Optimal control method and system for boiler thermal energy conversion efficiency

Abstract

This application discloses an optimization control method and system for boiler thermal energy conversion efficiency, relating to the field of data processing technology. The method effectively overcomes the technical problem of low thermal energy conversion efficiency caused by the pre-calibrated static parameter mapping method in existing technologies, which cannot adapt to dynamic conditions such as changes in ambient temperature, equipment wear, and fuel quality fluctuations, through a complete intelligent optimization control closed loop covering multi-source data acquisition, data preprocessing, random forest efficiency prediction, multi-objective Pareto search, optimal solution extraction, and closed-loop execution. The method replaces traditional table lookup with real-time prediction of thermal energy conversion efficiency under current operating conditions using a random forest regression model; replaces single-point calibration with a non-dominated sorting genetic algorithm to search for Pareto optimal solutions across multiple objectives; replaces manual experience-based setting with sorting to approximate ideal solutions to extract optimal balance parameters from multiple objectives; and replaces open-loop execution with closed-loop execution and effect verification.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an optimized control method and system for boiler thermal energy conversion efficiency. Background Technology

[0002] Industrial boilers, as important heat energy conversion equipment, are widely used in various industries such as petrochemicals, textile printing and dyeing, food processing, and district heating. Currently, more than 90% of straight-tube steam generators in China adopt forced circulation mode. Their core working principle is to automatically adjust the combustion load to meet steam demand based on the tracking pressure value set by the user.

[0003] In existing technologies, boiler optimization control generally employs a pre-calibrated static parameter mapping method. Specifically, before the boiler is put into operation, the equipment supplier presets operating parameters such as gas regulating valve opening, feedwater regulating valve opening, feedwater pressure, feedwater flow rate, and damper opening for nine typical load points: 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%. During actual operation, the system adjusts the combustion load by calling the preset parameters through a lookup table based on the deviation between the current pressure setpoint and the measured pressure.

[0004] However, this control method based on static table lookup has significant drawbacks: Low thermal energy conversion efficiency: The preset parameters are calibrated based on standard operating conditions and cannot adapt to the dynamic changes in actual operating conditions such as changes in ambient temperature, equipment wear, and fluctuations in fuel quality, causing the boiler to frequently operate in a region that deviates from the optimal operating conditions. The parameters need to be manually corrected regularly: As the equipment runs for a long time, factors such as scaling in the heat exchange pipeline, burner wear, and sensor drift will cause the original preset parameters to gradually deviate from the optimal state, requiring professional personnel to recalibrate them regularly, which results in high maintenance costs. Lack of real-time optimization capability: Traditional control systems can only execute preset static parameters and cannot dynamically adjust the control strategy based on real-time collected operating data, thus failing to achieve true real-time optimization. Low fuel utilization: Under variable load conditions, fixed parameters lead to incomplete fuel combustion (excessive air coefficient is too large) or unstable combustion (excessive air coefficient is too small), which wastes fuel and increases pollutant emissions. Summary of the Invention

[0005] The main objective of this application is to provide an optimized control method and system for boiler thermal energy conversion efficiency, so as to solve the problem of low boiler thermal energy conversion efficiency in the prior art.

[0006] To achieve the above objectives, this application provides the following technical solution: An optimization control method for boiler thermal energy conversion efficiency, the optimization control method comprising: Step S1: Collect the boiler's operating parameters, including fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure, and integrate them into a boiler operating parameter set. Step S2: Perform data preprocessing on the boiler operating parameter set to obtain a standardized boiler operating feature dataset; Step S3: Input the standardized boiler operating characteristic dataset into the random forest regression model and output the predicted value of thermal energy conversion efficiency under the current operating conditions; Step S4: Input the predicted thermal energy conversion efficiency value into the NSGA-Ⅲ non-dominated sorting genetic algorithm to perform multi-objective Pareto front iterative search and obtain the multi-objective Pareto solution set; Step S5: Extract the optimal solution of the multi-objective Pareto solution set by sorting the approximate ideal solution using TOPSIS, and obtain the optimal combination of control parameters based on gas, water supply, and dampers. Step S6: The optimal control parameter combination is sent to the corresponding execution device to complete the boiler thermal energy conversion efficiency optimization control.

[0007] Beneficial effects of steps S1 to S6: This method, through steps S1 to S6, forms a complete intelligent optimization control closed loop covering multi-source data acquisition, data preprocessing, random forest efficiency prediction, NSGA-III multi-objective Pareto search, TOPSIS optimal solution extraction, and closed-loop execution. It effectively overcomes the technical problem of low thermal energy conversion efficiency caused by pre-calibrated static parameter mapping methods in existing technologies, which cannot adapt to dynamic conditions such as changes in ambient temperature, equipment wear, and fuel quality fluctuations. This method replaces traditional table lookups with real-time prediction of thermal energy conversion efficiency under current operating conditions using a random forest regression model; replaces single-point calibration with NSGA-III non-dominated sorting genetic algorithm searching for Pareto optimal solutions across multiple objectives; replaces manual experience-based setting with TOPSIS approximation of ideal solutions sorting to extract optimal balance parameters from multiple objectives; and replaces open-loop execution with closed-loop execution and effect verification. This fundamentally solves the inherent defects of static control strategies deviating from optimal operating conditions under varying loads and requiring periodic recalibration by professional technicians.

[0008] Step S1 involves collecting nine operating parameters—fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure—and integrating them into an operating parameter set. This provides a complete multi-dimensional data foundation for subsequent analysis, overcoming the information limitations of traditional control systems that rely solely on a single pressure deviation signal for adjustment. Step S2 performs missing value imputation, outlier handling, and standardization preprocessing on the operating parameter set to eliminate the impact of sensor noise, sampling omissions, and dimensional differences on model accuracy, ensuring the stability and consistency of the input to the random forest regression model. Step S3 inputs the standardized feature dataset into the random forest regression model, using multiple decision trees to predict in parallel and take the average to output the predicted thermal energy conversion efficiency value, providing a basis for multi-objective optimization. Provide a reliable real-time efficiency benchmark; Step S4 aims to maximize thermal energy conversion efficiency, minimize pollutant emissions, and minimize fuel consumption. It uses the NSGA-Ⅲ algorithm to perform a multi-objective Pareto front iterative search, overcoming the technical deficiency of traditional single-objective optimization that cannot simultaneously consider efficiency and emission indicators; Step S5 uses the TOPSIS method to calculate the relative proximity of each solution to the positive and negative ideal solutions and extracts the optimal solution, achieving the best balance among multiple optimization objectives; Step S6 encapsulates the optimal control parameter combination according to the communication protocol and sends it to the gas regulating valve, feedwater pump, and damper controller. It receives execution feedback and verifies whether the adjusted efficiency has reached the preset target, forming a complete closed-loop adaptive optimization mechanism that can achieve continuous optimization across the entire operating range without manual intervention.

[0009] As a further improvement to this application, step S1 involves collecting the boiler's operating parameters, including fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure, and integrating them into a boiler operating parameter set, including: Step S1.1: Collect boiler operating parameters such as fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure through sensors, and integrate them into a boiler operation dataset. Step S1.2: Clean the boiler operation dataset to obtain a cleaned operation dataset; Step S1.3: Timestamp alignment is performed on the cleaned running dataset to obtain the timestamp-aligned running dataset; Step S1.4: After aligning the timestamps, perform field mapping and format unification on the operating parameters to obtain the boiler operating parameter set.

[0010] Beneficial effects of steps S1.1 to S1.4: This series of steps synchronously collects nine key operating parameters—fuel flow, steam production, condensate flow, feedwater flow, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure—using a sensor array. Data cleaning removes abnormal records caused by momentary sensor malfunctions. Timestamp alignment unifies the time base for parameters with different sampling frequencies. Field mapping and format unification eliminate data format differences between different sensors. Finally, these parameters are integrated into a consistent set of boiler operating parameters, providing a complete multi-dimensional data foundation for subsequent preprocessing and model prediction. Traditional control schemes rely solely on a single pressure deviation signal to adjust combustion load, resulting in insufficient information dimensions and an inability to reflect the full picture of the boiler's actual operating conditions. This series of steps, through comprehensive multi-source data acquisition, provides ample information input for intelligent optimization, ensuring a comprehensive, accurate, and consistent data foundation for subsequent preprocessing and model prediction, thus guaranteeing the reliability of the entire optimization control chain from the source.

[0011] Step S1.1 involves deploying sensor arrays, including gas flow meters, steam flow meters, electromagnetic flow meters, temperature thermocouples, oxygen content analyzers, and pressure transmitters, to simultaneously collect nine types of parameters at a 10-second sampling period, ensuring comprehensive coverage and timeliness of the data source. Step S1.2 removes abnormal data records caused by sensor interruptions, communication packet loss, or electromagnetic interference, filtering out invalid data from subsequent processes to avoid interference from abnormal samples to model training. Step S1.3 performs timestamp alignment on parameters with different sampling rates, integrating and aligning all parameters under the same time reference system to eliminate the impact of time sequence misalignment on multi-source data fusion analysis. Step S1.4 standardizes field names, unifies units of measurement, and standardizes data formats for each parameter, providing structurally unified and semantically clear data input for the preprocessing steps.

[0012] As a further improvement to this application, step S2 involves preprocessing the boiler operating parameter set to obtain a standardized boiler operating characteristic dataset, including: Step S2.1: Fill the missing values ​​in the boiler operating parameter set using interpolation to obtain the boiler operating parameter set after missing value filling; Step S2.2: Identify outliers in the boiler operating parameter set after the missing values ​​are filled, and process outliers that exceed the reasonable range to obtain the boiler operating parameter set after outlier processing. Step S2.3: Perform data standardization processing on the boiler operating parameter set after outlier processing to obtain a standardized boiler operating feature dataset.

[0013] Beneficial effects of steps S2.1 to S2.3: This series of steps eliminates incomplete data caused by sensor transients or communication interruptions by filling in missing values ​​through interpolation, removes outliers exceeding reasonable physical ranges by identifying and processing outliers, and eliminates the imbalanced impact of differences in the dimensions and numerical ranges of different parameters on model training through data standardization, transforming the boiler operating parameter set into a standardized boiler operating feature dataset. Traditional control schemes directly look up and retrieve static parameters based on raw measurements, where noise, missing values, and dimensional differences in the raw data are amplified at each stage in subsequent calculations, ultimately affecting control accuracy. This series of steps, through a systematic data preprocessing workflow, eliminates these interfering factors before the data enters the model, enabling the random forest regression model to extract true patterns from clean, uniformly scaled feature data, ensuring prediction accuracy while improving the convergence speed of model training.

[0014] In step S2.1, missing data caused by sensor momentary failure, communication interruption, or PLC sampling gap is filled using linear interpolation or spline interpolation. The missing values ​​are estimated based on the changing trends of adjacent valid data points, ensuring the continuity of time series while avoiding the introduction of systematic bias. Step S2.2 uses the 3σ criterion or box plot method to automatically identify outliers in parameters such as fuel flow rate, coil temperature, and flue gas oxygen content that exceed reasonable physical ranges or statistical distributions, and replaces them with the mean or median of adjacent time windows to avoid interference from outliers in the model's learning of normal operating conditions. Step S2.3 uses Min-Max normalization or Z-score standardization to map parameters with different dimensions, such as fuel flow rate (0-4000 Nm³ / h), coil temperature (0-1200℃), and steam pressure (0-2.5 MPa), to a unified numerical range, eliminating the imbalance of feature weights caused by differences in dimensions, and ensuring that multiple decision trees treat each dimension of features equally when splitting.

[0015] As a further improvement to this application, step S3 involves inputting the standardized boiler operating characteristic dataset into a random forest regression model and outputting a predicted value of the thermal energy conversion efficiency under the current operating condition, including: Step S3.1: Divide the standardized boiler operation feature dataset into a training dataset and a test dataset; Step S3.2: Train the random forest regression model using the training dataset, and construct multiple decision trees by minimizing the Gini coefficient to obtain the trained random forest regression model; Step S3.3: Use the test dataset to evaluate the performance of the trained random forest regression model to obtain the performance-evaluated random forest regression model. Step S3.4: Input the standardized boiler operation characteristic dataset collected in real time into the random forest regression model after performance evaluation, and obtain the predicted value of thermal energy conversion efficiency under the current operating condition by making parallel predictions through multiple decision trees and taking the average value.

[0016] Beneficial effects of steps S3.1 to S3.4: This series of steps establishes an end-to-end mapping relationship from multi-source operating parameters to efficiency predictions by dividing the standardized feature dataset into training and testing sets, training multiple decision trees to construct a random forest regression model based on the principle of minimizing the Gini coefficient, evaluating the model's generalization performance using the test set, and inputting real-time collected standardized feature data into the model for parallel prediction by multiple decision trees, taking the average value, and outputting the predicted value of heat conversion efficiency. Traditional solutions rely on looking up tables of pre-calibrated static parameters from equipment suppliers, which not only fails to reflect the current actual operating conditions of the equipment but also cannot predict unseen load points. This series of steps, through a data-driven ensemble learning method, enables the model to autonomously learn the nonlinear relationship between various parameters and efficiency from historical operating data, outputting reliable efficiency predictions under any operating condition, and providing real-time and accurate objective function input for NSGA-Ⅲ multi-objective optimization.

[0017] In step S3.1, the standardized boiler operation feature dataset is divided into training and test datasets in a 7:3 or 8:2 ratio to ensure that the training samples cover the operating characteristics of each load range while retaining independent samples for generalization performance evaluation. In step S3.2, multiple subsets are extracted from the training dataset with replacement using the Bootstrap sampling method. For each subset, a CART decision tree is recursively constructed by minimizing the Gini coefficient, with a tree depth of 8 layers and a feature subset size of 3. The risk of overfitting of a single tree is reduced by integrating multiple trees. In step S3.3, the root mean square error (RMSE) and coefficient of determination (R²) are used to evaluate the prediction bias on the test set. When R² is lower than 0.85, hyperparameter tuning is performed, such as adjusting the number of decision trees to 200 or limiting the minimum number of samples in the leaf nodes, until the model reaches the preset accuracy threshold. In step S3.4, the real-time standardized feature data is input into multiple decision trees, and the output values ​​of each tree are calculated in parallel and then the arithmetic mean is taken. The variance reduction principle of ensemble learning is used to make the prediction results have good noise resistance.

[0018] As a further improvement to this application, in step S4, the predicted thermal energy conversion efficiency value is input into the NSGA-Ⅲ non-dominated sorting genetic algorithm to perform a multi-objective Pareto front iterative search, obtaining a multi-objective Pareto solution set, including: Step S4.1: Construct a multi-objective optimization model with the objective functions of maximizing thermal energy conversion efficiency, minimizing pollutant emissions, and minimizing fuel consumption; Step S4.2: Input the predicted value of thermal energy conversion efficiency into the multi-objective optimization model, and use the NSGA-Ⅲ non-dominated sorting genetic algorithm to generate an initial population for the multi-objective optimization model to obtain the initial population; Step S4.3: Perform non-dominated sorting and reference point generation on the initialized population to obtain the Pareto front of the current population; Step S4.4: The offspring population is generated through selection, crossover, and mutation genetic operations, and the parent population and the offspring population are merged for environmental selection to obtain the environmentally selected population. Step S4.5: Repeat steps S4.3 to S4.4 to iteratively perform non-dominated sorting, reference point generation, and environment selection until the termination condition is met, and obtain the multi-objective Pareto solution set.

[0019] Beneficial effects of steps S4.1 to S4.5: This series of steps constructs a multi-objective optimization model with the optimization objectives of maximizing thermal energy conversion efficiency, minimizing pollutant emissions, and minimizing fuel consumption. Random forest predictions are used as model input. The NSGA-III non-dominated sorting genetic algorithm performs initial population generation, non-dominated sorting and reference point generation, genetic operations to produce offspring, environmental selection to preserve high-quality solutions, and iterative search until the termination condition is met, ultimately obtaining a uniformly distributed and convergent multi-objective Pareto solution set. Traditional static lookup table schemes only preset parameters with single pressure tracking as the objective, completely failing to consider the three conflicting optimization objectives of efficiency, emissions, and fuel consumption. This leads to improved boiler efficiency but worsened pollutant emissions or runaway fuel costs. This series of steps uses the NSGA-III algorithm to perform a global Pareto search in the three-dimensional objective space, utilizing a reference point mechanism to maintain the diversity distribution of the solution set in the three objective directions, ensuring that each solution on the Pareto front represents an optimal solution with different trade-offs among thermal efficiency, emissions, and fuel consumption.

[0020] Step S4.1 uses three objective functions: maximizing thermal energy conversion efficiency η, minimizing CO and NOx emission concentrations in flue gas ε, and minimizing natural gas consumption F. Constraints such as steam pressure fluctuation ±0.1 MPa, steam production deviation ±5%, and actuator opening upper limit not lower than lower limit not higher than safety boundary are set to clarify the boundaries and preference directions of the multi-objective optimization. Step S4.2 uses the current predicted efficiency value as a reference point and generates initial seeds using Latin hypercube sampling in a three-dimensional decision space with gas valve opening 20%-90%, water valve opening 30%-95%, and damper opening 25%-85%. The initial solution covers the entire feasible region. Step S4.3 performs a fast non-dominated sorting of the current population to divide the frontier hierarchy, and generates uniformly distributed reference points on the hyperplane to guide the search direction, taking into account the uniformity of coverage of each target dimension. Step S4.4 uses the tournament selection operator to select excellent parents, generates offspring through simulated binary crossover (SBX) and polynomial mutation, and then merges them with the parents to perform environment selection based on reference points, retaining elite solutions. Step S4.5 repeats the iteration until the maximum number of generations (50 generations) is reached or the Pareto front improvement is less than 1% for 10 consecutive generations, and then terminates to ensure the convergence and integrity of the solution set.

[0021] As a further improvement to this application, step S5 involves extracting the optimal solution from the multi-objective Pareto solution set by sorting the approximate ideal solutions using TOPSIS, thereby obtaining the optimal combination of control parameters based on gas, water supply, and dampers, including: Step S5.1: Calculate the positive ideal solution and negative ideal solution for each solution in the multi-objective Pareto solution set; Step S5.2: Calculate the distance from each solution to its corresponding positive ideal solution and negative ideal solution, and obtain the distance value pair for each solution; Step S5.3: Calculate the relative proximity value between each solution and the ideal solution based on the distance value of each solution; Step S5.4: Sort each solution in the multi-objective Pareto solution set according to all relative proximity values ​​from largest to smallest, and extract the solution ranked first as the optimal solution; Step S5.5: The optimal solution is analyzed into an optimal combination of control parameters based on gas, water supply, and dampers.

[0022] Beneficial effects of steps S5.1 to S5.5: This series of steps calculates the positive and negative ideal solutions for each objective on the Pareto solution set, quantifies the deviation of each solution from the positive and negative ideal solutions using Euclidean distance, calculates the relative proximity value based on the distance values, sorts the solutions in descending order of relative proximity value, and extracts the solution with the highest relative proximity value as the optimal solution. This optimal solution is then analyzed into specific values ​​for three control parameters: gas valve opening, water valve opening, and damper opening, completing the final decision output from the Pareto solution set to executable control parameters. In traditional schemes, operators select the operating parameters closest to the current load from a preset parameter table based on experience. This selection process lacks quantitative evaluation standards, and the results of different operators' selections may vary significantly, failing to guarantee that the selected parameters achieve an optimal balance between efficiency, emissions, and fuel consumption. This series of steps uses the TOPSIS method to perform multi-objective comprehensive evaluation on the Pareto front using positive and negative ideal solutions as a benchmark. It extracts the control parameter combination with the best overall performance based on relative proximity, ensuring that the final decision fully reflects the trade-offs between the importance of each optimization objective.

[0023] In step S5.1, the optimal extreme values ​​of each objective in the Pareto solution set are used to form the positive ideal solution vector Z+ and the nearest extreme value is used to form the negative ideal solution vector Z-, providing an objective benchmark reference for the positive and negative directions and avoiding preference bias caused by subjective weighting. In step S5.2, the distance di+ to the positive ideal solution and the distance di- to the negative ideal solution of each solution are calculated sequentially using the Euclidean distance formula, and the deviation of each solution from the ideal benchmark point in the three-dimensional target space is quantified by geometric distance. In step S5.3, the relative proximity value between each solution and the positive ideal solution is calculated. The closer the value is to 1, the closer the solution is to the ideal state. In step S5.4, the relative proximity values ​​of all solutions are sorted from largest to smallest, and the solution ranked first is extracted as the optimal solution. This solution achieves a globally optimal comprehensive balance in the three target dimensions of efficiency, emissions, and fuel consumption. In step S5.5, the three-dimensional decision variables of the optimal solution are mapped to the percentage opening of the gas regulating valve, the percentage opening of the water pump, and the percentage opening of the damper controller, forming a specific combination of control parameters that can be directly issued and executed.

[0024] As a further improvement to this application, step S6 involves sending the optimal control parameter combination to the corresponding execution device to complete the boiler thermal energy conversion efficiency optimization control, including: Step S6.1: Encapsulate the optimal control parameter combination into an instruction according to the corresponding execution device communication protocol to obtain the optimal control parameter combination after instruction encapsulation. Step S6.2: The optimal combination of control parameters after encapsulating the instruction is sent to the corresponding execution device; Step S6.3: Receive the corresponding optimal control parameters from each execution device, and receive the parameter adjustment completion signal fed back by each execution device after execution; Step S6.4: In response to the parameter adjustment completion signal fed back by all executing devices, collect the boiler's operating parameters after parameter adjustment and calculate the adjusted thermal energy conversion efficiency; Step S6.5: Determine whether the adjusted thermal energy conversion efficiency has reached the preset optimization target. If it has not been reached, trigger the re-optimization control process. If it has been reached, complete the boiler thermal energy conversion efficiency optimization control.

[0025] Beneficial effects of steps S6.1 to S6.5: This series of steps encapsulates the optimal control parameter combination into executable instructions according to the communication protocols of each executing device, and sends them to the gas regulating valve, feedwater pump, and damper controller respectively. It receives parameter adjustment completion signals from each device after execution, collects the adjusted boiler operating parameters, calculates the actual thermal energy conversion efficiency, and determines whether to trigger a re-optimization process or complete the optimized control through threshold judgment, forming a complete closed-loop adaptive optimization control loop. Traditional static schemes only calibrate parameters once in the preset stage, without any self-checking or effect feedback mechanisms during subsequent operation. Once the parameters deviate from the optimal state due to equipment wear, pipeline scaling, changes in fuel quality, etc., the control system continues to operate with inefficient parameters without awareness until professional inspection and recalibration. This series of steps, through a complete closed-loop mechanism of parameter issuance → execution feedback → effect calculation → condition judgment, automatically verifies the adjustment effect after each optimization and determines whether to re-trigger the optimization process, enabling the control system to have continuous adaptive capabilities and fundamentally eliminating the dependence on periodic manual correction.

[0026] Step S6.1 follows the communication protocol of the Siemens S7-1500 PLC or Modbus. The RTU protocol format converts the three control parameters—gas valve opening, feedwater valve opening, and damper opening—into standard register write instructions, ensuring the compatibility and executability of downlink instructions. Step S6.2 sends the encapsulated instructions as a 4-20mA current signal to each execution device via the PLC's analog output module, completing the physical implementation of the optimized parameters. Step S6.3 receives real-time opening signals and confirmation signals from each execution device, confirming that all parameter adjustments have been executed, providing a basis for subsequent effect verification. Step S6.4 re-collects sensor data after parameter adjustment and calculates the actual thermal energy conversion efficiency, comparing it with the efficiency before optimization to quantify the optimization effect. Step S6.5 sets an optimization improvement threshold, such as 2%. If the efficiency improvement after adjustment does not reach the threshold, the optimization process from S3 to S6 is re-executed. If the threshold is reached or exceeded, the current optimization cycle is complete, forming a continuously iterative adaptive control loop to ensure the boiler always operates in near-optimal condition throughout its entire lifecycle.

[0027] To achieve the above objectives, this application also provides the following technical solutions: An optimization control system for boiler thermal energy conversion efficiency, wherein the optimization control system is applied to the optimization control method described above, and the optimization control system includes: The boiler operating parameter acquisition module is used to collect boiler operating parameters such as fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure, and integrate them into a boiler operating parameter set. The boiler operating parameter preprocessing module is used to preprocess the boiler operating parameter set to obtain a standardized boiler operating feature dataset. The thermal energy conversion efficiency prediction module is used to input the standardized boiler operating characteristic dataset into the random forest regression model and output the predicted value of thermal energy conversion efficiency under the current operating conditions. The Pareto front iterative search module is used to input the predicted thermal energy conversion efficiency value into the NSGA-Ⅲ non-dominated sorting genetic algorithm to perform a multi-objective Pareto front iterative search and obtain a multi-objective Pareto solution set. The optimal control parameter combination acquisition module is used to extract the optimal solution of the multi-objective Pareto solution set by sorting the approximation ideal solution through TOPSIS, and obtain the optimal control parameter combination based on gas, water supply and damper. The boiler thermal energy conversion efficiency optimization control module is used to send the optimal control parameter combination to the corresponding execution device to complete the boiler thermal energy conversion efficiency optimization control.

[0028] To achieve the above objectives, this application also provides the following technical solutions: An electronic device includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the above-described method for optimizing the control of boiler thermal energy conversion efficiency.

[0029] To achieve the above objectives, this application also provides the following technical solutions: A computer-readable storage medium storing program instructions, which, when executed by a processor, enable the implementation of the above-described method for optimizing boiler thermal energy conversion efficiency. Attached Figure Description

[0030] Figure 1 This is a schematic flowchart illustrating the steps of an embodiment of the boiler thermal energy conversion efficiency optimization control method of this application; Figure 2 This is a schematic diagram of the functional modules of an embodiment of a boiler thermal energy conversion efficiency optimization control system according to this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the electronic device of this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the storage medium of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0032] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (e.g., as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive having. For example, a process, method, system, product, or device having a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0033] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be present in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0034] It should be noted that, due to the limited types and number of symbols or letters that can represent specific meanings, for embodiments with many formulas or codes, there may be situations where symbols or letters cannot meet the usage requirements. Therefore, the interpretation of formula symbols in the steps or sub-steps of the embodiments is only valid for the current step or sub-step.

[0035] If the same symbol has different interpretations in different steps or sub-steps, the interpretation in the current step or sub-step shall prevail; if the same symbol appears in different steps or sub-steps, but no interpretation is given in subsequent steps or sub-steps after its first appearance, the interpretation in the first step or sub-step shall be used.

[0036] For example, "i=1,2,...,n" is a writing convention and a well-known meaning. If "i=1,2,...,n" has already appeared once in an embodiment and needs to be used again in formulas with different scenarios and meanings in the future, then the meaning of "i=1,2,...,n" should be interpreted according to the corresponding formulas. If "i=1,2,...,n" is not used for the first time and is changed to a symbol or letter with a non-well-known meaning, such as "p=1,2,...,q", to avoid repetition, it is more likely to cause confusion, ambiguity, and unclear problems.

[0037] like Figure 1 As shown, this embodiment provides an optimized control method for boiler thermal energy conversion efficiency. In this embodiment, the optimized control method includes the following steps: Step S1: Collect the boiler's operating parameters, including fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure, and integrate them into a boiler operating parameter set.

[0038] Furthermore, step S1 specifically includes the following steps: Step S1.1: Collect boiler operating parameters such as fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure through sensors, and integrate them into a boiler operation dataset.

[0039] Preferably, the core of this step is to synchronously and in real-time collect multi-dimensional parameters of boiler operation. Through a high-precision sensor array deployed at various measuring points of the straight-tube steam generator, nine key parameters reflecting the boiler's operating status are acquired, including fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure. This provides a complete raw data foundation for subsequent data cleaning and integration.

[0040] For fuel flow acquisition, a Rosemount high-precision gas flow meter is installed in the natural gas inlet pipeline. The measurement accuracy is ±0.5%, and the standard 4-20mA current signal is output to the PLC analog input module. The range covers 0-4000Nm³ / h, and the sampling period is 10 seconds.

[0041] For the collection of steam production and condensate flow, an E+H steam flow meter with temperature and pressure compensation function is installed on the steam outlet pipeline, with a measurement accuracy of ±0.5%; a Yokogawa electromagnetic flow meter is installed on the sewage discharge pipeline, with a measurement accuracy of ±0.2%. Both are collected synchronously and output to the same data acquisition module.

[0042] For temperature parameter acquisition, a type K thermocouple is installed on the wall of the flame channel of the coil to measure the coil temperature, with a range of 0-1200℃ and a measurement accuracy of ±1℃; a similar thermocouple is installed in the flue gas duct to measure the flue gas temperature. Both are converted into standard signals by a temperature transmitter and then connected to the PLC.

[0043] For the collection of oxygen content in flue gas, a German Schutz zirconia oxygen content analyzer was installed in the flue gas duct with a measurement accuracy of ±1%, and the output signal was used for subsequent combustion efficiency analysis.

[0044] For pressure parameter acquisition, Rosemount pressure transmitters are installed on the steam outlet pipeline and the feedwater pump outlet pipeline, respectively, with a measurement accuracy of ±0.1% and measurement ranges of 0-2.5MPa and 0-5.0MPa, respectively.

[0045] For example, during a certain sampling period, the sensor array synchronously collected the following data: fuel flow rate of 2850 Nm³ / h, steam production of 22000 kg / h, condensate flow of 450 kg / h, feedwater flow of 23000 kg / h, coil temperature of 980℃, flue gas temperature of 185℃, flue gas oxygen content of 4.2%, steam pressure of 1.25 MPa, and feedwater pressure of 1.65 MPa. The above data was aggregated by the acquisition module to form a single raw record.

[0046] Step S1.2: Clean the boiler operation dataset to obtain a cleaned operation dataset.

[0047] Preferably, the core of this step is to perform quality checks and anomaly filtering on the raw collected data. By setting the physical reasonable range of each parameter and data integrity rules, data missing records caused by sensor interruption, communication packet loss or electromagnetic interference and abnormal records exceeding reasonable physical thresholds are eliminated, ensuring the data quality for subsequent processes.

[0048] Specifically, for data integrity checks, each of the nine types of parameters is checked to see if they all return valid values ​​within the current sampling period. If any parameter is missing or returns an empty value, the record is marked as invalid and removed to avoid incomplete data from misleading subsequent analysis.

[0049] For the physical rationality check, reasonable ranges are set for each parameter based on the equipment design parameters: fuel flow rate not exceeding 4000 Nm³ / h, coil temperature not exceeding 1200℃, and steam pressure not exceeding 2.5 MPa. Data records exceeding the above ranges are judged as sensor malfunctions or abnormal operating condition records and are rejected.

[0050] Among them, the logical consistency check examines the physical logical relationship between parameters. For example, the water supply should not be less than 80% of the sum of the steam production and the condensate, and the flue gas temperature should not be higher than the coil temperature. Data records that do not conform to the logical relationship are marked as questionable records and left for manual review.

[0051] For example, if a record shows a coil temperature of 1310℃, which exceeds the maximum range of 1200℃ for a type K thermocouple, it is determined to be abnormal data caused by a sensor malfunction. This record is then cleaned up and removed, and will not proceed to subsequent steps.

[0052] Step S1.3: Timestamp alignment is performed on the cleaned running dataset to obtain the timestamp-aligned running dataset.

[0053] Preferably, the core of this step is to complete the synchronization processing of different sensors in terms of sampling time. By standardizing the timestamps of the sampling time of each parameter, the timing misalignment caused by differences in sensor start-up time, communication delay, and PLC scanning cycle is eliminated, ensuring the comparability of each parameter under the same time reference system.

[0054] For timestamp standardization, the PLC's system clock is used as a unified time reference. The original timestamps of the data collected by each sensor are converted into the PLC system time format with millisecond precision, ensuring that all parameters are aligned on a unified time axis.

[0055] In the case of sampling frequency difference processing, considering the possible sampling rate differences of different sensors (such as gas flow meter sampling once per second and thermocouple sampling once every 2 seconds), a 10-second fixed window is used as the time reference. The nearest neighbor interpolation method is used to align each parameter within the 10-second time window, and the most recent sampled value within the window is taken as the representative value of the window.

[0056] For communication delay compensation, for sensors with long transmission distances (such as oxygen content analyzers at the exhaust duct, with a transmission distance of about 50 meters), compensation and correction are performed based on the average delay time measured on site, and the delayed sampling data is returned to the corresponding timestamp position.

[0057] For example, during a data acquisition process, the gas flow meter collects data of 2850 Nm³ / h at time T0, while the oxygen content analyzer arrives at the acquisition module at time T0+1.5 due to communication delay. After the timestamps are aligned, both are assigned to the 10-second window where T0 is located, ensuring the causal consistency of each parameter within the same time window.

[0058] Step S1.4: After aligning the timestamps, perform field mapping and format unification on the operating parameters to obtain the boiler operating parameter set.

[0059] Preferably, the core of this step is to complete the structured and unified processing of the output data from different sensors. By standardizing the field names of the original data, unifying the units of measurement, and ensuring the consistency of the data format, the scattered timestamps are aligned and the operating parameters are integrated into a set of boiler operating parameters with a consistent structure, providing structured data input that can be directly called for subsequent preprocessing steps.

[0060] In terms of field name standardization, custom field names in the raw data of each sensor are mapped to standard field names. For example, "Gas_Flow_Raw" is mapped to "Fuel Flow" and "Steam_Pressure_CH1" is mapped to "Steam Pressure", ensuring that all field names in the dataset are uniformly identified in Chinese.

[0061] Among these measures, the standardization of measurement units will convert parameters that may come from different equipment and have different range configurations into standard measurement units: fuel flow rate will be standardized to Nm³ / h, steam production to kg / h, coil temperature to ℃, and steam pressure to MPa, thus eliminating the risk of data misunderstanding caused by inconsistent units.

[0062] Specifically, for data format consistency, the numerical format of all parameters is unified to floating-point numbers with two decimal places, the timestamp is unified to the format "YYYY-MM-DDHH:MM:SS", and empty values ​​are uniformly filled with NULL markers, so that each record in the dataset is completely consistent in structure.

[0063] For example, in the original data, the output value of the gas flow meter is "2850.36" (unit: Nm³ / h, original sensor format), and the output value of the thermal resistor is "981.2" (unit: ℃, original floating point number). After format unification, the corresponding record in the boiler operating parameter set is "fuel flow: 2850.36 Nm³ / h; coil temperature: 981.20℃; ...".

[0064] Beneficial effects of steps S1.1 to S1.4: This series of steps synchronously collects nine key operating parameters—fuel flow, steam production, condensate flow, feedwater flow, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure—using a sensor array. Data cleaning removes abnormal records caused by momentary sensor malfunctions. Timestamp alignment unifies the time base for parameters with different sampling frequencies. Field mapping and format unification eliminate data format differences between different sensors. Finally, these parameters are integrated into a consistent set of boiler operating parameters, providing a complete multi-dimensional data foundation for subsequent preprocessing and model prediction. Traditional control schemes rely solely on a single pressure deviation signal to adjust combustion load, resulting in insufficient information dimensions and an inability to reflect the full picture of the boiler's actual operating conditions. This series of steps, through comprehensive multi-source data acquisition, provides ample information input for intelligent optimization, ensuring a comprehensive, accurate, and consistent data foundation for subsequent preprocessing and model prediction, thus guaranteeing the reliability of the entire optimization control chain from the source.

[0065] Step S1.1 involves deploying sensor arrays, including gas flow meters, steam flow meters, electromagnetic flow meters, temperature thermocouples, oxygen content analyzers, and pressure transmitters, to simultaneously collect nine types of parameters at a 10-second sampling period, ensuring comprehensive coverage and timeliness of the data source. Step S1.2 removes abnormal data records caused by sensor interruptions, communication packet loss, or electromagnetic interference, filtering out invalid data from subsequent processes to avoid interference from abnormal samples to model training. Step S1.3 performs timestamp alignment on parameters with different sampling rates, integrating and aligning all parameters under the same time reference system to eliminate the impact of time sequence misalignment on multi-source data fusion analysis. Step S1.4 standardizes field names, unifies units of measurement, and standardizes data formats for each parameter, providing structurally unified and semantically clear data input for the preprocessing steps.

[0066] Step S2: Perform data preprocessing on the boiler operating parameter set to obtain a standardized boiler operating characteristic dataset.

[0067] Furthermore, step S2 specifically includes the following steps: Step S2.1: Fill the missing values ​​in the boiler operating parameter set using interpolation to obtain the boiler operating parameter set after missing value filling.

[0068] Preferably, the core of this step is to fill in the missing data in the boiler operating parameter set caused by instantaneous sensor failure, communication interruption, or PLC sampling gap. The missing value is calculated by using the trend of adjacent valid data through linear interpolation, which ensures the continuity of time sequence while avoiding the introduction of systematic deviation, and provides a complete time sequence of data for subsequent outlier identification.

[0069] For handling single-point missing values, if a parameter is missing within a single sampling time window, linear interpolation is performed using the values ​​of the two valid sampling points before and after that parameter. Assuming the values ​​at times t0 and t1 are x0 and x1 respectively, the missing value x at time t between t0 and t1 is... t It is obtained by calculation using the following formula: Each parameter performs interpolation operations independently.

[0070] In the case of continuous multi-point missing data, if a parameter is missing for more than 3 sampling windows (i.e., no valid data for more than 30 seconds), it is determined to be a continuous sensor failure. No interpolation is performed, and all parameter records corresponding to that period are marked as invalid and removed to avoid data distortion caused by long-term interpolation.

[0071] Among them, for interpolation boundary constraints of different parameters, the fuel flow interpolation result is not lower than 0 Nm³ / h and not higher than 4000 Nm³ / h, the coil temperature does not exceed 1200℃, and the steam pressure does not exceed 2.5 MPa. Interpolation results that exceed the above physical ranges are forcibly corrected to boundary values ​​and marked as data with low confidence, and their weights are reduced in subsequent model training.

[0072] For example, if the gas flow meter loses fuel flow data at time T0+10 due to a momentary communication interruption, the effective value at time T0 is 2850 Nm³ / h and the effective value at time T0+20 is 2870 Nm³ / h. The fuel flow at time T0+10 is calculated to be 2850 + (2870 - 2850) × (10) / (20) = 2860 Nm³ / h by linear interpolation, and then filled into the missing position.

[0073] Step S2.2: Identify outliers in the boiler operating parameter set after the missing values ​​are filled, and process outliers that exceed the reasonable range to obtain the boiler operating parameter set after outlier processing.

[0074] Preferably, the core of this step is to automatically detect and process outlier samples in the data after missing value imputation. The 3σ criterion is used to identify data points that deviate from the statistical distribution of normal operating conditions and replace them with the mean or median of adjacent time windows to avoid the adverse effects of extreme outliers on subsequent standardization and model training.

[0075] For the outlier identification method, the mean μ and standard deviation σ within the current sliding time window are calculated for each operating parameter. Data points whose parameter values ​​exceed the interval [μ-3σ, μ+3σ] are identified as statistical outliers. At the same time, physical boundary constraints are used: parameter values ​​with fuel flow rate below 100 Nm³ / h or above 4000 Nm³ / h, coil temperature below 100℃ or above 1200℃, and steam pressure below 0.1 MPa or above 2.5 MPa are also identified as outliers.

[0076] Among them, for the outlier handling strategy, a single outlier is replaced by the median of the valid values ​​within the current sliding window of the parameter. The median is not sensitive to outliers, and the replacement will not cause any deviation or disturbance to the normal data distribution. If more than two consecutive outliers appear, a sensor fault warning is triggered and the sensor data is temporarily removed from the input features, and will be restored after confirmation by maintenance personnel.

[0077] For the judgment of special operating conditions, the 3σ criterion is not applied to the parameter values ​​during the boiler start-up stage (the process of fuel flow rising from 0 to a stable value) and the shutdown stage. The abnormal value detection is automatically skipped by the indicator that the fuel flow is lower than 100 Nm³ / h, so as to avoid misjudging the normal transition data during the start-up and shutdown stages as abnormal.

[0078] For example, in a certain record, the coil temperature suddenly dropped to 450℃ (under normal combustion conditions, the coil temperature is usually in the range of 850-1050℃). Compared with the statistical distribution of the data in the five windows before and after, the Z score of this point is -4.8, which is lower than -3, so it is judged as an outlier and replaced with the median of the effective temperature value in the sliding window, which is 985℃.

[0079] Step S2.3: Perform data standardization processing on the boiler operating parameter set after outlier processing to obtain a standardized boiler operating feature dataset.

[0080] Preferably, the core of this step is to unify the numerical scale of operating parameters with different dimensions. By using the Min-Max normalization method, parameters with different dimensions and numerical ranges, such as fuel flow rate (unit Nm³ / h, numerical range 0-4000), coil temperature (unit °C, numerical range 0-1200), and steam pressure (unit MPa, numerical range 0-2.5), are mapped to a unified numerical range. This eliminates the imbalance of feature weights in the random forest model caused by differences in dimensions and ensures that multiple decision trees treat features of each dimension equally when splitting.

[0081] Among them, the Min-Max normalization method is used instead of Z-score standardization because the distribution of boiler operating parameters does not have strict normality. Min-Max normalization linearly maps each parameter to the [0,1] interval, preserving the distribution shape of the original data.

[0082] The normalization boundaries for each parameter are set as follows: fuel flow rate is 0 Nm³ / h and 4000 Nm³ / h; steam production is 0 kg / h and 50000 kg / h; condensate flow rate is 0 kg / h and 2000 kg / h; feedwater flow rate is 0 kg / h and 50000 kg / h; coil temperature is 0℃ and 1200℃; flue gas temperature is 50℃ and 400℃; flue gas oxygen content is 0% and 21%; steam pressure is 0 MPa and 2.5 MPa; and feedwater pressure is 0 MPa and 5.0 MPa.

[0083] For the dynamic update of normalization parameters, the preset min and max boundaries are updated every 30 days based on the statistical extreme values ​​of newly accumulated operating data to adapt to parameter shifts caused by equipment wear and environmental changes. However, the magnitude of each update is limited to within 10% of the original boundary to avoid frequent changes in the normalization benchmark that could lead to instability in the model input distribution.

[0084] For example, in a certain record, the fuel flow rate is 2850 Nm³ / h, and its normalized calculation is 2850 / (4000-0)=0.7125; the coil temperature is 981℃, and its normalized value is 981 / (1200-0)=0.8175; the steam pressure is 1.25 MPa, and its normalized value is 1.25 / (2.5-0)=0.5000. After normalization, the above three parameters are all within the range of [0,1], thus eliminating the influence of dimensional differences on model training.

[0085] Beneficial effects of steps S2.1 to S2.3: This series of steps eliminates incomplete data caused by sensor transients or communication interruptions by filling in missing values ​​through interpolation, removes outliers exceeding reasonable physical ranges by identifying and processing outliers, and eliminates the imbalanced impact of differences in the dimensions and numerical ranges of different parameters on model training through data standardization, transforming the boiler operating parameter set into a standardized boiler operating feature dataset. Traditional control schemes directly look up and retrieve static parameters based on raw measurements, where noise, missing values, and dimensional differences in the raw data are amplified at each stage in subsequent calculations, ultimately affecting control accuracy. This series of steps, through a systematic data preprocessing workflow, eliminates these interfering factors before the data enters the model, enabling the random forest regression model to extract true patterns from clean, uniformly scaled feature data, ensuring prediction accuracy while improving the convergence speed of model training.

[0086] Specifically, step S2.1 fills in missing data caused by sensor momentary failures, communication interruptions, or PLC sampling gaps using linear interpolation or spline interpolation, estimating missing values ​​based on the changing trends of adjacent valid data points, ensuring temporal continuity while avoiding the introduction of systematic biases; step S2.2 automatically identifies outliers in parameters such as fuel flow rate, coil temperature, and flue gas oxygen content that exceed reasonable physical ranges or statistical distributions using the 3σ criterion or box plot method, replacing them with the mean or median of adjacent time windows to avoid interference from outliers in the model's learning of normal operating conditions; step S2.3 uses Min-Max normalization or Z-score standardization to map parameters with different dimensions, such as fuel flow rate (0-4000 Nm³ / h), coil temperature (0-1200℃), and steam pressure (0-2.5 MPa), to a unified numerical range, eliminating feature weight imbalances caused by dimensional differences, and ensuring that multiple decision trees treat each dimension of features equally during splitting.

[0087] Step S3: Input the standardized boiler operating characteristic dataset into the random forest regression model and output the predicted value of thermal energy conversion efficiency under the current operating conditions.

[0088] Furthermore, step S3 specifically includes the following steps: Step S3.1: Divide the standardized boiler operation feature dataset into a training dataset and a test dataset.

[0089] Preferably, the core of this step is to reasonably divide the standardized boiler operating feature dataset between model training and generalization evaluation. By dividing the dataset into training and test sets in an 8:2 ratio, the training samples are ensured to cover the typical operating conditions of the boiler in each load range, while independent test samples are reserved to evaluate the model's predictive ability for unseen operating conditions, thus avoiding the problem of model overfitting and evaluation distortion.

[0090] The reason for using an 8:2 ratio instead of the conventional 7:3 for the division ratio is that boiler operation data has strong temporal correlation. If the proportion of the test set is too low, it will not be enough to evaluate the generalization ability of the model. If it is too high, it will compress the training sample size and reduce the model's learning effect. The 8:2 ratio strikes a balance between the two.

[0091] In particular, the special handling of time series data partitioning is to avoid the problem of time leakage (i.e., unreasonable prediction of past working conditions by future data), the partitioning is carried out by time order rather than random sampling: the data is sorted by time sequence, the data in the first 80% of the time range is used as the training set, and the data in the last 20% of the time range is used as the test set, to ensure that the training set and the test set do not overlap on the time axis, and the test set is always future data that the model has never seen.

[0092] For data selection criteria, only data records of boilers operating under stable conditions (steam pressure fluctuation within ±0.05MPa and load change within ±2%) are selected as valid samples, while data of transitional conditions such as start-up, shutdown, and sudden load changes are excluded to ensure the quality of training data.

[0093] For example, a certain month's operational data contains 43,200 valid records. After being sorted in chronological order, the first 34,560 records (corresponding to approximately 24 days) are used as the training dataset, and the last 8,640 records (corresponding to approximately 6 days) are used as the test dataset. The training set and the test set are consecutive in time but do not overlap.

[0094] Step S3.2: Train the random forest regression model using the training dataset, and construct multiple decision trees by minimizing the Gini coefficient to obtain the trained random forest regression model.

[0095] Preferably, the core of this step is to learn the mapping relationship from standardized boiler operating characteristic data to predicted thermal energy conversion efficiency values. Multiple subsets are extracted from the training dataset with replacement using the Bootstrap sampling method. For each subset, a CART decision tree is recursively split using the Gini coefficient minimization criterion. The risk of overfitting of a single tree is reduced by ensemble of multiple trees, and finally a random forest regression model that can accurately predict thermal energy conversion efficiency under any operating condition is obtained.

[0096] For the hyperparameter configuration of the random forest, the number of decision trees is set to B=200, the feature subset size is set to m=(√9)=3 (that is, each tree randomly selects three from nine input features as candidate features when splitting), the maximum depth of a single decision tree is limited to 8 layers, and the minimum number of samples in a leaf node is 5. The above parameters are determined based on cross-validation tuning to achieve a balance between prediction accuracy and computational efficiency.

[0097] For the splitting criterion of minimizing the Gini coefficient, in each internal node of each decision tree, all candidate features in the feature subset and all possible splitting thresholds are traversed. The feature and threshold that maximize the decrease in the mean square error (MSE) of the samples in the child node are selected as the decision for this split. The above process is repeated for the split child nodes until the maximum depth is reached or the number of leaf node samples is lower than the minimum threshold.

[0098] In the Bootstrap bootstrap sampling process, N samples are randomly selected with replacement from the original training dataset (N samples) to form a training subset of a decision tree. Due to the sampling with replacement, each subset contains approximately 63.2% original samples and 36.8% duplicate samples. The approximately 36.8% of samples that are not selected are used as out-of-bag (OOB) data for unbiased error estimation during model training.

[0099] For example, during a certain training process, the Bootstrap subset of the k-th decision tree contains 34,560 training samples (with replacement). Three candidate features are randomly selected at the root node: normalized fuel flow rate, normalized coil temperature, and normalized flue gas oxygen content. After traversal calculation, it is found that the MSE of the child node decreases the most when the normalized coil temperature is 0.75 as the splitting threshold. Therefore, the root node is split based on this feature and the threshold.

[0100] Step S3.3: Use the test dataset to evaluate the performance of the trained random forest regression model to obtain the performance-evaluated random forest regression model.

[0101] Preferably, the core of this step is to quantitatively evaluate the predictive performance of the trained random forest regression model on unseen data. By inputting the test dataset into the model, the deviation index between the predicted efficiency value and the actual efficiency value is calculated. Based on the evaluation results, it is decided whether to accept the current model for online prediction or return to step S3.2 for hyperparameter tuning, so as to ensure that the model has sufficient prediction accuracy and generalization ability before actual deployment.

[0102] For performance evaluation metrics, two metrics, Root Mean Square Error (RMSE) and Coefficient of Determination (R²), are used for comprehensive evaluation: RMSE reflects the average deviation between predicted and true values, while R² reflects the model's ability to explain the variance of the data. The combination of the two metrics can comprehensively evaluate the model's predictive accuracy and goodness of fit.

[0103] Among them, the model acceptance criteria are set as follows: RMSE not exceeding 1.5% and R² not less than 0.90 as the qualified conditions for model deployment. If the evaluation results do not meet the above criteria, the process returns to step S3.2 to adjust the hyperparameters of the random forest. The adjustment strategies include: increasing the number of decision trees from 200 to 300 to reduce model variance, increasing the maximum depth from 8 layers to 10 layers to enhance model fitting ability, and increasing the feature subset size from 3 to 4 to increase the number of split candidates per tree.

[0104] For the auxiliary reference of OOB evaluation, the out-of-bag error OOB-RMSE is calculated based on the out-of-bag data of each decision tree during training. If the difference between OOB-RMSE and test set RMSE exceeds 0.5%, it indicates that the model has a tendency to overfit (test set RMSE is much higher than OOB-RMSE) or underfit (test set RMSE is much lower than OOB-RMSE), and the model complexity needs to be further adjusted.

[0105] For example, after a training iteration, 8640 test set data points were input into the random forest model. The RMSE between the predicted efficiency value and the actual efficiency value was 1.2%, and the R² was 0.93. Both metrics were better than the acceptance criteria, and the model could be directly deployed for online efficiency prediction.

[0106] Step S3.4: Input the standardized boiler operation characteristic dataset collected in real time into the random forest regression model after performance evaluation, and obtain the predicted value of thermal energy conversion efficiency under the current operating condition by making parallel predictions through multiple decision trees and taking the average value.

[0107] Preferably, the core of this step is to quickly predict the thermal energy conversion efficiency under real-time operating conditions. This is achieved by real-time acquisition and standardization of nine-dimensional feature vectors, which are then input into 200 decision trees. Each decision tree independently outputs its predicted efficiency value, and the arithmetic mean is taken. The variance reduction principle of ensemble learning is used to make the prediction results have good noise resistance and stability.

[0108] In the prediction process of a single decision tree, the real-time feature vector starts from the root node, compares and judges the splitting features and thresholds of the current node, and recursively moves to the downstream node until it reaches the leaf node. The average value of the training sample efficiency value in the leaf node is used as the prediction output of the tree. The prediction time complexity of a single decision tree is O(tree depth) = O(8), and the computational cost is extremely low.

[0109] In the ensemble process of multi-tree predictions, 200 decision trees independently and in parallel complete their respective predictions before being aggregated. The final prediction efficiency value is calculated using the following formula: .

[0110] in Let be the output value of the b-th decision tree for the same feature input p. The average of the 200 decision trees reduces the sensitivity of a single tree to noisy data.

[0111] The reliability assessment of the prediction results also includes calculating the standard deviation of the prediction values ​​from 200 decision trees. A smaller standard deviation indicates higher consistency in the predictions of each tree regarding the current operating condition, and higher reliability of the prediction results; when If the confidence level exceeds 2.0%, it indicates that the current operating condition is a rare condition that is not fully covered in the training data. The confidence level of the prediction result is low, and the system marks the prediction value for operators' reference rather than for direct use in optimization decisions.

[0112] For example, when a standardized feature vector p = [fuel flow rate 0.7125, coil temperature 0.8175, steam pressure 0.5000, ...] is input into a random forest model at a certain real-time sampling time, the predicted values ​​of the 200 decision trees are distributed between 91.2% and 93.8%, with a mean of 92.6% and a standard deviation of 0.8%, indicating good prediction consistency. The predicted value of the thermal energy conversion efficiency under the current operating condition is determined to be 92.6%.

[0113] Beneficial effects of steps S3.1 to S3.4: This series of steps establishes an end-to-end mapping relationship from multi-source operating parameters to efficiency predictions by dividing the standardized feature dataset into training and testing sets, training multiple decision trees to construct a random forest regression model based on the principle of minimizing the Gini coefficient, evaluating the model's generalization performance using the test set, and inputting real-time collected standardized feature data into the model for parallel prediction by multiple decision trees, taking the average value, and outputting the predicted value of heat conversion efficiency. Traditional solutions rely on looking up tables of pre-calibrated static parameters from equipment suppliers, which not only fails to reflect the current actual operating conditions of the equipment but also cannot predict unseen load points. This series of steps, through a data-driven ensemble learning method, enables the model to autonomously learn the nonlinear relationship between various parameters and efficiency from historical operating data, outputting reliable efficiency predictions under any operating condition, and providing real-time and accurate objective function input for NSGA-Ⅲ multi-objective optimization.

[0114] In step S3.1, the standardized boiler operation feature dataset is divided into training and test datasets in a 7:3 or 8:2 ratio to ensure that the training samples cover the operating characteristics of each load range while retaining independent samples for generalization performance evaluation. In step S3.2, multiple subsets are extracted from the training dataset with replacement using the Bootstrap sampling method. For each subset, a CART decision tree is recursively constructed by minimizing the Gini coefficient, with a tree depth of 8 layers and a feature subset size of 3. The risk of overfitting of a single tree is reduced by integrating multiple trees. In step S3.3, the root mean square error (RMSE) and coefficient of determination (R²) are used to evaluate the prediction bias on the test set. When R² is lower than 0.85, hyperparameter tuning is performed, such as adjusting the number of decision trees to 200 or limiting the minimum number of samples in the leaf nodes, until the model reaches the preset accuracy threshold. In step S3.4, the real-time standardized feature data is input into multiple decision trees, and the output values ​​of each tree are calculated in parallel and then the arithmetic mean is taken. The variance reduction principle of ensemble learning is used to make the prediction results have good noise resistance.

[0115] Step S4: Input the predicted thermal energy conversion efficiency value into the NSGA-Ⅲ non-dominated sorting genetic algorithm to perform multi-objective Pareto front iterative search and obtain the multi-objective Pareto solution set.

[0116] Furthermore, step S4 specifically includes the following steps: Step S4.1: Construct a multi-objective optimization model with the objective functions of maximizing thermal energy conversion efficiency, minimizing pollutant emissions, and minimizing fuel consumption.

[0117] Preferably, the core of this step is to complete the multi-objective formal definition of boiler optimization control, with the three optimization objectives being to maximize the thermal energy conversion efficiency η, minimize the pollutant emission concentration in the flue gas ε (the oxygen content in the flue gas indirectly represents the degree of combustion), and minimize the natural gas consumption F. The constrained three-objective optimization problem is constructed with steam pressure fluctuation, steam production deviation, and the safe opening degree of the actuator as constraints.

[0118] For the first objective function, the heat energy conversion efficiency, the efficiency calculation formula is as follows: .

[0119] in Steam flow rate (kg / h) The specific enthalpy of steam is obtained from the steam pressure table (kJ / kg). The water supply flow rate is (kg / h). The feedwater specific enthalpy is obtained from a table based on the feedwater temperature and pressure (kJ / kg). This represents the natural gas flow rate (Nm³ / h). The lower heating value of natural gas is taken as 35530 kJ / Nm³, and the objective is to maximize this value.

[0120] For the second objective function, pollutant emissions, the oxygen content (O2) in flue gas is used as an indirect indicator of combustion completeness and pollutant emission levels. Excessively high oxygen content (e.g., exceeding 8%) indicates an excessively high excess air coefficient, leading to increased heat loss and NOx formation. Conversely, excessively low oxygen content (e.g., below 2%) indicates incomplete combustion, resulting in increased CO formation. The objective function is based on a set target oxygen content value in the flue gas. Compared with the actual oxygen content of flue gas The absolute deviation is minimized in the direction, i.e. The ideal target oxygen content in flue gas is 3%-5%.

[0121] Among them, for the third objective function, fuel consumption, the natural gas flow rate is used as the criterion. As a direct indicator, under the premise of meeting the requirements of steam production and steam pressure, the less fuel consumed, the more efficient the combustion process and the lower the operating cost. The goal is to minimize this.

[0122] Among the constraints, the steam pressure P must satisfy... The steam production volume Q needs to meet The opening degree U of the three actuators must meet their respective safety boundaries: gas valve opening degree Water supply valve opening damper opening .

[0123] For example, under a specific operating condition, the steam pressure is set at 1.25 MPa and the steam production rate is set at 22,000 kg / h. The constraints are that the steam pressure is within the range of 1.15-1.35 MPa, the steam production rate is within the range of 20,900-23,100 kg / h, and the opening degree of the three actuators is within the range of 20%-90% for the gas valve, 30%-95% for the water supply valve, and 25%-85% for the damper.

[0124] Step S4.2: Input the predicted value of thermal energy conversion efficiency into the multi-objective optimization model, and use the NSGA-Ⅲ non-dominated sorting genetic algorithm to generate an initial population for the multi-objective optimization model to obtain the initial population.

[0125] Preferably, the core of this step is to complete the initial search space coverage of the multi-objective optimization process. Using the thermal energy conversion efficiency value predicted by the current random forest model as a reference, the initial population is generated by Latin hypercube sampling in the three-dimensional decision space of gas valve opening, water supply valve opening, and damper opening. This ensures that the initial solution is evenly distributed in the feasible region and avoids local concentration that leads to low search efficiency.

[0126] Among them, for the coding of decision variables, the gas valve opening degree Water supply valve opening Air damper opening All three decision variables are encoded with real numbers, with values ​​ranging from [20%, 90%] to [30%, 95%] and [25%, 85%], respectively. Each individual variable is represented as a three-dimensional real number vector. .

[0127] For the initial population generation method, Latin hypercube sampling is used instead of simple random sampling. The three-dimensional decision space is divided into N equal parts in each dimension (N is the population size). A value is randomly selected in each interval to ensure that all intervals in each dimension are covered by a solution. The total number of individuals in the initial population is set to 100.

[0128] In the process of verifying the legality of constraints for individuals in the initial population, each generated individual must satisfy all constraints, including steam pressure constraints, steam production constraints, and actuator opening safety boundary constraints. For individuals that do not meet the constraints, a new individual is generated nearby to replace them through random correction until all individuals in the population satisfy the constraints.

[0129] For example, the i-th individual in the initial population is generated by Latin hypercube sampling: The gas valve and water supply valve openings corresponding to this individual are both within their safety boundaries, and the damper opening is also within the legal range. After the constraint verification, it was included in the initial population.

[0130] Step S4.3: Perform non-dominated sorting and reference point generation on the initialized population to obtain the Pareto front of the current population.

[0131] Preferably, the core of this step is to complete the multi-objective performance evaluation and hierarchical ranking of all individuals in the current population. The population is divided into multiple Pareto front levels by using a fast non-dominated sorting algorithm. At the same time, uniformly distributed reference points are generated on the hyperplane to provide a guiding benchmark for maintaining the diversity of solution sets in subsequent environmental selection.

[0132] In the non-dominated sorting process, the objective function values ​​of any two individuals i and j in the population are compared in the three-objective space: if individual i is not inferior to individual j in all three objectives and is strictly superior to individual j in at least one objective, then individual i dominates individual j. Fast non-dominated sorting marks all individuals not dominated by any other individual as the first frontier F1. From the remaining individuals after removing F1, the non-dominated individuals are then identified and marked as the second frontier F2, and so on until all individuals are sorted.

[0133] In the reference point generation method, reference points are generated by uniformly sampling along each target direction in the three-dimensional target space composed of the three target dimensions. The number of segments in each target direction is set to p=4. Then the total number of reference points is C(3+p-1,p)=C(6,4)=15, and the sum of the coordinates of each reference point in the target space is 1.

[0134] In obtaining the Pareto front, after the non-dominated sorting is completed, the individuals in the first front F1 are the set of optimal solutions that no other individual in the current population can dominate. Each solution in this set constitutes the Pareto front of the current population in the three objective space. Each solution on the front represents a non-dominated solution with different preference orientations among efficiency, emissions and fuel consumption.

[0135] For example, in the initial population, individual A has three target values: [η=92.6%, O2 bias=1.2%], =2850], Individual B is [η=93.1%, O2 bias=0.8%, =2820], individual C is [η=91.8%, O2 bias=0.5%, =2900], after comparison, individual B is better than individual C in all objectives, and individual B dominates individual C; individual A and individual B do not dominate each other (individual B's efficiency and O2 deviation are better than individual A, but its fuel consumption index is not better than individual A in all aspects), so individual A and individual B belong to the first frontier F1.

[0136] Step S4.4: The offspring population is generated through selection, crossover, and mutation genetic operations. The parent population and the offspring population are then merged for environmental selection to obtain the environmentally selected population.

[0137] Preferably, the core of this step is to generate a new generation of offspring individuals from the current parent population through genetic operations, and then merge the parent and offspring and select the next generation of population through an environmental selection mechanism based on reference points. This process preserves excellent non-dominant individuals while maintaining broad coverage of the population in the three-objective space, thus avoiding excessive concentration of the population in a few local areas.

[0138] For the selection operation, a tournament selection method is used. Two individuals are randomly selected from the current population as a group for tournament comparison. Individuals with higher Pareto front levels are given priority (F1 is better than F2). If the levels are the same, individuals with smaller crowding distances after association with the reference point are selected. This process is repeated until 100 parent individuals are selected for subsequent genetic operations.

[0139] For the crossover operation, simulated binary crossover (SBX) is used to crossover and recombine selected parent pairs of individuals, with a crossover probability of 0.9 and a distribution index of 20. The three-dimensional real-valued encoded vectors of the two parent individuals P1 and P2 are crossovered by SBX to produce two offspring individuals C1 and C2. The decision variable values ​​of the offspring individuals are distributed in the vicinity of the parents, which can both retain the good characteristics of the parents and produce moderate variation.

[0140] In the mutation operation, multinomial mutation is used to fine-tune the offspring individuals generated by the crossover, with the mutation probability set to 1 / 3 (independent mutation probability for each decision variable) and the distribution index set to 20. The mutation operation causes small random perturbations in the offspring individuals around the characteristics of the parents, enhancing population diversity and avoiding premature convergence to local optima.

[0141] For example, parent individuals P1=[58.3%, 62.7%, 45.2%] and P2=[65.1%, 58.4%, 50.8%] undergo SBX crossover (crossover probability meets the condition) to produce offspring C1=[60.7%, 60.9%, 47.4%] and C2=[62.7%, 60.2%, 48.6%]. Offspring C1 then undergoes polynomial mutation on a third decision variable with a 1 / 3 probability to obtain [60.7%, 60.9%, 49.1%]. This offspring is then merged with the parent and enters environmental selection.

[0142] Step S4.5: Repeat steps S4.3 to S4.4 to iteratively perform non-dominated sorting, reference point generation, and environment selection until the termination condition is met, and obtain the multi-objective Pareto solution set.

[0143] Preferably, the core of this step is to complete the iterative loop of multi-objective optimization search. By repeatedly executing non-dominated sorting, reference point generation, genetic operations and environmental selection, the population gradually converges to the true Pareto front during the iteration process, while maintaining the solution set with broad coverage in the three-objective space, until the preset termination condition is met and the final multi-objective Pareto solution set is output.

[0144] Among them, for the iteration termination conditions, two conditions are set to terminate if either one is met: (1) the maximum number of generations reaches 50; (2) the improvement of the hypervolume index HV of the first Pareto front is less than 1% for 10 consecutive generations. Condition (1) ensures that the search will not continue indefinitely, and condition (2) terminates in time when the convergence rate saturates early to avoid invalid calculations.

[0145] Among them, regarding the population evolution trend between each iteration round, in the early iteration (generations 1-15), the individuals in the population are relatively dispersed in the target space, and the Pareto front gradually moves towards the direction of higher efficiency, lower emissions, and less fuel consumption; in the middle iteration (generations 16-35), the individuals in the population gradually gather in the region closer to the true Pareto front; in the late iteration (generations 36-50), the individuals in the population make minor adjustments near the Pareto front, and the solution set diversity and convergence tend to stabilize.

[0146] In the process of constraint handling during iteration, each new individual generated by crossover mutation needs to be re-verified to see if it meets the constraints set in step S4.1 (steam pressure fluctuation ±0.1MPa, steam production deviation ±5%, actuator opening safety boundary). Individuals that do not meet the constraints are subject to a target penalty function to reduce their fitness, so that they are naturally eliminated in the environmental selection.

[0147] For example, after 50 generations of iteration, the first frontier F1 in the final population contains 43 non-dominated Pareto optimal solutions. These solutions constitute a multi-objective Pareto solution set, which is distributed on an approximate contour line in the three-objective space. Each solution shows different trade-offs between thermal energy conversion efficiency (91.5%-94.2%), flue gas oxygen content deviation (0.3%-1.8%), and natural gas flow rate (2780-2950 Nm³ / h).

[0148] Beneficial effects of steps S4.1 to S4.5: This series of steps constructs a multi-objective optimization model with the optimization objectives of maximizing thermal energy conversion efficiency, minimizing pollutant emissions, and minimizing fuel consumption. Random forest predictions are used as model input. The NSGA-III non-dominated sorting genetic algorithm performs initial population generation, non-dominated sorting and reference point generation, genetic operations to produce offspring, environmental selection to preserve high-quality solutions, and iterative search until the termination condition is met, ultimately obtaining a uniformly distributed and convergent multi-objective Pareto solution set. Traditional static lookup table schemes only preset parameters with single pressure tracking as the objective, completely failing to consider the three conflicting optimization objectives of efficiency, emissions, and fuel consumption. This leads to improved boiler efficiency but worsened pollutant emissions or runaway fuel costs. This series of steps uses the NSGA-III algorithm to perform a global Pareto search in the three-dimensional objective space, utilizing a reference point mechanism to maintain the diversity distribution of the solution set in the three objective directions, ensuring that each solution on the Pareto front represents an optimal solution with different trade-offs among thermal efficiency, emissions, and fuel consumption.

[0149] Step S4.1 uses three objective functions: maximizing thermal energy conversion efficiency η, minimizing CO and NOx emission concentrations in flue gas ε, and minimizing natural gas consumption F. Constraints such as steam pressure fluctuation ±0.1 MPa, steam production deviation ±5%, and actuator opening upper limit not lower than lower limit not higher than safety boundary are set to clarify the boundaries and preference directions of the multi-objective optimization. Step S4.2 uses the current predicted efficiency value as a reference point and generates initial seeds using Latin hypercube sampling in a three-dimensional decision space with gas valve opening 20%-90%, water valve opening 30%-95%, and damper opening 25%-85%. The initial solution covers the entire feasible region. Step S4.3 performs a fast non-dominated sorting of the current population to divide the frontier hierarchy, and generates uniformly distributed reference points on the hyperplane to guide the search direction, taking into account the uniformity of coverage of each target dimension. Step S4.4 uses the tournament selection operator to select excellent parents, generates offspring through simulated binary crossover (SBX) and polynomial mutation, and then merges them with the parents to perform environment selection based on reference points, retaining elite solutions. Step S4.5 repeats the iteration until the maximum number of generations (50 generations) is reached or the Pareto front improvement is less than 1% for 10 consecutive generations, and then terminates to ensure the convergence and integrity of the solution set.

[0150] Step S5: Extract the optimal solution of the multi-objective Pareto solution set by sorting the approximate ideal solution using TOPSIS, and obtain the optimal combination of control parameters based on gas, water supply, and dampers.

[0151] Furthermore, step S5 specifically includes the following steps: Step S5.1: Calculate the positive ideal solution and negative ideal solution for each solution in the multi-objective Pareto solution set.

[0152] Preferably, the core of this step is to determine the benchmark reference points for the positive and negative directions of each objective in the multi-objective Pareto solution set. By taking the optimal extreme value of each objective in the Pareto solution set to form the positive ideal solution vector and the worst extreme value of each objective to form the negative ideal solution vector, an objective quantitative evaluation benchmark is provided for the positive and negative directions, avoiding preference bias caused by subjective weighting and ensuring the fairness and repeatability of subsequent ranking.

[0153] For the calculation of the positive ideal solution Z+, the optimal values ​​of each objective are taken from the Pareto solution set: the maximum value of the thermal energy conversion efficiency. Minimum deviation of oxygen content in flue gas (The smaller the deviation, the better), minimum natural gas flow rate The positive ideal solution vector is , representing the ideal optimal solution that may not exist in reality.

[0154] For the calculation of the negative ideal solution Z-, the worst-case values ​​of each objective are taken from the Pareto solution set: the minimum value of the thermal energy conversion efficiency. The maximum deviation of oxygen content in flue gas (The larger the deviation, the worse) Maximum natural gas flow rate The negative ideal solution vector is , representing the least desirable virtual worst solution.

[0155] Regarding the consideration of target weights, if decision-makers have different preference weights for each target (such as enterprises paying more attention to fuel conservation), the target values ​​can be weighted before calculating the positive and negative ideal solutions. By default, the three target weights are equal, and the weight coefficient of each target value is 1 / 3.

[0156] For example, if the Pareto solution has a concentrated thermal energy conversion efficiency ranging from 91.5% to 94.2%, a flue gas oxygen content deviation ranging from 0.3% to 1.8%, and a natural gas flow rate ranging from 2780 to 2950 Nm³ / h, then the positive ideal solution Z+ = [94.2%, 0.3%, 2780] and the negative ideal solution Z- = [91.5%, 1.8%, 2950].

[0157] Step S5.2: Calculate the distance from each solution to its corresponding positive ideal solution and negative ideal solution, and obtain the distance value pair for each solution.

[0158] Preferably, the core of this step is to quantify the geometric distance between each Pareto solution and the positive and negative ideal solutions in the three objective spaces. The distance di+ between each solution and the positive ideal solution Z+ and the negative ideal solution Z- is calculated sequentially using the Euclidean distance formula. This unified multidimensional spatial metric comprehensively reflects the degree to which each solution deviates from the ideal state in each objective dimension.

[0159] Specifically, regarding the definition of the distance formula, the Euclidean distance from the i-th solution to the positive ideal solution Z+ and the Euclidean distance to the negative ideal solution Z- are calculated using the following formulas: .

[0160] .

[0161] in , , These represent the thermal energy conversion efficiency, flue gas oxygen content deviation, and natural gas flow target value for the i-th solution, respectively.

[0162] In the target normalization in distance calculation, since the three targets have different dimensions and numerical scales (efficiency is a percentage, deviation is an absolute percentage, and flow rate is Nm³ / h), the target values ​​need to be first normalized in the Pareto solution set to map the numerical range of the three targets to the [0,1] interval before calculating the Euclidean distance, so as to avoid the target with larger dimensions from dominating the distance calculation.

[0163] Regarding the meaning of distance values, the smaller di+ is, the closer the solution is to the positive ideal solution, that is, the better the overall performance of each objective. The larger di- is, the further the solution is from the negative ideal solution, that is, the worse the overall performance of each objective. Ideally, the best solution should have the smallest di+ and the largest di-, but in reality, there is usually a trade-off between the two.

[0164] For example, the objective value of solution A in the Pareto solution set is [93.8%, 0.5%, 2810], and after normalization, the values ​​are [0.85, 0.13, 0.32]. The normalized positive ideal solution is [1.0, 0.0, 0.0], and the normalized negative ideal solution is [0.0, 1.0, 1.0]. The distance from solution A to the positive ideal solution is dA+=√[(0.85-1.0)²+(0.13-0.0)²+(0.32-0.0)²]=√0.1418=0.3766; the distance from solution A to the negative ideal solution is dA-=√[(0.85-0.0)²+(0.13-1.0)²+(0.32-1.0)²]=√1.9418=1.3935.

[0165] Step S5.3: Calculate the relative proximity value between each solution and the ideal solution based on the distance value of each solution.

[0166] Preferably, the core of this step is to quantitatively evaluate the relative closeness of each solution to the ideal solution, and calculate the relative closeness value of each solution using the following formula: The value is between 0 and 1. The closer it is to 1, the closer the solution is to the positive ideal solution and the further away it is from the negative ideal solution. In other words, a single score is used to comprehensively reflect the overall performance of each solution in multiple objective dimensions.

[0167] Regarding the physical meaning of the relative proximity formula, when Ci=1, it means that di+=0, and the solution completely coincides with the positive ideal solution, reaching the optimal value on all objectives, but such an ideal solution does not exist in the actual Pareto solution set; when Ci=0, it means that di-=0, and the solution completely coincides with the negative ideal solution, reaching the worst value on all objectives.

[0168] In special cases, if there is only one solution in the Pareto solution set (which may only appear under extreme constraints), then the solution is both a positive ideal solution and a negative ideal solution. In this case, di+=di-=0, and the solution is directly determined as the optimal solution without calculating the relative proximity.

[0169] Regarding the reliability of the relative proximity ranking, since the evaluation benchmark (positive and negative ideal solutions) of the TOPSIS method is determined entirely based on the internal statistical characteristics of the current Pareto solution set and does not rely on external prior knowledge or subjective scoring, the relative proximity ranking results are objective and repeatable.

[0170] For example, if the solution A has dA+=0.3766 and dA-=1.3935, then the relative closeness of solution A is CA=1.3935 / (0.3766+1.3935)=0.7872, which means that the relative closeness of solution A to the positive ideal solution is 78.72%.

[0171] Step S5.4: Sort each solution in the multi-objective Pareto solution set according to all relative proximity values ​​from largest to smallest, and extract the solution with the highest ranking as the optimal solution.

[0172] Preferably, the core of this step is to sort and extract the best solution from the Pareto solution set. The relative proximity values ​​of the solutions are arranged in descending order, and the solution with the highest relative proximity is the best solution in the multi-objective comprehensive evaluation. This quantitative sorting replaces manual experience judgment, making the optimal solution selection process objective and reproducible.

[0173] In the sorting process, the relative proximity values ​​of all 43 solutions in the Pareto solution set are arranged from largest to smallest to generate a sorting list. At the same time, the values ​​of the three-dimensional decision variables (gas valve opening, water supply valve opening, damper opening) corresponding to each solution are recorded to facilitate the subsequent parsing of the optimal solution into specific control parameter combinations.

[0174] In verifying the stability of the ranking results, the difference in relative proximity values ​​between the top three solutions is calculated. If the difference in relative proximity between the first and second solutions is less than 2%, it means that the solution ranked first does not have a statistically significant advantage. In this case, the average control parameter of the top three solutions is taken as the optimal solution to reduce the uncertainty of single-point selection.

[0175] Among them, for the rationality check after sorting, after extracting the solution ranked first, it is checked whether its specific values ​​on the three objectives are all better than the median of the corresponding objectives in the Pareto solution set. If an objective deviates significantly from the median (such as the efficiency value being lower than the median efficiency in the Pareto solution set), a manual review mechanism is triggered, and the operator judges whether the optimal solution is acceptable.

[0176] For example, after sorting the relative proximity values ​​of the 43 Pareto solutions in descending order, the top three solutions are CA=0.7872, CA=0.7745, and CA=0.7681, respectively. The difference between the first and second place solutions is 1.27%, which is less than the 2% threshold. Therefore, the average of the control parameters of the top three solutions is taken as the optimal solution.

[0177] Step S5.5: The optimal solution is analyzed into an optimal combination of control parameters based on gas, water supply, and dampers.

[0178] Preferably, the core of this step is to complete the final transformation from the multi-objective optimization result to the specific executable control parameters, mapping the three-dimensional decision variables of the optimal solution to the opening percentage of the gas regulating valve, the opening percentage of the water supply valve, and the opening percentage of the damper controller, respectively, forming a three-dimensional control parameter vector that can be directly issued to the execution equipment, thus completing the final connection from the multi-objective optimization decision to the physical execution command.

[0179] For decision variable restoration, the normalized values ​​of the decision variables in the optimal solution (within the [0,1] interval) need to be inversely normalized according to the actual range of each actuator, calculated using the following formula: .

[0180] .

[0181] .

[0182] All final results are rounded to integer percentages.

[0183] In terms of the parameter distribution format, the actual opening value is uniformly represented as an integer percentage. For example, a gas valve opening of 57% is represented as the integer 57. It is written into the corresponding area of ​​the register address according to the Modbus RTU communication protocol format, and each actuator reads the corresponding control parameter according to its own register address.

[0184] In the parameter rationality verification, before the release, the three parameters of gas valve opening, water supply valve opening, and damper opening are compared with their respective safety boundaries again to confirm that none of them exceed the limits; at the same time, the estimated thermal efficiency corresponding to the parameter combination is calculated to see if it is not lower than the efficiency before optimization. If it is lower, an emergency protection is triggered and the release is not executed.

[0185] For example, the normalized decision variable value of the optimal solution is After inverse normalization: gas valve opening = 0.53×(90-20)+20=57.1%, rounded to 57%; water valve opening = 0.48×(95-30)+30=61.2%, rounded to 61%; damper opening = 0.40×(85-25)+25=49%. Therefore, the optimal control parameter combination is [gas valve opening 57%, water valve opening 61%, damper opening 49%].

[0186] Beneficial effects of steps S5.1 to S5.5: This series of steps calculates the positive and negative ideal solutions for each objective on the Pareto solution set, quantifies the deviation of each solution from the positive and negative ideal solutions using Euclidean distance, calculates the relative proximity value based on the distance values, sorts the solutions in descending order of relative proximity value, and extracts the solution with the highest relative proximity value as the optimal solution. This optimal solution is then analyzed into specific values ​​for three control parameters: gas valve opening, water valve opening, and damper opening, completing the final decision output from the Pareto solution set to executable control parameters. In traditional schemes, operators select the operating parameters closest to the current load from a preset parameter table based on experience. This selection process lacks quantitative evaluation standards, and the results of different operators' selections may vary significantly, failing to guarantee that the selected parameters achieve an optimal balance between efficiency, emissions, and fuel consumption. This series of steps uses the TOPSIS method to perform multi-objective comprehensive evaluation on the Pareto front using positive and negative ideal solutions as a benchmark. It extracts the control parameter combination with the best overall performance based on relative proximity, ensuring that the final decision fully reflects the trade-offs between the importance of each optimization objective.

[0187] In step S5.1, the optimal extreme values ​​of each objective in the Pareto solution set are used to form the positive ideal solution vector Z+ and the nearest extreme value is used to form the negative ideal solution vector Z-, providing an objective benchmark reference for the positive and negative directions and avoiding preference bias caused by subjective weighting. In step S5.2, the distance di+ to the positive ideal solution and the distance di- to the negative ideal solution of each solution are calculated sequentially using the Euclidean distance formula, and the deviation of each solution from the ideal benchmark point in the three-dimensional target space is quantified by geometric distance. In step S5.3, the relative proximity value between each solution and the positive ideal solution is calculated. The closer the value is to 1, the closer the solution is to the ideal state. In step S5.4, the relative proximity values ​​of all solutions are sorted from largest to smallest, and the solution ranked first is extracted as the optimal solution. This solution achieves a globally optimal comprehensive balance in the three target dimensions of efficiency, emissions, and fuel consumption. In step S5.5, the three-dimensional decision variables of the optimal solution are mapped to the percentage opening of the gas regulating valve, the percentage opening of the water pump, and the percentage opening of the damper controller, forming a specific combination of control parameters that can be directly issued and executed.

[0188] Step S6: The optimal control parameter combination is sent to the corresponding execution device to complete the boiler thermal energy conversion efficiency optimization control.

[0189] Furthermore, step S6 specifically includes the following steps: Step S6.1: Encapsulate the optimal control parameter combination into instructions according to the corresponding execution device communication protocol to obtain the optimal control parameter combination after instruction encapsulation.

[0190] Preferably, the core of this step is to convert the multi-objective optimization decision results into execution instructions at the control system level. This is achieved by encapsulating the three control parameters—gas valve opening, water supply valve opening, and damper opening—into standard register write instructions according to the Modbus RTU communication protocol format of the S7-1500 PLC, ensuring the compatibility and resolvability of the downlink control instructions with the field execution equipment.

[0191] Regarding the choice of communication protocol, all field execution devices are controlled through the PLC's analog output module (8-channel 4-20mA current signal output). The PLC and the host computer use the Modbus RTU protocol for data exchange, with a baud rate of 19200bps, 8 data bits, 1 stop bit, and no parity.

[0192] For the instruction format definition, the encapsulated instruction for each control parameter includes five fields: device address, function code, register start address, data value, and CRC check code. The gas valve opening corresponds to Modbus address 40001, the water supply valve opening corresponds to 40002, and the damper opening corresponds to 40003. The data value is represented by a 16-bit unsigned integer, with a value range of 0-10000 (corresponding to an opening of 0.00%-100.00%).

[0193] Specifically, for the overall structure of instruction encapsulation, the encapsulated instructions of the three parameters are combined into an instruction frame sequence in a fixed order of gas valve, water supply valve, and damper. Each instruction frame is separated by an inter-frame interval of 3ms to ensure that the PLC receives and executes them in sequence, avoiding parsing conflicts caused by multiple instructions arriving at the same time.

[0194] For example, the Modbus RTU instructions corresponding to the optimal parameter combination [gas valve opening 57%, water valve opening 61%, damper opening 49%] are encapsulated as follows: gas valve instruction [device address 0x01, function code 0x06 (write single register), register address 0x00, data value 5700, CRC checksum xxxx]; water valve instruction [0x02, 0x06, 0x00, 6100, CRC]; damper instruction [0x03, 0x06, 0x00, 4900, CRC]. These three instructions are encapsulated into an instruction frame sequence at 3ms intervals.

[0195] Step S6.2: The optimal combination of control parameters after encapsulating the instruction is sent to the corresponding execution device.

[0196] Preferably, the core of this step is to accurately send the optimal control parameters from the control system to the physical execution equipment. The encapsulated Modbus RTU instructions are sent to the gas regulating valve actuator, water pump frequency converter and damper controller in the form of a 4-20mA standard current signal through the analog output module of the PLC, so as to realize the accurate implementation of the optimized parameter combination at the physical equipment level.

[0197] For the delivery path, the host computer software sends the instruction frame sequence to the S7-1500 PLC via the Modbus RTU protocol through the RS485 communication bus. After parsing the instruction, the PLC writes the target opening value into the corresponding analog output channel register. The analog output module converts the digital quantity in the register into a 4-20mA current signal and outputs it to the signal input terminal of each execution device.

[0198] For current signal conversion, the 4-20mA current signal received by each actuator has a linear relationship with the opening percentage. The conversion relationship between current and opening is as follows: .

[0199] 4mA corresponds to 0% opening, 20mA corresponds to 100% opening, and intermediate values ​​are obtained through linear interpolation. For example, the current value corresponding to 57% opening is 4 + 57% × 16 = 13.12mA.

[0200] In terms of execution sequence control, the instructions of the three actuators are issued in a fixed order of gas valve → water supply valve → damper. After each instruction is issued, a 500ms wait is made to confirm that the corresponding actuator is in place before issuing the next instruction. This avoids the simultaneous operation of the three devices, which could cause sudden changes in the boiler operating conditions and lead to safety hazards.

[0201] For example, after the PLC receives the instruction from the host computer that the gas valve opening is 57% and the corresponding current value is 13.12mA, the analog output channel 1 outputs a 13.12mA current signal to the gas regulating valve actuator. The actuator automatically adjusts the valve to the 57% opening position according to the current value, and feeds back the position signal after the adjustment is completed.

[0202] Step S6.3: Receive the corresponding optimal control parameters from each execution device, and receive the parameter adjustment completion signal fed back by each execution device after execution.

[0203] Preferably, the core of this step is to confirm the execution of control commands by the executing equipment. The limit switch signals and opening feedback signals of each executing equipment are received through the digital input module of the PLC to confirm that the gas regulating valve, water pump and damper controller have all executed the commands accurately, providing a status basis for subsequent effect verification.

[0204] Specifically, for the confirmation of execution completion, the limit switches of each actuator trigger a signal to the digital input module of the PLC when the valve / actuator reaches the target opening degree. The PLC simultaneously collects the analog feedback signal of the actual opening degree of each actuator and compares it with the target opening degree (the deviation is considered to be within ±1%). Only when both meet the requirements is the execution considered complete.

[0205] For execution timeout handling, the timer starts from the issuance of the instruction. If no confirmation signal of arrival is received from any execution device within 30 seconds, the device is determined to have timed out. The system automatically suspends the issuance of subsequent instructions and triggers an execution fault alarm. At the same time, the parameters of the already adjusted devices remain unchanged, waiting for maintenance personnel to intervene.

[0206] In terms of log recording during the execution process, the system automatically records detailed information such as the instruction issuance time, target value, actual value delivered, delivery time, and execution time for each parameter, forming a complete execution log for subsequent performance analysis and troubleshooting.

[0207] For example, the gas regulating valve actuator sends a feedback signal 8 seconds after the command is issued. The actual opening degree collected by the PLC is 56.8%, which is 0.2% less than the ±1% threshold and is therefore determined to be executed. The water pump sends a feedback signal 12 seconds later, with an actual opening degree of 60.5% and a deviation of 0.5%, which is also within the threshold. The damper controller sends a feedback signal 10 seconds later. After all three signals are confirmed, the system confirms that the command has been executed.

[0208] Step S6.4: In response to the parameter adjustment completion signal fed back by all executing devices, collect the boiler's operating parameters after parameter adjustment and calculate the adjusted thermal energy conversion efficiency.

[0209] Preferably, the core of this step is to quantify and verify the optimization control effect. After all the executing devices have been in place, the system continuously collects boiler operation data after parameter adjustment at 30 sampling points (i.e., 5 minutes) with a 10-second sampling period. The average value after stabilization is taken to calculate the adjusted heat energy conversion efficiency. The efficiency changes before and after optimization are compared with the quantitative data to objectively evaluate the actual effect of this round of optimization.

[0210] For the stability assessment of data acquisition, timing begins after parameter adjustment. The system continuously monitors the fluctuation of steam pressure. When the steam pressure fluctuation is stable within ±0.05MPa for 30 consecutive seconds, the system is determined to have entered a stable operating condition. Starting from the stable moment, the system begins to collect operating parameters from 30 valid sampling points.

[0211] The calculation of the adjusted heat conversion efficiency uses the same efficiency calculation formula as in step S4.1. The mean of the 28 points after removing the maximum and minimum values ​​from the 30 sampling points is taken as the adjusted thermal energy conversion efficiency value to improve the noise robustness of the calculation results.

[0212] Among them, the efficiency comparison before and after optimization is calculated. ,in The value represents the actual efficiency under the current operating conditions before optimization. If Δη > 0, it means that the optimization has effectively improved the thermal energy conversion efficiency. The value of Δη is the efficiency improvement margin of this round of optimization.

[0213] For example, before optimization, the actual efficiency of the boiler under the same operating conditions was 91.8%. After parameter adjustment, the system entered a stable operating condition after 25 seconds. Starting from the stable moment, the operating parameters of 30 sampling points were collected and the efficiency value of each point was calculated. After removing the maximum efficiency value of 93.2% and the minimum value of 91.9%, the average of the remaining 28 efficiency values ​​was 92.5%. Therefore, the heat energy conversion efficiency after adjustment was 92.5%, and the optimization improvement was Δη = 92.5% - 91.8% = 0.7 percentage points.

[0214] Step S6.5: Determine whether the adjusted thermal energy conversion efficiency has reached the preset optimization target. If it has not been reached, trigger the re-optimization control process. If it has been reached, complete the boiler thermal energy conversion efficiency optimization control.

[0215] Preferably, the core of this step is to complete the closed-loop decision of the optimization control process. By comparing and verifying the actual efficiency after adjustment with the efficiency before optimization, it is determined whether the current round of optimization has achieved the expected efficiency improvement target, and based on this, it is decided whether to end the current round of optimization or trigger a new round of S3 to S6 re-optimization process, forming a continuous adaptive iterative closed-loop control mechanism.

[0216] Specifically, the optimization target is set based on whether the improvement margin Δη reaches a preset threshold, which is set to Δη ≥ 1.0 percentage point. If Δη < 1.0%, it indicates that the actual effect of this round of optimization has not met expectations, and a re-optimization process needs to be triggered to fine-tune the model and parameters.

[0217] Among them, for the trigger mechanism of re-optimization process, if it is determined that re-optimization is required, the system will re-execute from step S3: re-input the latest collected running data (including data after parameter adjustment) into the random forest regression model to obtain the updated efficiency prediction value, and re-execute NSGA-Ⅲ search and TOPSIS ranking based on the updated prediction value to generate a new round of optimal control parameter combination.

[0218] Among them, the maximum number of re-optimization rounds is limited. The maximum number of re-execution rounds for a single trigger optimization control is set to 3 rounds. If the threshold of Δη≥1.0% is not reached in 3 consecutive rounds of optimization, it is considered that the optimization potential of the current working condition has been basically exhausted. The parameters of the last round executed are output as the final optimization result to avoid infinite loops that cause frequent operation of the execution equipment.

[0219] For example, after this round of optimization, Δη=0.7% is lower than the optimization improvement threshold of 1.0%, so the system triggers a re-optimization process: the latest set of operating parameters containing the adjusted new data is input into the random forest model to obtain an updated predicted value of 92.8%, the NSGA-Ⅲ search is re-executed to obtain a new Pareto solution set, and the new optimal control parameter combination [gas valve opening 61%, feedwater valve opening 58%, damper opening 52%] is extracted by TOPSIS sorting and re-issued for execution; after the second round of optimization, Δη=1.2% reaches the threshold of 1.0%, and the optimization of the boiler thermal energy conversion efficiency is determined to be successful, completing the optimization control of the boiler thermal energy conversion efficiency.

[0220] Beneficial effects of steps S6.1 to S6.5: This series of steps encapsulates the optimal control parameter combination into executable instructions according to the communication protocols of each executing device, and sends them to the gas regulating valve, feedwater pump, and damper controller respectively. It receives parameter adjustment completion signals from each device after execution, collects the adjusted boiler operating parameters, calculates the actual thermal energy conversion efficiency, and determines whether to trigger a re-optimization process or complete the optimized control through threshold judgment, forming a complete closed-loop adaptive optimization control loop. Traditional static schemes only calibrate parameters once in the preset stage, without any self-checking or effect feedback mechanisms during subsequent operation. Once the parameters deviate from the optimal state due to equipment wear, pipeline scaling, changes in fuel quality, etc., the control system continues to operate with inefficient parameters without awareness until professional inspection and recalibration. This series of steps, through a complete closed-loop mechanism of parameter issuance → execution feedback → effect calculation → condition judgment, automatically verifies the adjustment effect after each optimization and determines whether to re-trigger the optimization process, enabling the control system to have continuous adaptive capabilities and fundamentally eliminating the dependence on periodic manual correction.

[0221] In step S6.1, the three control parameters—gas valve opening, water supply valve opening, and damper opening—are converted into standard register writing instructions according to the Siemens S7-1500 PLC's communication protocol or Modbus RTU protocol format, ensuring the compatibility and executability of the downlink instructions. Step S6.2, the encapsulated instructions are sent to each execution device as a 4-20mA current signal via the PLC's analog output module, completing the physical implementation of the optimized parameters. Step S6.3, real-time opening signals and arrival confirmation signals from each execution device are received to confirm the parameters. All adjustments have been completed, providing a basis for subsequent effect verification. Step S6.4 involves re-collecting sensor data after parameter adjustment and calculating the actual thermal energy conversion efficiency to compare with the efficiency before optimization, using quantitative data to verify the optimization effect. Step S6.5 sets an optimization improvement threshold, such as 2%. If the efficiency improvement after adjustment does not reach the threshold, the optimization process from S3 to S6 is re-executed. If the threshold is reached or exceeded, the current round of optimization is completed, forming a continuously iterative adaptive control loop to ensure that the boiler always operates in a near-optimal state throughout its entire life cycle under all operating conditions.

[0222] In summary, the overall beneficial effects of steps S1 to S6 in this embodiment are as follows: This method, through steps S1 to S6, forms a complete intelligent optimization control closed loop covering multi-source data acquisition, data preprocessing, random forest efficiency prediction, NSGA-III multi-objective Pareto search, TOPSIS optimal solution extraction, and closed-loop execution. It effectively overcomes the technical problem of low thermal energy conversion efficiency caused by pre-calibrated static parameter mapping methods in existing technologies, which cannot adapt to dynamic conditions such as changes in ambient temperature, equipment wear, and fuel quality fluctuations. This method replaces traditional table lookups with real-time prediction of thermal energy conversion efficiency under current operating conditions using a random forest regression model; replaces single-point calibration with NSGA-III non-dominated sorting genetic algorithm searching for Pareto optimal solutions across multiple objectives; replaces manual experience-based setting with TOPSIS approximation of ideal solutions sorting to extract optimal balance parameters from multiple objectives; and replaces open-loop execution with closed-loop execution and effect verification. This fundamentally solves the inherent defects of static control strategies deviating from optimal operating conditions under varying loads and requiring periodic recalibration by professional technicians.

[0223] Step S1 involves collecting nine operating parameters—fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure—and integrating them into an operating parameter set. This provides a complete multi-dimensional data foundation for subsequent analysis, overcoming the information limitations of traditional control systems that rely solely on a single pressure deviation signal for adjustment. Step S2 performs missing value imputation, outlier handling, and standardization preprocessing on the operating parameter set to eliminate the impact of sensor noise, sampling omissions, and dimensional differences on model accuracy, ensuring the stability and consistency of the input to the random forest regression model. Step S3 inputs the standardized feature dataset into the random forest regression model, using multiple decision trees to predict in parallel and take the average to output the predicted thermal energy conversion efficiency value, providing a basis for multi-objective optimization. Provide a reliable real-time efficiency benchmark; Step S4 aims to maximize thermal energy conversion efficiency, minimize pollutant emissions, and minimize fuel consumption. It uses the NSGA-Ⅲ algorithm to perform a multi-objective Pareto front iterative search, overcoming the technical deficiency of traditional single-objective optimization that cannot simultaneously consider efficiency and emission indicators; Step S5 uses the TOPSIS method to calculate the relative proximity of each solution to the positive and negative ideal solutions and extracts the optimal solution, achieving the best balance among multiple optimization objectives; Step S6 encapsulates the optimal control parameter combination according to the communication protocol and sends it to the gas regulating valve, feedwater pump, and damper controller. It receives execution feedback and verifies whether the adjusted efficiency has reached the preset target, forming a complete closed-loop adaptive optimization mechanism that can achieve continuous optimization across the entire operating range without manual intervention.

[0224] like Figure 2 As shown, this embodiment provides an example of an optimized control system for boiler thermal energy conversion efficiency. In this embodiment, the optimized control system is applied to the optimized control method as described in the above embodiment.

[0225] Specifically, the optimized control system includes a boiler operating parameter acquisition module 1, a boiler operating parameter preprocessing module 2, a thermal energy conversion efficiency prediction module 3, a Pareto front iterative search module 4, an optimal control parameter combination acquisition module 5, and a boiler thermal energy conversion efficiency optimization control module 6, which are connected electrically or through communication in sequence.

[0226] The system comprises the following modules: Boiler Operation Parameter Acquisition Module 1 collects boiler operating parameters such as fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure, and integrates them into a boiler operation parameter set; Boiler Operation Parameter Preprocessing Module 2 preprocesses the boiler operation parameter set to obtain a standardized boiler operation feature dataset; Thermal Energy Conversion Efficiency Prediction Module 3 inputs the standardized boiler operation feature dataset into a random forest regression model and outputs a predicted thermal energy conversion efficiency value under the current operating condition; Pareto Front Iterative Search Module 4 inputs the predicted thermal energy conversion efficiency value into the NSGA-Ⅲ non-dominated sorting genetic algorithm for multi-objective Pareto front iterative search to obtain a multi-objective Pareto solution set; Optimal Control Parameter Combination Acquisition Module 5 extracts the optimal solution from the multi-objective Pareto solution set through TOPSIS approximation ideal solution sorting, obtaining the optimal control parameter combination based on gas, feedwater, and dampers; and Boiler Thermal Energy Conversion Efficiency Optimization Control Module 6 distributes the optimal control parameter combination to the corresponding execution equipment to complete the boiler thermal energy conversion efficiency optimization control.

[0227] like Figure 3 As shown, the electronic device 7 includes a processor 71 and a memory 72 coupled to the processor 71.

[0228] The memory 72 stores program instructions for implementing the optimized control method for boiler thermal energy conversion efficiency in any of the above embodiments.

[0229] The processor 71 is used to execute program instructions stored in the memory 72 to perform optimized control of boiler thermal energy conversion efficiency.

[0230] The processor 71 can also be referred to as a CPU (Central Processing Unit). The processor 71 may be an integrated circuit chip with signal processing capabilities. The processor 71 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0231] Furthermore, Figure 4 This is a schematic diagram of the structure of a storage medium according to an embodiment of this application. See also: Figure 4The storage medium 8 in this embodiment stores program instructions 81 capable of implementing all the above methods. These program instructions 81 can be stored in the storage medium as a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods in each embodiment of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0232] In the several embodiments provided in this application, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the optimized control of a unit is only a logical function optimization control; in actual implementation, there may be other optimized control methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between systems or units may be electrical, mechanical, signal, or other forms.

[0233] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for optimizing and controlling the thermal energy conversion efficiency of a boiler, characterized in that, The optimized control method includes: Step S1: Collect the boiler's operating parameters, including fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure, and integrate them into a boiler operating parameter set. Step S2: Perform data preprocessing on the boiler operating parameter set to obtain a standardized boiler operating feature dataset; Step S3: Input the standardized boiler operating characteristic dataset into the random forest regression model and output the predicted value of thermal energy conversion efficiency under the current operating conditions; Step S4: Input the predicted thermal energy conversion efficiency value into the NSGA-Ⅲ non-dominated sorting genetic algorithm to perform multi-objective Pareto front iterative search and obtain the multi-objective Pareto solution set; Step S5: Extract the optimal solution of the multi-objective Pareto solution set by sorting the approximate ideal solution using TOPSIS, and obtain the optimal combination of control parameters based on gas, water supply, and dampers. Step S6: The optimal control parameter combination is sent to the corresponding execution device to complete the boiler thermal energy conversion efficiency optimization control.

2. The optimized control method according to claim 1, characterized in that, Step S1: Collect boiler operating parameters including fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure, and integrate them into a boiler operating parameter set, including: Step S1.1: Collect boiler operating parameters such as fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure through sensors, and integrate them into a boiler operation dataset. Step S1.2: Clean the boiler operation dataset to obtain a cleaned operation dataset; Step S1.3: Timestamp alignment is performed on the cleaned running dataset to obtain the timestamp-aligned running dataset; Step S1.4: After aligning the timestamps, perform field mapping and format unification on the operating parameters to obtain the boiler operating parameter set.

3. The optimized control method according to claim 1, characterized in that, Step S2 involves preprocessing the boiler operating parameter set to obtain a standardized boiler operating characteristic dataset, including: Step S2.1: Fill the missing values ​​in the boiler operating parameter set using interpolation to obtain the boiler operating parameter set after missing value filling; Step S2.2: Identify outliers in the boiler operating parameter set after the missing values ​​are filled, and process outliers that exceed the reasonable range to obtain the boiler operating parameter set after outlier processing. Step S2.3: Perform data standardization processing on the boiler operating parameter set after outlier processing to obtain a standardized boiler operating feature dataset.

4. The optimized control method according to claim 1, characterized in that, Step S3: Input the standardized boiler operating characteristic dataset into the random forest regression model, and output the predicted value of thermal energy conversion efficiency under the current operating condition, including: Step S3.1: Divide the standardized boiler operation feature dataset into a training dataset and a test dataset; Step S3.2: Train the random forest regression model using the training dataset, and construct multiple decision trees by minimizing the Gini coefficient to obtain the trained random forest regression model; Step S3.3: Use the test dataset to evaluate the performance of the trained random forest regression model to obtain the performance-evaluated random forest regression model. Step S3.4: Input the standardized boiler operation characteristic dataset collected in real time into the random forest regression model after performance evaluation, and obtain the predicted value of thermal energy conversion efficiency under the current operating condition by making parallel predictions through multiple decision trees and taking the average value.

5. The optimized control method according to claim 1, characterized in that, Step S4: Input the predicted thermal energy conversion efficiency value into the NSGA-Ⅲ non-dominated sorting genetic algorithm to perform a multi-objective Pareto front iterative search, and obtain a multi-objective Pareto solution set, including: Step S4.1: Construct a multi-objective optimization model with the objective functions of maximizing thermal energy conversion efficiency, minimizing pollutant emissions, and minimizing fuel consumption; Step S4.2: Input the predicted value of thermal energy conversion efficiency into the multi-objective optimization model, and use the NSGA-Ⅲ non-dominated sorting genetic algorithm to generate an initial population for the multi-objective optimization model to obtain the initial population; Step S4.3: Perform non-dominated sorting and reference point generation on the initialized population to obtain the Pareto front of the current population; Step S4.4: The offspring population is generated through selection, crossover, and mutation genetic operations, and the parent population and the offspring population are merged for environmental selection to obtain the environmentally selected population. Step S4.5: Repeat steps S4.3 to S4.4 to iteratively perform non-dominated sorting, reference point generation, and environment selection until the termination condition is met, and obtain the multi-objective Pareto solution set.

6. The optimized control method according to claim 1, characterized in that, Step S5: Extract the optimal solution from the multi-objective Pareto solution set by sorting the approximate ideal solutions using TOPSIS, and obtain the optimal control parameter combination based on gas, water supply, and dampers, including: Step S5.1: Calculate the positive ideal solution and negative ideal solution for each solution in the multi-objective Pareto solution set; Step S5.2: Calculate the distance from each solution to its corresponding positive ideal solution and negative ideal solution, and obtain the distance value pair for each solution; Step S5.3: Calculate the relative proximity value between each solution and the ideal solution based on the distance value of each solution; Step S5.4: Sort each solution in the multi-objective Pareto solution set according to all relative proximity values ​​from largest to smallest, and extract the solution ranked first as the optimal solution; Step S5.5: The optimal solution is analyzed into an optimal combination of control parameters based on gas, water supply, and dampers.

7. The optimized control method according to claim 1, characterized in that, Step S6: The optimal control parameter combination is sent to the corresponding execution device to complete the boiler thermal energy conversion efficiency optimization control, including: Step S6.1: Encapsulate the optimal control parameter combination into an instruction according to the corresponding execution device communication protocol to obtain the optimal control parameter combination after instruction encapsulation. Step S6.2: The optimal combination of control parameters after encapsulating the instruction is sent to the corresponding execution device; Step S6.3: Receive the corresponding optimal control parameters from each execution device, and receive the parameter adjustment completion signal fed back by each execution device after execution; Step S6.4: In response to the parameter adjustment completion signal fed back by all executing devices, collect the boiler's operating parameters after parameter adjustment and calculate the adjusted thermal energy conversion efficiency; Step S6.5: Determine whether the adjusted thermal energy conversion efficiency has reached the preset optimization target. If it has not been reached, trigger the re-optimization control process. If it has been reached, complete the boiler thermal energy conversion efficiency optimization control.

8. An optimized control system for boiler thermal energy conversion efficiency, wherein the optimized control system is applied to the optimized control method as described in any one of claims 1 to 7, characterized in that, The optimized control system includes: The boiler operating parameter acquisition module is used to collect boiler operating parameters such as fuel flow rate, steam production, condensate flow rate, feedwater flow rate, coil temperature, flue gas temperature, flue gas oxygen content, steam pressure, and feedwater pressure, and integrate them into a boiler operating parameter set. The boiler operating parameter preprocessing module is used to preprocess the boiler operating parameter set to obtain a standardized boiler operating feature dataset. The thermal energy conversion efficiency prediction module is used to input the standardized boiler operating characteristic dataset into the random forest regression model and output the predicted value of thermal energy conversion efficiency under the current operating conditions. The Pareto front iterative search module is used to input the predicted thermal energy conversion efficiency value into the NSGA-Ⅲ non-dominated sorting genetic algorithm to perform a multi-objective Pareto front iterative search and obtain a multi-objective Pareto solution set. The optimal control parameter combination acquisition module is used to extract the optimal solution of the multi-objective Pareto solution set by sorting the approximation ideal solution through TOPSIS, and obtain the optimal control parameter combination based on gas, water supply and damper. The boiler thermal energy conversion efficiency optimization control module is used to send the optimal control parameter combination to the corresponding execution device to complete the boiler thermal energy conversion efficiency optimization control.

9. An electronic device, characterized in that, The method includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the optimized control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions that, when executed by a processor, enable the optimization control method as described in any one of claims 1 to 7.

Images