat-electric power allocation result and the control variable sequence. The dynamic coordination control execution module is used to convert the sequence of control variables into drive commands for the actuators. Based on the drive commands and the operation feedback data of the coal-fired boiler, it performs closed-loop regulation of the coal feed rate, air distribution rate, water feed rate, and the opening of the heating extraction steam valve. The adaptive correction module is used to adaptively correct the safety constraints based on the boiler heating surface slagging risk and turbine thermal stress accumulation updated during the closed-loop regulation process, and update the thermoelectric power allocation results and control variable sequence based on the corrected safety constraints.

9. A thermoelectric coupling control system for a coal-fired boiler according to claim 8, characterized in that, The dynamic coordination control execution module includes a coal feed rate feedforward-feedback controller, an air-coal ratio cross-limit controller, and a sliding pressure setpoint optimizer. The coal feed rate feedforward-feedback controller is used to take the coal feed rate sequence as a feedforward signal and determine the feedback correction amount based on the real-time deviation of the main steam pressure. The air-coal ratio cross-limit controller is used to determine the primary air volume and secondary air volume based on the coal feed rate and the excess air coefficient at the furnace outlet, and to limit the primary air volume and secondary air volume to a minimum. The sliding pressure setpoint optimizer is used to determine the main steam pressure setpoint in a rolling manner based on the unit load command and the turbine regulating valve opening, using a model predictive control method.

10. A thermoelectric coupling control system for a coal-fired boiler according to claim 8, characterized in that, It also includes a human-computer interaction and edge computing gateway module, which is connected to a multi-source data fusion and preprocessing module, a load forecasting and operating condition identification module, a health status assessment and adaptive correction module, a hybrid intelligent optimization decision-making module, and a dynamic coordination control execution module, respectively. The human-computer interaction and edge computing gateway module is equipped with an interpretable artificial intelligence component, which is used to calculate the contribution of each input feature to the output control variable in the hybrid intelligent optimization decision module, and generate corresponding decision basis data. The human-computer interaction and edge computing gateway module supports OPC UA and MQTT protocols, which are used to upload the operating parameters, equipment health status parameters and system self-test status of coal-fired boilers to the cloud production management system and receive power grid dispatch information.

Images