A method and device for multi-objective collaborative optimization of coal blending and blending combustion in a thermal power plant

Abstract

The application discloses a kind of multi-objective collaborative optimization method and device of coal blending of thermal power plant, belong to energy management and intelligent control technical field.Integrating the coal quality of thermal power plant, operation, emission and scheduling data, construct unified feature library, and based on this, set the initial weight of economic, safety, environmental protection target, establish multi-objective optimization model;Adopt parallel computing architecture to solve model, generate multiple candidate blending scheme;Subsequently, combined with the artificial adjustment result of operator and actual combustion feedback data, carry out secondary prediction and comparative analysis;Finally, through continuous learning adjustment behavior and operation effect, drive model automatically update weight and optimization algorithm combination, form closed loop self-learning mechanism, to realize multi-objective collaborative optimization and adaptive decision of thermal power plant.The application improves the intelligence, economy and environmental protection of system operation.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and energy management technology, and in particular to a multi-objective collaborative optimization method, apparatus, equipment and storage medium for coal blending in thermal power plants. Background Technology

[0002] With the increasing demands for energy conservation, emission reduction, and ultra-low emissions in thermal power plants, the refined and intelligent management of fuel systems has become crucial for ensuring the economical and efficient operation of units and achieving environmental compliance. Currently, coal blending decisions in thermal power plants heavily rely on the experience of operators or employ static calculation models based on historical data. These traditional methods exhibit increasing limitations when facing complex and variable coal source structures (such as widely blended high-alkali and high-sulfur coals). On the one hand, they cannot perceive and respond in real time to dynamic fluctuations in coal quality, changes in equipment status, and the dispatching demands of the electricity market. On the other hand, in scenarios with significant differences in coal quality characteristics and complex blending ratio calculations, experience-based judgments or simple models struggle to accurately balance multiple objectives such as combustion economy, equipment safety (e.g., preventing slagging and corrosion), and environmental emissions, easily leading to reduced boiler efficiency and increased operational risks.

[0003] In recent years, some enterprises have attempted to introduce information technology, establishing coal quality databases and constructing blending calculation models to assist decision-making. However, existing systems still have several significant shortcomings: First, most models are essentially static mathematical models based on fixed formulas or linear relationships, with limited predictive accuracy and a lack of adaptability to nonlinear and strongly coupled operational constraints. Second, even when optimization algorithms are introduced, they are mostly limited to solving a single objective (such as minimizing cost), lacking a collaborative optimization and dynamic weight balancing mechanism for multi-dimensional objectives such as economy, safety, and environmental protection. Furthermore, although the system integrates diverse data such as coal quality, equipment operation, environmental monitoring, and economic dispatch, the data is often scattered and poorly correlated, failing to form a unified, interconnected, high-quality feature knowledge and data foundation, thus hindering global optimization analysis. Crucially, existing systems generally lack intelligent feedback learning capabilities. When operators adjust recommended solutions based on actual on-site conditions, the system cannot effectively capture and learn these human interventions and their subsequent combustion effect feedback, leading to lagging model iteration and optimization, difficulty in continuously improving decision-making accuracy, and an inability to form a closed-loop evolution that aligns with the actual operating patterns of power plants.

[0004] In summary, existing coal blending technologies in thermal power plants generally face bottlenecks such as data silos, simplistic algorithms, rigid models, and a lack of self-learning capabilities, making it difficult to support dynamic, collaborative, and optimal decision-making needs under complex coal sources, variable operating conditions, and multiple constraints. Therefore, there is an urgent need to invent an intelligent blending system and method that deeply integrates artificial intelligence technology, enabling intelligent processing throughout the entire process, from dynamic perception of coal quality characteristics and equipment operating conditions, multi-objective intelligent optimization calculations, scheme verification and adjustment to closed-loop self-learning, thereby systematically improving the scientific nature and overall efficiency of fuel management in thermal power plants. Summary of the Invention

[0005] The present invention aims to at least partially solve one of the technical problems in the related art.

[0006] To address this, this invention proposes a multi-objective collaborative optimization method for coal blending in thermal power plants. This method integrates coal quality characteristics, equipment operating parameters, emission monitoring data, and economic dispatch information to construct a unified feature database. Based on this database, initial weights are set for economic, safety, and environmental objectives, establishing a multi-objective dynamic optimization model. The model is solved using a multi-algorithm parallel computing architecture, generating multiple sets of candidate blending schemes with different weights and constraints. Subsequently, secondary prediction and comparative analysis are performed by combining operator adjustments with actual combustion performance data. Finally, by continuously recording adjustment behavior and operational feedback, the model is driven to automatically update objective weights and optimize algorithm combinations, thus forming a closed-loop self-learning mechanism to achieve dynamic collaborative optimization of multiple objectives and continuous improvement of system performance.

[0007] Another objective of this invention is to propose a multi-objective collaborative optimization device for coal blending in thermal power plants.

[0008] The third objective of this invention is to provide a computer device.

[0009] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0010] To achieve the above objectives, this invention proposes a multi-objective collaborative optimization method for coal blending in thermal power plants, comprising: S1 acquires coal quality characteristics, equipment operating parameters, emission monitoring data and economic dispatch information of thermal power plants, and generates a unified feature database through data cleaning and standardization. S2. Based on a unified feature database, initial weights are set for three categories of objectives: economy, safety, and environmental protection, and a multi-objective optimization model is established. S3 utilizes a multi-algorithm parallel computing architecture to execute a multi-objective optimization model, generating multiple sets of candidate blending schemes containing different weight combinations and operational constraints. S4. Based on the operator's manual adjustments to the candidate blending schemes, combined with actual combustion effect data, a secondary prediction is made and a comparative analysis report is generated. S5, by recording human adjustment behavior and combustion feedback data, drives the AI ​​model to automatically update weight parameters and optimize algorithm combinations, forming a closed-loop self-learning mechanism.

[0011] The multi-objective collaborative optimization method for coal blending in thermal power plants according to an embodiment of the present invention may also have the following additional technical features: In one embodiment of the present invention, the step of acquiring coal quality characteristics, equipment operating parameters, emission monitoring data, and economic dispatch information of thermal power plants, and generating a unified feature database through data cleaning and standardization processing, includes: S11, Data cleaning includes performing integrity verification, duplicate item removal, anomaly detection, and time series repair on data collected from coal quality testing systems, DCS system interfaces, emission monitoring interfaces, and economic data interfaces; S12, the standardization process adopts the Z-Score standardization and normalization method to uniformly encode and transform the dimensions of multi-source heterogeneous data, ensuring the consistency of feature input.

[0012] In one embodiment of the present invention, the step of setting initial weights for three categories of objectives—economic efficiency, safety, and environmental friendliness—based on a unified feature database and establishing a multi-objective optimization model includes: S21, the initial weights are set based on historical combustion data and real-time coal quality fluctuations, and the allocation ratio of economic, safety and environmental protection weights is dynamically adjusted through a reinforcement learning model. S22, the multi-objective optimization model includes constraints, specifically the upper limit of boiler load, the combustion temperature threshold, and the boundary values ​​of equipment operating parameters.

[0013] In one embodiment of the present invention, the step of using a multi-algorithm parallel computing architecture to execute a multi-objective optimization model and generate multiple sets of candidate blending schemes containing different weight combinations and operational constraints includes: S31, a multi-algorithm parallel computing architecture, includes a genetic algorithm module, a particle swarm optimization module, and a reinforcement learning module, and can dynamically switch algorithm combinations through an algorithm management interface; S32, the generation of candidate blending schemes is based on parallel computing results. Each scheme corresponds to different weight combinations and running constraints, and the results are compared through the algorithm management interface.

[0014] In one embodiment of the present invention, the step of performing secondary prediction and generating a comparative analysis report based on the operator's manual adjustment results of the candidate blending scheme and actual combustion effect data includes: S41, the results of the manual adjustment include fine-tuning parameters for blending ratio, air-coal ratio, air supply temperature and coal mill current; S42, the comparative analysis report includes a comparison of the deviations in combustion efficiency, emission concentration and economic indicators before and after the adjustment, and generates parameter optimization suggestions based on sensitivity analysis.

[0015] In one embodiment of the present invention, it further includes: S6 generates a 3D visualization model and sensitivity analysis curves, displays the blending scheme, combustion status and boiler operating conditions in an interactive graphical interface, and updates the sensitivity analysis results in real time based on user-adjusted weights.

[0016] To achieve the above objectives, another aspect of the present invention proposes a multi-objective synergistic optimization device for coal blending in thermal power plants, comprising: The data acquisition and preprocessing module is used to acquire coal quality characteristics, equipment operating parameters, emission monitoring data and economic dispatch information of thermal power plants, and generate a unified feature database through data cleaning and standardization. The weight configuration and modeling module is used to set the initial weights for three categories of objectives—economic efficiency, safety, and environmental protection—based on a unified feature database, and to establish a multi-objective optimization model. The multi-algorithm parallel computing module is used to execute a multi-objective optimization model using a multi-algorithm parallel computing architecture, generating multiple sets of candidate blending schemes containing different weight combinations and running constraints. The manual adjustment and secondary prediction module is used to perform secondary prediction and generate a comparative analysis report based on the operator's manual adjustment results of the candidate blending scheme and the actual combustion effect data. The closed-loop self-learning control module is used to drive the AI ​​model to automatically update weight parameters and optimize algorithm combinations by recording manual adjustment behavior and combustion feedback data, thus forming a closed-loop self-learning mechanism.

[0017] In one embodiment of the present invention, it further includes: The visualization and sensitivity analysis module is used to generate three-dimensional visualization models and sensitivity analysis curves. It displays the blending scheme, combustion status and boiler operating conditions in an interactive graphical interface, and updates the sensitivity analysis results in real time based on user-adjusted weights.

[0018] This invention discloses a multi-objective collaborative optimization method and apparatus for coal blending in thermal power plants. By constructing a multi-objective dynamic optimization model based on a unified feature database and employing a multi-algorithm parallel computing architecture and a closed-loop self-learning mechanism, it effectively solves the problems of insufficient optimization accuracy and poor adaptability in existing technologies caused by rigid models, singular objectives, and lack of feedback iteration. It achieves closed-loop optimization throughout the entire process, from multi-source data fusion, multi-objective dynamic weight allocation, intelligent scheme generation, manual interactive feedback to autonomous updating of model parameters. This significantly improves the economic, safety, and environmental synergistic optimization level of blending schemes, and enhances the system's adaptive decision-making ability and long-term operational efficiency when facing complex coal quality and variable operating conditions.

[0019] To achieve the above objectives, a third aspect of this application provides a computer device, including a processor and a memory; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, for implementing a multi-objective collaborative optimization method for coal blending in thermal power plants as described in the first aspect embodiment.

[0020] To achieve the above objectives, the fourth aspect of this application proposes a non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements a multi-objective collaborative optimization method for coal blending in thermal power plants as described in the first aspect embodiment.

[0021] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a multi-objective collaborative optimization method for coal blending in thermal power plants according to an embodiment of the present invention; Figure 2 This is a diagram of a multi-objective collaborative optimization system architecture for coal blending in a thermal power plant, according to an embodiment of the present invention. Figure 3 This is a flowchart of a multi-objective collaborative optimization system for coal blending in a thermal power plant according to an embodiment of the present invention. Figure 4 This is a schematic diagram of a multi-objective collaborative optimization device for coal blending in a thermal power plant according to an embodiment of the present invention; Figure 5 It is a computer device according to an embodiment of the present invention. Detailed Implementation

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] The following description, with reference to the accompanying drawings, describes a multi-objective collaborative optimization method, apparatus, equipment, and storage medium for coal blending in thermal power plants, according to embodiments of the present invention.

[0026] The core idea of ​​this invention is to construct a unified feature database by integrating coal quality characteristics, equipment operating parameters, emission monitoring data, and economic dispatch information from thermal power plants, laying a data foundation for subsequent optimization. Based on this database, initial weights for economic, safety, and environmental objectives are set, and a multi-objective dynamic optimization model is established. A multi-algorithm parallel computing architecture is used to solve the model, generating multiple candidate blending schemes that integrate different weight combinations and operational constraints. Furthermore, a human-computer interaction and feedback mechanism is introduced. Based on operator adjustments to the schemes, secondary prediction and comparative analysis are performed using real-time combustion effect data, achieving a fusion verification of theoretical calculations and field experience. Finally, by recording human intervention behaviors and their corresponding combustion feedback data, the system drives the model to automatically update target weight parameters and dynamically optimize algorithm combinations, thus forming a continuously evolving, self-adaptive closed-loop self-learning mechanism. This invention transforms traditional static calculation and experience-based decision-making into a data-driven, multi-objective collaborative, human-computer interactive, and continuously optimized intelligent decision-making system, significantly improving the overall economic, safety, and environmental performance of thermal power plants under complex coal sources and variable operating conditions.

[0027] Example 1 To achieve the above invention, embodiments of the present invention provide a multi-objective collaborative optimization method for coal blending in thermal power plants, such as... Figure 1 As shown, it includes: S1 acquires coal quality characteristics, equipment operating parameters, emission monitoring data, and economic dispatch information from thermal power plants, and generates a unified feature database through data cleaning and standardization.

[0028] Specifically, this step is the foundation of the entire intelligent blending system, and its technical implementation relies on the real-time acquisition of multi-source heterogeneous data, protocol compatibility processing, and data quality assurance mechanisms.

[0029] Specifically, the system collects coal quality data from fuel laboratories or online analysis equipment (such as near-infrared spectrometers and X-ray fluorescence spectrometers) through a coal quality testing system, including ash content, volatile matter, total sulfur, moisture, calorific value, ash fusion temperature, and sodium oxide content. ) and potassium oxides ( Key indicators such as [list of indicators]. The DCS system interface uses industrial communication protocols such as OPC-UA or Modbus-TCP to acquire real-time operating parameters such as coal mill current, output, inlet temperature, air-coal ratio, and main steam pressure, achieving millisecond-level synchronization with the boiler combustion status. The emission monitoring interface connects to the online monitoring system (CEMS) to collect [data / information]. , Dust concentration, fly ash alkali metal content, flue gas temperature and The system collects emission data such as concentrations and uses timestamp alignment technology to establish a causal relationship between combustion and emission data. The economic data interface integrates scheduling information such as electricity prices, fuel procurement costs, inventory levels, and load curves, providing economic input for subsequent multi-objective optimization.

[0030] Furthermore, the system employs a multi-level cleaning strategy, including integrity verification, duplicate item removal, outlier detection (such as Z-Score and IQR methods), and time series restoration (such as linear interpolation and moving average). The data standardization module supports normalization (Min-Max), Z-Score standardization, and time smoothing filtering (such as Savitzky-Golay filters) to ensure consistency in scale, precision, and temporal resolution across data sources. Finally, all data is uniformly encoded and stored in a feature database (204), forming structured and standardized feature vectors that provide high-quality input for the AI ​​algorithm engine.

[0031] Specifically, this step plays a crucial role in the system, and its output directly affects the accuracy and stability of subsequent blending calculations. By establishing a unified data hub, the system achieves the fusion and feature encoding of multi-source data, providing data support for multi-objective optimization, combustion prediction, and self-learning mechanisms. This is the first step in realizing a closed-loop system of "algorithm-driven + data-supported + interactive decision-making."

[0032] Furthermore, S1 includes: S11, data cleaning includes performing integrity verification, duplicate item removal, anomaly detection, and time series repair on data collected from coal quality testing systems, DCS system interfaces, emission monitoring interfaces, and economic data interfaces.

[0033] Specifically, in some implementations, the data cleaning module first connects to the raw data streams collected by the coal quality testing system, DCS system interface, emission monitoring interface, and economic data interface, and performs operations such as integrity verification, duplicate item removal, anomaly detection, and time series repair.

[0034] Specifically, the system ensures that each record contains key parameters, such as ash content, volatile matter, total sulfur, and calorific value in coal, and equipment operating parameters, such as mill current and air-fuel ratio, by checking the non-empty status of data fields and the continuity of timestamps. If a field is missing or the timestamp interval exceeds a set threshold (e.g., 30 seconds), the system will trigger a data completion mechanism, using linear interpolation or regression prediction based on historical data to repair the data.

[0035] Furthermore, duplicate item removal uses hash comparisons based on unique identifiers (such as timestamps + device IDs) to identify and delete duplicate records, avoiding data redundancy from interfering with model training. Outlier detection employs statistical methods (such as Z-Score detection and IQR detection) and AI-based outlier identification algorithms to mark and isolate data points exceeding set thresholds (such as ash content > 40% or NOx concentration > 300 mg / Nm³), ensuring the reliability of the dataset.

[0036] Specifically, in terms of time series restoration, the system resamples and interpolates time series data with breakpoints or discontinuities, using Spline interpolation or ARIMA models for trend fitting to ensure the continuity and predictability of the time series data in subsequent algorithm calculations. This step plays a crucial role in the data hub and feature analysis layer, providing high-quality, standardized data input for feature extraction and multi-objective optimization, thereby improving the computational accuracy and decision reliability of the AI ​​algorithm engine.

[0037] S12, the standardization process adopts the Z-Score standardization and normalization method to uniformly encode and transform the dimensions of multi-source heterogeneous data, ensuring the consistency of feature input.

[0038] Specifically, this step is a key preprocessing step for achieving efficient modeling and optimization calculations in the AI ​​algorithm engine layer. Its technical implementation principle is based on statistics and data preprocessing methods, aiming to eliminate differences in numerical range, unit system, distribution pattern, etc. of different data sources, thereby improving the convergence speed and prediction accuracy of the model.

[0039] Furthermore, Z-Score standardization transforms the original data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Its mathematical expression is: ,in For the original data points, The sample mean. denoted as the sample standard deviation. This method is suitable for datasets with obvious normal distribution characteristics, especially in coal quality testing data (such as ash content, volatile matter, total sulfur, calorific value, etc.), effectively eliminating dimensional differences between different coal types and improving the model's robustness to coal quality fluctuations. The normalization method linearly maps the data to the [0,1] interval, and its formula is: It is suitable for data that is not normally distributed or has extreme values, such as real-time operating parameters like coal mill current and air supply temperature.

[0040] Furthermore, the system automatically selects the standardization method based on the data type. For example, coal quality characteristic data typically uses Z-Score standardization to preserve its distribution characteristics, while equipment operating parameters may be normalized to adapt to the input requirements of AI models such as neural networks. After standardization, all feature variables are uniformly encoded into standardized vectors and input into the feature database for use by the AI ​​algorithm engine layer.

[0041] Specifically, on the one hand, standardization eliminates dimensional differences between multi-source data, ensuring comparability of data from different sources at the model input end. On the other hand, standardized data helps improve the training efficiency and generalization ability of AI models (such as genetic algorithms and reinforcement learning), thus providing a high-quality input foundation for subsequent multi-objective optimization calculations. Furthermore, the standardization module supports time-series smoothing, filtering dynamic features such as coal quality fluctuations and equipment status changes, further enhancing the system's adaptability to complex operating conditions.

[0042] S2. Based on a unified feature database, initial weights are set for three categories of objectives: economy, safety, and environmental protection, and a multi-objective optimization model is established.

[0043] Specifically, this step is a key link in achieving multi-dimensional collaborative optimization of the entire intelligent blending system. Its technical implementation involves the construction of the objective function, weight allocation strategy, integration of constraints, and initial configuration of the multi-objective optimization algorithm.

[0044] Furthermore, the system first extracts standardized coal quality feature vectors, combustion feature parameters, and equipment constraint indicators from the data hub and feature analysis layer, which serve as input variables for the multi-objective optimization model. The economic objective is typically based on the unit power generation cost (…). ) or fuel cost ( Using as a metric, safety targets include the boiler slagging risk index (BSE). ), corrosion tendency ( ) and combustion stability ( Environmental targets are based on the concentration of major pollutants emitted (such as...). , Dust concentration) and carbon emission intensity ( The core parameter is ). The system uses a pre-defined weight allocation mechanism to map the three types of objectives into a unified optimization objective function, which takes the form: ; in, , , The initial weighting coefficients are respectively for economy, safety, and environmental protection, satisfying... ,and The weights can be initialized based on historical operating data, expert experience, or the power plant's current dispatch strategy. For example, during an environmentally prioritized operation phase, the weights can be... Set to 0.4, It is 0.3. It is 0.3.

[0045] Specifically, the system employs multi-objective optimization algorithms (such as NSGA-II, MOEA / D, etc.) to solve for the blending ratio, while simultaneously combining a reinforcement learning mechanism to dynamically adjust the weights. The initial weight settings must meet the power plant's operational constraints, such as the boiler load range (…). Combustion temperature range ( (e.g., ensuring the optimization model operates within the physically feasible domain).

[0046] Specifically, this step is widely used in practical applications for coal blending decisions in thermal power plants at different operating stages. For example, economic efficiency is prioritized during high-load operation, safety weight is strengthened when coal quality fluctuates significantly, and environmental weight is increased when environmental policies become stricter. By establishing a multi-objective optimization model, the system can achieve a global optimal search for blending schemes, avoiding the local optimum trap caused by single-objective optimization, thereby improving the operating efficiency and scientific nature of thermal power plants under complex operating conditions.

[0047] Furthermore, S2 includes: S21, the initial weights are set based on historical combustion data and real-time coal quality fluctuations, and the allocation ratio of economic, safety and environmental protection weights is dynamically adjusted through a reinforcement learning model.

[0048] Specifically, this step first constructs a multi-objective optimization function, quantifying economic efficiency, safety, and environmental friendliness into calculable indicators, such as unit power generation cost. Boiler slagging risk index and total pollutant emissions The system extracts key features from a historical combustion database, including coal quality parameters (such as ash content). Total sulfur volatile matter calorific value Combustion efficiency Emissions data (such as) , Dust concentration (e.g., constructing multidimensional feature vectors) .

[0049] Furthermore, initial weights The settings must meet the normalization condition, that is... The system dynamically adjusts the weights based on the priority distribution of each objective in historical data, combined with the power plant's operating strategy (such as whether it is in the period of environmental protection limit enforcement, whether economic dispatch is given priority, etc.), using reinforcement learning algorithms (such as DQN, PPO, etc.). The reinforcement learning model uses a state space... (Including real-time data such as current coal quality, load, and emissions) and action space The mapping relationship between (weight allocation ratios) is used to learn the optimal weight allocation strategy to maximize the long-term reward function. ,in These are preset weighting coefficients that reflect the relative importance of different objectives.

[0050] Specifically, this step is applicable to thermal power plants operating under conditions of frequent coal quality fluctuations, complex load changes, and stringent environmental protection requirements, enabling dynamic optimization of blending strategies. For example, in high-sulfur coal blending scenarios, the system can automatically increase the environmental weight to reduce... Emission risks; during low-load operation, the economic weight can be appropriately increased to optimize fuel costs.

[0051] Specifically, this step introduces a reinforcement learning mechanism to achieve adaptive adjustment of multi-objective weights, effectively solving the problem that traditional static weight allocation cannot cope with complex working conditions. The reinforcement learning model continuously accumulates experience during long-term operation, improving the rationality of weight allocation and the execution effect of the blending scheme, thereby enhancing the system's intelligence level and operational stability.

[0052] S22, the multi-objective optimization model includes constraints, specifically the upper limit of boiler load, the combustion temperature threshold, and the boundary values ​​of equipment operating parameters.

[0053] Specifically, this step is the core of the entire intelligent blending system. Its technical implementation is based on a multi-algorithm parallel optimization mechanism, which generates and screens multiple candidate blending schemes through the blending algorithm module in the AI ​​algorithm engine layer.

[0054] Furthermore, this optimization model employs a multi-objective collaborative optimization strategy, using economy, safety, and environmental protection as objective functions, while introducing constraints to ensure the feasibility of the blending scheme in actual operation. The upper limit constraint on boiler load is typically set at 80% to 110% of the unit's rated load, i.e. This prevents the boiler from operating under overload or with insufficient output. The combustion temperature threshold is set according to the boiler design and coal characteristics, and is generally controlled within a certain range. Within a certain range, to avoid abnormal combustion phenomena such as slagging and corrosion. Equipment operating parameter boundary values ​​include coal mill current, air-to-coal ratio, and air supply temperature. For example, the coal mill current must not exceed 110% of its rated value, i.e. The ratio of wind power to coal needs to meet the requirements. This is to ensure combustion efficiency and equipment safety.

[0055] Furthermore, the system uses coal quality feature vectors, combustion feature parameters, and equipment constraint indicators extracted from the data hub and feature analysis layer as inputs, and combines real-time operating data with historical feedback data to construct dynamic boundaries for the constraints. For example, the combustion temperature threshold can be based on the ash fusion point of the current coal type (e.g., ...). The system is dynamically adjusted to adapt to the combustion characteristics of different coal types.

[0056] Specifically, this step is widely used in real-time optimization decision-making in thermal power plants under complex coal type scenarios such as blending high-alkali coal and high-sulfur coal. After receiving coal quality detection, DCS operating status and emission data, the system immediately starts a multi-objective optimization model. Combining the current load demand and equipment status, it selects blending schemes that meet the operating boundary conditions, ensuring that the combustion process achieves the optimal balance in terms of safety, economy and environmental protection.

[0057] Specifically, by introducing multi-dimensional constraints, combustion accidents caused by coal quality fluctuations or abnormal operating parameters are effectively prevented, while the feasibility and stability of blending schemes are improved. Combined with a multi-algorithm parallel computing mechanism, the system can quickly generate multiple candidate schemes under different constraints, providing a high-quality decision-making basis for subsequent human-machine collaboration and self-learning closed loop, thereby realizing intelligent and refined fuel management in thermal power plants.

[0058] S3 utilizes a multi-algorithm parallel computing architecture to execute a multi-objective optimization model, generating multiple sets of candidate blending schemes with different weight combinations and operational constraints.

[0059] Specifically, this step is the core of the entire intelligent blending system. Its technical implementation is based on a multi-algorithm collaborative optimization mechanism, which combines the dynamic weight allocation of the three major objectives of economy, safety and environmental protection to achieve optimal decision support for complex coal blending scenarios in thermal power plants.

[0060] Specifically, The algorithm engine layer integrates various optimization algorithm modules, including genetic algorithm (GA), particle swarm optimization (PSO), and reinforcement learning (RL). These modules are scheduled in parallel and their results are compared through an algorithm management interface. The system first takes the coal quality feature vector (after feature extraction and standardization), combustion feature parameters, and equipment constraint indicators as input to construct a multi-objective optimization problem. The objective function can be expressed as: ; in, Indicates the first The economic cost of the blending scheme Indicates a risk of combustion (such as slagging, corrosion, etc.). Indicates the level of pollutant emissions. , , Let be the weight coefficients corresponding to the target, and satisfy . The weighting coefficients can be dynamically adjusted based on the power plant's current operating strategy, for example, by increasing them when environmental policies tighten. Increase when fuel costs fluctuate significantly .

[0061] Furthermore, the system supports user-defined target weight ratios and incorporates a weight self-learning module based on historical combustion data, automatically optimizing weight allocation through a feedback learning mechanism. During parallel computation, each algorithm module runs independently, solving for the optimal blending ratio. Finally, the results are aggregated and sorted through the algorithm management interface, outputting multiple candidate schemes that meet operational constraints (such as boiler load limits, combustion temperature ranges, and equipment output limits).

[0062] Specifically, this step is widely applicable in practical applications to dynamic optimization decisions in thermal power plants under complex coal blending scenarios, such as blending high-alkali coal and high-sulfur coal. Through parallel computation of multiple algorithms, the system can quickly generate multiple feasible blending schemes under different operating conditions (such as load changes and coal quality fluctuations) for operators to evaluate and select, thereby improving the robustness and adaptability of the blending strategy.

[0063] Specifically, this step effectively solves the problems of single algorithm and slow response in traditional coal blending systems, realizes multi-objective collaborative optimization and comparison of multi-algorithm results, provides a high-quality candidate solution foundation for subsequent human-machine collaboration and model self-learning, and significantly improves the intelligence level and operational economy of fuel blending in thermal power plants.

[0064] Furthermore, S3 includes: S31, a multi-algorithm parallel computing architecture, includes a genetic algorithm module, a particle swarm optimization module, and a reinforcement learning module, and dynamically switches the algorithm combination through an algorithm management interface.

[0065] Specifically, this architecture is built on a distributed computing framework (such as Spark or Flink), supporting parallel execution and result comparison of various algorithm modules. The genetic algorithm module adopts a population evolution strategy based on a fitness function, iteratively optimizing the mating scheme through crossover, mutation, and selection operations. The fitness function is defined as... ,in Indicates economic indicators (such as unit power generation cost). Indicates safety indicators (such as boiler slagging risk). Indicates environmental protection indicators (such as NOx emission concentration). Let be the weight coefficients corresponding to the target, and satisfy . .

[0066] Furthermore, the particle swarm optimization module updates the particle position and velocity using formulas. To achieve a global search, where For inertial weights, As a learning factor, It is a random number. This is the optimal solution for the individual. This is the globally optimal solution. This represents the current particle position. The reinforcement learning module employs a deep Q-network (DQN) structure, optimizing the blending strategy through a state-action-reward mechanism. The state space includes coal quality feature vectors, equipment operating parameters, and historical combustion data; the action space is for blending ratio adjustment; and the reward function is designed as follows: ,in Due to cost changes, For efficiency changes, For changes in emissions, This is a dynamically adjusted reward coefficient.

[0067] Specifically, the algorithm management interface implements algorithm scheduling through an API interface, supporting user-defined algorithm combinations and weight allocation. The system automatically selects the optimal algorithm set based on current operating conditions (such as load changes and coal quality fluctuations), ensuring the robustness and optimality of the blending scheme under complex operating conditions. This step realizes multi-algorithm collaborative optimization and dynamic decision-making capabilities in the intelligent blending system of thermal power plants, significantly improving the system's overall performance in terms of economy, safety, and environmental protection.

[0068] S32, the generation of candidate blending schemes is based on parallel computing results. Each scheme corresponds to different weight combinations and running constraints, and the results are compared through the algorithm management interface.

[0069] Specifically, this step involves parallel optimization calculations performed by multiple algorithm modules (such as genetic algorithm module, particle swarm optimization module, reinforcement learning module, etc.) in the algorithm engine layer to generate multiple candidate blending schemes, each of which corresponds to a specific set of weight combinations and operating constraints.

[0070] Furthermore, the algorithm engine layer constructs a multi-objective optimization model based on standardized coal quality feature vectors, combustion feature parameters, and equipment constraint indicators provided by a unified data hub. The optimization objectives include economic efficiency (e.g., fuel cost, power generation revenue), safety (e.g., boiler slagging risk, combustion stability), and environmental friendliness (e.g., ...). , (Dust emission concentration). The generation of each candidate scheme depends on a weighted combination of objective functions, which can be expressed as: ; in, For the first The overall loss function of the blending scheme, The weighting coefficients are respectively for economic efficiency, safety, and environmental friendliness, and they meet the following requirements. The algorithm module runs under different weight combinations, and accelerates the solution process through parallel computing frameworks (such as Spark, MPI, or GPU clusters) to ensure that multiple feasible solutions can be generated quickly even under complex coal types and operating conditions.

[0071] Furthermore, the system supports user-defined weight allocation, such as setting weights during high-load operation. , , Prioritizing both economic efficiency and safety. Meanwhile, operational constraints include the boiler's maximum load (such as...). ), the upper limit of coal mill output (such as ), combustion temperature range (e.g.) (etc.) to ensure that the generated blending scheme is within the limits of physical and equipment constraints.

[0072] Specifically, this step applies to dynamic blending decisions made by thermal power plants in response to fluctuations in coal quality, load adjustments, or changes in environmental policies. The system can respond in real time to changes in coal sources and operational needs, generating multiple blending ratio suggestions for operators or automated decision-making modules to select the optimal solution.

[0073] Specifically, through parallel computation of multiple algorithms and dynamic adjustment of weights, the system can achieve collaborative optimization of multiple objectives, improve the adaptability and robustness of blending schemes, provide a high-quality candidate set for subsequent human-machine collaboration and model self-learning, thereby enhancing the intelligent decision-making capabilities of thermal power plants in complex operating environments.

[0074] S4. Based on the operator's manual adjustments to the candidate blending schemes, combined with actual combustion performance data, a secondary prediction is made and a comparative analysis report is generated.

[0075] Specifically, the system achieves manual intervention and secondary prediction functions through a human-machine collaboration mechanism. Its technical implementation principle is based on real-time data feedback and the dynamic response mechanism of the AI ​​model. In the human-machine collaboration window at the decision-making and visualization layer, operators can fine-tune the proportions or modify operating parameters of candidate blending schemes generated by the AI ​​algorithm engine layer, such as adjusting key combustion control parameters like the air-coal ratio, air supply temperature, and blended coal type ratio. Upon receiving a manual adjustment command, the system triggers a recalculation process in the blending algorithm module and the combustion prediction algorithm module through the algorithm management interface, ensuring that the adjusted scheme meets boiler operating constraints (such as main steam pressure). Combustion temperature Equipment output limit Under the premise of ), a second prediction is made.

[0076] Furthermore, the system adopts an event-driven mechanism. After the operator modifies and submits the parameters in the human-machine collaboration window, the system automatically modifies the adjusted parameters with the current coal quality feature vector. Device state vector Operating load The input variables are then recombined to form new input samples. These samples are then fed into the combustion prediction algorithm module, where combustion efficiency is determined using an AI regression model and a reinforcement learning model. Pollutant emission concentration (e.g.) , Dust concentration The system predicts and calculates key indicators such as [list of indicators]. Simultaneously, the optimization algorithm module optimizes the adjusted operating parameters to minimize operating costs. And maximize combustion stability .

[0077] Furthermore, the system supports adjusting the blending ratio. ( (Coal type number), air-coal ratio Combustion temperature Ash melting point Sodium oxide content Potassium oxide content Real-time adjustments and predictions are made. The secondary prediction results are compared with the original AI-suggested solution to generate a comparative analysis report that includes dimensions such as economy, environmental protection, and safety. The report can show the relative deviations of each indicator. It also provides visual trend charts and sensitivity analysis curves.

[0078] Specifically, in practical applications, this step is widely used in thermal power plants for dynamic blending optimization when coal quality fluctuates, load adjusts, or equipment status changes. For example, after adjusting the blending ratio of high-sulfur coal, the system can instantly predict... Emissions changes, and environmental limits The comparison helps operators determine whether adjustments are feasible. Furthermore, this step uses a manual intervention recording module to link operator adjustments with actual combustion data, forming a closed-loop learning path of "manual intervention—combustion feedback—model correction," significantly improving the AI ​​model's adaptability and prediction accuracy under complex operating conditions.

[0079] Furthermore, S4 includes: S41, the results of the manual adjustment include fine-tuning parameters for blending ratio, air-coal ratio, air supply temperature, and coal mill current.

[0080] Specifically, the technical implementation of this step is based on the human-machine collaboration window and manual intervention recording module in the human-machine collaboration and feedback learning layer. By binding the graphical interface with real-time data, it achieves fine control of the combustion process.

[0081] Furthermore, operators can manually adjust the blending ratio (e.g., from 15% to 12% for high-sulfur coal), the air-to-coal ratio (e.g., from 1.8 kg / kg to 1.6 kg / kg), the air supply temperature (e.g., from 280℃ to 260℃), and the mill current (e.g., from 120A to 115A) based on historical operating experience or current equipment status. Upon receiving these manually adjusted parameters, the system immediately calls the blending algorithm module and the combustion prediction algorithm module through the algorithm management interface to perform secondary prediction calculations on the adjusted parameters, generating new prediction results for combustion efficiency, emission levels, and economic indicators.

[0082] Furthermore, the system supports millisecond-level response and parameter sensitivity analysis, ensuring that adjusted parameters take effect quickly in the combustion model. For example, when the mill current decreases, the system will combine the mill output model... (in This refers to the output power of the coal mill. For current, The effects of pulverized coal fineness and combustion stability were reassessed (as a constant for the equipment). Simultaneously, adjustments to the air-to-coal ratio will affect the combustion air supply, and the system utilizes a combustion prediction model. Make real-time corrections.

[0083] Specifically, this step is widely used in practical applications for refined adjustments in thermal power plants under conditions of coal quality fluctuations, equipment aging, or special operating conditions. For example, during the co-firing of high-alkali coal, operators can adjust the air supply temperature based on historical slagging data to reduce the risk of furnace slagging. In addition, this step binds the operator's adjustment behavior with actual combustion data through the manual intervention recording module (502), forming a "manual-data-feedback" closed loop, providing key training samples for the subsequent self-learning mechanism of the AI ​​model, thereby improving the system's adaptability and optimization accuracy under complex operating conditions.

[0084] S42, the comparative analysis report includes a comparison of the deviations in combustion efficiency, emission concentration and economic indicators before and after the adjustment, and generates parameter optimization suggestions based on sensitivity analysis.

[0085] Specifically, the system performs manual intervention and secondary prediction, and its technical implementation is based on a real-time response mechanism and a data closed-loop feedback structure using a multi-objective optimization model. Operators can fine-tune the blending ratio or operating parameters (such as air-coal ratio, air supply temperature, and coal mill current) output by the AI ​​algorithm engine layer in the human-machine collaborative window at the decision-making and visualization layer. The system receives the manually adjusted parameter input through the algorithm management interface and uses it as new constraints to re-trigger the calculation processes of the blending algorithm module and the combustion prediction algorithm module.

[0086] Specifically, the system employs a multi-algorithm parallel mechanism, automatically selecting one or more of genetic algorithms, particle swarm optimization algorithms, or reinforcement learning algorithms for secondary optimization based on the current operating conditions (such as boiler load, coal quality fluctuations, and equipment status). During the optimization process, the system compares manually adjusted parameters with the original AI suggestions, calculating their impact on combustion efficiency, emission concentration, and economic indicators. For example, the formula for calculating combustion efficiency is: ; in, To ensure the boiler effectively outputs heat, This refers to the total calorific value of the input fuel. Emission concentrations are collected via an online monitoring system (CEMS). , Real-time data such as dust concentration are quantitatively evaluated, and their rate of change can be expressed as: .

[0087] Furthermore, after completing the secondary prediction, the system generates a comparative analysis report. The report includes a comparison of the deviations of various indicators before and after the adjustment, such as combustion efficiency deviation not exceeding ±2%, emission concentration deviation controlled within ±5%, and economic indicators (such as unit power generation cost) deviation not exceeding ±1.5%. These indicators are set according to the "Guidelines for Feasible Technologies for Pollution Prevention and Control in Thermal Power Plants" (HJ 2301-2018) and the "Energy Efficiency Standard for Coal-fired Power Generating Units" (GB / T 30310-2013) to ensure that the system output complies with industry standards.

[0088] Specifically, in practical applications, this step is often used to address scenarios such as sudden changes in coal quality, equipment malfunctions, or adjustments to operating strategies. For example, when the proportion of high-sulfur coal blending increases, the operator can manually reduce the air-coal ratio to control the combustion temperature, and the system will then re-predict based on the adjusted parameters. The system analyzes emission trends and generates comparative reports for decision-making reference. Through this step, the system achieves a closed-loop optimization process of "algorithm suggestion - manual correction - effect verification," significantly improving the adaptability and execution accuracy of blending strategies and providing key support for intelligent decision-making in thermal power plants under complex coal types and operating conditions.

[0089] S5, by recording human adjustment behavior and combustion feedback data, drives the AI ​​model to automatically update weight parameters and optimize algorithm combinations, forming a closed-loop self-learning mechanism.

[0090] Specifically, the core technical principle of this step lies in using combustion feedback data and human intervention records to build an online learning framework for the model, thereby continuously optimizing the weight parameters of the AI ​​model and improving its adaptability and prediction accuracy under complex working conditions.

[0091] Specifically, the system first obtains actual combustion performance data from the operation feedback and comparison module, including boiler combustion temperature distribution and pollutant emission concentrations (such as...). , Key indicators such as dust concentration, ash accumulation and slagging status, main steam pressure, and air-coal ratio are used. These data are aligned with timestamps and compared with the combustion results predicted in the AI ​​algorithm engine layer (30) to calculate the prediction error. Error assessment can use standard indicators such as mean square error (MSE) or weighted average absolute error (MAE), for example: ; in, Indicates the first The loss value of the second optimization iteration. These are the model's predicted values. These are actual measured values. The number of samples is given. Based on this loss function, the system uses machine learning optimization methods such as gradient descent and Adam optimizer to update the weight matrix of the AI ​​model, thereby improving the model's fitting ability and generalization performance.

[0092] Furthermore, the system supports configuring hyperparameters such as learning rate, batch size, and number of epochs to adapt to the operating rhythm and data update frequency of different power plants. For example, the learning rate can be set to... to Between these, the batch size can be set to or To balance computational efficiency with model convergence speed.

[0093] Specifically, in practical applications, this step can be deployed in the fuel management system of a thermal power plant, and interface with systems such as DCS and CEMS in real time to achieve online training of the model and updating of strategies. Especially in operating environments with frequent fluctuations in coal quality, complex load changes, and strict environmental protection requirements, the system can automatically adjust the blending strategy through continuous learning, reduce the frequency of manual intervention, and improve combustion efficiency and emission control accuracy.

[0094] Specifically, this closed-loop self-learning mechanism significantly enhances the adaptive capability of the AI ​​model, enabling it to continuously approach the optimal solution during long-term operation and achieve a dynamic balance between economic efficiency, safety, and environmental protection. Simultaneously, through automatic updating of model parameters, the system possesses stronger robustness and generalization ability, providing solid technical support for intelligent fuel management in thermal power plants.

[0095] S6 generates a 3D visualization model and sensitivity analysis curves, displays the blending scheme, combustion status and boiler operating conditions in an interactive graphical interface, and updates the sensitivity analysis results in real time based on user-adjusted weights.

[0096] Specifically, the technical implementation of this step is based on the output of a multi-objective optimization model, combined with data visualization and dynamic response mechanisms, to achieve multi-dimensional perception and intelligent feedback of the coal blending process in thermal power plants.

[0097] Specifically, the system first maps multiple blending schemes output by the AI ​​algorithm engine layer into visual elements, including coal quality prediction curves, combustion efficiency trend charts, emission concentration change curves, and economic indicator comparison charts. The 3D visualization module uses graphics engines such as Unity or WebGL to construct a 3D model of the boiler combustion chamber. Through dynamic shading and particle simulation technology, it intuitively displays the combustion temperature field, pollutant diffusion paths, and slagging risk areas under different blending ratios. The sensitivity analysis module, based on the gradient information of the multi-objective optimization model, calculates the sensitivity of each parameter (such as ash content, sulfur content, volatile matter, air-coal ratio, etc.) to economic, safety, and environmental objectives, forming a... The loss function and weight mapping relationship shown are as follows: Indicates the first One blending parameter, For the first Each target weight.

[0098] Furthermore, the system allows users to dynamically adjust target weights within the interactive interface, for example, increasing the economic weight from 0.4 to 0.6 while maintaining the total weight of safety and environmental friendliness at 1. The sensitivity analysis module calculates the partial derivatives of each parameter with respect to the objective function in real time to quantify their impact. The system uses Z-score standardization to process input variables, ensuring comparability of parameters with different dimensions in sensitivity analysis. In addition, the output accuracy of the combustion prediction model is required to be within ±2%, and the update delay of the sensitivity curve is controlled within 500ms to meet real-time interaction requirements.

[0099] Specifically, this step is widely used in scenarios such as fuel scheduling, combustion optimization, and environmental control in thermal power plants. Operators can adjust the weights using drag-and-drop controls or numerical input boxes based on the system's recommended blending schemes. The system provides real-time feedback on the changing trends of key indicators such as combustion efficiency, NOx emissions, and operating costs, assisting them in making decisions that are more in line with on-site conditions.

[0100] Specifically, through an interactive graphical interface and real-time sensitivity analysis, the interpretability of the blending scheme and the efficiency of operator decision-making are significantly improved. Simultaneously, by recording user adjustment behaviors and sensitivity responses, the system provides training samples for the AI ​​model's weight self-learning mechanism, thereby achieving closed-loop optimization and improving the model's adaptability and prediction accuracy under complex operating conditions.

[0101] This invention discloses a multi-objective collaborative optimization method for coal blending in thermal power plants. By constructing a unified feature database and establishing a multi-objective dynamic optimization model, it generates candidate blending schemes using a multi-algorithm parallel computing architecture. Combined with a human-computer interaction feedback mechanism and a real-time data-driven self-learning strategy, it effectively solves the problems of insufficient optimization accuracy and poor adaptability in existing technologies caused by rigid models, singular objectives, and a lack of closed-loop iteration. It achieves collaborative decision-making throughout the entire process, from multi-source data fusion, dynamic weight optimization, intelligent scheme generation to autonomous updating of model parameters. This significantly improves the overall collaborative optimization level of blending schemes in terms of economic efficiency, safety, and environmental protection, and enhances the system's adaptive decision-making ability and long-term operational efficiency when facing complex coal quality and variable operating conditions.

[0102] Example 2 To achieve the above invention, embodiments of the present invention also provide a multi-objective collaborative optimization system for coal blending in thermal power plants, comprising: The purpose of this invention is to provide a system that, through comprehensive analysis of coal quality characteristics, coal mill performance, and unit operating constraints, can automatically or assistedly calculate the optimal blending ratio and combustion strategy, forming a closed-loop system of "algorithm-driven + data-supported + interactive decision-making". This system aims to solve the practical problems of existing thermal power plant blending systems, which often suffer from scattered data, simple algorithms, lack of self-learning capabilities, and inability to support dynamic decision-making needs under complex operating conditions.

[0103] To achieve the above objectives, the embodiments of the present invention adopt the following solutions: The system architecture consists of five layers of modules (such as...) Figure 2 As shown in the figure, it includes a data acquisition and preprocessing layer, a data hub and feature analysis layer, an AI+ algorithm engine layer, a decision-making and visualization layer, and a human-machine collaboration and feedback learning layer.

[0104] Specifically, the data acquisition and preprocessing layer (01) consists of a coal quality testing system (011), a DCS system interface (012), an emission monitoring interface (013), and an economic data interface (014). It is responsible for collecting coal quality, equipment, emission, and economic data, and transmitting the data to the data center through internationally standardized protocols such as TCP / IP.

[0105] Specifically, the data hub and feature analysis layer (02) includes a data cleaning module (021), a feature extraction module (022), a data standardization module (023), and a feature database (024). The key output variables include: coal quality feature vector, combustion feature parameters, equipment constraint indicators, etc., to realize feature extraction and correlation analysis of the collected data, set the weights of the three major objectives of economy, safety and environmental protection, ensure multi-source data fusion and feature encoding, and facilitate subsequent data utilization.

[0106] Specifically, the AI+ algorithm engine layer (03) is the core algorithm module of the system, including multi-algorithm and multi-objective weight optimization; the functional algorithms include: coal blending quality algorithm (031), blending algorithm (032), combustion prediction algorithm (033), and operation optimization algorithm (034); the algorithm module includes a genetic algorithm module, a particle swarm optimization module, a reinforcement learning module, and a weight self-learning module; each module achieves switching, parallel computing, and weight comparison through the algorithm management interface (035). The AI+ algorithm engine automatically performs parallel optimization, calculates multiple sets of candidate blending schemes, and each set of schemes corresponds to different weight combinations and operating constraints.

[0107] Specifically, the decision-making and visualization layer (04) includes a prediction display module (041), a sensitivity analysis module (042), a three-dimensional visualization module (043), and a result evaluation and reporting module (044). The system displays multi-dimensional prediction results such as economic efficiency, emissions, and safety, as well as coal blending prediction results through the interface, and generates blending trend curves and comparison charts of economic benefits or other optional targets in real time.

[0108] Specifically, the human-machine collaboration and feedback learning layer (05) includes a human-machine collaboration window (051), where operators can adjust blending schemes or operating parameters based on system suggestions and experience. The system records the results of human intervention and compares them with actual combustion data, automatically learning operational patterns through AI + database to form an experience database. This ensures that the model's adaptive accuracy continuously improves after long-term operation, forming a closed-loop learning system of "algorithm-human-feedback".

[0109] Furthermore, the core logical steps of the system (such as...) Figure 3 As shown below: Step S101: Data acquisition and synchronization. The system automatically collects information from coal quality, equipment and emission data sources, and stores it in a unified data center after cleaning.

[0110] Step S102: Feature analysis and weight initialization. Perform feature extraction and correlation analysis on the collected data to prepare for data input.

[0111] Step S103: Set blending targets, and set the weights of the three major targets: economy, safety, and environmental protection.

[0112] Step S104: AI+ Intelligent Blending Calculation. The algorithm engine automatically performs parallel optimization, calculating multiple candidate blending schemes based on multiple algorithms. Each scheme corresponds to different weight combinations and running constraints.

[0113] Step S105: Result prediction and sensitivity analysis. The system outputs trend curves of coal quality prediction, combustion efficiency, emission level and economic indicators. At the same time, it provides a sensitivity analysis interface. When the user adjusts the weights (for example, increase the economic weight), the system displays the trend changes of various optimization objectives in real time, and finally forms the experience of the impact of parameters on different objectives, and optimizes the AI ​​self-learning closed-loop system.

[0114] Step S106: Manual intervention and secondary prediction. Users can fine-tune the blending ratio or operating parameters. The system will recalculate the predicted values ​​in real time and generate a comparative analysis report.

[0115] Step S107: Result Evaluation and Export. The system generates a comprehensive report including coal quality prediction, combustion results, operational optimization control, manual experience adjustments, pollution control, and economic analysis. The report is exported as a PDF or Excel file, and includes a scoring system and blending recommendations. Step S108: AI weight self-learning and continuous optimization. The system automatically updates the AI ​​model weights based on actual combustion effect, unit operating parameters, emission feedback, and ash accumulation status, realizing self-learning and rolling model optimization.

[0116] This invention discloses a multi-objective collaborative optimization system for coal blending in thermal power plants. By constructing a multi-layered collaborative architecture integrating multi-source data acquisition, unified feature analysis, multi-algorithm intelligent optimization, interactive decision-making, and feedback learning, and employing dynamic weight adjustment and closed-loop self-learning mechanisms, it effectively solves the problems of insufficient optimization accuracy and poor dynamic decision-making capabilities in existing technologies caused by data silos, single algorithms, and a lack of adaptive iteration. It achieves closed-loop intelligent decision-making throughout the entire process, from data perception, feature extraction, multi-objective collaborative optimization to autonomous updating of model parameters. This significantly improves the overall collaborative optimization level and execution effect of blending schemes in terms of economic, safety, and environmental objectives, and enhances the system's adaptability, robustness, and long-term operational efficiency under complex coal quality and variable operating conditions.

[0117] Example 3 To achieve the above invention, embodiments of the present invention also provide an application scenario for a multi-objective collaborative optimization system for coal blending in thermal power plants, including: Specifically, a multi-objective collaborative optimization system for coal blending in thermal power plants achieves automatic or assisted calculation of the optimal blending ratio and combustion strategy by comprehensively analyzing coal quality characteristics, coal mill performance, and unit operation constraints. This forms a closed-loop system of "algorithm-driven + data-supported + interactive decision-making," addressing the practical problems of existing thermal power plant blending systems, such as scattered data, single algorithms, lack of self-learning capabilities, and inability to support dynamic decision-making needs under complex operating conditions.

[0118] Specifically, a multi-objective collaborative optimization system for coal blending in thermal power plants has a system architecture consisting of five modules, including a data acquisition and preprocessing layer, a data hub and feature analysis layer, an AI+algorithm engine layer, a decision-making and visualization layer, and a human-machine collaboration and feedback learning layer.

[0119] Specifically, the data acquisition and preprocessing layer (01) consists of a coal quality testing system (011), a DCS system interface (012), an emission monitoring interface (013), and an economic data interface (014). It is responsible for collecting coal quality, equipment, emission, and economic data, and transmitting the data to the data center through internationally standardized protocols such as TCP / IP.

[0120] Furthermore, the coal quality testing system (011) collects data from fuel laboratories or online coal quality analyzers, including industrial and elemental analysis indicators of coal, such as ash content, volatile matter, total sulfur, moisture, calorific value, ash fusion point, sodium, and potassium content (Na2O, K2O). Data sources can be laboratory LIMS systems or online near-infrared spectrometers, X-ray fluorescence spectrometers (XRF), etc. This module has automatic data synchronization and anomaly detection functions; when out-of-limit or missed data occurs, data verification and default value repair can be triggered.

[0121] Furthermore, the DCS system interface (012) is used to collect real-time operating data of the boiler and coal mill system, including coal mill current, output, inlet temperature, air supply temperature, air-to-coal ratio, main steam pressure, and combustion temperature distribution. Data acquisition is achieved through industry standard protocols such as OPC-UA or Modbus-TCP, maintaining millisecond-level synchronization with the plant's DCS (Distributed Control System). The system supports data stream buffering and breakpoint resume to ensure data integrity is maintained even during network fluctuations.

[0122] Furthermore, the emission monitoring interface (013) collects emission data provided by the online monitoring system (CEMS), including SO2 and NO. x The system tracks data such as dust concentration, fly ash alkali metal content, flue gas temperature, and O2 concentration. It automatically timestamps emission data with coal quality and combustion status data, enabling combustion-emission correlation tracking. The system incorporates data smoothing and anomaly removal algorithms to eliminate the impact of sensor drift or short-term disturbances on model training.

[0123] Furthermore, the economic data interface (014) collects economic dispatch data such as electricity price, fuel procurement cost, inventory quantity, fuel supply plan, and unit output load curve; it supports data docking with the fuel management system to provide real-time parameter input for subsequent economic target optimization.

[0124] Specifically, the data hub and feature analysis layer (02) includes a data cleaning module (021), a feature extraction module (022), a data standardization module (023), and a feature database (024). The key output variables include: coal quality feature vector, combustion feature parameters, equipment constraint indicators, etc., to realize feature extraction and correlation analysis of the collected data, set the weights of the three major objectives of economy, safety and environmental protection, ensure multi-source data fusion and feature encoding, and facilitate subsequent data utilization.

[0125] Furthermore, the data cleaning module (021) is responsible for verifying and filtering data from the coal quality detection system, DCS, CEMS, and economic management system in the previous module, performing operations such as data integrity verification, duplicate item removal, outlier detection, and time series repair. Outlier detection automatically identifies out-of-limit, breakpoint, and drifting data, and stores the abnormal data for manual review. Data cleaning supports multi-level cleaning strategies.

[0126] Furthermore, the feature extraction module (022) is responsible for extracting key influencing factors for algorithm calculation and AI modeling from the cleaned data, performing feature transformation on the coal quality data, and calculating derived features such as ash fusion point. Various dynamic features are also extracted from equipment operation data, emission data, etc.

[0127] Furthermore, the data standardization module (023) performs unified encoding and dimensional conversion on multi-source heterogeneous data to ensure consistency of feature inputs. It supports normalization, Z-score standardization, and temporal smoothing filtering.

[0128] Furthermore, the feature database (024) is used to store high-quality datasets after feature processing, providing standardized input data for algorithm computation and AI+ algorithms.

[0129] Specifically, the AI+ algorithm engine layer (03) is the core algorithm module of the system, including multi-algorithm and multi-objective weight optimization; the functional algorithms include: coal blending quality algorithm (031), blending algorithm (032), combustion prediction algorithm (033), and operation optimization algorithm (034); the algorithm module includes a genetic algorithm module, a particle swarm optimization module, a reinforcement learning module, and a weight self-learning module; each module achieves switching, parallel computing, and weight comparison through the algorithm management interface (035). The AI+ algorithm engine automatically performs parallel optimization, calculates multiple sets of candidate blending schemes, and each set of schemes corresponds to different weight combinations and operating constraints.

[0130] Furthermore, the coal blending algorithm module (031) includes linear coal blending calculation and nonlinear coal blending calculation, and the blending algorithm module provides the input coal quality feature basis.

[0131] Furthermore, the blending algorithm module (032) includes various blending algorithms, including genetic algorithm, particle swarm optimization algorithm, reinforcement learning algorithm, etc., to establish a multi-objective optimization combination model, and to screen feasible solutions by combining constraints (boiler load, combustion temperature, equipment upper limit).

[0132] Furthermore, the combustion prediction algorithm module (033) predicts the combustion performance of the boiler after blending based on AI regression and reinforcement learning models. The model inputs include coal quality characteristics, combustion parameters (air-coal ratio, furnace temperature, coal mill output, etc.) and historical emission data, and support sensitivity analysis and parameter inversion.

[0133] Furthermore, the operation optimization algorithm module (034) refines and optimizes the unit operating parameters under the candidate blending scheme. It adopts a reinforcement learning algorithm and trains the strategy network through historical combustion feedback to gradually approach the optimal operating parameter set, which can maintain the boiler combustion stability and efficiency under different loads and coal quality changes.

[0134] Furthermore, the algorithm management interface module (035) is used to coordinate the running order and data interaction of each algorithm module, realize multi-algorithm parallelism and result comparison, and simultaneously realize algorithm switching, parallel computation and weight comparison. After reinforcement learning, the optimal algorithm combination can be automatically selected according to the running scenario.

[0135] Specifically, the decision-making and visualization layer (04) includes a prediction display module (041), a sensitivity analysis module (042), a three-dimensional visualization module (043), and a result evaluation and reporting module (044). The system displays multi-dimensional prediction results such as economic efficiency, emissions, and safety, as well as coal blending prediction results through the interface, and generates blending trend curves and comparison charts of economic benefits or other optional targets in real time.

[0136] Furthermore, the prediction display module (041) is used to display the prediction results of the blending scheme calculated by the AI ​​algorithm engine, including multi-dimensional data such as coal quality indicators, combustion efficiency, pollutant emission concentration and operating cost, and supports real-time curve display, tabular output and trend comparison functions.

[0137] Furthermore, the sensitivity analysis module (042) is used to analyze the impact of different target weights (economic efficiency, safety, and environmental friendliness) on the blending results. It supports displaying sensitivity curves of parameter changes on cost, efficiency, and emissions, forming a visualized sensitivity analysis curve; Furthermore, the three-dimensional visualization module (043) displays the blending scheme, combustion status and boiler operating conditions in a three-dimensional model and adopts an interactive graphical interface, allowing users to view the combustion and emission prediction results under different blending ratios; Furthermore, the results evaluation and reporting module (044) comprehensively scores and assesses the risks of multiple blending schemes output by the AI ​​algorithm engine. The evaluation dimensions include: combustion stability, emission compliance, economic benefits, and implementation feasibility. The report can be exported as PDF, Excel, or image format for internal technical archiving and decision recording.

[0138] Specifically, the human-machine collaboration and feedback learning layer (05) includes a human-machine collaboration window (051), a manual intervention recording module (052), and an operation feedback and comparison module (053). Operators can adjust the blending scheme or operating parameters based on experience and system suggestions. The system records the results of manual intervention and compares them with actual combustion data. Through AI + database, it automatically learns the operating rules and forms an experience database. This ensures that the model's adaptive accuracy continuously improves after long-term operation, forming a closed-loop learning system of "algorithm-manual-feedback".

[0139] Furthermore, the human-machine collaboration window (051) provides operators with an intuitive interactive interface for viewing the blending suggestions and combustion prediction results output by the AI ​​algorithm engine. Operators can make manual adjustments based on their own operating experience or on-site monitoring data, and can also conduct human-machine interactive Q&A based on historical data and knowledge graphs.

[0140] Furthermore, the manual intervention recording module (052) can record the operator's relevant adjustment operations, including adjustment time, parameter change value, adjustment reason and result number, and bind it to the current status, establish a correlation with the corresponding blending scheme, operating cycle and combustion data, and form a complete data chain of "manual intervention - result verification - feedback correction".

[0141] Furthermore, the operation feedback and comparison module (053) compares the adjusted combustion effect and operation effect, and compares the deviation of the prediction results. If the deviation is significant, the model self-learning mechanism is triggered in combination with parameter sensitivity analysis and the model parameters under the corresponding load conditions are updated.

[0142] Specifically, a multi-objective collaborative optimization system for coal blending in thermal power plants comprises the following core logical steps: Step S101: Data Acquisition and Synchronization. The system automatically collects information from coal quality, equipment, and emission data sources, cleans it, and stores it in a unified data center. Step S102: Feature Analysis and Weight Initialization. Feature extraction and correlation analysis are performed on the collected data to prepare for data input. Step S103: Blending Target Setting. The weights of three major objectives—economic efficiency, safety, and environmental protection—are set. Step S104: AI+ Intelligent Blending Calculation. The algorithm engine automatically executes parallel optimization, calculating multiple candidate blending schemes based on multiple algorithms. Each scheme corresponds to different weight combinations and operating constraints. Step S105: Result Prediction and Sensitivity Analysis. The system outputs trend curves for coal quality prediction, combustion efficiency, emission levels, and economic indicators. It also provides a sensitivity analysis interface. When the user adjusts the weights (e.g., increasing the economic weight), the system displays the real-time trend changes of each optimization objective, ultimately forming experience on the impact of parameters on different objectives and optimizing the AI ​​self-learning closed-loop system. Step S106: Manual Intervention and Secondary Prediction. Users can fine-tune the blending ratio or operating parameters. The system instantly recalculates the predicted values ​​and generates a comparative analysis report. Step S107: Result Evaluation and Export. The system generates a comprehensive report including coal quality prediction, combustion results, operational optimization control, manual experience adjustments, pollution control, and economic analysis. The report is exported as a PDF or Excel file and includes a scoring system and blending recommendation levels. Step S108: AI Weight Self-Learning and Continuous Optimization. Based on actual combustion effects, unit operating parameters, emission feedback, and ash accumulation / slagging status, the system automatically updates the AI ​​model weights, achieving self-learning and rolling model optimization.

[0143] This invention discloses an application scenario of a multi-objective collaborative optimization system for coal blending in thermal power plants. By constructing a multi-layered collaborative architecture integrating multi-source data acquisition, unified feature analysis, multi-algorithm intelligent optimization, interactive decision-making, and feedback learning, and employing dynamic weight adjustment and a closed-loop self-learning mechanism, it effectively solves the problems of insufficient optimization accuracy and dynamic decision-making difficulties caused by scattered data, single algorithms, and lack of self-learning capabilities in existing technologies. It achieves closed-loop intelligent decision-making throughout the entire process, from multi-source data fusion, feature extraction, multi-objective collaborative optimization to autonomous updating of model parameters, forming a collaborative optimization system of "algorithm-driven + data-supported + interactive decision-making." This significantly improves the overall collaborative optimization level and execution accuracy of blending schemes in terms of economic, safety, and environmental objectives, and enhances the system's adaptive decision-making ability, robustness, and long-term operational efficiency under complex coal quality and variable operating conditions.

[0144] Example 4 To achieve the above invention, such as Figure 4 As shown, this embodiment also provides a multi-objective collaborative optimization device 10 for coal blending in thermal power plants. The device 10 includes: The data acquisition and preprocessing module 100 is used to acquire coal quality characteristics, equipment operating parameters, emission monitoring data and economic dispatch information of thermal power plants, and generate a unified feature database through data cleaning and standardization processing.

[0145] The weight configuration and modeling module 200 is used to set the initial weights for three types of objectives—economic efficiency, safety, and environmental protection—based on a unified feature database, and to establish a multi-objective optimization model.

[0146] The multi-algorithm parallel computing module 300 is used to execute a multi-objective optimization model using a multi-algorithm parallel computing architecture to generate multiple sets of candidate blending schemes containing different weight combinations and running constraints.

[0147] The manual adjustment and secondary prediction module 400 is used to perform secondary prediction and generate a comparative analysis report based on the operator's manual adjustment results of the candidate blending scheme and the actual combustion effect data.

[0148] The closed-loop self-learning control module 500 is used to drive the AI ​​model to automatically update weight parameters and optimize algorithm combinations by recording manual adjustment behavior and combustion feedback data, thus forming a closed-loop self-learning mechanism.

[0149] In one embodiment of the present invention, it further includes: a visualization and sensitivity analysis module, used to generate a three-dimensional visualization model and sensitivity analysis curves, display the blending scheme, combustion state and boiler operating conditions in an interactive graphical interface, and update the sensitivity analysis results in real time based on user-adjusted weights.

[0150] This invention discloses a multi-objective collaborative optimization device for coal blending in thermal power plants. By integrating data acquisition and preprocessing, weight configuration and modeling, parallel computation of multiple algorithms, manual adjustment and secondary prediction, and a closed-loop self-learning control module, it constructs a complete technology chain from data fusion, multi-objective dynamic optimization, intelligent scheme generation to autonomous model updating. The device employs a multi-algorithm parallel architecture and a human-machine collaborative feedback mechanism, effectively solving the problems of insufficient optimization accuracy and decision lag caused by isolated data, rigid models, and lack of adaptive iteration in existing technologies. It achieves deep fusion of multi-source information, collaborative optimization of economic, safety, and environmental objectives, and closed-loop self-evolution of system parameters, significantly improving the overall accuracy of blending decisions, the system's adaptability under complex operating conditions, and long-term operational efficiency.

[0151] To implement the methods of the above embodiments, the present invention also provides a computer device, such as... Figure 5As shown, the computer device 600 includes a memory 601 and a processor 602; wherein, the processor 602 reads the executable program code stored in the memory 601 to run a program corresponding to the executable program code, so as to implement the various steps of the multi-objective collaborative optimization method for coal blending in thermal power plants described above.

[0152] To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a multi-objective collaborative optimization method for coal blending in thermal power plants as described in the foregoing embodiments.

[0153] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0154] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A multi-objective collaborative optimization method for coal blending in thermal power plants, characterized in that, include: S1 acquires coal quality characteristics, equipment operating parameters, emission monitoring data and economic dispatch information of thermal power plants, and generates a unified feature database through data cleaning and standardization. S2. Based on a unified feature database, initial weights are set for three categories of objectives: economy, safety, and environmental protection, and a multi-objective optimization model is established. S3 utilizes a multi-algorithm parallel computing architecture to execute a multi-objective optimization model, generating multiple sets of candidate blending schemes containing different weight combinations and operational constraints. S4. Based on the operator's manual adjustments to the candidate blending schemes, combined with actual combustion effect data, a secondary prediction is made and a comparative analysis report is generated. S5, by recording human adjustment behavior and combustion feedback data, drives the AI ​​model to automatically update weight parameters and optimize algorithm combinations, forming a closed-loop self-learning mechanism.

2. The method as described in claim 1, characterized in that, The process of acquiring coal quality characteristics, equipment operating parameters, emission monitoring data, and economic dispatch information from thermal power plants, and generating a unified feature database through data cleaning and standardization, includes: S11, Data cleaning includes performing integrity verification, duplicate item removal, anomaly detection, and time series repair on data collected from coal quality testing systems, DCS system interfaces, emission monitoring interfaces, and economic data interfaces; S12, the standardization process adopts the Z-Score standardization and normalization method to uniformly encode and transform the dimensions of multi-source heterogeneous data, ensuring the consistency of feature input.

3. The method as described in claim 1, characterized in that, The method establishes a multi-objective optimization model by setting initial weights for three categories of objectives—economic efficiency, safety, and environmental protection—based on a unified feature database, including: S21, the initial weights are set based on historical combustion data and real-time coal quality fluctuations, and the allocation ratio of economic, safety and environmental protection weights is dynamically adjusted through a reinforcement learning model. S22, the multi-objective optimization model includes constraints, specifically the upper limit of boiler load, the combustion temperature threshold, and the boundary values ​​of equipment operating parameters.

4. The method as described in claim 1, characterized in that, The method utilizes a multi-algorithm parallel computing architecture to execute a multi-objective optimization model, generating multiple sets of candidate blending schemes containing different weight combinations and operational constraints, including: S31, a multi-algorithm parallel computing architecture, includes a genetic algorithm module, a particle swarm optimization module, and a reinforcement learning module, and can dynamically switch algorithm combinations through an algorithm management interface; S32, the generation of candidate blending schemes is based on parallel computing results. Each scheme corresponds to different weight combinations and running constraints, and the results are compared through the algorithm management interface.

5. The method as described in claim 1, characterized in that, The process of generating a comparative analysis report based on the operator's manual adjustments to the candidate blending schemes, combined with actual combustion performance data, includes: S41, the results of the manual adjustment include fine-tuning parameters for blending ratio, air-coal ratio, air supply temperature and coal mill current; S42, the comparative analysis report includes a comparison of the deviations in combustion efficiency, emission concentration and economic indicators before and after the adjustment, and generates parameter optimization suggestions based on sensitivity analysis.

6. The method as described in claim 1, characterized in that, Also includes: S6 generates a 3D visualization model and sensitivity analysis curves, displays the blending scheme, combustion status and boiler operating conditions in an interactive graphical interface, and updates the sensitivity analysis results in real time based on user-adjusted weights.

7. A multi-objective synergistic optimization device for coal blending in thermal power plants, characterized in that, include: The data acquisition and preprocessing module is used to acquire coal quality characteristics, equipment operating parameters, emission monitoring data and economic dispatch information of thermal power plants, and generate a unified feature database through data cleaning and standardization. The weight configuration and modeling module is used to set the initial weights for three categories of objectives—economic efficiency, safety, and environmental protection—based on a unified feature database, and to establish a multi-objective optimization model. The multi-algorithm parallel computing module is used to execute a multi-objective optimization model using a multi-algorithm parallel computing architecture, generating multiple sets of candidate blending schemes containing different weight combinations and running constraints. The manual adjustment and secondary prediction module is used to perform secondary prediction and generate a comparative analysis report based on the operator's manual adjustment results of the candidate blending scheme and the actual combustion effect data. The closed-loop self-learning control module is used to drive the AI ​​model to automatically update weight parameters and optimize algorithm combinations by recording manual adjustment behavior and combustion feedback data, thus forming a closed-loop self-learning mechanism.

8. The apparatus as claimed in claim 7, characterized in that, Also includes: The visualization and sensitivity analysis module is used to generate three-dimensional visualization models and sensitivity analysis curves. It displays the blending scheme, combustion status and boiler operating conditions in an interactive graphical interface, and updates the sensitivity analysis results in real time based on user-adjusted weights.

9. An electronic device, comprising: processor; The memory stores executable instructions; when the processor executes the instructions, it implements the multi-objective collaborative optimization method for coal blending in thermal power plants as described in any one of claims 1-6.

10. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements a multi-objective collaborative optimization method for coal blending in thermal power plants as claimed in any one of claims 1-6.

Images