Thermal power unit intelligent combustion adjustment method and system based on multi-objective optimization

Abstract

The application provides a thermal power unit intelligent combustion adjustment method and system based on multi-objective optimization, relates to the technical field of multi-objective optimization, and comprises the following steps: acquiring working condition parameters, predicting a load demand trajectory and an emission limit value change curve, constructing a time sequence associated multi-stage optimization target, and coupling combustion efficiency, pollutant emission and equipment load into a multi-objective function through time attenuation weight; a Pareto frontier solution set is obtained by solving, and a combustion adjustment parameter is selected and issued; feedback data is collected, and a dynamic weight adjustment mechanism is used to adaptively update weight configuration. The application realizes intelligent regulation and control of combustion parameters, improves unit operation efficiency, and reduces pollutant emission.

Description

Technical Field

[0001] This invention relates to multi-objective optimization technology, and more particularly to a method and system for intelligent combustion adjustment of thermal power units based on multi-objective optimization. Background Technology

[0002] The combustion process of thermal power units is a complex system involving the interaction of multiple factors, and its operating efficiency and emission levels are directly related to energy utilization and environmental protection.

[0003] The main defects and shortcomings of existing technologies are reflected in the following aspects: Existing combustion adjustment methods lack the foresight to anticipate future load changes and emission limit adjustments. They often rely solely on static optimization based on current operating conditions, which fails to address the complex and ever-changing grid dispatching demands and dynamic environmental constraints, resulting in delayed system response and poor adjustment effectiveness.

[0004] Existing optimization methods typically employ multi-objective functions with fixed weights, making it difficult to dynamically adjust the priority of each sub-objective based on actual operating conditions. When faced with conflicts between load demand and emission control, they cannot achieve optimal resource allocation, thus reducing the system's adaptability and robustness. Summary of the Invention

[0005] This invention provides a method and system for intelligent combustion adjustment of thermal power units based on multi-objective optimization, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides a method for intelligent combustion adjustment of thermal power units based on multi-objective optimization, comprising: Obtain the operating parameters of the combustion system of the thermal power unit; Based on the trend of the operating parameters, the load demand trajectory and emission limit change curve within the future time window are predicted. A time-series related multi-stage optimization objective is constructed based on the load demand trajectory and emission limit change curve. The multi-stage optimization objective cascades and couples the combustion efficiency, pollutant emissions and equipment load at each moment within the future time window into a multi-objective function through time decay weights. The Pareto front solution set is obtained by solving the multi-objective function based on the current weight configuration. Target combustion adjustment parameters are selected from the Pareto front solution set and sent to the actuators of the combustion system. The feedback data after collection and execution is used to adaptively update the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism calculates the weight increment based on the ratio of load response deviation to emission compliance margin in the feedback data, and sends the updated weight configuration back to the next round of solution.

[0007] The steps for predicting the load demand trajectory and emission limit change curve within a future time window based on the aforementioned operating condition parameter change trends include: Obtain the operating parameters of the combustion system of the thermal power unit; extract the load change rate and emission concentration change rate from the operating parameters as time series features, and perform time series extension on the time series features to generate the load evolution sequence and emission evolution sequence within the future time window; A load demand trajectory is constructed based on the load evolution sequence, and an emission limit change curve is constructed based on the emission evolution sequence and environmental emission standards.

[0008] For each time node, the load demand value is extracted as the equipment load sub-target based on the load demand trajectory, the emission limit is extracted as the pollutant emission sub-target based on the emission limit change curve, and the combustion efficiency sub-target is calculated based on the equipment load sub-target and the pollutant emission sub-target. The absolute value of the slope of the load demand trajectory at each time node is calculated as the load change severity, and the proximity of the emission limit change curve to the current emission value at each time node is calculated as the emission constraint tension. A time decay weight is constructed based on the load change severity and the emission constraint tension. The time decay weights are applied to the combustion efficiency sub-objective, pollutant emission sub-objective, and equipment load sub-objective at the corresponding time nodes, and then summed to form a multi-objective function.

[0009] The steps for constructing time decay weights include: A base time decay function is defined, the value of which decreases monotonically as the time interval between the time node and the current time increases; For each time point within the future time window, a conflict situation vector is constructed in two-dimensional space based on the intensity of load change and the tension of emission constraints at that time. The magnitude of the conflict situation vector is then calculated as the comprehensive criticality of that time point. The moment nodes whose comprehensive criticality exceeds a preset comprehensive threshold are identified as potential conflict moments. A buffer time window is set before each conflict moment. The weight values ​​of the basic time decay function at each moment node within the buffer time window are proportionally transferred to the potential conflict moment, so that the time decay weight of the potential conflict moment is increased and the time decay weight of each moment node in the preceding buffer window decreases synchronously. The time decay weight sequence after weight transfer is normalized to obtain the final time decay weight.

[0010] The steps of solving the multi-objective function based on the current weight configuration to obtain the Pareto front solution set, and selecting target combustion adjustment parameters from the Pareto front solution set and issuing them to the actuators of the combustion system include: The weight values ​​of each sub-objective in the current weight configuration are used as weight coefficients for multi-objective optimization. The multi-objective function is solved to obtain the Pareto front solution set. The Pareto front solution set contains multiple sets of non-dominated solutions. Each set of non-dominated solutions corresponds to a set of combustion adjustment parameters and a sequence of target achievement at each time in the future time window. Calculate the fluctuation amplitude of the target achievement sequence of each group of non-dominated solutions within a future time window, calculate the transition smoothness between adjacent moments in the target achievement sequence, and evaluate the temporal stability index of each group of non-dominated solutions based on the fluctuation amplitude and transition smoothness. The combustion adjustment parameter corresponding to the non-dominated solution with the best time-series stability index is selected as the target combustion adjustment parameter, and the target combustion adjustment parameter is sent to the actuator of the combustion system.

[0011] The steps of adaptively updating the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism and then feeding back the updated weight configuration for the next round of solving include: Collect feedback data after execution, extract actual load response value and actual emission value, calculate the load response deviation between the actual load response value and the load demand value, and calculate the emission compliance margin between the actual emission value and the emission limit. Within a short time window, the transient fluctuation characteristics of historical feedback data are statistically analyzed, and within a long time window, the steady-state deviation characteristics of historical feedback data are statistically analyzed. The transient fluctuation characteristics and steady-state deviation characteristics are then fused into a deviation trend correction amount and superimposed on the load response deviation and emission compliance margin of the current cycle. Calculate the ratio of the corrected load response deviation to the emission compliance margin, calculate the initial weight increment based on the ratio, analyze the trade-off between the equipment load sub-target and the pollutant emission sub-target in the Pareto front solution to obtain the target conflict degree, and scale the initial weight increment based on the target conflict degree to obtain the weight increment. The weight increment is added to the current weight configuration to obtain the updated weight configuration, which is then sent back for the next round of solving.

[0012] The steps of analyzing the trade-off between the equipment load sub-objective and the pollutant emission sub-objective in the Pareto front solution to obtain the target conflict degree, and scaling the initial weight increment based on the target conflict degree to obtain the weight increment include: The optimal solutions for the equipment load sub-objective and the pollutant emission sub-objective are selected from the Pareto front solution set as two sets of boundary solutions. The load change and emission change between the two sets of boundary solutions are calculated. The target sensitivity is calculated based on the load change and emission change. The Pareto front curve is obtained by curve fitting of all non-dominated solutions in the Pareto front solution set in the two-dimensional space formed by the equipment load sub-objective and the pollutant emission sub-objective, and the average curvature of the Pareto front curve is calculated. Calculate the variance of the Euclidean distance between adjacent non-dominated solutions in the Pareto front solution set; The target conflict degree is obtained by weighted fusion of the target sensitivity, the average curvature and the Euclidean distance variance; a scaling factor is constructed based on the target conflict degree, the scaling factor decreases as the target conflict degree increases, and the scaling factor is applied to the initial weight increment to obtain the weight increment.

[0013] A second aspect of the present invention provides an intelligent combustion adjustment system for thermal power units based on multi-objective optimization, comprising: The operating condition parameter acquisition module is used to acquire the operating condition parameters of the combustion system of the thermal power unit; The prediction and target construction module is used to predict the load demand trajectory and emission limit change curve within the future time window based on the change trend of the operating parameters, and construct a time-series related multi-stage optimization target. The multi-stage optimization target cascades and couples the combustion efficiency, pollutant emissions and equipment load at each moment within the future time window into a multi-objective function through time decay weight. The solution and parameter distribution module is used to solve the multi-objective function based on the current weight configuration to obtain the Pareto front solution set, select target combustion adjustment parameters from the Pareto front solution set, and distribute them to the actuators of the combustion system. The feedback and weight update module is used to collect feedback data after execution and adaptively update the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism calculates the weight increment based on the ratio of load response deviation to emission compliance margin in the feedback data and sends the updated weight configuration back for the next round of solving.

[0014] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0016] This invention constructs multi-stage optimization objectives with temporal correlations and uses time decay weights to cascade and couple combustion efficiency, pollutant emissions, and equipment load within future windows. This achieves proactive control of thermal power unit operation, effectively improving the pre-adaptive capability to load changes and reducing combustion efficiency losses caused by operating condition fluctuations. It innovatively introduces a dynamic weight adjustment mechanism based on feedback data, calculating weight increments through the ratio of load response deviation to emission compliance margin. This enables adaptive updates of the weights of each sub-objective in multi-objective optimization, significantly enhancing the adaptive adjustment capability under different operating conditions and allowing the combustion adjustment process to intelligently adjust optimization strategies according to actual operating conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the intelligent combustion adjustment method for thermal power units based on multi-objective optimization, according to an embodiment of the present invention. Figure 2 Build a flowchart for multi-objective functions. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0020] Figure 1 This is a flowchart illustrating the intelligent combustion adjustment method for thermal power units based on multi-objective optimization, as described in an embodiment of the present invention. Figure 1 As shown, the method includes: Obtain the operating parameters of the combustion system of the thermal power unit; Based on the trend of the operating parameters, the load demand trajectory and emission limit change curve within the future time window are predicted. A time-series related multi-stage optimization objective is constructed based on the load demand trajectory and emission limit change curve. The multi-stage optimization objective cascades and couples the combustion efficiency, pollutant emissions and equipment load at each moment within the future time window into a multi-objective function through time decay weights. The Pareto front solution set is obtained by solving the multi-objective function based on the current weight configuration. Target combustion adjustment parameters are selected from the Pareto front solution set and sent to the actuators of the combustion system. The feedback data after collection and execution is used to adaptively update the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism calculates the weight increment based on the ratio of load response deviation to emission compliance margin in the feedback data, and sends the updated weight configuration back to the next round of solution.

[0021] In one optional implementation, the step of predicting the load demand trajectory and emission limit change curve within a future time window based on the trend of the operating parameters includes: Obtain the operating parameters of the combustion system of the thermal power unit; extract the load change rate and emission concentration change rate from the operating parameters as time series features, and perform time series extension on the time series features to generate the load evolution sequence and emission evolution sequence within the future time window; A load demand trajectory is constructed based on the load evolution sequence, and an emission limit change curve is constructed based on the emission evolution sequence and environmental emission standards.

[0022] For example, operating parameters are acquired through a sensor network distributed across key components of the thermal power unit, including but not limited to parameters such as boiler load, steam pressure, steam temperature, feedwater flow rate, air volume, fuel quantity, oxygen content, nitrogen oxide emission concentration, sulfur dioxide emission concentration, and particulate matter concentration. Key parameters such as load and emission concentration are collected once per second, while other parameters can be collected once per minute. The collected raw data undergoes preprocessing such as filtering and noise reduction before being stored in a time-series database, forming a reliable historical operating parameter dataset.

[0023] The load change rate is calculated as the difference between the current load value and the previous load value, divided by the sampling time interval. The emission concentration change rate is calculated as the difference between the current emission concentration and the previous emission concentration, divided by the sampling time interval. Taking nitrogen oxides (NOx) as an example, if the NOx concentration at the current time t is NOx(t), the concentration at the previous time t-1 is NOx(t-1), and the sampling interval is Δt, then the NOx concentration change rate is [NOx(t) - NOx(t-1)] / Δt. Extracted features are standardized to eliminate the influence of different dimensions, facilitating subsequent analysis.

[0024] Time series extension is achieved using a linear extrapolation method based on historical trends. Specifically, statistical analysis is performed on the load change rate series within historical time windows. The weighted average of the load change rate within recent time windows (e.g., the last 30 minutes) is calculated as the short-term trend, and the arithmetic mean of the load change rate within longer time windows (e.g., the last 2 hours) is calculated as the long-term trend. The short-term and long-term trends are then combined with a 7:3 weighting to obtain a comprehensive trend coefficient. This comprehensive trend coefficient is used as the initial predicted value for the load change rate within future time windows, and the predicted value is corrected based on the fluctuation characteristics of similar operating conditions in historical data, thereby generating the load change rate series for each predicted moment within the future time window. The emission concentration change rate series is extended and predicted using the same method.

[0025] Based on the predicted rate of change sequence, the load and emission concentration values ​​at future time points are recovered through integral calculation. Specifically, if the starting point of the prediction time window is t0 and the current load is Load(t0), the predicted load at the i-th time point (t0+i·Δt) is Load(t0)+∑(load change rate j·Δt), where j ranges from 1 to i. Similarly, the emission concentration values ​​are recovered using the same method, thus forming a complete future load evolution sequence and emission evolution sequence.

[0026] The load evolution sequence is constructed into a load demand trajectory after demand mapping. The demand mapping process considers grid dispatch instructions, unit ramp-up rate constraints, and load reserve margins, converting the predicted load evolution values ​​into target load demand values ​​that the units should meet. The emission evolution sequence is constructed into an emission limit change curve by combining environmental emission standards. By comparing the predicted emission concentration values ​​with the statutory emission limits at the corresponding time, potential exceedance risk periods are identified and dynamic limit constraint curves are generated.

[0027] The load demand trajectory is represented as a two-dimensional time-load curve, with time on the horizontal axis and load demand value on the vertical axis. To enhance visualization, different colors can be added to the trajectory to distinguish different load change rate ranges: green indicates a stable range (change rate less than 0.5% per minute), yellow indicates a moderate change range (change rate between 0.5% and 1% per minute), and red indicates a rapid change range (change rate greater than 1% per minute). Furthermore, key time points, such as the expected peak load time and load change inflection points, are marked on the trajectory to provide operators with intuitive references.

[0028] The construction of emission limit change curves is achieved in conjunction with environmental emission standards. First, applicable environmental emission standards are obtained, such as national or local concentration limits for pollutants like nitrogen oxides, sulfur dioxide, and particulate matter. Considering that different emission limits may apply to different load ranges, the emission standards are combined with load demand trajectories to generate dynamic emission limit curves. Simultaneously, considering adjustment rules for emission standards during special operating conditions such as start-up and shutdown, the limits are corrected within the corresponding time periods. The final emission limit change curve is overlaid with the predicted emission concentration change curve, clearly showing the risk areas where emissions may exceed standards within future time windows, enabling operators to take preventative measures.

[0029] In practical applications, the forecast time window can be set to 4 hours, with a forecast step size of 5 minutes. This generates load demand points and emission limit points every 5 minutes for the next 4 hours, forming a continuous load demand trajectory and emission limit change curve. This setting satisfies operators' needs for grasping short- and medium-term trends while ensuring that forecast accuracy remains within an acceptable range.

[0030] This invention provides decision support for the operation of thermal power units, helping to ensure emissions meet standards while satisfying load demands, thus achieving a balance between economic efficiency and environmental protection.

[0031] In one optional implementation, the step of constructing a time-series correlated multi-stage optimization objective based on the load demand trajectory and emission limit change curve, wherein the multi-stage optimization objective cascades and couples combustion efficiency, pollutant emissions, and equipment load at each time point within a future time window and fuses them into a multi-objective function through time decay weights, includes: For each time node, the load demand value at that time is extracted as the equipment load sub-target based on the load demand trajectory, and the emission limit at that time is extracted as the pollutant emission sub-target based on the emission limit change curve. The combustion efficiency sub-target at that time is calculated based on the equipment load sub-target and the pollutant emission sub-target. The absolute value of the slope of the load demand trajectory at each time node is calculated as the load change severity, and the proximity of the emission limit change curve to the current emission value at each time node is calculated as the emission constraint tension. A time decay weight is constructed based on the load change severity and the emission constraint tension. The time decay weights are applied to the combustion efficiency sub-objective, pollutant emission sub-objective, and equipment load sub-objective at the corresponding time nodes, and then summed to form a multi-objective function.

[0032] Combination Figure 2The flowchart for constructing a multi-objective function is illustrated below. For example, for each time node within a future time window, the load demand value for that time moment is extracted from the load demand trajectory. This value serves as the equipment load sub-objective, reflecting the output level that the unit should achieve at that time. Simultaneously, the emission limit value corresponding to that time moment is extracted from the emission limit change curve, serving as the pollutant emission sub-objective, characterizing the maximum allowable pollutant concentration at that time.

[0033] The combustion efficiency sub-target is determined by the comprehensive relationship between the equipment load sub-target and the pollutant emission sub-target. Specifically, the standard combustion efficiency benchmark value under the load level is obtained by querying the historical operating condition database based on the equipment load sub-target. Then, the benchmark value is corrected according to the pollutant emission sub-target: when the ratio of the emission limit at that moment to the historical average emission limit is less than 0.9, the emission requirements are considered to be relatively strict. At this time, the standard combustion efficiency benchmark value is multiplied by the correction coefficient k to obtain the corrected combustion efficiency sub-target. The value of the correction coefficient k is determined according to the strictness of the emission limit. The smaller the ratio of the emission limit, the closer the correction coefficient is to the lower limit of 0.95. The closer the ratio of the emission limit is to 1, the closer the correction coefficient is to the upper limit of 1.05. The specific correction coefficient value is obtained by linear interpolation, so that the combustion efficiency sub-target reflects both load demand and emission constraints.

[0034] The slope of the load demand trajectory at each time point is calculated by dividing the difference in load demand values ​​between two adjacent time points by the time interval, i.e., the load change severity S(t) = |[Load(t) - Load(t-1)] ÷ Δt|, where the time interval Δt is the prediction step size (300 seconds for a 5-minute interval), and Load is the load demand value. A larger absolute slope indicates a more drastic load change, reflecting the urgency of the unit's adjustment speed requirement. Emission constraint tension is obtained by dividing the actual emission concentration at the current time by the emission limit at that time, i.e., T(t) = E(t) ÷ E_{limit}(t), where E(t) is the actual emission concentration and E_{limit}(t) is the emission limit. A ratio close to 1 indicates that the emission constraint is approaching the limit and becoming increasingly tense; a ratio less than 0.7 is considered lenient; a ratio between 0.7 and 0.9 is considered moderately tense; and a ratio greater than 0.9 is considered highly tense.

[0035] The time decay weights are constructed based on the combined effects of the drastic load changes and the severity of emission constraints. The basic time decay function adopts an exponential decay form. Where W0 is the initial weight value of 1, Δt is the time interval, and τ is the decay time constant. The basic decay time constant τ0 is set to 3600 seconds, or 60 minutes, by default.

[0036] Based on the basic decay function, the time decay weights are adjusted according to the severity of load changes and the intensity of emission constraints. When the severity of load changes increases or the intensity of emission constraints rises, the weights of the corresponding time points and subsequent time points are increased relative to the basic decay weights to enhance focus on critical periods. This adjustment can be achieved through dynamically adjusting the decay time constant or through a weight transfer mechanism.

[0037] When using a dynamic adjustment of the decay time constant, the base decay time constant is multiplied by an adjustment factor, i.e., τ = τ0 × (1 + α × S(t) + β × [T(t) - 0.7]), where α is the load change severity coefficient (1.5) and β is the emission constraint tension coefficient (0.2). The upper limit of the adjusted decay time constant is set to 120 minutes and the lower limit to 30 minutes to avoid excessively fast or slow decay leading to optimization imbalance.

[0038] When weighting the combustion efficiency sub-objective, a forward weighting method is used, meaning the weights are directly multiplied to maximize efficiency. Specifically, the weighted value equals the combustion efficiency sub-objective multiplied by the time decay weight.

[0039] When weighting pollutant emission sub-targets, negative weighting is used to aim for emissions to be as low as possible below the limit. The sub-target value is represented as the difference between the emission limit and the actual emission concentration, which is then multiplied by the weight. Specifically, the weighted value equals the emission limit minus the actual emission concentration, and the result is then multiplied by the time decay weight.

[0040] When weighting the sub-targets of equipment load, a deviation weighting method is used to pursue load tracking accuracy. Specifically, the difference between the load demand value and the predicted load value is first calculated, and its absolute value is taken. Then, it is multiplied by the time decay weight to obtain the weighted value.

[0041] The weighted combustion efficiency sub-objectives for all time points within the future time window are summed to form the combustion efficiency objective function component. The weighted pollutant emission sub-objectives for all time points are summed to form the pollutant emission objective function component. The weighted equipment load sub-objectives for all time points are summed to form the equipment load objective function component.

[0042] The three components are normalized and then combined into a multi-objective function according to set weighting coefficients. The normalization method is to divide each component by its maximum value in historical data. The multi-objective function consists of the sum of three parts: the first weighting coefficient multiplied by the normalized combustion efficiency objective function component, the second weighting coefficient multiplied by the normalized pollutant emission objective function component, and the third weighting coefficient multiplied by the normalized equipment load objective function component. The default weighting coefficients are set as follows: first weighting coefficient 0.4, second weighting coefficient 0.35, and third weighting coefficient 0.25. The weighting coefficient allocation can be adjusted according to the actual operating strategy.

[0043] This invention achieves differentiated attention to different time periods in the future through time decay weighting. It automatically enhances the influence weight of long-term time periods during periods of drastic load changes and tight emission constraints, effectively balancing multiple objectives such as improving combustion efficiency, achieving pollutant emission standards, and accurately tracking load, and providing accurate and reliable decision-making basis for the dynamic optimization operation of thermal power units.

[0044] In one alternative implementation, the step of constructing the time decay weights includes: A base time decay function is defined, the value of which decreases monotonically as the time interval between the time node and the current time increases; For each time point within the future time window, a conflict situation vector is constructed in two-dimensional space based on the intensity of load change and the tension of emission constraints at that time. The magnitude of the conflict situation vector is then calculated as the comprehensive criticality of that time point. The moment nodes whose comprehensive criticality exceeds a preset comprehensive threshold are identified as potential conflict moments. A buffer time window is set before each conflict moment. The weight values ​​of the basic time decay function at each moment node within the buffer time window are proportionally transferred to the potential conflict moment, so that the time decay weight of the potential conflict moment is increased and the time decay weight of each moment node in the preceding buffer window decreases synchronously. The time decay weight sequence after weight transfer is normalized to obtain the final time decay weight.

[0045] For example, as described above, the base time decay function uses an exponential decay form. The base time decay function ensures that the weight value decreases monotonically with the time interval. When the time interval is 0, the weight value is 1.0, and when the time interval reaches 3 times the base decay time constant, i.e., 10800 seconds, the weight value decays to about 0.05. This function provides a baseline distribution for subsequent weight adjustments.

[0046] A conflict situation vector is constructed for each time point within the future time window to quantify the overall criticality at that time. The future time window length is adjustable from 180 to 360 minutes, with a default of 240 minutes. The sampling interval for each time point is consistent with the prediction step size of the load demand trajectory, typically set to 5 minutes. The conflict situation vector is defined in a two-dimensional plane, with the horizontal axis representing the load change intensity at that time and the vertical axis representing the emission constraint tension at that time. The load change intensity is measured as a percentage load change rate per minute, and the emission constraint tension is a dimensionless ratio between 0 and 1. To ensure the comparability of the two dimensions in the two-dimensional space, the load change intensity is normalized by dividing the current value by the 95th percentile of the indicator in historical statistics, with a historical statistical window of at least 30 days of operating data. The normalized load change intensity and emission constraint tension constitute the two components of the two-dimensional vector, and the magnitude of this vector is calculated as the overall criticality C= , where x is the normalized load change intensity and y is the emission constraint tension.

[0047] The preset comprehensive threshold is flexibly set according to the operating strategy, with a default threshold of 0.8. This threshold corresponds to scenarios where both the normalized load change intensity and emission constraint tension reach a medium-to-high level. When the comprehensive criticality of a certain time node exceeds 0.8, that time node is marked as a potential conflict time, indicating that there is dual pressure of rapid load adjustment and tightening emission constraints at that time. A buffer time window is set before each potential conflict time. The length of the buffer time window is dynamically determined according to the comprehensive criticality: the buffer time window length is 30 minutes when the comprehensive criticality is between 0.8 and 1.0, 45 minutes when the comprehensive criticality is between 1.0 and 1.5, and 60 minutes when the comprehensive criticality exceeds 1.5. The starting time of the buffer time window is the potential conflict time minus the buffer time window length, and the number of time nodes included in the window = buffer time window length ÷ sampling interval.

[0048] The weight transfer mechanism proportionally transfers the weight values ​​of nodes within a buffer window to potential conflict moments. The transfer ratio is determined by the distance between the nodes within the buffer window and the potential conflict moment; nodes closer to each other have a higher transfer ratio. Specifically, a transfer coefficient is calculated for each node within the buffer window. First, the time interval between that moment and the potential conflict moment is calculated. This time interval is divided by the buffer window length to obtain the distance ratio. Then, this distance ratio is subtracted from 1 to obtain a distance normalized value. The distance normalized value ranges from 0 to 1, with the normalized value at the beginning of the buffer window close to 0, and the normalized value for nodes immediately adjacent to the conflict moment close to 1. The square of the distance normalized value is used as the final transfer coefficient. This squaring operation gives nodes closer to the conflict moment a higher transfer weight and makes the distribution of transfer weights exhibit a non-linear increasing characteristic. The actual migration amount of each time node within the buffer window is calculated by multiplying the node's base time decay weight by the migration coefficient and then by the migration intensity factor. The migration intensity factor is set to 0.3 by default, indicating that a maximum of 30% of the weight will be migrated to the conflict time. The value of this factor is adjustable between 0.2 and 0.5, with a larger value indicating more aggressive weight migration. The weight increment at a potential conflict time is the sum of the actual migration amounts of all time nodes within the buffer window. Simultaneously, the weight value of each time node within the buffer window is updated to its original weight minus its actual migration amount, ensuring a unidirectional transfer of weight from the buffer window to the conflict time. The final weight at a potential conflict time is its base time decay weight plus the weight increment aggregated from all nodes within the buffer window.

[0049] After weight migration, the time-decayed weight sequence needs to be normalized to ensure consistency in the total weight. Normalization uses a linear scaling method, calculating the sum of weight values ​​at all time points after migration, then dividing the weight value at each time point by this sum and multiplying by the total weights before migration. The total weights before migration are obtained by summing the weight values ​​of the base time-decay function at all time points. Normalization ensures that weight migration does not change the dimensions of the overall weight allocation, only adjusting the distribution of weights along the time axis. The resulting time-decayed weights exhibit a clear peak characteristic at potential conflict moments, with the peak height positively correlated with the overall criticality at that moment, while the weights in the buffer zone before the peak show a decreasing trend.

[0050] This invention achieves dynamic adjustment of time decay weights through conflict situation vectors and weight transfer mechanisms, automatically strengthening the influence weights at critical moments during periods of drastic load changes and tense emission constraints.

[0051] In one optional implementation, the step of solving the multi-objective function based on the current weight configuration to obtain the Pareto front solution set, and selecting target combustion adjustment parameters from the Pareto front solution set and issuing them to the actuators of the combustion system includes: The weight values ​​of each sub-objective in the current weight configuration are used as weight coefficients for multi-objective optimization. The multi-objective function is solved to obtain the Pareto front solution set. The Pareto front solution set contains multiple sets of non-dominated solutions. Each set of non-dominated solutions corresponds to a set of combustion adjustment parameters and a sequence of target achievement at each time in the future time window. Calculate the fluctuation amplitude of the target achievement sequence of each group of non-dominated solutions within a future time window, calculate the transition smoothness between adjacent moments in the target achievement sequence, and evaluate the temporal stability index of each group of non-dominated solutions based on the fluctuation amplitude and transition smoothness. The combustion adjustment parameter corresponding to the non-dominated solution with the best time-series stability index is selected as the target combustion adjustment parameter, and the target combustion adjustment parameter is sent to the actuator of the combustion system.

[0052] For example, the process of solving the multi-objective function based on the current weight configuration is implemented using a non-dominated sorting genetic algorithm. The current weight configuration is stored as a weight vector, containing three components: combustion efficiency sub-objective weight, pollutant emission sub-objective weight, and equipment load sub-objective weight. The initial weight configuration defaults to 0.4, 0.35, and 0.25, and the sum of the three components equals 1. The solver population size is set to 100 to 200 individuals, adjustable with a default of 150. Each individual represents a set of combustion adjustment parameters, including control variables such as primary air volume, secondary air volume, burner sway angle, and coal feed rate. The number of generations is set to 50 to 100, adjustable with a default of 80 generations. The crossover probability is set to 0.8 to 0.9, the mutation probability is set to 0.05 to 0.1, and the elite retention strategy retains the top 10% of individuals in the current population to directly enter the next generation.

[0053] Non-dominated ranking stratifies individuals in a population according to their dominance relationships. For any two individuals, if one individual is no worse than the other in all objectives and is strictly better than the other in at least one objective, then the former is said to dominate the latter. The Pareto front solution set consists of all individuals not dominated by any other individual, typically ranging from 20 to 50 sets of solutions. Each set of non-dominated solutions contains a set of combustion adjustment parameters and the objective achievement status at each time point within a future time window.

[0054] The target achievement sequence is calculated by substituting the combustion adjustment parameters corresponding to each non-dominated solution into a set of multiple regression equations based on historical operating data. The inputs to the regression equations are primary air volume, secondary air volume, burner tilt angle, coal feed rate, and current operating parameters. The outputs are the predicted combustion efficiency, predicted pollutant emissions, and predicted equipment load for each time point within the future time window. The equation coefficients are fitted from no fewer than 10,000 sets of historical operating data using the least squares method, with the fitting residuals controlled to a combustion efficiency prediction error of less than 1%, a pollutant emission prediction error of less than 5%, and an equipment load prediction error of less than 2%. The target achievement degree for the combustion efficiency sub-target is calculated as: Predicted combustion efficiency ÷ Ideal combustion efficiency, where the ideal combustion efficiency is typically extracted from the historical best operating conditions and is usually between 92% and 95%. The target achievement degree for the pollutant emission sub-target is calculated as: (Emission limit - Predicted emission value) ÷ Emission limit. The target achievement degree for the equipment load sub-target is calculated as: 1 - (Load response deviation ÷ Load demand value). The target achievement degree for each of the three sub-targets is calculated for each non-dominated solution at each time point within the future time window, and these are arranged in a time sequence to form the target achievement sequence.

[0055] The volatility is calculated using a combination of the range and standard deviation of the target achievement sequence. The range is the difference between the maximum and minimum values ​​in the sequence. Volatility = Range × 0.6 + Standard Deviation × 0.4; a smaller value indicates better temporal stability. Transition smoothness is calculated by summing the absolute values ​​of the differences in target achievement at adjacent time points. The absolute value of the difference between the value at time t and the value at time t-1 in the target achievement sequence is subtracted, and the sum of these absolute values ​​over all adjacent time points yields the total transition change. Transition smoothness = 1 - Total transition change ÷ (Number of time points - 1); a value closer to 1 indicates a smoother transition.

[0056] The temporal stability index is calculated as follows: Transition smoothness × 0.6 - Normalized fluctuation amplitude × 0.4, where the normalized fluctuation amplitude is the current fluctuation amplitude divided by the maximum fluctuation amplitude among all non-dominated solutions. The temporal stability index ranges from -0.4 to 1; a higher value indicates better stability of the combustion adjustment scheme within the future time window. The temporal stability index is calculated by iterating through all non-dominated solutions in the Pareto front solution set, and the non-dominated solution with the highest temporal stability index is selected as the optimal solution. If multiple non-dominated solutions have the same maximum temporal stability index, the overall objective function values ​​of these solutions are further compared, and the solution with the largest objective function value is selected.

[0057] The target combustion adjustment parameters are extracted from the optimal solution and include primary air volume setpoints, secondary air volume setpoints, burner sway angle setpoints, and coal feed rate setpoints. These parameters are sent to the actuators via the combustion control system's communication interface. The data frame contains the device address, function code, register address, parameter value, and checksum. The actuators drive the fan inverter to adjust the air volume, the sway mechanism to adjust the burner angle, and the coal feeder inverter to adjust the coal feed rate. Parameter transmission employs either a step-like or a ramp-like mode. The step-like mode directly sets the parameter value to the target value, suitable for scenarios with drastic load changes. The ramp-like mode gradually transitions the parameter from the current value to the target value at a set rate, suitable for steady-state adjustment scenarios. The transition rate defaults to a parameter change of 1% to 3% per second.

[0058] This invention uses a time-series stability index to screen the Pareto front solution set to ensure that the combustion adjustment scheme has good execution continuity and stability within future time windows.

[0059] In one optional implementation, the step of adaptively updating the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism, and then feeding back the updated weight configuration for the next round of solving includes: Collect feedback data after execution, extract actual load response value and actual emission value, calculate the load response deviation between the actual load response value and the load demand value, and calculate the emission compliance margin between the actual emission value and the emission limit. Within a short time window, the transient fluctuation characteristics of historical feedback data are statistically analyzed, and within a long time window, the steady-state deviation characteristics of historical feedback data are statistically analyzed. The transient fluctuation characteristics and steady-state deviation characteristics are then fused into a deviation trend correction amount and superimposed on the load response deviation and emission compliance margin of the current cycle. Calculate the ratio of the corrected load response deviation to the emission compliance margin, calculate the initial weight increment based on the ratio, analyze the trade-off between the equipment load sub-target and the pollutant emission sub-target in the Pareto front solution to obtain the target conflict degree, and scale the initial weight increment based on the target conflict degree to obtain the weight increment. The weight increment is added to the current weight configuration to obtain the updated weight configuration, which is then sent back for the next round of solving.

[0060] For example, feedback data is acquired through a distributed sensor network, including actual unit output measured by load sensors, actual pollutant emission concentrations measured by flue gas analyzers, and operating parameters such as furnace temperature and superheated steam temperature measured by temperature sensors. The data acquisition cycle is adjustable from 5 to 30 seconds, with a default of 10 seconds. The acquired raw data is filtered and outlier-removed before being stored in a historical database. The actual load response value is extracted from the load sensor data, representing the actual unit output level after implementing combustion adjustment parameters, in megawatts. The actual emission value is extracted from the flue gas analyzer data, representing the actual pollutant emission concentration after implementation, in milligrams per standard cubic meter.

[0061] Load response deviation = (Actual load response value - Load demand value). The load demand value is extracted from the aforementioned load demand trajectory to represent the target load at the current moment. A positive load response deviation indicates that the actual output is higher than the demand, indicating over-adjustment; a negative value indicates that the actual output is lower than the demand, indicating under-adjustment. The allowable range for the absolute value of the deviation is usually within ±3% of the load demand value. Emission compliance margin = (Emission limit - Actual emission value) ÷ Emission limit. The emission limit is extracted from the aforementioned emission limit change curve to represent the emission constraint at the current moment. A positive emission compliance margin indicates that the actual emission is lower than the limit, indicating an emission margin; the larger the value, the more sufficient the margin. A negative value indicates that the actual emission exceeds the limit, indicating an exceedance. The safe range for the emission compliance margin is usually required to be greater than 0.1, i.e., the actual emission is more than 10% lower than the limit.

[0062] The statistical analysis of transient fluctuation characteristics is conducted within a short-term time window, with the window length adjustable from 10 to 30 minutes, defaulting to 20 minutes. Within this window, the load response deviation sequence and emission compliance margin sequence from the most recent several rounds are extracted, and the standard deviation of each sequence is calculated as the transient fluctuation intensity index. The transient fluctuation intensity of the load response deviation reflects the degree of fluctuation in load tracking within the short term, while the transient fluctuation intensity of the emission compliance margin reflects the stability of emission control within the short term. Transient fluctuation characteristic = (current round deviation value - mean deviation within the short-term window) × (transient fluctuation intensity ÷ standard deviation of deviation), which quantifies the degree of abnormality of the current deviation relative to the short-term mean.

[0063] The steady-state offset characteristic is statistically analyzed within a long-term time window, with the window length adjustable from 60 to 180 minutes, defaulting to 120 minutes. Historical load response deviation and emission compliance margin sequences are extracted within this window, and a trend line is calculated using a moving average method. The moving average window length is one-fifth of the long-term time window length. The average value of the data within the preceding and following moving windows is calculated for each data point in the sequence to obtain a smoothed trend line. The steady-state offset characteristic is calculated as (the value of the trend line at the end of the long-term window - the value at the beginning of the window). This characteristic reflects the direction and magnitude of the deviation's drift over a long-term time scale. A positive steady-state offset characteristic indicates a continuously increasing deviation trend, while a negative value indicates a continuously decreasing deviation trend.

[0064] The deviation trend correction is calculated as follows: Transient fluctuation characteristic × 0.3 + Steady-state deviation characteristic × 0.7. This formula assigns higher weight to the steady-state deviation characteristic to focus on long-term trends. The corrected load response deviation is calculated as: Current cycle load response deviation + Deviation trend correction of the load response deviation. The corrected emission compliance margin is calculated as: Current cycle emission compliance margin + Deviation trend correction of the emission compliance margin. The correction process incorporates the impact of short-term fluctuations and long-term drift into the current deviation assessment, enhancing the foresight and robustness of the weight adjustments.

[0065] The initial weight increment is calculated based on the ratio of the corrected load response deviation to the emission compliance margin. A positive ratio indicates that the load deviation and emission margin change in the same direction, while a negative ratio indicates that they change in opposite directions. The initial weight increment uses a piecewise linear mapping rule. When the ratio is greater than 1, the initial weight increment for the equipment load sub-target is 0.05, and the initial weight increment for the pollutant emission sub-target is -0.05, indicating that the load tracking weight needs to be increased while the emission weight needs to be reduced. When the ratio is less than -1, the initial weight increment for the equipment load sub-target is -0.05, and the initial weight increment for the pollutant emission sub-target is 0.05, indicating that the load weight needs to be reduced while the emission weight needs to be increased. When the ratio is between -1 and 1, the initial weight increment = ratio × 0.05. The initial weight increment for the combustion efficiency sub-target is -(initial weight increment for the equipment load sub-target + initial weight increment for the pollutant emission sub-target), ensuring that the sum of the weights of the three sub-targets remains 1.

[0066] The target conflict degree is calculated based on the trade-off between the equipment load sub-target and the pollutant emission sub-target in the Pareto front solution. This index comprehensively reflects the intensity of the conflict between the two targets. The larger the value, the more severe the conflict and the more cautious the weight adjustment should be.

[0067] The weight increment is obtained by applying a scaling factor based on the target conflict degree to the initial weight increment. The scaling factor decreases as the target conflict degree increases. The scaling operation avoids overly aggressive weight adjustments that could lead to optimization oscillations when there is significant target conflict. The updated weight configuration is obtained by adding the weight increment to the current weight configuration. The updated weight for the equipment load sub-target = current weight + weight increment of the equipment load sub-target. The weights for the pollutant emission sub-target and the combustion efficiency sub-target are updated in the same way. The updated weight configuration must satisfy the constraint that all weight values ​​are between 0.1 and 0.6 and the sum of the three weights equals 1. If the updated weights exceed the constraint range, they are normalized proportionally to the constraint range.

[0068] This invention achieves closed-loop optimization by integrating short-term transient fluctuations and long-term steady-state offset characteristics and adaptively adjusting the weight configuration based on the target conflict degree.

[0069] In one optional implementation, the step of analyzing the trade-off between the equipment load sub-target and the pollutant emission sub-target in the Pareto front solution to obtain the target conflict degree, and scaling the initial weight increment according to the target conflict degree to obtain the weight increment includes: The optimal solutions for the equipment load sub-objective and the pollutant emission sub-objective are selected from the Pareto front solution set as two sets of boundary solutions. The load change and emission change between the two sets of boundary solutions are calculated. The target sensitivity is calculated based on the load change and emission change. The Pareto front curve is obtained by curve fitting of all non-dominated solutions in the Pareto front solution set in the two-dimensional space formed by the equipment load sub-objective and the pollutant emission sub-objective, and the average curvature of the Pareto front curve is calculated. Calculate the variance of the Euclidean distance between adjacent non-dominated solutions in the Pareto front solution set; The target conflict degree is obtained by weighted fusion of the target sensitivity, the average curvature and the Euclidean distance variance; a scaling factor is constructed based on the target conflict degree, the scaling factor decreases as the target conflict degree increases, and the scaling factor is applied to the initial weight increment to obtain the weight increment.

[0070] For example, by traversing all non-dominated solutions in the Pareto front solution set, comparing the equipment load sub-objective values ​​of each solution, and selecting the solution with the largest sub-objective value as the optimal solution for the equipment load sub-objective, the same logic applies. Similarly, by traversing all non-dominated solutions and comparing the pollutant emission sub-objective values, the solution with the largest sub-objective value is selected as the optimal solution for the pollutant emission sub-objective. These two sets of boundary solutions represent the two extremes of the Pareto front: the optimal solution for the equipment load sub-objective implies optimal load tracking performance but may sacrifice emission control, while the optimal solution for the pollutant emission sub-objective implies optimal emission control but may sacrifice load response.

[0071] The load change is calculated by subtracting the optimal value of the equipment load sub-target from the optimal value of the pollutant emission sub-target. This difference reflects the improvement in the equipment load sub-target when switching from optimal emission to optimal load. Similarly, the emission change is calculated by subtracting the optimal value of the pollutant emission sub-target from the optimal value of the equipment load sub-target. This difference also reflects the improvement in the pollutant emission sub-target when switching from optimal load to optimal emission. Both load and emission changes are non-negative; larger values ​​indicate a wider adjustable range for the target at the Pareto front.

[0072] The normalized value of load change = load change ÷ maximum value of equipment load sub-targets in the Pareto front solution set; the normalized value of emissions change = emissions change ÷ maximum value of pollutant emissions sub-targets in the Pareto front solution set. Normalization eliminates the influence of differences in the dimensions and numerical ranges of the two targets, making them comparable. Target sensitivity = normalized value of load change ÷ normalized value of emissions change; this ratio reflects the load performance cost required for a unit improvement in emissions performance. A higher target sensitivity indicates a more unequal trade-off between the two targets, requiring greater caution when adjusting weights.

[0073] The Pareto front curve fitting is achieved by fitting all non-dominated solutions in the Pareto front solution set to a two-dimensional coordinate space defined by the equipment load sub-objective and the pollutant emission sub-objective. The horizontal axis represents the equipment load sub-objective value, and the vertical axis represents the pollutant emission sub-objective value. Each non-dominated solution corresponds to a point in the two-dimensional space. After sorting these points in ascending order of their horizontal coordinates, a cubic spline interpolation method is used for curve fitting. The cubic spline interpolation constructs a piecewise cubic polynomial between each data point, ensuring that the curve passes through all data points and that the first and second derivatives are continuous at the connection points. The fitted Pareto front curve is a continuously differentiable and smooth curve, and its shape reflects the trade-off between the two objectives.

[0074] The average curvature is calculated by averaging the local curvatures of the Pareto front curve at various points. The curvature at any point on the curve is defined as the rate of change of the tangent angle with respect to the arc length at that point. A larger absolute value of curvature indicates a more pronounced curvature at that point. Curvature calculation employs a discretization method, uniformly sampling the Pareto front curve. The number of sampling points is set to 3 to 5 times the number of curve data points, adjustable to 4 times by default. For each sampling point, the angle between the two straight lines formed by its two adjacent sampling points is calculated. Dividing this angle by the average length of the two straight lines yields the approximate curvature of that sampling point. The average curvature is obtained by summing the absolute values ​​of curvature at all sampling points and dividing by the number of sampling points. The average curvature reflects the overall curvature of the Pareto front curve; a larger value indicates a more complex trade-off between objectives and a more significant difference in the trade-off proportions across different regions.

[0075] The calculation of Euclidean distance variance is used to evaluate the uniformity of the distribution of non-dominated solutions in the Pareto front solution set. All non-dominated solutions in the Pareto front solution set are sorted in ascending order of equipment load sub-objective values, and the Euclidean distance between any two adjacent non-dominated solutions in the two-dimensional objective space is calculated. Euclidean distance = (square of the difference between adjacent solutions' equipment load sub-objective values ​​+ square of the difference between adjacent solutions' pollutant emission sub-objective values). The Euclidean distance sequence for all adjacent solutions is calculated, and the variance of this sequence is taken as the Euclidean distance variance. The variance is calculated as the average of the squares of the differences between each Euclidean distance and its mean. A smaller Euclidean distance variance indicates a more uniform distribution of non-dominated solutions on the Pareto front, better solution diversity, and more comprehensive coverage of alternative solutions for decision-making.

[0076] The target conflict degree is calculated as follows: Target Sensitivity × 0.5 + Mean Curvature × 0.3 + Euclidean Distance Variance × 0.2, with the sum of the three weighting coefficients being 1. This fusion formula comprehensively considers the asymmetry of target trade-offs, the complexity of trade-off relationships, and the distribution characteristics of the solution set, thus fully characterizing the degree of conflict in multi-objective optimization problems. The target conflict degree typically ranges from 0.1 to 2.0, with a higher value indicating more severe target conflict.

[0077] The scaling factor = 1 ÷ (1 + target conflict level × 0.8), where 0.8 is an adjustment factor controlling the scaling intensity, adjustable between 0.5 and 1.0. When the target conflict level is 0.1, the scaling factor is approximately 0.93; when the target conflict level is 2.0, the scaling factor is approximately 0.38. The scaling factor ranges from 0.3 to 1.0. The weight increment = initial weight increment × scaling factor. The scaling operation adaptively adjusts the weight change magnitude based on the target conflict level of the current optimization problem, avoiding overly aggressive adjustments in high-conflict scenarios that could degrade optimization performance.

[0078] This invention constructs a target conflict degree by fusing target sensitivity, average curvature, and Euclidean distance variance, and adaptively scales the weight increment accordingly, thereby achieving a robust weight adjustment strategy.

[0079] A second aspect of the present invention provides an intelligent combustion adjustment system for thermal power units based on multi-objective optimization, comprising: The operating condition parameter acquisition module is used to acquire the operating condition parameters of the combustion system of the thermal power unit; The prediction and target construction module is used to predict the load demand trajectory and emission limit change curve within the future time window based on the change trend of the operating parameters, and construct a time-series related multi-stage optimization target. The multi-stage optimization target cascades and couples the combustion efficiency, pollutant emissions and equipment load at each moment within the future time window into a multi-objective function through time decay weight. The solution and parameter distribution module is used to solve the multi-objective function based on the current weight configuration to obtain the Pareto front solution set, select target combustion adjustment parameters from the Pareto front solution set, and distribute them to the actuators of the combustion system. The feedback and weight update module is used to collect feedback data after execution and adaptively update the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism calculates the weight increment based on the ratio of load response deviation to emission compliance margin in the feedback data and sends the updated weight configuration back for the next round of solving.

[0080] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0081] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0082] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0083] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent combustion adjustment of thermal power units based on multi-objective optimization, characterized in that, include: Obtain the operating parameters of the combustion system of the thermal power unit; Based on the trend of the change of the operating parameters, the load demand trajectory and emission limit change curve within the future time window are predicted, and a multi-stage optimization objective with time-series correlation is constructed. The multi-stage optimization objective cascades and couples the combustion efficiency, pollutant emissions and equipment load at each moment within the future time window into a multi-objective function through time decay weight. The Pareto front solution set is obtained by solving the multi-objective function based on the current weight configuration. Target combustion adjustment parameters are selected from the Pareto front solution set and sent to the actuators of the combustion system. The feedback data after collection and execution is used to adaptively update the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism calculates the weight increment based on the ratio of load response deviation to emission compliance margin in the feedback data, and sends the updated weight configuration back to the next round of solution.

2. The method according to claim 1, characterized in that, The steps for predicting the load demand trajectory and emission limit change curve within a future time window based on the aforementioned operating condition parameter change trends include: The load change rate and emission concentration change rate in the operating parameters are extracted as time series features, and the time series features are extended to generate load evolution sequence and emission evolution sequence within the future time window. A load demand trajectory is constructed based on the load evolution sequence, and an emission limit change curve is constructed based on the emission evolution sequence and environmental emission standards.

3. The method according to claim 1, characterized in that, The steps for constructing a time-series-related multi-stage optimization objective, wherein the multi-stage optimization objective cascades and couples combustion efficiency, pollutant emissions, and equipment load at each time point within a future time window and fuses them into a multi-objective function through time decay weights, include: For each time node, the load demand value is extracted as the equipment load sub-target based on the load demand trajectory, the emission limit is extracted as the pollutant emission sub-target based on the emission limit change curve, and the combustion efficiency sub-target is calculated based on the equipment load sub-target and the pollutant emission sub-target. The absolute value of the slope of the load demand trajectory at each time node is calculated as the load change severity, and the proximity of the emission limit change curve to the current emission value at each time node is calculated as the emission constraint tension. A time decay weight is constructed based on the load change severity and the emission constraint tension. The time decay weights are applied to the combustion efficiency sub-objective, pollutant emission sub-objective, and equipment load sub-objective at the corresponding time nodes, and then summed to form a multi-objective function.

4. The method according to claim 3, characterized in that, The steps for constructing time decay weights include: A base time decay function is defined, the value of which decreases monotonically as the time interval between the time node and the current time increases; For each time point within the future time window, a conflict situation vector is constructed in two-dimensional space based on the intensity of load change and the tension of emission constraints at that time. The magnitude of the conflict situation vector is then calculated as the comprehensive criticality of that time point. The moment nodes whose comprehensive criticality exceeds a preset comprehensive threshold are identified as potential conflict moments. A buffer time window is set before each conflict moment. The weight values ​​of the basic time decay function at each moment node within the buffer time window are proportionally transferred to the potential conflict moment, so that the time decay weight of the potential conflict moment is increased and the time decay weight of each moment node in the preceding buffer window decreases synchronously. The time decay weight sequence after weight transfer is normalized to obtain the final time decay weight.

5. The method according to claim 1, characterized in that, The steps of solving the multi-objective function based on the current weight configuration to obtain the Pareto front solution set, and selecting target combustion adjustment parameters from the Pareto front solution set and issuing them to the actuators of the combustion system include: The weight values ​​of each sub-objective in the current weight configuration are used as weight coefficients for multi-objective optimization. The multi-objective function is solved to obtain the Pareto front solution set. The Pareto front solution set contains multiple sets of non-dominated solutions. Each set of non-dominated solutions corresponds to a set of combustion adjustment parameters and a sequence of target achievement at each time in the future time window. Calculate the fluctuation amplitude of the target achievement sequence of each group of non-dominated solutions within a future time window, calculate the transition smoothness between adjacent moments in the target achievement sequence, and evaluate the temporal stability index of each group of non-dominated solutions based on the fluctuation amplitude and transition smoothness. The combustion adjustment parameters corresponding to the non-dominated solution with the best time-series stability index are selected as the target combustion adjustment parameters and sent to the actuator of the combustion system.

6. The method according to claim 1, characterized in that, The steps of adaptively updating the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism and then feeding back the updated weight configuration for the next round of solving include: Collect feedback data after execution, extract actual load response value and actual emission value, calculate the load response deviation between the actual load response value and the load demand value, and calculate the emission compliance margin between the actual emission value and the emission limit. Within a short time window, the transient fluctuation characteristics of historical feedback data are statistically analyzed, and within a long time window, the steady-state deviation characteristics of historical feedback data are statistically analyzed. The transient fluctuation characteristics and steady-state deviation characteristics are then fused into a deviation trend correction amount and superimposed on the load response deviation and emission compliance margin of the current cycle. Calculate the ratio of the corrected load response deviation to the emission compliance margin, calculate the initial weight increment based on the ratio, analyze the trade-off between the equipment load sub-target and the pollutant emission sub-target in the Pareto front solution to obtain the target conflict degree, and scale the initial weight increment based on the target conflict degree to obtain the weight increment. The weight increment is added to the current weight configuration to obtain the updated weight configuration, which is then sent back for the next round of solving.

7. The method according to claim 6, characterized in that, The steps of analyzing the trade-off between the equipment load sub-objective and the pollutant emission sub-objective in the Pareto front solution to obtain the target conflict degree, and scaling the initial weight increment based on the target conflict degree to obtain the weight increment include: The optimal solutions for the equipment load sub-objective and the pollutant emission sub-objective are selected from the Pareto front solution set as two sets of boundary solutions. The load change and emission change between the two sets of boundary solutions are calculated. The target sensitivity is calculated based on the load change and emission change. The Pareto front curve is obtained by curve fitting of all non-dominated solutions in the Pareto front solution set in the two-dimensional space formed by the equipment load sub-objective and the pollutant emission sub-objective, and the average curvature of the Pareto front curve is calculated. Calculate the variance of the Euclidean distance between adjacent non-dominated solutions in the Pareto front solution set; The target conflict degree is obtained by weighted fusion of the target sensitivity, the average curvature and the Euclidean distance variance; a scaling factor is constructed based on the target conflict degree, the scaling factor decreases as the target conflict degree increases, and the scaling factor is applied to the initial weight increment to obtain the weight increment.

8. A multi-objective optimization-based intelligent combustion adjustment system for thermal power units, used to implement the method of any one of claims 1-7, characterized in that, include: The operating condition parameter acquisition module is used to acquire the operating condition parameters of the combustion system of the thermal power unit; The prediction and target construction module is used to predict the load demand trajectory and emission limit change curve within the future time window based on the change trend of the operating parameters, and construct a time-series related multi-stage optimization target. The multi-stage optimization target cascades and couples the combustion efficiency, pollutant emissions and equipment load at each moment within the future time window into a multi-objective function through time decay weight. The solution and parameter distribution module is used to solve the multi-objective function based on the current weight configuration to obtain the Pareto front solution set, select target combustion adjustment parameters from the Pareto front solution set, and distribute them to the actuators of the combustion system. The feedback and weight update module is used to collect feedback data after execution and adaptively update the weight configuration of each sub-objective in the multi-objective function by establishing a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism calculates the weight increment based on the ratio of load response deviation to emission compliance margin in the feedback data and sends the updated weight configuration back for the next round of solving.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images