Simulation data processing system for coal gasification device based on multi-source data fusion

Abstract

The invention discloses a multi-source data fusion coal gasification device simulation data processing system, and relates to the technical field of chemical process simulation optimization, the system comprises a simulation management platform, the simulation management platform is in communication connection with the following modules: a data acquisition and processing module, which is used for collecting multi-source data in the operation process of a coal gasification device, comprising simulation data from a traditional physical simulation model and real-time operation data of a real device, and the collected multi-source data is preprocessed and integrated to form a comprehensive data set. According to the multi-source data fusion coal gasification device simulation data processing system, the problem of dynamic mismatch between a traditional simulation model and a real coal gasification device is effectively solved through a collaborative optimization mechanism of a deep Q network and a fruit fly algorithm, and a virtual-real difference penalty term is added in a reward function by using the deep Q network; and the difference between simulation and reality is fully considered in the model learning process.

Description

Technical Field

[0001] This invention relates to the field of chemical process simulation and optimization technology, specifically to a multi-source data fusion simulation data processing system for coal gasification plants. Background Technology

[0002] Coal gasification is a chemical process that converts coal into usable energy sources such as syngas (mainly composed of hydrogen, carbon monoxide, carbon dioxide, and methane). The reactors in coal gasification plants typically operate under extreme high temperature and high pressure conditions. Coal gasification is a highly complex chemical reaction process. Through simulation data processing, the reaction process under different operating conditions can be simulated, and the changes in reactants and products can be analyzed. Based on the analysis of simulation data, key parameters such as reaction temperature, pressure, and gas composition can be optimized.

[0003] In existing technologies, traditional physical simulation models (Aspen Plus) cannot reflect the dynamic degradation of coal gasification units in real time, such as coking and catalyst deactivation. This leads to a significant decrease in efficiency of the optimal strategy trained in the simulation in the real unit, resulting in a dynamic mismatch between the simulation model and the real unit and poor transferability of efficiency optimization strategies. Therefore, how to combine deep Q-networks and the fruit fly algorithm to solve the dynamic mismatch and improve the overall efficiency of coal gasification units is the problem that this invention aims to solve. To this end, a multi-source data fusion coal gasification unit simulation data processing system is proposed. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention is implemented through the following technical solution: a multi-source data fusion simulation data processing system for a coal gasification device, comprising a simulation management platform, wherein the simulation management platform is communicatively connected to the following modules:

[0005] The data acquisition and processing module is used to collect multi-source data during the operation of the coal gasification unit, including simulation data from traditional physical simulation models, covering simulated parameters of temperature, pressure and flow rate; and real-time operating data of the actual unit, covering the calorific value of syngas collected by sensors and the actual parameters of equipment operating status. The module performs preprocessing operations such as cleaning, denoising and normalizing on the collected multi-source data to remove outliers and noise interference, and integrates them into a comprehensive dataset.

[0006] A digital twin reward shaping module is used to combine deep Q-networks and the fruit fly algorithm to analyze the deviation between simulation data and real data, and to initialize the dual population of the fruit fly algorithm.

[0007] The collaborative optimization module, based on the co-evolutionary mechanism of the fruit fly algorithm, performs collaborative optimization of virtual and real parameters and uses the latest real data to quickly calibrate the simulation model;

[0008] The strategy generation and output module, based on the optimized deep Q-network model and combined with parameters that have undergone co-evolution, generates an efficiency optimization strategy for the coal gasification unit and outputs it to the control system of the coal gasification unit to guide the operation of the coal gasification unit.

[0009] The efficiency evaluation module is used to evaluate the operating efficiency of the coal gasification unit after applying the generation efficiency optimization strategy. By comparing the efficiency indicators before and after applying the efficiency optimization strategy, including syngas production and energy consumption, the module analyzes the effect of the efficiency optimization strategy on the efficiency of the coal gasification unit, so as to understand the actual effect of the current strategy and make further improvements.

[0010] Preferably, the data acquisition and processing module specifically includes:

[0011] Simulation parameters covering temperature, pressure, and flow rate are obtained from the traditional physical simulation model (Aspen Plus) as simulation data. At the same time, actual parameters of syngas calorific value and equipment operating status are collected by sensors during actual plant operation to ensure comprehensive data sources.

[0012] The collected multi-source data is preprocessed by data cleaning to remove outliers and noise interference, and denoising to further reduce interference components. The data is also normalized to unify data of different dimensions and ranges to the same standard, thereby ensuring the accuracy and reliability of the data.

[0013] The pre-processed multi-source data is integrated and aggregated into a comprehensive dataset by following a unified data format and structure, bringing together data from different sources and of different types.

[0014] Preferably, the digital twin reward shaping module includes a deep Q-network model optimization unit and a fruit fly algorithm initialization unit;

[0015] The deep Q-network model optimization unit is used to add a virtual-real difference penalty term to the reward function of the deep Q-network, that is, the deviation of the syngas calorific value between the simulation and the real device. By analyzing the deviation between the simulation data and the real data, it is added as a penalty term to the reward function.

[0016] The fruit fly algorithm initialization unit is used to initialize the dual populations of the fruit fly algorithm. Population 1 is used to optimize the reward weights of the deep Q network, i.e. the virtual-real difference penalty coefficient. By continuously adjusting the reward weights, the deep Q network can learn more effectively under the guidance of the reward function. Population 2 is used to optimize the compensation parameters of the real device. According to the operation of the real device, the compensation parameters are dynamically adjusted to better match the simulation model.

[0017] Preferably, the deep Q-network model optimization unit specifically includes:

[0018] Extract the simulated calorific value of syngas from the simulation data of the coal gasification unit from the comprehensive dataset, and the actual calorific value of syngas collected by the sensors of the real unit, and calculate the deviation between the simulation data and the real data.

[0019] Based on the deviation analysis results, combined with the operating characteristics and optimization objectives of the coal gasification unit, quantitative rules for the penalty items for virtual and real differences are set, clarifying the correspondence between the magnitude of the deviation and the intensity of the penalty, and ensuring that the penalty items can effectively reflect the impact of the deviation;

[0020] By incorporating the pre-defined virtual-real difference penalty term into the original reward function of the deep Q-network, a new reward function system is constructed, enabling the deep Q-network to fully consider the differences between simulation and reality during the learning process.

[0021] Preferably, the fruit fly algorithm initialization unit specifically includes:

[0022] Initialize the dual populations of the fruit fly algorithm, determine the size and dimension parameters of the dual populations, and set the variable range for population 1 to optimize the reward weights of the deep Q network according to the problem requirements; set the variable range for population 2 to optimize the compensation parameters of the real device, and randomly generate the initial dual populations.

[0023] Individuals from population 1 are fed into a deep Q-network, and the reward weights are evaluated based on the network's learning performance to calculate fitness. Individuals from population 2 are applied to a real device, and the compensation parameters are evaluated based on the operational matching degree to calculate fitness, thereby determining the quality of the individuals.

[0024] Select superior individuals based on fitness, update the dual population positions through the taste-finding operation of the fruit fly algorithm, adjust the reward weight of population 1 to help the network learn, and optimize the compensation parameter matching simulation of population 2. Repeat this process until the optimization conditions are met.

[0025] Preferably, the collaborative optimization module includes a fruit fly algorithm optimization unit and a model calibration unit;

[0026] The fruit fly algorithm optimization unit is used to utilize the co-evolution mechanism of the fruit fly algorithm, with population 1 optimizing the reward weights of the deep Q network and population 2 optimizing the compensation parameters of the real device. Through the odor concentration sharing mechanism, the virtual and real parameters are co-optimized to reduce efficiency loss after strategy migration and improve the operating efficiency of the coal gasification device in the real environment.

[0027] The model calibration unit activates the fruit fly algorithm every 10 minutes to quickly calibrate the simulation model using the latest real data. At the same time, it uses a deep Q-network real-time update strategy to solve the dynamic mismatch problem and ensure the dynamic consistency between the simulation model and the real device.

[0028] Preferably, the fruit fly algorithm optimization unit specifically includes:

[0029] Population 1 and Population 2 are constructed based on the fruit fly algorithm. Population 1 is used to optimize the reward weight of the deep Q network, and Population 2 optimizes the compensation parameters of the real device. According to the problem requirements, the range of variables for the two populations is set, and the initial dual population individuals are randomly generated.

[0030] Individuals from population 1 are fed into a deep Q-network to evaluate the reward weights based on the learning effect, which includes convergence speed and accuracy. Individuals from population 2 are used in a real device, and the compensation parameters are evaluated based on the operational matching degree. Then, the individual fitness is calculated, and information is exchanged through an odor concentration sharing mechanism.

[0031] Based on the fitness-first principle, select superior individuals from the dual populations, carry the optimal solution information, use the fruit fly algorithm for taste-finding operations, collaboratively update the positions of the dual populations, optimize reward weights and compensation parameters, and repeat until the optimization conditions are met, thereby improving the efficiency of operation in real-world environments.

[0032] Preferably, the model calibration unit specifically includes:

[0033] The system automatically collects the latest real data every 10 minutes, triggers the model calibration process, and starts the fruit fly algorithm, using the latest real data as input.

[0034] The fruit fly algorithm runs based on the latest real data, optimizes and adjusts key parameters in the simulation model, and quickly completes the calibration of the simulation model by iteratively searching for the optimal parameter combination, thus narrowing the gap between it and the real situation. The deep Q network updates the control strategy in real time based on the calibrated simulation model and real-time feedback to solve the dynamic mismatch problem, so that the control strategy can better adapt to changes and ensure that the simulation model and the real device always maintain dynamic consistency.

[0035] Preferably, the strategy generation output module specifically includes:

[0036] Based on the optimized deep Q-network model and combined with the parameters of co-evolution, an efficiency optimization strategy for the coal gasification unit is generated. The deep Q-network uses the calibrated model and real-time data to output the optimal control strategy, ensuring the scientific validity and effectiveness of the strategy.

[0037] The generated optimal control strategy is verified to ensure its feasibility and safety under the current operating conditions. Based on the verification results, the optimal control strategy is fine-tuned to better adapt to the actual operating environment and improve the reliability of the strategy.

[0038] The optimized control strategy is output to the control system of the coal gasification unit to guide the operation of the unit. The control system adjusts the operating parameters according to the strategy to ensure that the coal gasification unit operates in a high-efficiency and stable state, thereby improving the overall efficiency.

[0039] Preferably, the efficiency evaluation module specifically includes:

[0040] Relevant operational data before and after the application of efficiency optimization strategies are obtained from the coal gasification unit, including efficiency index data of syngas production and energy consumption, and then screened and cleaned to form a basic dataset for comparative analysis.

[0041] The efficiency index data before and after the efficiency optimization strategy are compared one by one. By calculating the increase in output and the reduction in energy consumption, the changes in the operating efficiency of the coal gasification unit after the implementation of the strategy are analyzed, and the direction of the impact of the efficiency optimization strategy on each efficiency index is clarified.

[0042] Based on the comparative analysis of efficiency indicators, the effect of efficiency optimization strategies on improving the operating efficiency of coal gasification units is quantified, and an efficiency evaluation report is generated to present the actual effectiveness of the efficiency optimization strategies.

[0043] This invention provides a multi-source data fusion simulation data processing system for coal gasification plants. It has the following beneficial effects:

[0044] (i) The multi-source data fusion coal gasification unit simulation data processing system effectively solves the dynamic mismatch problem between traditional simulation models and real coal gasification units through the collaborative optimization mechanism of deep Q network and fruit fly algorithm. The deep Q network adds a virtual-real difference penalty term to the reward function, prompting the model to fully consider the difference between simulation and reality during the learning process. The fruit fly algorithm further adjusts the simulation model to better match the operating characteristics of the real unit by optimizing the reward weight and compensation parameters, so that the simulation model can reflect the dynamic changes of the real unit in real time and significantly improve the consistency between the simulation model and the real unit.

[0045] (II) The multi-source data fusion simulation data processing system for coal gasification equipment utilizes the dual-population optimization strategy of the fruit fly algorithm. Population 1 optimizes the reward weights of the deep Q network, while population 2 optimizes the compensation parameters of the real equipment. Through the odor concentration sharing mechanism, the system achieves collaborative optimization of virtual and real parameters, which can effectively reduce the efficiency loss after strategy migration. This allows the optimal strategy trained in the simulation environment to be more stably transferred to the real equipment, thereby enhancing the adaptability and stability of strategy migration and reducing the risk of production efficiency decline and equipment failure caused by strategy mismatch.

[0046] (III) The multi-source data fusion coal gasification unit simulation data processing system can comprehensively collect simulation data and real data during the operation of the coal gasification unit through multi-source data fusion, and conduct in-depth analysis and optimization. The efficiency optimization strategy generated by the deep Q network combined with the optimized parameters can accurately guide the operation of the coal gasification unit, realize the fine control of key parameters, give full play to the performance potential of the coal gasification unit, and significantly improve the overall operating efficiency of the coal gasification unit. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the workflow of a coal gasification device simulation data processing system based on multi-source data fusion according to the present invention.

[0048] Figure 2 This is a data flow diagram of a multi-source data fusion simulation data processing system for a coal gasification device according to the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: a multi-source data fusion simulation data processing system for a coal gasification device, comprising a simulation management platform, the simulation management platform having the following communication connections:

[0051] As the core management and coordination hub of the system, the simulation management platform is responsible for communicating with each module, ensuring smooth data interaction between modules, coordinating the operation process of the entire system, and ensuring that each module works collaboratively according to the established logic to realize the overall function of the system.

[0052] The data acquisition and processing module is used to collect multi-source data during the operation of the coal gasification unit. This includes simulation data from traditional physical simulation models, covering simulated parameters of temperature, pressure, and flow rate; and real-time operating data from the actual unit, covering syngas calorific value collected by sensors and actual parameters of equipment operating status. The module performs preprocessing operations such as cleaning, denoising, and normalization on the collected multi-source data to remove outliers and noise interference, integrating the data into a comprehensive dataset. Simulation data, including temperature, pressure, and flow rate, is obtained from the traditional physical simulation model (Aspen Plus). Simultaneously, real-time parameters of syngas calorific value and equipment operating status are collected from sensors during actual unit operation, ensuring comprehensive data sources. The collected multi-source data undergoes preprocessing, including data cleaning to remove outliers and noise interference, further reducing interference components, and data normalization to unify data of different dimensions and ranges to the same standard, thus ensuring data accuracy and reliability. Finally, the preprocessed multi-source data is integrated according to a unified data format and structure, aggregating data from different sources and of different types to form a comprehensive dataset.

[0053] The specific tasks of the data acquisition and processing module are as follows: During the operation of the coal gasification unit, it comprehensively collects multi-source data. On the one hand, it obtains simulation parameters covering temperature, pressure, and flow rate from the traditional physical simulation model Aspen Plus as simulation data. This simulation data is generated based on theoretical models and preset conditions, reflecting the operating characteristics of the coal gasification unit under ideal conditions. On the other hand, it uses sensors to collect real-time parameters of the syngas calorific value and equipment operating status during actual unit operation, directly sensing the actual operating conditions of the coal gasification unit and truly reflecting its working status in the real environment. By simultaneously acquiring simulation data and actual parameters, the comprehensiveness of the data sources is ensured. After completing the collection of multi-source data, the collected data undergoes preprocessing. First, data cleaning is performed, and outliers are detected and removed to ensure the rationality and consistency of the data. Outliers originate from sensor malfunctions, data transmission errors, or simulation model errors. Deviations and outliers can mislead subsequent analysis and optimization. Noise reduction is performed to further reduce interference and improve data purity. Noise originates from random sensor errors, environmental interference, or approximation errors in simulation models. Finally, data normalization is applied to unify data of different dimensions and ranges to the same standard, eliminating dimensional and numerical range differences and ensuring accuracy and reliability. Following a unified data format and structure, data from different sources and of different types are aggregated and integrated. The preprocessed multi-source data is then integrated to form a comprehensive dataset. This comprehensive dataset not only preserves the complete information of both simulation and real data but also improves data operability and usability through a unified data format and structure.

[0054] The digital twin reward shaping module is used to combine deep Q-networks and the fruit fly algorithm to analyze the deviation between simulation data and real data, and to initialize the dual population of the fruit fly algorithm. The digital twin reward shaping module includes a deep Q-network model optimization unit and a fruit fly algorithm initialization unit.

[0055] The Deep Q-Network Model Optimization Unit adds a virtual-to-real difference penalty term to the reward function of the Deep Q-Network, namely the deviation of the syngas calorific value between the simulation and the real device. By analyzing the deviation between the simulation data and the real data, this deviation is added to the reward function as a penalty term. The simulated syngas calorific value of the coal gasification unit in the simulation data and the actual syngas calorific value collected by the real device through sensors are extracted from the comprehensive dataset. The deviation between the simulation data and the real data is calculated. Based on the deviation analysis results, combined with the operating characteristics of the coal gasification unit and the optimization objectives, the quantitative rules of the virtual-to-real difference penalty term are set, clarifying the correspondence between the magnitude of the deviation and the intensity of the penalty, ensuring that the penalty term can effectively reflect the impact of the deviation. The set virtual-to-real difference penalty term is integrated into the original reward function of the Deep Q-Network to construct a new reward function system, so that the Deep Q-Network can fully consider the difference between simulation and reality during the learning process.

[0056] The specific tasks of the deep Q-network model optimization unit are as follows: Firstly, key data is extracted from the comprehensive dataset. On one hand, simulated syngas calorific value from the simulation data of the coal gasification unit is selected. This simulated value is generated based on the theoretical model and preset conditions, reflecting the syngas calorific value of the unit under ideal conditions. On the other hand, the actual syngas calorific value collected by sensors from the real unit is extracted simultaneously. This actual value directly reflects the operating status of the unit in a complex real environment. After extraction, the deviation between the simulation data and the real data is calculated. Based on the deviation analysis results, combined with the operating characteristics of the coal gasification unit and the established optimization objectives, quantitative rules for the virtual-real difference penalty term are set. During the operation of the coal gasification unit, different deviation ranges have varying degrees of impact on the unit's efficiency and stability. The correspondence between the magnitude of the deviation and the penalty intensity is clarified to ensure that the penalty term effectively reflects the impact of the deviation. When the deviation is small, the penalty is relatively light to avoid over-adjustment affecting model stability; when the deviation is large, the penalty is increased to prompt the model to correct quickly and move closer to reality. By setting quantization rules, the penalty term plays an appropriate guiding role in model optimization, improving the model's adaptability to real device operation. After completing the quantization rules for the virtual-real difference penalty term, it is integrated into the original reward function of the deep Q network to construct a new reward function system. The original reward function is mainly based on the model's own target setting. After incorporating the penalty term, the deep Q network analyzes the difference between simulation and reality during the learning process. The new reward function system focuses on the model's performance under ideal conditions and analyzes the applicability of the model's output strategy in real devices. During training, the deep Q network automatically adjusts its learning direction and strategy based on the feedback from the penalty term, making the generated strategy more in line with actual needs.

[0057] The Fruit Fly Algorithm Initialization Unit initializes the dual populations for the Fruit Fly Algorithm. Population 1 optimizes the reward weights of the Deep Q-Network, i.e., the virtual-to-real difference penalty coefficient. By continuously adjusting the reward weights, the Deep Q-Network learns more effectively under the guidance of the reward function. Population 2 optimizes the compensation parameters of the real device. Based on the operation of the real device, the compensation parameters are dynamically adjusted to better match the simulation model. The initialization of the dual populations determines the size and dimension parameters of the dual populations. According to the problem requirements, the variable range for optimizing the reward weights of the Deep Q-Network is set for Population 1; the variable range for optimizing the compensation parameters of the real device is set for Population 2. An initial dual population is randomly generated. Individuals from Population 1 are substituted into the Deep Q-Network, and the reward weights are evaluated based on the network learning effect to calculate the fitness. Individuals from Population 2 are applied to the real device, and the compensation parameters are evaluated based on the running matching degree to calculate the fitness. Based on this, the quality of the individuals is judged, and excellent individuals are selected. The dual population positions are updated through the taste-seeking operation of the Fruit Fly Algorithm. Population 1 adjusts the reward weights to assist network learning, and Population 2 optimizes the compensation parameters to match the simulation. This process is repeated until the optimization conditions are met.

[0058] The specific tasks of the initialization unit for the fruit fly algorithm are as follows: First, determine the size and dimensionality parameters of the dual populations. The population size affects the search capability and computational efficiency of the fruit fly algorithm; too small a size leads to insufficient searching and difficulty in finding the global optimum; too large a size increases computational load and reduces the running speed of the fruit fly algorithm. The dimensionality parameter is related to the number of variables in the optimization problem and should be precisely set according to the specific problem. Based on the problem requirements, set the variable range for optimizing the reward weights of the deep Q-network for population 1. This range determines the possible value range of the reward weights during the optimization process. Set the variable range for optimizing the compensation parameters of the real device for population 2, ensuring that the compensation parameters can be adjusted within a reasonable range. After completing the parameter setting, randomly generate the initial dual populations, with each individual representing a possible solution. After generating the initial dual populations, evaluate the quality of individuals in the populations. For population 1, its individuals are substituted into the deep Q-network, and the learning effect of the network is used as the standard for evaluating the quality of the reward weights. The learning effect of the deep Q-network is measured by the convergence speed during training and the accuracy of the final model. Then, calculate the quality of each individual in population 1. The fitness of each individual is calculated, with higher fitness indicating that the reward weight corresponding to that individual is more conducive to network learning. For population 2, its individuals are applied to the real device, and the quality of the compensation parameters is evaluated based on the running matching degree. The running matching degree is used to analyze the deviation between the actual output and the expected output of the real device, and the fitness of individuals in population 2 is calculated to determine the quality of individuals. Based on the calculated fitness, excellent individuals are selected from the dual populations. The selection of excellent individuals follows the principle of fitness priority, that is, individuals with higher fitness have a greater probability of being selected. Excellent individuals carry better solution information. After selecting excellent individuals, the position of the dual populations is updated through the taste-finding operation of the fruit fly algorithm. For population 1, the reward weight is adjusted to help the network learn, so that the deep Q network can better adapt to the problem requirements and improve the learning effect. For population 2, the compensation parameters are optimized to match the simulation and reduce the difference between the real device and the simulation model. The process of fitness calculation, excellent individual selection and population position update is repeated until the optimization conditions are met, that is, the preset number of iterations is reached or the fitness reaches the preset threshold.

[0059] The expression for the fitness of population 1 is as follows:

[0060] ;

[0061] In the formula: The fitness of population 1 The reward weight is the variable for optimization. This represents the convergence loss value of the deep Q-network; the smaller the better. The maximum possible loss value is used to normalize the loss value to the [0,1] interval; The final accuracy of the deep Q-network should be as high as possible; and The first and second weighting balancing factors are used to adjust the weights of loss and accuracy in fitness. The smaller the value (i.e., the lower the loss), The higher, when When the value is larger (i.e., the higher the accuracy), The higher;

[0062] The expression for the fitness of population 2 is as follows:

[0063] ;

[0064] In the formula: For the fitness of population 2, For compensation parameters; This indicates the deviation between the actual output and the expected output of the real device; This represents the maximum possible deviation value, used for normalization; The smaller the fluctuation in the actual device's operation, the better; This represents the maximum possible fluctuation value, used for normalization. and These are the first and second compensation balance factors, used to adjust the weights of bias and volatility in fitness. The smaller the value (i.e., the lower the deviation), The higher, when The smaller the value (i.e., the lower the fluctuation), The higher;

[0065] The collaborative optimization module, based on the co-evolutionary mechanism of the fruit fly algorithm, performs collaborative optimization of virtual and real parameters and uses the latest real data to quickly calibrate the simulation model;

[0066] The strategy generation and output module, based on the optimized deep Q-network model and combined with parameters that have undergone co-evolution, generates an efficiency optimization strategy for the coal gasification unit and outputs it to the control system of the coal gasification unit to guide the operation of the coal gasification unit.

[0067] The efficiency evaluation module is used to evaluate the operating efficiency of the coal gasification unit after applying the generation efficiency optimization strategy. By comparing the efficiency indicators before and after applying the efficiency optimization strategy, including syngas production and energy consumption, the module analyzes the effect of the efficiency optimization strategy on the efficiency of the coal gasification unit, so as to understand the actual effect of the current strategy and make further improvements.

[0068] Example 2, as Figure 1 , Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: the collaborative optimization module includes a fruit fly algorithm optimization unit and a model calibration unit;

[0069] The fruit fly algorithm optimization unit utilizes the co-evolution mechanism of the fruit fly algorithm. Population 1 optimizes the reward weights of the deep Q network, while population 2 optimizes the compensation parameters of the real device. Through the odor concentration sharing mechanism, it performs collaborative optimization of virtual and real parameters, reducing efficiency loss after strategy migration and improving the operating efficiency of the coal gasification unit in the real environment. Based on the fruit fly algorithm, populations 1 and 2 are constructed. Population 1 is used to optimize the reward weights of the deep Q network, while population 2 optimizes the compensation parameters of the real device. According to the problem requirements, the range of variables for the two populations is set, and initial dual population individuals are randomly generated. Individuals from population 1 are substituted into the deep Q network to evaluate the merits of the reward weights, covering the learning effect of convergence speed and accuracy. Individuals from population 2 are used in the real device, and the compensation parameters are evaluated according to the operating matching degree. Then, the individual fitness is calculated. Information is shared through the odor concentration sharing mechanism. Excellent individuals are selected from the dual populations according to the fitness priority principle to carry the optimal solution information. Using the fruit fly algorithm's odor-finding operation, the positions of the dual populations are updated collaboratively, and the reward weights and compensation parameters are optimized. This process is repeated until the optimization conditions are met, thereby improving the operating efficiency in the real environment.

[0070] The specific work of the fruit fly algorithm optimization unit is as follows: Based on the fruit fly algorithm, two populations are constructed: Population 1 and Population 2. Population 1 is responsible for optimizing the reward weights of the deep Q-network, while Population 2 focuses on optimizing the compensation parameters of the real device. According to the characteristics and requirements of the specific problem, the variable ranges of the two populations are precisely set. The variable ranges determine the feasible interval for parameter optimization. After setting the variable ranges, individuals are randomly generated for the initial dual populations. Each individual represents a potential solution. Individuals in Population 1 correspond to different combinations of reward weights, and individuals in Population 2 correspond to different combinations of compensation parameters. Individuals from Population 1 are then substituted into the deep Q-network, and the learning effect of the deep Q-network is used as an indicator to evaluate the quality of the reward weights. The learning effect includes both convergence speed and accuracy. Reward weight combinations with fast convergence speed and high accuracy are considered superior. For individuals from Population 2, they are applied to the real device, and the compensation is evaluated based on the operational matching degree. The quality of parameters and the operational matching degree are determined by analyzing the deviation between the actual output and the expected output of the real device. Based on the above evaluation results, the fitness of each individual in the dual population is calculated. At the same time, with the help of the odor concentration sharing mechanism, information exchange between the dual populations is realized, so that the two populations can learn from each other's effective information in the optimization process. According to the principle of fitness priority, excellent individuals are selected from the dual populations. Using the odor-finding operation of the fruit fly algorithm, the positions of the dual populations are updated collaboratively. For population 1, the reward weight is adjusted to help the deep Q network adapt to the problem requirements and improve its learning performance. For population 2, the compensation parameters are optimized to reduce the difference between the real device and the expected state. Individual evaluation, selection of excellent individuals and population position updates are continuously performed in a loop until the preset optimization conditions are met, that is, the preset number of iterations is reached, or the fitness of population 1 and population 2 respectively reach the preset threshold, thereby improving the system's operating efficiency in the real environment.

[0071] The expressions for the odor concentrations of population 1 and population 2 are as follows:

[0072] ;

[0073] ;

[0074] In the formula: The odor concentration is the average fitness of all individuals in population 1. The odor concentration is the average fitness of all individuals in population 2. The size of population 1, that is, the number of individuals in population 1; The size of population 2 is the number of individuals in population 2. The first in population 1 The fitness of an individual represents the quality of the reward weight combination corresponding to that individual; The second in population 2 The fitness of an individual represents the quality of the combination of compensation parameters corresponding to that individual; The first in population 1 The reward weighting combination for each individual; The second in population 2 The combination of compensation parameters for each individual;

[0075] For population 1, the reward weights are adjusted to improve learning performance, and the expression is as follows:

[0076] ;

[0077] In the formula: For the updated population 1, the first The reward weighting combination for each individual; The weight learning rate is used to control the size of the update step. The reward weight combination is the individual with the highest fitness in population 1, which is the global optimal solution;

[0078] For population 2, the compensation parameters are optimized to reduce the difference between the actual device and the expected state, and the expression is as follows:

[0079] ;

[0080] In the formula: For the updated population 2, the The combination of compensation parameters for each individual; It is used to compensate for the learning rate and control the size of the update step. The combination of compensation parameters is the combination of the individuals with the highest fitness in population 2, i.e., the global optimal solution;

[0081] The model calibration unit activates the fruit fly algorithm every 10 minutes to quickly calibrate the simulation model using the latest real data. Simultaneously, it uses a deep Q-network to update the strategy in real time to solve the dynamic mismatch problem and ensure the dynamic consistency between the simulation model and the real device. Every 10 minutes, the unit automatically collects the latest real data, triggers the model calibration program, and starts the fruit fly algorithm. The fruit fly algorithm runs based on the latest real data, optimizes and adjusts the key parameters in the simulation model, and quickly completes the calibration of the simulation model by iteratively searching for the optimal parameter combination, thus narrowing the gap between the simulation model and the real situation. The deep Q-network updates the control strategy in real time based on the calibrated simulation model and real-time feedback to solve the dynamic mismatch problem, enabling the control strategy to better adapt to changes and ensuring that the simulation model and the real device always maintain dynamic consistency.

[0082] The specific tasks of the model calibration unit are as follows: The model calibration unit sets up a timed task to periodically trigger the model calibration program. Every 10 minutes, it automatically collects the latest real data, directly reflecting the actual operating status of the coal gasification unit at the current moment. Once the data collection is complete, the fruit fly algorithm is immediately started, using the latest real data as input for subsequent simulation model optimization and adjustment. After acquiring the latest real data, the fruit fly algorithm officially starts running, using it as a guide to deeply optimize and adjust key parameters in the simulation model. By simulating the taste-seeking behavior of fruit flies, it performs efficient iterative searches in the parameter space. In each iteration, it evaluates the merits of different parameter combinations based on the fitness function. By approximating the optimal solution through multiple iterations, the deep Q-network quickly finds the parameter combination that makes the simulation model output closest to the real data, thereby completing the precise calibration of the simulation model, effectively narrowing the gap between the simulation model and the real situation, and improving the accuracy and reliability of the model. Based on the calibrated simulation model and real-time feedback information, the deep Q-network updates the control strategy in real time. The calibrated simulation model reflects the operating characteristics of the real device. Through interaction with the simulation model and combined with real-time feedback data, the deep Q-network continuously adjusts its decision logic to solve the dynamic mismatch problem, so that the control strategy can better adapt to changes in the real environment and ensure that the simulation model and the real device always maintain a high degree of consistency.

[0083] The strategy generation and output module specifically includes: generating an efficiency optimization strategy for the coal gasification unit based on the optimized deep Q-network model and combined with co-evolutionary parameters; the deep Q-network using the calibrated model and real-time data to output the optimal control strategy, ensuring the scientific validity and effectiveness of the strategy; verifying the generated optimal control strategy to ensure its feasibility and safety under the current operating conditions; fine-tuning the optimal control strategy based on the verification results to better adapt to the actual operating environment and improve the reliability of the strategy; and outputting the optimized optimal control strategy to the control system of the coal gasification unit to guide the operation of the unit. The control system adjusts the operating parameters according to the strategy to ensure that the coal gasification unit operates in a highly efficient and stable state, thereby improving the overall efficiency.

[0084] The specific tasks of the strategy generation and output module are as follows: Based on the optimized deep Q-network model and combined with the parameters obtained from co-evolution, the strategy generation process is initiated. The calibrated model provides the deep Q-network model with information that closely reflects the actual operating characteristics of the real device. Real-time data reflects the current actual state of the real device. Through learning and simulation, the deep Q-network analyzes and makes decisions in the parameter space, outputting the optimal control strategy. Based on the current operating state of the gasification unit, the staff conducts a comprehensive evaluation of the feasibility and safety of the optimal control strategy. Simulated operation is used to check whether the optimal control strategy will cause device malfunctions or comply with safety regulations during implementation. If, during the verification process, the optimal control strategy is found to be insufficient or incompatible with the actual operating environment, adjustments are made to the optimal strategy according to the specific problems. The control strategy is precisely fine-tuned, continuously optimizing the parameters of the optimal control strategy to better adapt to the actual operating environment, thereby improving the reliability of the strategy and ensuring its stable operation in practical applications. The verified and fine-tuned optimal control strategy is output to the control system of the coal gasification unit. As the executor of the strategy, the control system receives the strategy and responds by adjusting the operating parameters of the unit, including temperature, pressure, and flow rate, according to the strategy requirements. During the adjustment process, the control system monitors the operating status of the coal gasification unit in real time to ensure that the parameter adjustment does not have an adverse effect on the unit. At the same time, the control system feeds back the operating data to the deep Q network, forming a closed-loop optimization system. The strategy is further optimized based on the actual operating conditions to continuously improve the operating efficiency of the unit.

[0085] The efficiency assessment module specifically includes: acquiring relevant operational data from the coal gasification unit before and after the application of the efficiency optimization strategy, including efficiency index data of syngas production and energy consumption, and screening and cleaning the data to form a basic dataset for comparative analysis. The efficiency index data before and after the efficiency optimization strategy are compared one by one. By calculating the increase in production and the reduction in energy consumption, the changes in the operating efficiency of the coal gasification unit after the implementation of the strategy are analyzed, the direction of the impact of the efficiency optimization strategy on each efficiency index is clarified, and based on the comparative analysis results of the efficiency index, the effect of the efficiency optimization strategy on the improvement of the operating efficiency of the coal gasification unit is quantified, and an efficiency assessment report is generated to present the actual effectiveness of the efficiency optimization strategy.

[0086] The specific tasks of the efficiency evaluation module are as follows: First, acquire relevant operational data from the coal gasification unit before and after applying the efficiency optimization strategy. Focus on efficiency indicators such as syngas production and energy consumption to measure the unit's operating efficiency. After acquiring the data, perform screening and cleaning. Screening removes irrelevant, duplicate, or erroneous data to ensure the data entering the analysis is targeted and accurate. Cleaning focuses on handling missing values ​​and outliers in the data. By filling in missing data and correcting outliers, the data meets the analytical requirements, thus forming the basic dataset for comparative analysis. With the basic dataset, compare the efficiency indicators before and after the efficiency optimization strategy one by one, calculate the increase in production and the reduction in energy consumption, and analyze the changes in the operating efficiency of the coal gasification unit after the implementation of the efficiency optimization strategy. The increase in production reflects the efficiency... The optimization strategy's positive impact on syngas production capacity is demonstrated by the reduction in energy consumption, reflecting the improved energy efficiency of the optimization strategy. The changes in the increase in output and the reduction in energy consumption are clearly identified, and the direction of the optimization strategy's impact on various efficiency indicators—whether it positively improves or negatively inhibits—is determined, thus providing guidance for further optimization. Based on the comparative analysis of efficiency indicators, the improvement effect of the efficiency optimization strategy on the operating efficiency of the coal gasification unit is quantified, the benefits brought by the optimization strategy are measured, and a detailed efficiency assessment report is generated. This report comprehensively presents the actual effectiveness of the efficiency optimization strategy, including the specific changes in various efficiency indicators and the degree to which the optimization strategy improves the overall operating efficiency of the coal gasification unit. This assists decision-makers in judging the effectiveness of the efficiency optimization strategy and promotes the continuous improvement of the operating efficiency of the coal gasification unit.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0088] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A simulation data processing system for a coal gasification unit that integrates multi-source data, comprising a simulation management platform, characterized in that: The simulation management platform has the following communication connection modules: The data acquisition and processing module is used to collect multi-source data during the operation of the coal gasification unit, including simulation data from traditional physical simulation models and real-time operation data of the actual unit, and to preprocess the collected multi-source data and integrate it into a comprehensive dataset. A digital twin reward shaping module is used to combine deep Q-networks and the fruit fly algorithm to analyze the deviation between simulation data and real data, and to initialize the dual population of the fruit fly algorithm. The collaborative optimization module, based on the co-evolutionary mechanism of the fruit fly algorithm, performs collaborative optimization of virtual and real parameters and uses the latest real data to quickly calibrate the simulation model; The strategy generation and output module, based on the optimized deep Q-network model and combined with parameters that have undergone co-evolution, generates an efficiency optimization strategy for the coal gasification unit and outputs it to the control system of the coal gasification unit to guide the operation of the coal gasification unit. The efficiency evaluation module is used to evaluate the operating efficiency of the coal gasification unit after applying the generation efficiency optimization strategy, and to analyze the effect of the efficiency optimization strategy on the efficiency improvement of the coal gasification unit.

2. The multi-source data fusion simulation data processing system for coal gasification equipment according to claim 1, characterized in that: The data acquisition and processing module specifically includes: Simulation parameters covering temperature, pressure, and flow rate are obtained from traditional physical simulation models as simulation data. At the same time, actual parameters of syngas calorific value and equipment operating status are collected by sensors during the actual operation of the device, and multi-source data is collected. The collected multi-source data is preprocessed by data cleaning to remove outliers and noise interference, and then normalized. The pre-processed multi-source data is integrated and aggregated into a comprehensive dataset by following a unified data format and structure, bringing together data from different sources and of different types.

3. The multi-source data fusion simulation data processing system for coal gasification equipment according to claim 1, characterized in that: The digital twin reward shaping module includes a deep Q-network model optimization unit and a fruit fly algorithm initialization unit; The deep Q-network model optimization unit is used to add a virtual-real difference penalty term to the reward function of the deep Q-network, that is, the deviation of the syngas calorific value between the simulation and the real device. By analyzing the deviation between the simulation data and the real data, it is added as a penalty term to the reward function. The fruit fly algorithm initialization unit is used to initialize the dual populations of the fruit fly algorithm. Population 1 is used to optimize the reward weights of the deep Q network, i.e., the virtual-real difference penalty coefficient; Population 2 is used to optimize the compensation parameters of the real device, and dynamically adjusts the compensation parameters according to the operation of the real device.

4. The multi-source data fusion simulation data processing system for coal gasification equipment according to claim 3, characterized in that: The deep Q-network model optimization unit specifically includes: Extract the simulated calorific value of syngas from the simulation data of the coal gasification unit from the comprehensive dataset, and the actual calorific value of syngas collected by the sensors of the real unit, and calculate the deviation between the simulation data and the real data. Based on the deviation analysis results, combined with the operating characteristics and optimization objectives of the coal gasification unit, quantitative rules for the penalty items for virtual and real differences are set, and the correspondence between the magnitude of the deviation and the intensity of the penalty is clarified. The pre-defined virtual-real difference penalty term is integrated into the original reward function of the deep Q-network to construct a new reward function system.

5. The multi-source data fusion simulation data processing system for coal gasification equipment according to claim 3, characterized in that: The fruit fly algorithm initialization unit specifically includes: Initialize the dual populations of the fruit fly algorithm, determine the size and dimension parameters of the dual populations, and set the variable range for population 1 to optimize the reward weights of the deep Q network according to the problem requirements; set the variable range for population 2 to optimize the compensation parameters of the real device, and randomly generate the initial dual populations. Individuals from population 1 are fed into a deep Q-network, and the reward weights are evaluated based on the network's learning performance to calculate fitness. Individuals from population 2 are applied to a real device, and the compensation parameters are evaluated based on the operational matching degree to calculate fitness, thereby determining the quality of the individuals. Select superior individuals based on fitness, update the dual population positions through the taste-finding operation of the fruit fly algorithm, adjust the reward weight of population 1 to help the network learn, and optimize the compensation parameter matching simulation of population 2. Repeat this process until the optimization conditions are met.

6. The multi-source data fusion simulation data processing system for coal gasification equipment according to claim 3, characterized in that: The collaborative optimization module includes a fruit fly algorithm optimization unit and a model calibration unit; The fruit fly algorithm optimization unit is used to utilize the co-evolution mechanism of the fruit fly algorithm, with population 1 optimizing the reward weights of the deep Q network and population 2 optimizing the compensation parameters of the real device, and co-optimizing the virtual and real parameters through the odor concentration sharing mechanism. The model calibration unit activates the fruit fly algorithm every 10 minutes to quickly calibrate the simulation model using the latest real data, while simultaneously updating the strategy in real time using a deep Q-network.

7. The multi-source data fusion simulation data processing system for coal gasification equipment according to claim 6, characterized in that: The fruit fly algorithm optimization unit specifically includes: Population 1 and Population 2 are constructed based on the fruit fly algorithm. Population 1 is used to optimize the reward weight of the deep Q network, and Population 2 optimizes the compensation parameters of the real device. According to the problem requirements, the range of variables for the two populations is set, and the initial dual population individuals are randomly generated. Individuals from population 1 are fed into a deep Q-network to evaluate the merits of the reward weights based on the learning performance, which includes convergence speed and accuracy; individuals from population 2 are used in a real device to evaluate the compensation parameters based on the operational matching degree, and then calculate the individual fitness. Selecting superior individuals from the dual population based on the fitness priority principle, carrying the optimal solution information, and using the fruit fly algorithm for taste-finding operations, collaboratively updating the positions of the dual populations, optimizing reward weights and compensation parameters, and repeating the process until the optimization conditions are met.

8. The multi-source data fusion simulation data processing system for coal gasification equipment according to claim 6, characterized in that: The model calibration unit specifically includes: The system automatically collects the latest real data every 10 minutes, triggers the model calibration process, and starts the fruit fly algorithm, using the latest real data as input. The fruit fly algorithm runs on the latest real data, optimizes and adjusts key parameters in the simulation model, and quickly completes the calibration of the simulation model by iteratively searching for the optimal parameter combination, thus narrowing the gap between it and the real situation. The deep Q network updates the control strategy in real time based on the calibrated simulation model and real-time feedback.

9. The multi-source data fusion simulation data processing system for coal gasification equipment according to claim 6, characterized in that: The strategy generation output module specifically includes: Based on the optimized deep Q-network model and combined with the parameters of co-evolution, an efficiency optimization strategy for the coal gasification unit is generated. The deep Q-network uses the calibrated model and real-time data to output the optimal control strategy. The generated optimal control strategy is verified to ensure its feasibility and safety under the current operating conditions, and the optimal control strategy is fine-tuned based on the verification results; The optimized control strategy is output to the control system of the coal gasification unit to guide the operation of the unit. The control system adjusts the operating parameters according to the strategy.

10. A multi-source data fusion simulation data processing system for coal gasification equipment according to claim 9, characterized in that: The efficiency evaluation module specifically includes: Relevant operational data before and after the application of efficiency optimization strategies are obtained from the coal gasification unit, including efficiency index data of syngas production and energy consumption, and then screened and cleaned to form a basic dataset for comparative analysis. The efficiency index data before and after the efficiency optimization strategy are compared one by one. By calculating the increase in output and the reduction in energy consumption, the changes in the operating efficiency of the coal gasification unit after the implementation of the strategy are analyzed, and the direction of the impact of the efficiency optimization strategy on each efficiency index is clarified. Based on the comparative analysis of efficiency indicators, the effect of efficiency optimization strategies on improving the operating efficiency of coal gasification units is quantified, and an efficiency evaluation report is generated.

Images