Intelligent optimization method and system for power plant boiler combustion control

Abstract

The application discloses an intelligent optimization method and system for power plant boiler combustion control, and relates to the technical field of combustion control. The method comprises the following steps: constructing a potential function model of a power plant boiler combustion process; defining a potential energy field on a combustion state space by analyzing current operation state data of the power plant boiler, and outputting a control form occurrence field; calculating a gradient vector of the potential energy value relative to a current control variable; determining an adjustment direction and an adjustment amount of the control variable according to the gradient vector, and generating a control instruction; and issuing the control instruction to an executing mechanism of the power plant boiler, and completing the optimization control of the combustion process. The technical problem that the boiler combustion control in the prior art relies on experience rules, is difficult to dynamically optimize and adjust under multiple working conditions, and results in low combustion efficiency, large emission fluctuation and insufficient control stability is solved, and the technical effects of improving the combustion stability and combustion efficiency of the boiler and reducing the emission fluctuation of pollutants are achieved.

Description

Technical Field

[0001] This invention relates to the field of combustion control technology, and more specifically to an intelligent optimization method and system for combustion control of power plant boilers. Background Technology

[0002] As a core piece of equipment in thermal power generation systems, the stability and control precision of the combustion process in power plant boilers directly affect the unit's power generation efficiency, operational safety, and pollutant emission levels. The boiler combustion process is characterized by strong nonlinearity, multivariate coupling, and frequent changes in operating conditions. Even minor changes in fuel quality, load demand, air distribution ratio, and furnace conditions can cause significant fluctuations in combustion status. Therefore, implementing refined and dynamic control of the combustion process is of great importance. Existing power plant boiler combustion control methods are mostly based on empirical rules, manually tuned parameters, or fixed control curves, typically relying on threshold judgments and local feedback control for single or a small number of operating parameters. In actual operation, when the boiler is under multiple operating conditions such as start-up / shutdown, load changes, or changes in fuel characteristics, the above control methods are insufficient to comprehensively characterize the combustion state, and the coupling relationships between control variables are difficult to effectively handle. This leads to control response lag, unclear adjustment direction, and easily causes problems such as decreased combustion efficiency, increased fluctuations in furnace operating conditions, and unstable pollutant emissions. Summary of the Invention

[0003] This application provides an intelligent optimization method and system for combustion control of power plant boilers, which solves the technical problems in the prior art where boiler combustion control relies on empirical rules, making it difficult to dynamically optimize and adjust under multiple operating conditions, resulting in low combustion efficiency, large emission fluctuations, and insufficient control stability.

[0004] The first aspect of this application provides an intelligent optimization method for combustion control of power plant boilers, the method comprising:

[0005] A potential function model of the combustion process in a power plant boiler is constructed. This model analyzes the collected current operating state data of the power plant boiler, defines a potential energy field in the combustion state space, and outputs a control mode generation field. The level of the potential energy value indicates the degree to which the current state deviates from the ideal combustion state. Based on the potential energy value in the control mode generation field, the gradient vector of the potential energy value relative to the current control variable is calculated. The adjustment direction and amount of the control variable are determined according to the gradient vector, and control commands are generated. These control commands are then sent to the actuators of the power plant boiler to complete the optimized control of the combustion process.

[0006] A second aspect of this application provides an intelligent optimization system for combustion control of power plant boilers, the system comprising:

[0007] Data Analysis Component: Constructs a potential function model of the power plant boiler combustion process. This model analyzes the collected current operating state data of the power plant boiler, defines a potential energy field in the combustion state space, and outputs a control state generation field. The level of the potential energy value indicates the degree to which the current state deviates from the ideal combustion state. Gradient Calculation Component: Calculates the gradient vector of the potential energy value relative to the current control variable based on the potential energy value in the control state generation field. Command Generation Component: Determines the adjustment direction and amount of the control variable based on the gradient vector and generates control commands. Optimization Control Component: Sends the control commands to the actuators of the power plant boiler to complete the optimized control of the combustion process.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] First, a potential function model of the power plant boiler combustion process is constructed. This model analyzes and collects current operating state data of the power plant boiler, defines a potential energy field in the combustion state space, and outputs a control state generation field. The level of the potential energy value indicates the degree to which the current state deviates from the ideal combustion state. Next, based on the potential energy value in the control state generation field, the gradient vector of the potential energy value relative to the current control variable is calculated. Then, the adjustment direction and amount of the control variable are determined according to the gradient vector, generating control commands. Finally, the control commands are sent to the actuators of the power plant boiler to complete the optimized control of the combustion process. This solves the technical problems of existing boiler combustion control technologies that rely on empirical rules, making dynamic optimization under multiple operating conditions difficult, resulting in low combustion efficiency, large emission fluctuations, and insufficient control stability. It achieves the technical effect of improving boiler combustion stability and efficiency, and reducing pollutant emission fluctuations. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A schematic diagram of the intelligent optimization method for combustion control of power plant boilers provided in an embodiment of this application;

[0012] Figure 2 This is a schematic diagram of the intelligent optimization system structure for combustion control of power plant boilers provided in an embodiment of this application.

[0013] Figure labeling: Data analysis component 11, gradient calculation component 12, instruction generation component 13, optimization control component 14. Detailed Implementation

[0014] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0015] Example 1, as Figure 1 As shown, this application provides an intelligent optimization method for combustion control of power plant boilers, wherein the method includes:

[0016] A potential function model of the combustion process of a power plant boiler is constructed. The potential function model defines a potential energy field in the combustion state space by analyzing the collected current operating state data of the power plant boiler and outputs a control mode generation field. The level of potential energy indicates the degree to which the current state deviates from the ideal combustion state.

[0017] In this embodiment, sensors distributed at key locations in the feeding system, air supply system, and furnace are used to collect real-time operating status data of the biomass boiler in the power plant. This operating status data includes at least operating condition variables and control variables. Operating condition variables include fuel feed rate and its mixing ratio, bed temperature, furnace outlet oxygen content, main steam pressure, and pollutant emission concentration. Control variables include feeder speed, primary air volume, secondary air volume, secondary air ratio, and induced draft fan speed. The collected operating status data undergoes time synchronization, outlier removal, and normalization to form a current state vector. Based on this current state vector, each operating condition variable and control variable is mapped to a predefined combustion state space, which is a multi-dimensional continuous space composed of these variables. Within this combustion state space, a potential energy value is calculated for any state point using a potential function model. This potential energy value characterizes the degree of deviation of the state point from the ideal combustion state; a larger potential energy value indicates a greater deviation between the current combustion state and the ideal combustion state. The potential function model is constructed using historical operating data as training samples. By labeling the historical operating states and corresponding performance indicators with potential energy, the potential function model can learn the mapping relationship between combustion state and potential energy value. During real-time operation, the current state vector is input into the trained potential function model, and the potential energy value of the current state point in the combustion state space is output, thereby forming a continuously distributed potential energy field in the combustion state space.

[0018] Based on the potential energy field calculated by the potential function model in the combustion state space, a control mode generation field is constructed. The control mode generation field is used to characterize the influence relationship of each control variable on the potential energy change under different combustion states, and provides a basis for subsequent calculation of the adjustment direction and adjustment amount of control variables based on the potential energy gradient.

[0019] Furthermore, constructing a potential function model for the combustion process in a power plant boiler includes:

[0020] Historical operating data of the power plant boiler is collected, including operating condition variables, control variables, and corresponding performance indicators; potential energy is labeled for each state in the historical operating data according to the performance indicators; the potential function model is trained using the operating condition variables and control variables in the historical operating data as inputs and the labeled potential energy values ​​as outputs.

[0021] Furthermore, the control variables include feeder speed, primary air volume, secondary air volume, secondary air ratio, and induced draft fan speed; the operating condition variables include the feed rate and mixing ratio of various fuels, bed temperature, furnace outlet oxygen content, main steam pressure, and pollutant emission concentration.

[0022] Preferably, historical operating data of the boiler under different load levels, fuel ratios, and operating stages are collected from the power plant's historical operating database. The historical operating data is organized in time slices, with each time slice corresponding to a historical state. Each historical state includes at least operating condition variables, control variables, and performance indicators measured or statistically obtained under that state. The operating condition variables include fuel feed rate and its mixing ratio, bed temperature, furnace outlet oxygen content, main steam pressure, and pollutant emission concentration. The control variables include feeder speed, primary air volume, secondary air volume, secondary air ratio, and induced draft fan speed. The performance indicators include at least one of boiler thermal efficiency, combustion stability indicators, and pollutant emission indicators. Potential energy is labeled for each historical state in the historical operating data based on performance indicators. Specifically, each performance indicator is dimensionless and weighted and fused according to preset weights to obtain a comprehensive performance evaluation value corresponding to that historical state. The comprehensive performance evaluation value is compared with a pre-set ideal combustion performance benchmark, and the corresponding potential energy label value is calculated according to the degree of performance deviation. The closer the performance is to the ideal combustion performance benchmark, the lower the corresponding potential energy label value; the greater the performance deviation, the higher the corresponding potential energy label value, thus generating a unique potential energy label value for each historical state. The state vector composed of operating condition variables and control variables in the historical operating data is used as the model input, and the corresponding potential energy label value is used as the supervision output to train the potential function model. During the training process, the model parameters are gradually adjusted by minimizing the error between the model output potential energy value and the potential energy label value, so that the potential function model learns the mapping relationship between combustion state and potential energy value. After training, the potential function model can perform potential energy inference output for any given boiler operating state, mapping the state to the corresponding potential energy value, which is used to characterize the degree of deviation of the state from the ideal combustion state.

[0023] Furthermore, based on the performance metrics, potential energy is labeled for each state in the historical operating data, including:

[0024] A multidimensional performance space is defined based on the performance indicators, and an ideal attractor point is defined in the multidimensional performance space. The ideal attractor point is a set of preset optimal values ​​of each performance indicator. For each historical state point in the historical operation data, the Mahalanobis distance between the coordinate value in the multidimensional performance space and the ideal attractor point is calculated to obtain the potential energy label value corresponding to each historical state point.

[0025] Preferably, a multi-dimensional performance space is constructed based on performance indicators. This multi-dimensional performance space includes at least three dimensions: efficiency indicators, pollutant emission indicators, and safe and stable operation indicators. The efficiency indicators characterize the boiler's energy utilization level under corresponding historical conditions; the pollutant emission indicators characterize the emission intensity or emission deviation under corresponding conditions; and the safe and stable operation indicators characterize combustion fluctuations, over-limit risks, or operational margins. Within the multi-dimensional performance space, an ideal attractor point is predefined based on boiler design parameters, operating procedures, or historical best-performing statistical results. This ideal attractor point is a vector set composed of the optimal values ​​of each performance indicator. Specifically, the efficiency indicator takes the maximum allowable or target efficiency value, the pollutant emission indicator takes the minimum allowable or target emission value, and the safe and stable operation indicator takes the center value of the optimal stable interval, used to characterize the ideal combustion operation state. For each historical state point in the historical operating data, its corresponding efficiency index, pollutant emission index, and safe and stable operation index are mapped to coordinate vectors in the multidimensional performance space. Based on the statistical analysis of the historical operating data, the covariance matrix between each performance index is obtained. The difference between the coordinate vector and the ideal attractor point is normalized and corrected for correlation. The Mahalanobis distance between the historical state point and the ideal attractor point is calculated, and this Mahalanobis distance is used as the potential energy label value corresponding to the historical state point. The smaller the Mahalanobis distance value, the closer the historical state is to the ideal attractor point in the multidimensional performance space, and the lower the corresponding potential energy value, indicating that the combustion state is closer to the ideal operating state. Conversely, the larger the Mahalanobis distance value, the greater the deviation of the historical state from the ideal operating state, and the higher the corresponding potential energy value. Through this method, a unified potential energy labeling is achieved for each state in the historical operating data under the comprehensive meaning of efficiency, pollutant emission, and safe and stable operation, providing supervised samples for subsequent potential function model training.

[0026] Furthermore, a potential function model is constructed using a deep kernel learning framework, and physical conservation laws are incorporated as soft constraints into the loss function to train the potential function model.

[0027] Preferably, a deep kernel learning model structure is constructed, and the state vector composed of operating condition variables and control variables is used as the model input. The input state is feature extracted through a multi-layer nonlinear feature mapping network to obtain the latent space feature representation. A kernel function is introduced into the latent space to model the feature similarity, so as to characterize the nonlinear relationship between different combustion states, thereby outputting the corresponding potential energy value, so that the potential function model can calculate the potential energy of any boiler operating state in the combustion state space.

[0028] During model training, a joint loss function, comprising a data fitting term and a physical constraint term, is constructed to optimize the potential function model. The data fitting term measures the error between the potential energy value output by the potential function model and the potential energy labeling value obtained from historical operational data. The physical constraint term is constructed based on the physical conservation laws in the combustion process of power plant boilers. These physical conservation laws include at least mass conservation, energy conservation, and air-fuel ratio constraints. By penalizing the gradient change of the model's output potential energy in the state space, unreasonable potential energy abrupt changes in directions that violate physical laws are restricted. The physical constraint term is incorporated into the joint loss function as a soft constraint; that is, without forcing the model output to satisfy exact physical equations, a weighted penalty is applied to the degree of deviation from physical conservation relationships, guiding the potential function model to maintain consistency with the combustion physics mechanism while satisfying the historical data fitting accuracy. Through iterative optimization training of the joint loss function, the potential function model forms a potential energy field in the combustion state space that conforms to historical operational patterns and satisfies physical interpretability.

[0029] Furthermore, training the potential function model also includes:

[0030] Within the state space comprised of all operating condition variables and control variables, a systematic search is performed on the trained potential function model to identify all local minima. These local minima are then matched with high-density state regions in the historical operating dataset to select a set of potential well points. Spatial and performance clustering is performed on this set of potential well points, merging multiple neighboring potential wells representing the same stable combustion mode into a representative potential well, and calculating the representative center and depth. Using the center and depth of the representative potential well as constraint targets, the potential function model is then fine-tuned and regularized.

[0031] Within the combustion state space comprised of all operating and control variables, a systematic search is performed on the trained potential function model to identify local minima of the potential function in the state space. This systematic search involves sampling from multiple starting points within the state space and iteratively calculating along the descent direction of the potential energy gradient. When the potential energy gradient norm is less than a preset convergence threshold and the potential energy value changes less than a preset stability threshold over several consecutive iterations, the current state point is determined to be a local minimum. Each identified local minimum is matched with the state distribution in the historical operating data. By statistically analyzing the density distribution of historical operating data in the state space, local minimums falling into high-density state regions and whose corresponding operating performance meets the stable combustion criterion are selected as the potential well point set. The high-density state region characterizes the stable combustion state that frequently occurs and can be sustained during long-term boiler operation, thereby avoiding misidentification of accidental or unsustainable minimums as effective potential wells. Subsequently, the set of potential well points is subjected to joint spatial and performance clustering. The clustering is based on the distance between potential well points in the state space and the similarity of their corresponding efficiency, pollutant emissions, and safety stability performance indicators. Multiple adjacent potential well points representing the same stable combustion mode are merged into a representative potential well. The center position of the representative potential well is calculated based on the clustering results. The center position is the weighted average state vector of the potential well points in the corresponding cluster. At the same time, the depth of the representative potential well is calculated. The depth is determined by the difference between the potential energy value at the center of the representative potential well and the average potential energy value of its surrounding state points, which is used to characterize the attraction intensity of the stable combustion mode. Finally, using the center position and depth of the representative potential well as the constraint target, the potential function model is fine-tuned and regularized. By introducing a potential well preservation term into the model loss function, the potential energy pattern in the neighborhood of the representative potential well is restricted from changing drastically. At the same time, the generation of excessively deep or unreasonable new potential wells in non-historically stable regions is suppressed. Thus, the optimized potential function model can maintain its ability to characterize historically stable combustion modes while improving the overall smoothness, interpretability, and control stability of the potential energy field.

[0032] The preset convergence threshold for determining local minimum points is determined based on the statistical results of historical stable combustion conditions. Specifically, it involves: statistically analyzing the gradient change level corresponding to the potential energy change under historical stable operating conditions, using the average value of the potential energy gradient change amplitude in the historical stable samples as a benchmark, and preferably selecting 5% to 10% of this average value as the potential energy gradient convergence threshold to determine whether the search process has entered a region where the potential energy decreases at a slower pace; simultaneously, considering the change in potential energy value during continuous search iterations, when the potential energy change amplitude is less than 1 to 2 times the standard deviation of historical stable potential energy fluctuations, and this change characteristic is maintained in at least 10 consecutive iterations, it is used as the potential energy convergence criterion.

[0033] The preset stability threshold is used to characterize the stability of the operating condition variables and control variables during the search iteration process. It is set based on the natural fluctuation range of each operating condition variable and control variable within a unit time window under historical stable combustion conditions. Specifically, the fluctuation amplitude of each operating condition variable and control variable within a unit time window during the historical stable operation phase is statistically analyzed. The mean or median value of the fluctuation amplitude is used as a benchmark, and preferably 1 to 1.5 times the benchmark is taken as the preset stability threshold. When the change amplitude of each operating condition variable and control variable is less than the preset stability threshold during continuous search iteration, and the potential energy change remains within the steady-state fluctuation range, the corresponding local minimum point is determined to meet the stability requirements.

[0034] The high-density state region is determined based on the distribution characteristics of historical operating data in the combustion state space. By statistically analyzing the frequency of occurrence of historical state points, regions with a state point density 1.5 to 2.0 times higher than the median value of the historical state density distribution are selected as high-density state regions to exclude short-term abnormal operating conditions or occasional operating conditions. Only when a local minimum point falls into the high-density state region, and its corresponding efficiency indicators, pollutant emission indicators, and safe and stable operation indicators all meet the allowable range of the power plant operation regulations, is it included in the potential well point set.

[0035] When performing spatial and performance-based clustering on the potential well point set, the spatial clustering scale is determined based on the standard deviation range of each operating condition variable and control variable in historical stable operation data, preferably 0.5 to 1.0 times the historical standard deviation of each variable; the performance clustering is determined based on the comprehensive performance differences of efficiency, pollutant emissions and safe and stable operation indicators. When the difference in comprehensive performance evaluation values ​​between different potential well points does not exceed 10%, they are determined to represent the same stable combustion mode and are merged into the same representative potential well.

[0036] The center position of the representative potential well is obtained by weighted averaging of the state vectors corresponding to each potential well point in the same cluster. The depth of the representative potential well is determined by comparing the potential energy value at the center position with the average potential energy value of the surrounding historical stable operating state points. The selection range of the surrounding state points is determined based on the average distribution scale of the historical stable operating states in the state space, preferably 1 to 1.5 times the average expansion range of the historical stable state points relative to their cluster centers.

[0037] When fine-tuning and regularizing the potential function model using the center and depth of the representative potential wells as constraint targets, different regularization constraint strengths are set according to the frequency and duration of each representative potential well in historical operating data. Among them, the representative potential wells with higher historical frequency and longer continuous operating time have larger corresponding regularization weights, so as to ensure that the potential function model maintains the stability of the potential energy structure near the high-confidence stable combustion mode, while suppressing the formation of unreasonable deep potential wells in low-frequency or atypical operating condition regions.

[0038] Furthermore, after collecting the current operating status data of the power plant boiler, it also includes:

[0039] Establish an abnormal operating condition case library; perform parameter abnormal change analysis on the operating condition variables in the current operating status data; when the change value of any operating condition variable is detected to be greater than the preset change threshold, match similar historical abnormal cases in the abnormal operating condition case library; call the expert intervention plan that has been verified and effective in the similar historical abnormal cases; after the expert intervention plan is executed and the abnormal change is resolved, adjust the control variables.

[0040] After collecting the current operating status data of the power plant boiler, the process also includes abnormal operating condition identification and safety intervention. Specifically, this includes: First, establishing an abnormal operating condition case library. This library stores records of abnormal operating conditions that have been manually confirmed or system-verified during the boiler's historical operation, along with corresponding handling plans. Each abnormal operating condition case includes at least the abnormality type, triggering conditions, abnormal characteristic parameters, intervention operation steps, and post-intervention effect evaluation results. Second, performing parameter abnormal change analysis on each operating condition variable in the current operating status data. This involves obtaining the instantaneous change of each operating condition variable by performing differential calculations or sliding window statistics on the operating condition variables at continuous sampling times. When the change of any operating condition variable exceeds a corresponding preset change threshold, or its rate of change exceeds the threshold for a preset time limit, the current operating status is determined to have entered an abnormal operating condition, and an abnormal handling process is triggered. The preset time limit is set to 3-10 consecutive sampling cycles, preferably 5 sampling cycles, to distinguish between instantaneous noise disturbances and actual operating condition anomalies. When the rate of change continuously exceeds the change threshold within the preset time limit, it is determined to be an abnormal change event with engineering significance. Subsequently, the current abnormal operating condition is matched for similarity based on abnormal feature parameters in the abnormal operating condition case library. The similarity matching is based on a comprehensive similarity calculation of at least the magnitude, direction, and timing of the change of the operating condition variable and the state of the associated control variable, thereby filtering out one or more historical abnormal cases that are most similar to the current abnormal operating condition. Validated expert intervention plans are retrieved from the similar historical abnormal cases, and the corresponding expert intervention operations are executed according to the operation sequence and parameter range defined in the plan to suppress or eliminate the abnormal jumps in the operating condition variable. During the execution of the expert intervention plan, the changes in the operating condition variables are continuously monitored. When the abnormal jump is detected to be resolved and the operating condition variables return to the preset safe range, the anomaly handling is deemed complete. Based on this, the control variable optimization and adjustment process based on the potential function model is reactivated. The preset safe range is jointly determined by the design allowable range of each operating condition variable, the long-term stable operation statistical range, and the safety redundancy margin. Specifically, it is a control safe range that is 5% to 15% narrowed inward from the design allowable upper and lower limits to ensure sufficient operational margin during the anomaly resolution determination stage. If the abnormal jump is not resolved within the preset intervention time, or the operating condition variables continue to deviate from the safe range, an anomaly warning mechanism is triggered, an anomaly warning message is output, and automatic control variable adjustment is suspended to avoid performing unsafe optimization control operations under abnormal operating conditions.

[0041] Preferably, the historical change sequence of each operating condition variable under stable combustion conditions is statistically analyzed to calculate the mean μ and standard deviation σ of the change within a unit sampling period. The preset change threshold is set to μ+kσ, where k is 2~3, to cover more than 95% of the normal fluctuation range. When the operating condition variable is a continuous variable such as temperature, pressure or concentration, the change threshold is further limited to not exceeding the maximum instantaneous change rate allowed in the equipment operation procedure, so as to avoid misjudging parameter changes caused by normal adjustment.

[0042] Based on the potential energy value in the control mode generation field, calculate the gradient vector of the potential energy value relative to the current control variable.

[0043] Specifically, the operating condition variables under the current operating state are kept unchanged, and only the control variables are selected to form the control vector. The control vector and the corresponding operating condition variables are used as the state input and input into the trained potential function model to obtain the potential energy value corresponding to the current state point. Subsequently, in the neighborhood of the current control vector, a small perturbation is constructed for each control variable. While keeping other control variables unchanged, positive and negative perturbations are applied to a single control variable, and the perturbed state vector is input into the potential function model to calculate the corresponding perturbation potential energy value. Based on the change in potential energy before and after the perturbation and the perturbation amplitude, the first-order partial derivative of the potential energy with respect to the control variable is calculated using a numerical difference method. Repeat the above perturbation and difference calculation process for each control variable in the control vector to obtain the combination of the first-order partial derivatives of the potential energy value with respect to each control variable, thereby forming the gradient vector of the potential energy with respect to the current control variable; wherein, the sign of each component of the gradient vector is used to characterize the direction of the influence of the corresponding control variable on the change of potential energy, and the absolute value of the component is used to characterize the sensitivity of the control variable to the change of potential energy, providing a basis for subsequently determining the adjustment direction and adjustment amount of the control variable.

[0044] The adjustment direction and amount of the control variable are determined based on the gradient vector, and control commands are generated.

[0045] Furthermore, based on the gradient vector, the adjustment direction and amount of the control variable are determined, and control commands are generated, including:

[0046] The control adjustment direction for potential energy reduction is determined based on the gradient vector; based on the potential function model, a set of adjustment step sizes from small to large is preset, and a linear search is performed along the control adjustment direction to simulate and calculate the adjusted potential energy value. The step size with the largest potential energy reduction ratio is selected to generate a preliminary adjustment command; the preliminary adjustment command is input into a pre-built forward dynamic simulation module to predict the operating condition variables within a preset time window. If the prediction result meets the safety constraints, the control command is generated based on the preliminary adjustment command.

[0047] Preferably, the control adjustment direction for the decrease in potential energy is determined based on the gradient vector of the potential energy value relative to the control variable. The negative direction of the potential energy gradient is used as the adjustment direction of the control variable to ensure that the control adjustment can drive the combustion state towards the ideal attraction region along the direction of decreasing potential energy. Based on the potential function model, a set of discrete adjustment step sizes, ranging from small to large, is preset for the control adjustment direction. This set of discrete adjustment step sizes is a sequence of control increments pre-set based on the minimum adjustable resolution of the boiler control system actuator and historical stable adjustment experience. Preferably, the set of adjustment step sizes includes multiple candidate step sizes incremented by 0.5%, 1%, 2%, 3%, and 5% of the current control variable's rated value, used to balance convergence speed while ensuring adjustment precision. The potential energy reduction ratio is the ratio of the potential energy difference before and after adjustment to the potential energy value before adjustment, used to measure the relative contribution of different step sizes to the combustion state optimization effect. A linear search is performed sequentially along the control adjustment direction for each adjustment step size; that is, while keeping the operating condition variable constant, the corresponding step size is superimposed on the current control variable to construct multiple candidate control states. Each candidate control state is input into the potential function model to simulate and calculate the corresponding potential energy value, and the potential energy reduction ratio of each candidate step size relative to the current state is calculated based on the potential energy change. The step size with the largest potential energy reduction ratio is selected from the set of adjustment step sizes to generate the corresponding preliminary adjustment command.The preliminary adjustment command is input into a pre-built forward dynamic simulation module to predict the evolution of operating condition variables within a preset time window. This forward dynamic simulation module characterizes the impact of changes in control variables on the dynamic response of the boiler combustion process. The preset time window is determined based on the dynamic response characteristics of the boiler combustion process and the sampling period of the control system, preferably set to 10s~60s to cover the main response process of the combustion state to changes in control variables. Based on the simulation prediction results, the stability of the operating condition and safety constraints are judged. When the prediction results indicate that the changes in operating condition variables are gradual, without violent oscillations, and do not trigger preset safety hard constraints (including furnace temperature exceeding limits, furnace negative pressure exceeding limits, etc.), the preliminary adjustment command is determined to be safe and effective, and the final control command is generated accordingly. The stability determination is based on the magnitude and rate of change of each key operating condition variable within the time window. Specifically, when the fluctuation amplitudes of bed temperature, furnace outlet oxygen content, and main steam pressure do not exceed ±2% of their corresponding set values ​​and there is no continuous oscillation trend, the operating condition stability is determined to meet the requirements. The safety hard constraints are determined based on the boiler design safety parameters and operating procedures, including but not limited to: furnace temperature not exceeding the design allowable upper limit, furnace negative pressure maintained within the preset safety range, and the operating parameters of the induced draft fan, forced draft fan, and feeding system not exceeding the rated limits of the equipment. When the prediction result does not meet the safety constraints, the current preliminary adjustment command is abandoned, and the step size with the second largest potential energy decrease ratio is reselected. The above simulation and determination process is repeated until a control command that meets the safety constraints is obtained or an abnormal handling process is triggered.

[0048] Furthermore, if the prediction results do not meet the safety constraints, including:

[0049] Select the step size with the second largest potential energy decrease ratio, reduce the step size and repeat the simulation until the optimal adjustment amount that satisfies the safety constraint is found, and then generate the control command.

[0050] When the prediction result output by the forward dynamic simulation module does not meet the safety constraints, a safety rollback and reselection process for the control adjustment is executed. Specifically: First, it is determined that the prediction result indicates that the operating condition fluctuation exceeds the allowable range or triggers at least one safety hard constraint. The safety hard constraints include at least furnace temperature exceeding the limit, furnace negative pressure exceeding the limit, or combustion instability risk index exceeding the limit. Subsequently, among the multiple candidate adjustment step sizes generated, rollback selection is performed sequentially in descending order of potential energy decrease ratio. When the prediction result corresponding to the currently selected adjustment step size does not meet the safety constraints, the adjustment step size is abandoned, and the adjustment step size with the second largest potential energy decrease ratio is selected instead. Based on this adjustment step size, a preliminary adjustment command is reconstructed and input into the forward dynamic simulation module for re-prediction. The above rollback and reselection process is repeated until the prediction result simultaneously satisfies the condition that the dynamic process is stable and no safety hard constraints are triggered. At this point, the corresponding adjustment step size is determined as the optimal adjustment amount, and the control command is generated based on the optimal adjustment amount. If no adjustment amount that satisfies the safety constraints can be found after traversing all candidate adjustment step sizes, an exception handling or early warning process is triggered, and the automatic optimization adjustment of the control variables is suspended.

[0051] The control commands are sent to the actuators of the power plant boiler to achieve optimized control of the combustion process.

[0052] Based on the generated control instructions, the control instructions are formatted and verified. The control instructions at least include the target setpoint or adjustment increment for each control variable, execution priority, and effective timestamp. The control instructions are sent to the boiler distributed control system via an industrial communication network. The distributed control system parses the control instructions and distributes them to the corresponding actuators. The actuators, based on the received control instructions, physically adjust the corresponding control variables. The actuators include at least a feeder drive device, primary and secondary air regulating valves, and an induced draft fan speed control device, used to adjust fuel supply, air distribution ratio, and furnace negative pressure, respectively. During execution, the actual changes in the control variables are subject to rate and amplitude limits to prevent equipment shock or combustion instability caused by sudden changes in instructions.

[0053] After the control command is executed, the boiler's real-time operating status data is continuously collected, and the collected operating status data is compared and analyzed with the status before the control command was issued to determine the effect of the control command on the combustion status. If the combustion status is detected to evolve in the direction of decreasing potential energy without triggering any safety constraints, the execution of this control command is deemed effective, and the next control cycle is entered. If the combustion status is detected to be not improved as expected or to show an abnormal trend, the control correction or abnormal handling process is triggered, thereby forming a closed loop of combustion process optimization control based on the potential function model.

[0054] In summary, the embodiments of this application have at least the following technical effects:

[0055] First, a potential function model of the power plant boiler combustion process is constructed. This model analyzes and collects current operating state data of the power plant boiler, defines a potential energy field in the combustion state space, and outputs a control state generation field. The level of the potential energy value indicates the degree to which the current state deviates from the ideal combustion state. Next, based on the potential energy value in the control state generation field, the gradient vector of the potential energy value relative to the current control variable is calculated. Then, the adjustment direction and amount of the control variable are determined according to the gradient vector, generating control commands. Finally, the control commands are sent to the actuators of the power plant boiler to complete the optimized control of the combustion process. This solves the technical problems of existing boiler combustion control technologies that rely on empirical rules, making dynamic optimization under multiple operating conditions difficult, resulting in low combustion efficiency, large emission fluctuations, and insufficient control stability. It achieves the technical effect of improving boiler combustion stability and efficiency, and reducing pollutant emission fluctuations.

[0056] Example 2, based on the same inventive concept as the intelligent optimization method for power plant boiler combustion control in the foregoing examples, such as... Figure 2 As shown, this application provides an intelligent optimization system for combustion control of power plant boilers, wherein the system includes:

[0057] Data Analysis Component 11: Constructs a potential function model of the power plant boiler combustion process. This model analyzes the collected current operating state data of the power plant boiler, defines a potential energy field in the combustion state space, and outputs a control state generation field. The level of the potential energy value indicates the degree to which the current state deviates from the ideal combustion state. Gradient Calculation Component 12: Calculates the gradient vector of the potential energy value relative to the current control variable based on the potential energy value in the control state generation field. Command Generation Component 13: Determines the adjustment direction and amount of the control variable based on the gradient vector and generates control commands. Optimization Control Component 14: Sends the control commands to the actuators of the power plant boiler to complete the optimized control of the combustion process.

[0058] In this embodiment, the data analysis component 11, gradient calculation component 12, instruction generation component 13, and optimization control component 14 are sequentially connected via an internal data bus or function call interface to form a closed-loop optimization control process for power plant boiler combustion control. The data interaction relationship between the components is as follows: First, the data analysis component 11 is communicatively connected to the boiler site data acquisition system and historical operation database to receive the current operating status data and historical operating data of the power plant boiler. The data analysis component 11 preprocesses the received current operating status data and, based on the trained potential function model, maps the current operating status to the corresponding potential energy value, constructs the potential energy field and control mode generation field in the combustion state space, and outputs the potential energy value and control mode generation field as analysis results to the gradient calculation component 12. Secondly, the gradient calculation component 12 establishes a data connection with the data analysis component 11 to receive the potential energy value and corresponding control variable information in the control mode generation field. While keeping the operating conditions constant, the gradient calculation component 12 performs disturbance analysis on each control variable, calculates the first-order partial derivative of the potential energy value with respect to each control variable, generates a potential energy gradient vector, and sends the gradient vector to the instruction generation component 13. Subsequently, the instruction generation component 13 establishes a data interaction relationship with the gradient calculation component 12 and the data analysis component 11. Based on the received potential energy gradient vector, the instruction generation component 13 determines the adjustment direction of the control variable and calls the potential function model to simulate the potential energy changes corresponding to different adjustment step sizes, generating a preliminary adjustment instruction. Simultaneously, the instruction generation component 13 inputs the preliminary adjustment instruction to the forward dynamic simulation module to predict the operating conditions within a future preset time window, and determines whether the safety constraints are met based on the prediction results, ultimately generating a control instruction that meets the safety constraints. Finally, the optimization control component 14 is communicatively connected to the instruction generation component 13 and the boiler actuators, and is used to receive the control instructions and send them to the feeder, air distribution system and induced draft system and other actuators to drive the boiler combustion process to be adjusted according to the control instructions. During the execution of the control instructions, the optimization control component 14 continuously collects execution feedback and the latest operating status data, and sends the operating status data back to the data analysis component 11 for potential energy calculation and optimization analysis in the next control cycle, thereby forming a combustion process optimization control closed loop based on the potential function model.

[0059] Furthermore, when the data analysis component 11 detects an abnormal jump in the operating condition variable in the current operating status data, it generates a corresponding expert intervention plan call signal by matching it with the abnormal operating condition case library, and synchronizes the abnormal handling status to the instruction generation component 13; before the abnormality is resolved, the instruction generation component 13 suspends the generation of control instructions based on potential energy gradient, and resumes the normal optimization control process after the abnormality is handled.

[0060] Furthermore, the data analysis component 11 is used to perform the following methods:

[0061] Historical operating data of the power plant boiler is collected, including operating condition variables, control variables, and corresponding performance indicators; potential energy is labeled for each state in the historical operating data according to the performance indicators; the potential function model is trained using the operating condition variables and control variables in the historical operating data as inputs and the labeled potential energy values ​​as outputs.

[0062] Furthermore, the data analysis component 11 is used to perform the following methods:

[0063] A multidimensional performance space is defined based on the performance indicators, and an ideal attractor point is defined in the multidimensional performance space. The ideal attractor point is a set of preset optimal values ​​of each performance indicator. For each historical state point in the historical operation data, the Mahalanobis distance between the coordinate value in the multidimensional performance space and the ideal attractor point is calculated to obtain the potential energy label value corresponding to each historical state point.

[0064] Furthermore, the data analysis component 11 is used to perform the following methods:

[0065] A potential function model is constructed using a deep kernel learning framework, and the physical conservation law is added as a soft constraint to the loss function to train the potential function model.

[0066] Furthermore, the data analysis component 11 is used to perform the following methods:

[0067] Within the state space comprised of all operating condition variables and control variables, a systematic search is performed on the trained potential function model to identify all local minima. These local minima are then matched with high-density state regions in the historical operating dataset to select a set of potential well points. Spatial and performance clustering is performed on this set of potential well points, merging multiple neighboring potential wells representing the same stable combustion mode into a representative potential well, and calculating the representative center and depth. Using the center and depth of the representative potential well as constraint targets, the potential function model is then fine-tuned and regularized.

[0068] Furthermore, the data analysis component 11 is used to perform the following methods:

[0069] The control variables include feeder speed, primary air volume, secondary air volume, secondary air ratio, and induced draft fan speed; the operating condition variables include the feed rate and mixing ratio of various fuels, bed temperature, furnace outlet oxygen content, main steam pressure, and pollutant emission concentration.

[0070] Furthermore, the instruction generation component 13 is used to perform the following method:

[0071] The control adjustment direction for potential energy reduction is determined based on the gradient vector; based on the potential function model, a set of adjustment step sizes from small to large is preset, and a linear search is performed along the control adjustment direction to simulate and calculate the adjusted potential energy value. The step size with the largest potential energy reduction ratio is selected to generate a preliminary adjustment command; the preliminary adjustment command is input into a pre-built forward dynamic simulation module to predict the operating condition variables within a preset time window. If the prediction result meets the safety constraints, the control command is generated based on the preliminary adjustment command.

[0072] Furthermore, the instruction generation component 13 is used to perform the following method:

[0073] Select the step size with the second largest potential energy decrease ratio, reduce the step size and repeat the simulation until the optimal adjustment amount that satisfies the safety constraint is found, and then generate the control command.

[0074] Furthermore, the data analysis component 11 is used to perform the following methods:

[0075] Establish an abnormal operating condition case library; perform parameter abnormal change analysis on the operating condition variables in the current operating status data; when the change value of any operating condition variable is detected to be greater than the preset change threshold, match similar historical abnormal cases in the abnormal operating condition case library; call the expert intervention plan that has been verified and effective in the similar historical abnormal cases; after the expert intervention plan is executed and the abnormal change is resolved, adjust the control variables.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent optimization method for combustion control of power plant boilers, characterized in that, The method includes: A potential function model of the combustion process of a power plant boiler is constructed. The potential function model defines a potential energy field in the combustion state space by analyzing the current operating state data of the power plant boiler and outputs a control mode generation field. The level of potential energy value indicates the degree to which the current state deviates from the ideal combustion state. Based on the potential energy value in the control mode generation field, calculate the gradient vector of the potential energy value relative to the current control variable; The adjustment direction and amount of the control variable are determined based on the gradient vector, and control commands are generated. The control commands are sent to the actuators of the power plant boiler to achieve optimized control of the combustion process.

2. The intelligent optimization method for combustion control of power plant boilers as described in claim 1, characterized in that, Constructing a potential function model for the combustion process in a power plant boiler, including: Collect historical operating data of the power plant boiler, including operating condition variables, control variables, and corresponding performance indicators; Based on the performance metrics, potential energy is labeled for each state in the historical operating data; The potential function model is trained by taking the operating condition variables and control variables from the historical operating data as inputs and the labeled potential energy value as output.

3. The intelligent optimization method for combustion control of power plant boilers as described in claim 2, characterized in that, Based on the performance metrics, potential energy is labeled for each state in the historical operating data, including: A multidimensional performance space is defined based on the performance indicators, and an ideal attractor point is defined in the multidimensional performance space. The ideal attractor point is a set of preset optimal values ​​of each performance indicator. For each historical state point in the historical operating data, the Mahalanobis distance between the coordinate value in the multidimensional performance space and the ideal attractor point is calculated to obtain the potential energy label value corresponding to each historical state point.

4. The intelligent optimization method for combustion control of power plant boilers as described in claim 2, characterized in that, A potential function model is constructed using a deep kernel learning framework, and the physical conservation law is added as a soft constraint to the loss function to train the potential function model.

5. The intelligent optimization method for combustion control of power plant boilers as described in claim 3, characterized in that, Training the potential function model further includes: Within the state space comprised of all operating condition variables and control variables, a systematic search is performed on the trained potential function model to identify all local minima. The local minimum points are matched with high-density state regions in the historical running dataset to filter out a set of potential well points; Spatial and performance clustering is performed on the set of potential well points to merge multiple neighboring potential wells that represent the same stable combustion mode into a representative potential well, and the representative center and depth are calculated. Using the center and depth of the potential well as the constraint targets, the potential function model is fine-tuned and regularized in a directional manner.

6. The intelligent optimization method for combustion control of power plant boilers as described in claim 2, characterized in that, The controlled variables include feeder speed, primary air volume, secondary air volume, secondary air ratio, and induced draft fan speed; The operating variables include the feed rate and mixing ratio of various fuels, bed temperature, furnace outlet oxygen content, main steam pressure, and pollutant emission concentration.

7. The intelligent optimization method for combustion control of power plant boilers as described in claim 1, characterized in that, Based on the gradient vector, the adjustment direction and amount of the control variable are determined, and control commands are generated, including: The control adjustment direction for the potential energy decrease is determined based on the gradient vector; Based on the potential function model, a set of adjustment step sizes from small to large is preset, and a linear search is performed along the control adjustment direction to simulate and calculate the adjusted potential energy value. The step size with the largest potential energy decrease ratio is selected to generate a preliminary adjustment command. The preliminary adjustment command is input into the pre-built forward dynamic simulation module to predict the operating condition variables within a preset time window. If the prediction result meets the safety constraints, the control command is generated based on the preliminary adjustment command.

8. The intelligent optimization method for combustion control of power plant boilers as described in claim 7, characterized in that, If the prediction result does not meet the safety constraints, including: Select the step size with the second largest potential energy decrease ratio, reduce the step size and repeat the simulation until the optimal adjustment amount that satisfies the safety constraint is found, and then generate the control command.

9. The intelligent optimization method for combustion control of power plant boilers as described in claim 1, characterized in that, After collecting the current operating status data of the power plant boiler, the following is also included: Establish an abnormal operating condition case library; Perform parameter anomaly jump analysis on the operating condition variables in the current operating status data. When the jump value of any operating condition variable is detected to be greater than the preset change threshold, match similar historical anomaly cases in the abnormal operating condition case library. The expert intervention plan that has been verified and effective in similar historical anomaly cases is invoked. After the expert intervention plan is executed and the jump anomaly is resolved, the control variables are adjusted.

10. An intelligent optimization system for combustion control of power plant boilers, characterized in that, The system is used to implement the intelligent optimization method for combustion control of power plant boilers according to any one of claims 1-9, the system comprising: Data analysis component: Constructs a potential function model of the combustion process of a power plant boiler. The potential function model defines a potential energy field in the combustion state space by analyzing the collected current operating state data of the power plant boiler and outputs a control mode generation field. The level of potential energy indicates the degree to which the current state deviates from the ideal combustion state. Gradient calculation component: Based on the potential energy value in the control mode generation field, calculate the gradient vector of the potential energy value relative to the current control variable; Command generation component: Determines the adjustment direction and amount of the control variable based on the gradient vector, and generates control commands; Optimized control components: The control commands are sent to the actuators of the power plant boiler to complete the optimized control of the combustion process.

Images