A coal-based polypropylene full-process process optimization method based on multi-source data

Abstract

The application discloses a coal-based polypropylene full-process process optimization method based on multi-source data, and belongs to the field of coal chemical process optimization. The method integrates full-process multi-source data, and constructs an accurate model system covering material balance and characteristics of each device. Global optimization instructions are generated through reverse optimization and collaborative game calculation to drive each device to perform parallel local optimization and issue execution. The system uses operation data to correct model parameters online, forming a self-learning closed loop. Through the complete technical system of data fusion, collaborative optimization and feedback learning, the application solves the problems of material imbalance and high energy consumption caused by traditional single unit independent optimization, and realizes the collaborative stable operation and continuous improvement of economic benefits of the full process.

Description

Technical Field

[0001] This invention relates to the field of chemical process control and optimization technology, and in particular to a method for optimizing the entire coal-based polypropylene process based on multi-source data. Background Technology

[0002] Current solutions for coal-based polypropylene process optimization based on multi-source data primarily focus on improving the independent performance of single process units. The core idea is to optimize the equipment structure, catalysts, reaction conditions, or control strategies locally using process data such as temperature, pressure, and key components from individual units like gasifiers, methanol synthesis, methanol-to-olefins (MTO), or polypropylene synthesis, through the establishment of mechanistic or data-driven models. The optimization objective of this method is to improve the process stability, product yield, or energy efficiency of the unit itself, and its optimization boundary typically stops at the material inlet and outlet of the unit.

[0003] Existing technologies suffer from a core problem: process fragmentation, namely the unknowability and lack of coordination across the entire process chain. The system cannot perceive or coordinate the cascading effects of upstream unit operating conditions on downstream units. A typical example is how fluctuations in the gasifier load directly impact the feed stability of the methanol synthesis unit. This results in an isolated and static optimization process, with downstream units passively adapting to upstream fluctuations and forced to make frequent and delayed local adjustments. The ultimate consequence is that local optimizations of individual units cannot converge into optimal overall efficiency; instead, supply-demand imbalances between units may lead to decreased overall process energy efficiency, product quality fluctuations, or even unplanned shutdowns. Essentially, this system can only guarantee local stability within individual units, but cannot achieve dynamic coordination across the entire process or maximize overall efficiency.

[0004] The technical problem to be solved by this invention is that the core problem of the existing technology is the lack of coordination in the whole process, namely, the isolated optimization of each process unit leads to the imbalance of supply and demand in the upstream and downstream, the fluctuation of upstream operating conditions causes downstream anomalies and high unit consumption, and there is a lack of a global collaborative optimization mechanism driven by multi-source data, resulting in high overall energy efficiency and operating costs. To this end, we propose a coal-based polypropylene whole process optimization method based on multi-source data. Summary of the Invention

[0005] The technical problem to be solved by this invention is the core issue of the lack of coordination in the entire process in the existing technology, namely, the independent optimization of each process unit leads to the imbalance of supply and demand of upstream and downstream materials and high operating costs. To address this, we propose a coal-based polypropylene whole-process optimization method based on multi-source data.

[0006] To achieve the above objectives, this application adopts the following technical solution: a method for optimizing the entire coal-based polypropylene process based on multi-source data, characterized by comprising the following steps: S1 collects and integrates multi-source process data covering the entire process of gasification, synthesis, conversion and polymerization to form a unified process dataset.

[0007] S2 builds and runs a full-process material and energy balance model, and generates global optimization settings for upstream processes based on end-product targets.

[0008] S3. Based on the process dataset, construct device control models corresponding to each process unit, wherein the upstream device model is a mechanism model and the terminal aggregation device model is a data-driven model.

[0009] S4, under the constraint of the global optimization setting value, run each device control model in parallel to generate the operating parameter sequence of each process device.

[0010] S5, the operation parameter sequence is sent to the corresponding device for execution, and the actual response data of the device is monitored and collected.

[0011] S6, the actual response data is fed back to the full-process model and each of the device models to perform online model correction and parameter updates, forming a closed loop.

[0012] Preferably, step S1 specifically includes the following steps: S11, construct a hierarchical flow traceability model, abstract the entire process material flow into a virtual flow pipeline, and define its source device, inflow device and element composition vector.

[0013] S12 maps multi-source heterogeneous data to corresponding virtual flow channels based on measurement location and physical meaning through a probabilistic graphical model, and uses dynamic time warping algorithm and rule base to achieve time synchronization and dimensional unification.

[0014] S13 employs a rule engine and an isolated forest model for two-stage bad value detection and removal, outputting a unified process dataset that is time-aligned, dimensionally consistent, and flow-related.

[0015] The method as described in claim 1, wherein step S12 specifically performs the following operations: A hierarchical flow traceability model based on material flow tracking is constructed. All physical material flows transmitted through pipelines and equipment in the entire process are abstracted into virtual flow pipelines with unique identifiers. For each flow pipeline, its source device, inlet device, and chemical element composition vector are defined. For heterogeneous data items from different data sources, they are mapped to the corresponding virtual flow pipelines according to their measurement location and physical meaning. A probabilistic graphical model with confidence weights is established for this mapping relationship. Timestamp synchronization is solved by introducing a dynamic time warping algorithm with equipment dwell time and transmission delay as parameters. Dimensional unification is automatically completed by an intelligent conversion rule base integrating international standard units, factory custom units, and online analyzer ranges. A two-stage outlier detection and elimination process based on rule engine and machine learning is adopted. The first stage uses the physical limit change rate threshold and material balance constraints set by the process knowledge base for fast filtering. The second stage uses a trained isolated forest model to identify and eliminate hidden outliers from the overall distribution of multidimensional associated data, and outputs a unified process dataset with time alignment and consistent dimensions.

[0016] Preferably, step S2 specifically includes the following steps: S21, construct the chemical element-species association matrix, and define the input and output species vectors of each process unit and the global total species flow vector.

[0017] S22 introduces the conversion rate matrix determined by the reaction mechanism of the device and the diversion and mixing coefficient matrix determined by the pipeline connection to construct a steady-state material balance model for the entire process.

[0018] S23, the product quality requirements of the polypropylene synthesis unit are transformed into constraints on the equation set, and real-time measurement data is integrated as boundary conditions.

[0019] S24, Numerically solve the steady-state material balance model to obtain the baseline flow distribution that satisfies the global material conservation and quality requirements.

[0020] Furthermore, the characteristic is that step S21 specifically performs the following operations: Based on all chemical reactions involved in the entire coal-based polypropylene process, an extended chemical element species association matrix is ​​constructed, covering the entire process from raw coal to final polypropylene products. The rows of the matrix correspond to the three core tracking elements (carbon, hydrogen, and oxygen), and the columns correspond to all key chemical species in the process. Input and output species vectors for each process unit are defined, and a global species total flow vector representing all material streams in the entire process is constructed based on the actual pipeline connection topology. A conversion rate matrix determined by the reaction mechanism, catalyst characteristics, and operating conditions of each unit, as well as a splitting and mixing coefficient matrix determined by pipeline design and control logic, are introduced. The chemical reaction conversion within the unit and the physical transport and distribution processes between units are uniformly represented as a simultaneous steady-state material balance model of the entire process, including constraints from nonlinear equations and inequalities. The key requirements for product quality of downstream polypropylene synthesis units are transformed into constraints on the polypropylene-related flow and properties in this equation set. Measurement data from a real-time database are integrated as boundary conditions for some known variables or equations. Solving the equation set yields a baseline flow distribution that satisfies global material conservation and product quality requirements.

[0021] Furthermore, step S23 specifically involves performing the following operations: Based on the system structure provided by the full-process material balance model and the target output and melt flow index of the polypropylene unit, a reverse optimization problem is constructed with minimizing the output deviation of the final product as the core, while considering the operating costs and stability of the upstream units. This problem is expressed as a constrained nonlinear programming problem with the upstream unit operating parameter vector as the decision variable. The objective function is the weighted sum of squares of the terminal output error, and the constraints cover the operating boundaries of each unit and the quality requirements of intermediate products. To solve this optimization problem, the adjoint method is introduced to calculate the full-process gradient information of the objective function with respect to the upstream operating parameters. This gradient information is obtained by backpropagating the adjoint vector and using the local sensitivity matrix of each unit, thus avoiding the direct calculation of the chain product of the Jacobian matrix of the full-process model. An iterative optimization algorithm based on this gradient is used to drive the update of the upstream operating parameters, so that the terminal output obtained from the full-process chain simulation starting from the updated parameters continuously approaches the preset target until convergence. The optimal setpoints of the operating parameters of each upstream unit that satisfy the global optimum are decoupled in reverse. The core formula of the reverse solution process is: , Define the adjoint vector equation as: , The formula for calculating the gradient of the objective function is: , This is a vector sum of the operating parameters of all upstream units, including gasifiers, methanol synthesis, and MTO. This is a full-process chain-integrated model, with upstream operation parameters as input. The output is a vector predicting the yield and quality of the final polypropylene unit. This model is a mechanism model for each device. , , , Function composition, Let be the output and quality target vector for the polypropylene unit. This is a diagonal positive definite weight matrix used to adjust the relative importance of yield and quality deviation in the objective function. This is a regularization term used to penalize deviations or drastic fluctuations in operating parameters, improving the engineering applicability of the solution. , The adjoint vector has the same dimension as... Similarly, it encapsulates all the indirect, chain-like effects of changes in the terminal target on changes in upstream operating parameters, making it an efficient method for calculating gradients. The key, For the full Jacobian matrix of the upstream operating parameters in the whole process model, its relationship with the vector is efficiently calculated by automatic differentiation in the inverse mode or by inverse solution of the sub-device. The operational relationships.

[0022] Furthermore, step S24 specifically involves performing the following operations: Based on the optimal material allocation scheme and composition control range of the whole-process material balance model, a collaborative optimization framework based on Nash bargaining game theory is constructed. The three upstream process units—gasifier, methanol synthesis, and MTO—are defined as game participants, and each participant's strategy set is a vector of independently adjustable operating parameters. This vector includes the oxygen-to-coal ratio and furnace temperature of the gasifier, the inlet hydrogen-to-carbon ratio and reaction pressure of the methanol synthesis unit, and the reaction temperature and methanol feed space velocity of the MTO unit. A composite utility function is designed for each participant, comprehensively considering both individual operational economics and the overall contribution to the entire process. The downstream polypropylene synthesis unit's feed material quantity is also considered. The rigid requirements for quality are transformed into mandatory coupling constraints in the game. By integrating high-fidelity mechanism models of various devices with real-time collected disturbance data as the state transition and payoff calculation engine for the game, an iterative negotiation algorithm is used to solve the Nash equilibrium solution of this cooperative game online. This equilibrium solution is a Pareto optimal combination of operating parameters that enables the upstream devices to operate collaboratively under the premise of satisfying individual rationality and collective rationality. After the obtained equilibrium operating parameter combination is verified and smoothed by safe operating boundary, a time synchronization global optimization setpoint instruction that can be directly issued to the control systems of each corresponding device is generated. The formula for the composite utility function is: , in, for For the Composite utility function of upstream devices For the first The operational variable vector of each device To exclude the first The vector of operational variables of other devices besides this device. To act on the first Real-time disturbance vector of each device For the first The operating economic index function of each device, and These represent the worst and best historical values ​​of the indicator within the feasible region, respectively. For the first The key quality indicator function of the output material of each device This is the target value for this indicator issued by the full-process model. The vector representing the coupling effect function between devices. , , These are the preference coefficients for economy, quality compliance, and overall synergy, respectively. This is the parameter for tolerance to coupling effects.

[0023] Furthermore, the iterative negotiation algorithm employs an alternating direction multiplier method based on augmented Lagrange multipliers to achieve distributed solution, specifically including the following operations: Each upstream device constructs and solves a local optimization subproblem in parallel based on the received global setpoint instruction and the current Lagrange multiplier. The objective function of the subproblem consists of the device's composite utility function and the quadratic penalty term corresponding to the downstream common constraint. The central coordinator collects the locally optimal operating parameters obtained by each device, calculates the actual violation of the downstream common constraint, and updates the Lagrange multiplier according to the violation and the preset multiplier update step size. The updated multiplier and the strategies of each device are broadcast to all participants for the next iteration. The above operations are repeated until the changes in the operating parameters of each device and the violation of the downstream common constraint are both lower than the set threshold. The resulting set of strategies is the Nash equilibrium solution.

[0024] Preferably, step S4 specifically includes the following steps: S41, each device's control model receives the global optimization setpoint and transforms it into the device's boundary constraints and tracking targets.

[0025] S42, each device uses its own model as the core to construct a local constraint optimization problem that incorporates upstream and downstream collaboration requirements.

[0026] S43 employs a distributed iterative strategy based on augmented Lagrange multipliers and alternating direction multipliers to solve local problems in parallel and coordinate the consistency of predicted values ​​by exchanging coupling variables.

[0027] S44, after the algorithm converges, outputs the optimal operating parameter sequence that satisfies the material balance of the entire process.

[0028] Furthermore, step S42 specifically involves performing the following operations: Each device's control model receives and parses the global optimization setpoint instruction from step S24, transforming it into the device's boundary constraints and tracking targets. Using the device's mechanism model or data-driven model as the core, it constructs a local constraint optimization problem that includes its own operational optimization target and forcibly embeds upstream and downstream coordination requirements. By adopting a distributed iterative solution strategy based on augmented Lagrange multipliers and alternating direction multipliers, each device can solve its own local problem independently and in parallel, while exchanging prediction information of coupling variables and coordinating consistency. Ultimately, it converges to a set of operating parameter sequences that meet the overall material balance and quality connection requirements of the entire process and are locally optimal.

[0029] Preferably, step S5 specifically includes the following steps: S51 sends the optimal operating parameter sequence of each device to the field actuators through the advanced process control system.

[0030] S52, collects actual operating data and online product quality analysis data after the acquisition device is executed.

[0031] S53 monitors the deviation between actual data and model expected values, and triggers an alert when the deviation exceeds a safety threshold.

[0032] Preferably, step S6 specifically includes the following steps: S61, compare the actual response data of each device with the model prediction value to generate a multi-dimensional prediction error vector.

[0033] S62 employs a hierarchical collaborative adaptive filtering algorithm, using export test data as observations, to synchronously estimate and correct the core mechanism parameters of the entire process model and each device model online.

[0034] S63 feeds back the corrected model parameters to the next round of optimization calculation.

[0035] Furthermore, the characteristic is that step S61 specifically performs the following operations: The actual response data of each process unit collected in real time in step S5 is compared with the current predicted value of the corresponding model. The actual response data includes the actual values ​​of key operating variables, the actual values ​​of output material quality and flow rate, and auxiliary variables of equipment status. A multi-dimensional prediction error vector is generated. Based on the prediction error vector, a hierarchical collaborative adaptive filtering algorithm is used to synchronously estimate and dynamically correct the key transfer coefficients in the whole process material balance model and the core mechanism parameters of each unit control model. The core mechanism parameters include reaction kinetic constants and catalyst activity factors. The corrected and updated model parameters are fed back and integrated into the next round of optimization calculation cycle starting from step S2.

[0036] The technical effects and advantages of this invention are as follows: 1. This invention fundamentally reconstructs the traditional unit-independent optimization paradigm by constructing a data fusion model based on hierarchical flow traceability and a full-process steady-state material balance model. The system models and optimizes the entire process of gasification, synthesis, conversion, and polymerization as a whole, realizing synchronous and coordinated control of the operating parameters of upstream units from a global perspective. This solves the core problems of upstream and downstream material supply and demand imbalance, operating condition fluctuation transmission, and overall energy efficiency limitations caused by process fragmentation in traditional methods.

[0037] 2. The dual-layer optimization architecture and distributed collaborative algorithm employed in this invention achieve a dynamic balance between global collaboration and precise local control through a Nash bargaining game framework and the adjoint gradient method. The end-to-end optimization engine generates global instructions, ensuring seamless material and energy supply; each unit-level optimizer searches for optimization in parallel within the instruction boundaries, fully exploring local potential. An integrated high-fidelity mechanism model and data-driven model provide an accurate prediction and calculation engine for optimization, enabling the system to absorb raw material fluctuations, track catalyst decay, and achieve continuous optimal decision-making under complex operating conditions. Method verification shows that this system can significantly improve the stability of the entire process operation and product consistency, achieving increased production capacity and reduced overall energy consumption.

[0038] 3. This invention constructs an intelligent optimization system with self-learning and adaptive capabilities through a closed-loop design encompassing multi-source data perception, end-to-end collaborative optimization, instruction issuance and execution, and online model calibration. This method significantly reduces the reliance on human experience in the optimization process. Through continuous data feedback and model updates, the system can dynamically adapt to changes in the production process, continuously improving optimization accuracy and reliability. This system provides an advanced solution for the digital transformation and deep energy conservation and consumption reduction of coal-based polypropylene and similar process industries, combining a global perspective, collaborative intelligence, and engineering practicality. Attached Figure Description

[0039] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts: Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a data flow diagram of the present invention; Figure 3 This is a system architecture diagram of the present invention. Detailed Implementation

[0040] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0041] Reference Figure 1 , Figure 2 As shown, the present invention provides a technical solution: a method for optimizing the entire process of coal-based polypropylene based on multi-source data, characterized by comprising the following steps: S1 collects and integrates multi-source process data covering the entire process of gasification, synthesis, conversion and polymerization to form a unified process dataset.

[0042] S2 builds and runs a full-process material and energy balance model, and generates global optimization settings for upstream processes based on end-product targets.

[0043] S3. Based on the process dataset, construct device control models corresponding to each process unit, wherein the upstream device model is a mechanism model and the terminal aggregation device model is a data-driven model.

[0044] S4, under the constraint of the global optimization setting value, run each device control model in parallel to generate the operating parameter sequence of each process device.

[0045] S5, the operation parameter sequence is sent to the corresponding device for execution, and the actual response data of the device is monitored and collected.

[0046] S6, the actual response data is fed back to the full-process model and each of the device models to perform online model correction and parameter updates, forming a closed loop.

[0047] In this embodiment, step S1 specifically includes the following steps: S11, construct a hierarchical flow traceability model, abstract the entire process material flow into a virtual flow pipeline, and define its source device, inflow device and element composition vector.

[0048] S12 maps multi-source heterogeneous data to corresponding virtual flow channels based on measurement location and physical meaning through a probabilistic graphical model, and uses dynamic time warping algorithm and rule base to achieve time synchronization and dimensional unification.

[0049] S13 employs a rule engine and an isolated forest model for two-stage bad value detection and removal, outputting a unified process dataset that is time-aligned, dimensionally consistent, and flow-related.

[0050] The method as described in claim 1, wherein step S12 specifically performs the following operations: A hierarchical flow traceability model based on material flow tracking is constructed. All physical material flows transmitted through pipelines and equipment in the entire process are abstracted into virtual flow pipelines with unique identifiers. For each flow pipeline, its source device, inlet device, and chemical element composition vector are defined. For heterogeneous data items from different data sources, they are mapped to the corresponding virtual flow pipelines according to their measurement location and physical meaning. A probabilistic graphical model with confidence weights is established for this mapping relationship. Timestamp synchronization is solved by introducing a dynamic time warping algorithm with equipment dwell time and transmission delay as parameters. Dimensional unification is automatically completed by an intelligent conversion rule base integrating international standard units, factory custom units, and online analyzer ranges. A two-stage outlier detection and elimination process based on rule engine and machine learning is adopted. The first stage uses the physical limit change rate threshold and material balance constraints set by the process knowledge base for fast filtering. The second stage uses a trained isolated forest model to identify and eliminate hidden outliers from the overall distribution of multidimensional associated data, and outputs a unified process dataset with time alignment and consistent dimensions.

[0051] In this embodiment, step S2 specifically includes the following steps: S21, construct the chemical element-species association matrix, and define the input and output species vectors of each process unit and the global total species flow vector.

[0052] S22 introduces the conversion rate matrix determined by the reaction mechanism of the device and the diversion and mixing coefficient matrix determined by the pipeline connection to construct a steady-state material balance model for the entire process.

[0053] S23, the product quality requirements of the polypropylene synthesis unit are transformed into constraints on the equation set, and real-time measurement data is integrated as boundary conditions.

[0054] S24, Numerically solve the steady-state material balance model to obtain the baseline flow distribution that satisfies the global material conservation and quality requirements.

[0055] Furthermore, the characteristic is that step S21 specifically performs the following operations: Based on all chemical reactions involved in the entire coal-based polypropylene process, an extended chemical element species association matrix is ​​constructed, covering the entire process from raw coal to final polypropylene products. The rows of the matrix correspond to the three core tracking elements (carbon, hydrogen, and oxygen), and the columns correspond to all key chemical species in the process. Input and output species vectors for each process unit are defined, and a global species total flow vector representing all material streams in the entire process is constructed based on the actual pipeline connection topology. A conversion rate matrix determined by the reaction mechanism, catalyst characteristics, and operating conditions of each unit, as well as a splitting and mixing coefficient matrix determined by pipeline design and control logic, are introduced. The chemical reaction conversion within the unit and the physical transport and distribution processes between units are uniformly represented as a simultaneous steady-state material balance model of the entire process, including constraints from nonlinear equations and inequalities. The key requirements for product quality of downstream polypropylene synthesis units are transformed into constraints on the polypropylene-related flow and properties in this equation set. Measurement data from a real-time database are integrated as boundary conditions for some known variables or equations. Solving the equation set yields a baseline flow distribution that satisfies global material conservation and product quality requirements.

[0056] Furthermore, step S23 specifically involves performing the following operations: Based on the system structure provided by the full-process material balance model and the target output and melt flow index of the polypropylene unit, a reverse optimization problem is constructed with minimizing the output deviation of the final product as the core, while considering the operating costs and stability of the upstream units. This problem is expressed as a constrained nonlinear programming problem with the upstream unit operating parameter vector as the decision variable. The objective function is the weighted sum of squares of the terminal output error, and the constraints cover the operating boundaries of each unit and the quality requirements of intermediate products. To solve this optimization problem, the adjoint method is introduced to calculate the full-process gradient information of the objective function with respect to the upstream operating parameters. This gradient information is obtained by backpropagating the adjoint vector and using the local sensitivity matrix of each unit, thus avoiding the direct calculation of the chain product of the Jacobian matrix of the full-process model. An iterative optimization algorithm based on this gradient is used to drive the update of the upstream operating parameters, so that the terminal output obtained from the full-process chain simulation starting from the updated parameters continuously approaches the preset target until convergence. The optimal setpoints of the operating parameters of each upstream unit that satisfy the global optimum are decoupled in reverse. The core formula of the reverse solution process is: , Define the adjoint vector equation as: , The formula for calculating the gradient of the objective function is: , This is a vector sum of the operating parameters of all upstream units, including gasifiers, methanol synthesis, and MTO. This is a full-process chain-integrated model, with upstream operation parameters as input. The output is a vector predicting the yield and quality of the final polypropylene unit. This model is a mechanism model for each device. , , , Function composition, Let be the output and quality target vector for the polypropylene unit. This is a diagonal positive definite weight matrix used to adjust the relative importance of yield and quality deviation in the objective function. This is a regularization term used to penalize deviations or drastic fluctuations in operating parameters, improving the engineering applicability of the solution. , The adjoint vector has the same dimension as... Similarly, it encapsulates all the indirect, chain-like effects of changes in the terminal target on changes in upstream operating parameters, making it an efficient method for calculating gradients. The key, For the full Jacobian matrix of the upstream operating parameters in the whole process model, its relationship with the vector is efficiently calculated by automatic differentiation in the inverse mode or by inverse solution of the sub-device. The operational relationships.

[0057] Specifically, in this embodiment, a quasi-Newton iterative optimization algorithm based on the aforementioned gradient is used for solving the problem, and a rolling optimization framework is employed to enhance the system's anti-interference capability, including: Rolling optimization form: In each optimization cycle, consider a future finite time domain. The prediction and objective within the scope of the problem, and the optimization problem, are extended to: , in, , Weighting coefficients to suppress operational fluctuations.

[0058] Integrated real-time perturbation: Integrates perturbation data such as real-time fluctuations in raw material properties and catalyst activity decay. As input to the model, construct a structure of the form The disturbance perception model enables optimized instructions to proactively compensate for measurable disturbances.

[0059] Adaptive weight adjustment: weight matrix It can be dynamically adjusted according to the production stage strategy. For example, when pursuing high load, increase the weight of output, and when ensuring product brand switching, increase the weight of quality indicators.

[0060] Furthermore, step S24 specifically involves performing the following operations: Based on the optimal material allocation scheme and composition control range of the whole-process material balance model, a collaborative optimization framework based on Nash bargaining game theory is constructed. The three upstream process units—gasifier, methanol synthesis, and MTO—are defined as game participants, and each participant's strategy set is a vector of independently adjustable operating parameters. This vector includes the oxygen-to-coal ratio and furnace temperature of the gasifier, the inlet hydrogen-to-carbon ratio and reaction pressure of the methanol synthesis unit, and the reaction temperature and methanol feed space velocity of the MTO unit. A composite utility function is designed for each participant, comprehensively considering both individual operational economics and the overall contribution to the entire process. The downstream polypropylene synthesis unit is also included. The rigid requirements for the quantity and quality of the feed materials are transformed into mandatory coupling constraints in the game. By integrating high-fidelity mechanism models of various devices and real-time collected disturbance data as the state transition and payoff calculation engine of the game, an iterative negotiation algorithm is used to solve the Nash equilibrium solution of the cooperative game online. The equilibrium solution is a Pareto optimal combination of operating parameters that enables the upstream devices to operate in coordination under the premise of satisfying individual rationality and collective rationality. After the obtained equilibrium operating parameter combination is verified and smoothed by safe operating boundary, a time synchronization global optimization setpoint instruction that can be directly issued to the control systems of each corresponding device is generated.

[0061] Specifically, the formula for the composite utility function includes: Standardized economic items: , in, For device The real-time production cost function, For the first The unit cost of a resource The first calculation based on the device mechanism model Consumption of various resources , This refers to the extreme values ​​of a sliding window that are dynamically updated based on historical data.

[0062] Quality compliance item: , in, The key quality indicators calculated by the unit's mechanistic model are: effective gas content for the gasifier, methanol selectivity for methanol synthesis, and propylene-to-ethylene ratio for MTO. This refers to the real-time target value issued by the full-process model.

[0063] Global collaboration items: , in, , For device For the The coupling effect of each associated device For device For the device Jacobian sensitivity matrix, This is the coupling tolerance parameter, set according to the buffer capacity between devices.

[0064] Furthermore, the final form of the composite utility function is: , Constraints: .

[0065] Furthermore, the iterative negotiation algorithm employs an alternating direction multiplier method based on augmented Lagrange multipliers to achieve distributed solution, specifically including the following operations: Each upstream device constructs and solves a local optimization subproblem in parallel based on the received global setpoint instruction and the current Lagrange multiplier. The objective function of the subproblem consists of the device's composite utility function and the quadratic penalty term corresponding to the downstream common constraint. The central coordinator collects the locally optimal operating parameters obtained by each device, calculates the actual violation of the downstream common constraint, and updates the Lagrange multiplier according to the violation and the preset multiplier update step size. The updated multiplier and the strategies of each device are broadcast to all participants for the next iteration. The above operations are repeated until the changes in the operating parameters of each device and the violation of the downstream common constraint are both lower than the set threshold. The resulting set of strategies is the Nash equilibrium solution.

[0066] In this embodiment, step S3 specifically includes the following steps: S31, Construct a mechanism model of the gasifier device, including the conservation of element mass, heterogeneous gasification reaction kinetics, and energy conservation equations.

[0067] S32, construct a mechanism model of a methanol synthesis unit to describe the macroscopic kinetics and catalyst deactivation mechanics of the synthesis of methanol by hydrogenation of CO and CO2.

[0068] S33, construct a mechanism model of a methanol-to-olefins unit, and describe the reaction network and catalyst cycle loss dynamics of methanol to olefins based on the hydrocarbon pool mechanism.

[0069] S34, based on historical operation and product quality data, trains a data-driven model as the data-driven model for the polypropylene synthesis unit.

[0070] Furthermore, in step S3, the core of the mechanism model constructed for the gasification furnace device includes: The mass conservation equation is used to balance the distribution of carbon, hydrogen, and oxygen in coal gas, ash, and tar.

[0071] The reaction kinetic equations, which include coal pyrolysis and heterogeneous gasification reactions of carbon with oxygen, water vapor, and carbon dioxide respectively, are described by intrinsic reaction rate equations. Key parameters include pre-exponential factor and activation energy.

[0072] The energy conservation equation is used to calculate the heat input and output of the gasifier.

[0073] The model's inputs are raw coal quality data, oxygen-to-coal ratio, steam-to-coal ratio, and furnace pressure, while the outputs are syngas composition, temperature, and carbon conversion rate.

[0074] Furthermore, in step S3, the core of the mechanism model constructed for the methanol synthesis unit includes: The reaction rate equations, considering both CO hydrogenation and CO2 hydrogenation reactions, are described using a macroscopic kinetic model that takes into account the internal diffusion efficiency factor. Key parameters include the intrinsic kinetic constant, the reaction equilibrium constant, and the catalyst effectiveness factor.

[0075] Catalyst deactivation function describes the decline in catalyst activity over time or cumulative yield.

[0076] The model's inputs are syngas composition, temperature, pressure, and space velocity, and its outputs are outlet gas composition and methanol yield.

[0077] Furthermore, in step S3, a mechanistic model for a methanol-to-olefins unit is constructed, the core of which includes: The reaction network kinetic equations describe the complex network of methanol conversion into olefins and byproducts based on the hydrocarbon pool mechanism. A lumped kinetic model is used, and key parameters include the rate constants of each reaction pathway.

[0078] Catalyst circulation and deactivation dynamics describe the activity decay and regeneration dynamics of catalysts in a reactor due to carbon deposition.

[0079] The model's inputs are methanol feed purity, reaction temperature, and methanol-to-ethanol ratio, and its outputs are the yield distributions of products such as ethylene and propylene.

[0080] Furthermore, in step S3, the data-driven model constructed for the polypropylene synthesis apparatus is a data-driven model, and its construction process includes: Input characteristic variables include: propylene concentration in the feed, hydrogen concentration, catalyst feed rate, and temperature and pressure of each bed in the reactor.

[0081] The target variables for output include: polypropylene yield and melt index.

[0082] The model architecture adopts a long short-term memory network or a spatiotemporal convolutional network structure, uses historical operation data and corresponding product quality test data for supervised training, and adapts to changes in working conditions through an online rolling update mechanism.

[0083] In this embodiment, step S4 specifically includes the following steps: S41, each device's control model receives the global optimization setpoint and transforms it into the device's boundary constraints and tracking targets.

[0084] S42, each device uses its own model as the core to construct a local constraint optimization problem that incorporates upstream and downstream collaboration requirements.

[0085] S43 employs a distributed iterative strategy based on augmented Lagrange multipliers and alternating direction multipliers to solve local problems in parallel and coordinate the consistency of predicted values ​​by exchanging coupling variables.

[0086] S44, after the algorithm converges, outputs the optimal operating parameter sequence that satisfies the material balance of the entire process.

[0087] Furthermore, step S42 specifically involves performing the following operations: Each device's control model receives and parses the global optimization setpoint instruction from step S24, transforming it into the device's boundary constraints and tracking targets. Using the device's mechanism model or data-driven model as the core, it constructs a local constraint optimization problem that includes its own operational optimization target and forcibly embeds upstream and downstream coordination requirements. By adopting a distributed iterative solution strategy based on augmented Lagrange multipliers and alternating direction multipliers, each device can solve its own local problem independently and in parallel, while exchanging prediction information of coupling variables and coordinating consistency. Ultimately, it converges to a set of operating parameter sequences that meet the overall material balance and quality connection requirements of the entire process and are locally optimal.

[0088] like Figure 3 As shown, step S5 specifically includes the following steps: S51 sends the optimal operating parameter sequence of each device to the field actuators through the advanced process control system.

[0089] S52, collects actual operating data and online product quality analysis data after the acquisition device is executed.

[0090] S53 monitors the deviation between actual data and model expected values, and triggers an alert when the deviation exceeds a safety threshold.

[0091] In this embodiment, step S6 specifically includes the following steps: S61, compare the actual response data of each device with the model prediction value to generate a multi-dimensional prediction error vector.

[0092] S62 employs a hierarchical collaborative adaptive filtering algorithm, using export test data as observations, to synchronously estimate and correct the core mechanism parameters of the entire process model and each device model online.

[0093] S63 feeds back the corrected model parameters to the next round of optimization calculation.

[0094] Furthermore, the characteristic is that step S61 specifically performs the following operations: The actual response data of each process unit collected in real time in step S5 is compared with the current predicted value of the corresponding model. The actual response data includes the actual values ​​of key operating variables, the actual values ​​of output material quality and flow rate, and auxiliary variables of equipment status. A multi-dimensional prediction error vector is generated. Based on the prediction error vector, a hierarchical collaborative adaptive filtering algorithm is used to synchronously estimate and dynamically correct the key transfer coefficients in the whole process material balance model and the core mechanism parameters of each unit control model. The core mechanism parameters include reaction kinetic constants and catalyst activity factors. The corrected and updated model parameters are fed back and integrated into the next round of optimization calculation cycle starting from step S2.

[0095] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for optimizing the entire coal-based polypropylene process based on multi-source data, characterized in that, Includes the following steps: S1 collects and integrates multi-source process data covering the entire process of gasification, synthesis, conversion and polymerization to form a unified process dataset; S2, builds and runs a full-process material and energy balance model, and generates global optimization settings for upstream processes based on end-product targets; S3. Based on the process dataset, construct the device control model corresponding to each process unit; S4, under the constraint of the global optimization setting value, run each of the device control models in parallel to generate the operating parameter sequence of each process device; S5, the operation parameter sequence is sent to the corresponding device for execution, and the actual response data of the device is monitored and collected; S6, the actual response data is fed back to the full-process model and each of the device models to perform online model calibration and parameter updates.

2. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 1, characterized in that, S2 includes the following steps: S21, construct the chemical element-species correlation matrix, and define the input and output species vectors of each process unit and the global total species flow vector; S22, introduce the conversion rate matrix determined by the reaction mechanism of the device and the diversion and mixing coefficient matrix determined by the pipeline connection to construct a steady-state material balance model for the entire process; S23, the product quality requirements of the polypropylene synthesis unit are transformed into constraints on the equation set, and real-time measurement data is integrated as boundary conditions; S24, Numerically solve the steady-state material balance model to obtain the baseline flow distribution that satisfies the global material conservation and quality requirements.

3. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 2, characterized in that, Step S21 specifically involves the following operations: Based on all chemical reactions involved in the entire coal-based polypropylene process, an extended chemical element species association matrix is ​​constructed, covering the entire process from raw coal to final polypropylene products. The rows of the matrix correspond to the three core tracking elements (carbon, hydrogen, and oxygen), and the columns correspond to all key chemical species in the process. Input and output species vectors for each process unit are defined, and a global species total flow vector representing all material streams in the entire process is constructed based on the actual pipeline connection topology. A conversion rate matrix determined by the reaction mechanism, catalyst characteristics, and operating conditions of each unit, as well as a splitting and mixing coefficient matrix determined by pipeline design and control logic, are introduced. The chemical reaction conversion within the unit and the physical transport and distribution processes between units are uniformly represented as a simultaneous steady-state material balance model of the entire process, including constraints from nonlinear equations and inequalities. The key requirements for product quality of downstream polypropylene synthesis units are transformed into constraints on the polypropylene-related flow and properties in this equation set. Measurement data from a real-time database are integrated as boundary conditions for some known variables or equations. Solving the equation set yields a baseline flow distribution that satisfies global material conservation and product quality requirements.

4. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 2, characterized in that, Step S23 specifically involves the following operations: Based on the system structure provided by the full-process material balance model and the target output and melt flow index of the polypropylene unit, a reverse optimization problem is constructed with minimizing the output deviation of the final product as the core, while considering the operating costs and stability of the upstream units. This problem is expressed as a constrained nonlinear programming problem with the operating parameter vector of the upstream units as the decision variables. The objective function is the weighted sum of squares of the terminal output error, and the constraints cover the operating boundaries of each unit and the quality requirements of intermediate products. To solve this optimization problem, the adjoint method is introduced to calculate the full-process gradient information of the objective function with respect to the upstream operating parameters. This gradient information is obtained by backpropagating the adjoint vector and using the local sensitivity matrix of each unit. An iterative optimization algorithm based on this gradient is used to drive the update of the upstream operating parameters, so that the terminal output obtained from the full-process chain simulation starting from the updated parameters continuously approaches the preset target until convergence. The optimal operating parameter settings of each upstream unit that satisfy the global optimum are decoupled in reverse. The core formula of the reverse solution process is: ， Define the adjoint vector equation as: ， The formula for calculating the gradient of the objective function is: ， in, This is an aggregate vector of operating parameters for all upstream devices. It is a full-process chain integration model. Let be the output and quality target vector for the polypropylene unit. It is a diagonal positive definite weight matrix. For regularization terms, For the adjoint vector, For the full Jacobian matrix of the upstream operating parameters in the whole process model, its relationship with the vector is efficiently calculated by automatic differentiation in the inverse mode or by inverse solution of the sub-device. The operational relationships.

5. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 2, characterized in that, Step S24 specifically involves the following operations: Based on the optimal material allocation scheme and composition control range of the whole-process material balance model, a collaborative optimization framework based on Nash bargaining game theory is constructed. The three upstream process units—gasifier, methanol synthesis, and MTO—are defined as game participants, and each participant's strategy set is its independently adjustable operational parameter vector. A composite utility function is designed for each participant, comprehensively considering both individual operational economy and global contribution to the entire process. The rigid demand of the downstream polypropylene synthesis unit for the quantity and quality of feed materials is transformed into mandatory coupling constraints of the game. By integrating high-fidelity mechanism models of each unit and real-time acquired disturbance data as the game's state transition and payoff calculation engine, an iterative negotiation algorithm is used to solve the Nash equilibrium solution of this cooperative game online. This equilibrium solution is a Pareto optimal combination of operational parameters that enables the upstream units to operate collaboratively while satisfying individual and collective rationality. After the obtained equilibrium operational parameter combination is verified and smoothed by safe operating boundaries, a time-synchronized global optimization setpoint instruction that can be directly issued to the control systems of each corresponding unit is generated. The composite utility function formula is: ， in, for For the Composite utility function of upstream devices For the first The operational variable vector of each device To exclude the first The vector of operational variables of other devices besides this device. To act on the first Real-time disturbance vector of each device For the first The operating economic index function of each device, and These represent the worst and best historical values ​​of the indicator within the feasible region, respectively. For the first The key quality indicator function of the output material of each device This is the target value for this indicator issued by the full-process model. The vector representing the coupling effect function between devices. , , These are the preference coefficients for economy, quality compliance, and overall synergy, respectively. This is the parameter for tolerance to coupling effects.

6. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 5, characterized in that, The iterative negotiation algorithm employs an augmented Lagrange-based alternating direction multiplier method to achieve distributed solution, specifically including the following operations: Each upstream device constructs and solves a local optimization subproblem in parallel based on the received global setpoint instruction and the current Lagrange multiplier. The objective function of the subproblem consists of the device's composite utility function and the quadratic penalty term corresponding to the downstream common constraint. The central coordinator collects the locally optimal operating parameters obtained by each device, calculates the actual violation of the downstream common constraint, and updates the Lagrange multiplier according to the violation and the preset multiplier update step size. The updated multiplier and the strategies of each device are broadcast to all participants for the next iteration. The above operations are repeated until the changes in the operating parameters of each device and the violation of the downstream common constraint are both lower than the set threshold. The resulting set of strategies is the Nash equilibrium solution.

7. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 1, characterized in that, Step S4 includes: S41, Each device control model receives the global optimization setpoint and transforms it into the boundary constraints and tracking targets of its own device; S42, each device uses its own model as the core to construct a local constraint optimization problem that embeds upstream and downstream collaboration requirements; S43 employs a distributed iterative strategy based on augmented Lagrange multiplier and alternating direction multiplier method to solve local problems in parallel and coordinate the consistency of predicted values ​​by exchanging coupling variables. S44, after the algorithm converges, outputs the optimal operating parameter sequence that satisfies the material balance of the entire process.

8. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 7, characterized in that, Step S42 specifically involves the following operations: Each unit's control model receives and parses the global optimization setpoint instruction, transforming it into the unit's boundary constraints and tracking targets. Using the unit's mechanism model or data-driven model as the core, it constructs a local constraint optimization problem that includes its own operational optimization target and forcibly embeds upstream and downstream coordination requirements. By adopting a distributed iterative solution strategy based on augmented Lagrange multipliers and alternating direction multipliers, each unit can solve its own local problem independently and in parallel. By exchanging the prediction information of coupled variables and coordinating consistency, it ultimately converges to a set of operating parameter sequences that meet the overall material balance and quality connection requirements of the entire process and are locally optimal.

9. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 1, characterized in that, Step S6 includes: S61, compare the actual response data of each device with the model prediction values ​​to generate a multi-dimensional prediction error vector; S62 uses a hierarchical collaborative adaptive filtering algorithm, with the export test data as the observation value, to synchronously estimate and correct the core mechanism parameters of the whole process model and each device model online. S63 feeds back the corrected model parameters to the next round of optimization calculation.

10. The method for optimizing the entire coal-based polypropylene process based on multi-source data according to claim 9, characterized in that, Step S61 specifically involves the following operations: The actual response data is compared with the current predicted value of the corresponding model to generate a multi-dimensional prediction error vector. A hierarchical collaborative adaptive filtering algorithm is used to synchronously estimate and dynamically correct the key transfer coefficients in the whole process material balance model and the core mechanism parameters of each unit control model. The corrected and updated model parameters are fed back and integrated into the next round of optimization calculation cycle starting from step S2.

Images