Shell device multi-scene sequential control integrated system based on PKS system

Abstract

The invention relates to the technical field of process industry automation, and discloses a shell device multi-scene sequential control integrated system based on a PKS system. The real-time working condition sensing module dynamically extracts working condition characteristic parameters; the self-adaptive strategy generation module carries out intelligent decision making and generates a dynamic strategy packet containing program branch selection and delay parameters; the sequence control dynamic execution module intervenes in a running sequence control program in real time through a PKS control interface; and the execution efficiency evaluation module collects the data after execution, and completes closed-loop inspection through multi-dimensional efficiency evaluation. By constructing a full-link intelligent sequence control system of sensing, decision making, execution and evaluation, the industrial problem that the static logic design of the sequence control program is not matched with the dynamic working condition change is solved, the technical spanning from static execution to dynamic adaptation is realized, and the safety, stability and economical efficiency of the device under the complex working condition are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of process industry automation technology, and in particular to a Shell unit multi-scenario sequential control integrated system based on the PKS system. Background Technology

[0002] Existing technologies for Shell's multi-scenario sequential control integration system based on the PKS system primarily rely on sequential control programs configured on the Honeywell Experion PKS platform. These programs, built using Sequential Function Chart or Phase Logic, arrange a series of equipment operation steps according to a preset logical sequence and fixed time delays, thereby achieving automated operation under specific conditions. Typical applications include automatic start-up and shutdown of equipment units, switching of material conveying paths, and setpoint adjustment of single process parameters such as the oxygen-coal ratio. The core technology transforms validated manual operating procedures into standardized, repeatable machine instructions. Through the powerful control algorithm library and graphical configuration tools of the PKS system, stable and reliable program control is achieved, significantly reducing the reliance on human experience and the risk of misoperation in routine operations.

[0003] The main problem with the existing technology is that the sequential control program uses static logic design and fixed parameter settings, which cannot perceive and adapt to the dynamic changes in boundary conditions such as coal quality, equipment performance, and load requirements during the production process. This leads to adjustment failure or poor operation effect under complex working conditions, and still requires frequent operator intervention.

[0004] Publication number CN218383718U discloses a control system for a bulk fly ash loading machine in a thermal power plant. It uses a microprocessor to execute preset sequential operations such as lifting and lowering the bulk loading machine and starting and stopping the feeding valve, which solves the problems of low efficiency and dust overflow in manual operation. However, it does not solve the problem that the corresponding sequential control program cannot automatically adjust the operation sequence and parameters according to dynamic factors such as material characteristics and equipment wear.

[0005] Publication No. CN119126688A discloses an operation control system for a pulverized coal gasifier. It provides optimization suggestions such as oxygen-coal ratio through data balance verification and parameter optimization modules, which solves the problem of excessive pulverized coal consumption caused by inaccurate manual control. However, it does not solve the problem of how to convert the above-mentioned optimization conclusions into dynamic execution instructions of the sequential control program in real time and safely. Its optimization module is disconnected from the underlying sequential control execution engine. Summary of the Invention

[0006] The technical problem to be solved by this invention is that the existing sequential control program adopts static logic design and fixed parameter settings, which cannot perceive and adapt to the dynamic changes of boundary conditions such as coal quality, equipment performance, and load requirements during the production process. To this end, we propose a Shell unit multi-scenario sequential control integrated system based on the PKS system.

[0007] To achieve the above objectives, this application adopts the following technical solution: a Shell device multi-scenario sequential control integrated system based on the PKS system, comprising: The real-time operating condition perception module receives operating condition sensor data in real time, removes outliers through data cleaning, processes multi-source data using time series alignment, extracts coal quality fluctuation index and equipment health index, and outputs operating condition feature vectors to the adaptive strategy generation module.

[0008] The adaptive strategy generation module receives the operating condition feature vector, identifies the operating mode through a machine learning classifier, retrieves the operation strategy of similar historical cases from the historical case library, uses a reinforcement learning algorithm to optimize the current strategy parameters, generates a dynamic strategy package containing program branch selection and delay parameters, and outputs it to the sequential control dynamic execution module.

[0009] The sequential control dynamic execution module acquires dynamic strategy packages, parses and executes instructions, modifies the transfer conditions and timer settings of the running sequential control program through the PKS control interface, and outputs equipment control instructions to the field actuators.

[0010] The performance evaluation module collects process parameters and equipment status data after execution, calculates performance scores through a multi-objective evaluation model, generates case data, and outputs it to the adaptive strategy generation module to complete closed-loop optimization.

[0011] Preferably, the real-time operating condition sensing module includes: The data cleaning unit receives real-time data from operating condition sensors, including coal quality data, equipment operating parameters, and load commands. It applies limit checks and rate change checks to identify and remove abnormal data points, outputs verified valid data, and sends it to the data synchronization unit.

[0012] The data synchronization unit receives the verified multi-source data, uses a timestamp alignment algorithm to unify data with different sampling frequencies to the same time base, outputs a time-synchronized data sequence, and outputs it to the feature calculation unit.

[0013] The feature calculation unit receives the synchronized data sequence, calculates dynamic feature parameters in real time using the coal quality impact factor calculation model and equipment performance evaluation algorithm, including but not limited to the coal quality fluctuation index and equipment health index, and outputs a set of dynamic feature parameters to the vector assembly unit.

[0014] The vector assembly unit receives a set of feature parameters, performs normalization and vectorization encapsulation, and outputs a high-dimensional working condition feature vector, which is then fed to the adaptive strategy generation module.

[0015] Preferably, the adaptive strategy generation module includes: The working condition classification unit receives the working condition feature vector, performs real-time pattern recognition through a pre-trained neural network model, outputs the current working condition type identifier, and sends it to the case retrieval unit.

[0016] The case retrieval unit receives the working condition type identifier and feature vector, performs similarity matching in the historical case database, outputs relevant historical strategies, and sends them to the strategy optimization unit.

[0017] The strategy optimization unit receives historical strategies, operating condition feature vectors, and operating condition type identifiers. It performs strategy effect deduction through a simplified process model, adjusts strategy parameters by comprehensively considering economic and safety indicators, and outputs the optimized strategy parameters to the strategy verification unit.

[0018] The strategy verification unit receives the optimized strategy, performs logical consistency and security checks, and outputs the final strategy instruction package that has passed verification, which includes program branch selection and delay parameters, and outputs it to the sequential control dynamic execution module.

[0019] Furthermore, the working condition classification unit specifically performs the following operations: The system receives feature vectors from the operating condition perception module, performs standardized preprocessing and feature dimensionality reduction on the input data, calls a pre-trained deep neural network classification model to perform operating condition pattern recognition, classifies the current operating state into a preset typical operating condition category, calculates the classification confidence and outputs the category probability distribution, and packages the operating condition category identifier and related feature vectors and outputs them to the case retrieval unit and strategy optimization unit.

[0020] Furthermore, the strategy optimization unit specifically performs the following operations: The current operating condition feature vector and the retrieved historical strategies together constitute a reinforcement learning state space. The strategy effect is deduced by simplifying the process model to maximize long-term cumulative reward. A constrained Markov decision algorithm is used for strategy search, where the reward function comprehensively considers economic and safety indicators, and the constraints are embedded in the algorithm to ensure that all exploration strategies meet the safe operating window of key process parameters. The output is the strategy parameters that converge after hundreds of simulation iterations and have the highest expected return. The strategy search formula is: , ; in, Strategies to be optimized The expected value of the following mathematical expression This is a mapping from state to parameter adjustment. For time step The state at any given time is composed of operating condition characteristics and historical strategy characteristics. For time step The actions taken at that time, namely, the fine-tuning of the sequential control program parameters, The reward function is calculated by weighting economic indicators, safety indicators, and operational penalty. As a discount factor, it weighs the importance of current and future rewards. For example, the difference between the predicted temperature and the safe temperature.

[0021] Preferably, the sequential control dynamic execution module includes: The instruction parsing unit receives dynamic policy packets, parses the target program identifier and parameter modification requirements, outputs structured execution instructions, and outputs them to the interface adaptation unit.

[0022] The interface adaptation unit receives structured execution instructions, converts them into control commands that the PKS system can recognize, modifies the transfer conditions and timer settings of the running sequential control program through the PKS control interface, and outputs standardized control commands to the conflict detection unit.

[0023] The conflict detection unit receives control commands, checks multiple sequential control programs for control requests to the same device, outputs conflict detection results, and sends them to the execution scheduling unit.

[0024] The execution scheduling unit receives control commands and conflict detection results, sorts the commands according to preset priorities, outputs ordered equipment control commands, and outputs them to the field actuators.

[0025] Preferably, the performance evaluation module includes: The data acquisition unit receives post-execution process parameters and equipment status data from the operating condition data, performs data quality checks, outputs standardized evaluation data, and sends it to the performance evaluation unit.

[0026] The performance evaluation unit receives standardized evaluation data, calculates and outputs a comprehensive performance score through a multi-objective evaluation model, including economic indicators and stability indicators, and outputs it to the case generation unit.

[0027] The case generation unit receives performance scores and operating condition data, automatically extracts effective features, outputs standardized case data, and sends it to the knowledge base management unit.

[0028] The knowledge base management unit receives case data, performs deduplication and quality screening of similar cases, updates the historical case library, and outputs the results to the adaptive strategy generation module.

[0029] The adaptive learning unit receives long-term operating data, including historical operating condition data, historical strategy data, historical performance data, and long-term performance degradation trends of the equipment. It adjusts the evaluation model parameters and weight coefficients, outputs the optimized model parameters, and sends them to the performance evaluation unit.

[0030] Furthermore, the performance evaluation unit specifically performs the following operations: receiving post-execution process parameters and equipment status data collected by the real-time operating condition perception module, including key performance indicators such as gasifier effective gas production, carbon conversion rate, and oxygen consumption; calling benchmark data from the historical case library for comparative analysis; calculating the economic index score using a multi-objective evaluation model; and combining the stability index of equipment safety operation data and environmental emission data, weighting and summing the two to obtain a comprehensive performance score; generating a detailed evaluation report containing the comprehensive performance score, performance analysis of each indicator, and improvement suggestions; outputting the report to the case generation unit as a case quality criterion; and feeding it back to the adaptive strategy generation module.

[0031] Furthermore, the case generation unit specifically performs the following operations: The system receives the execution performance score output by the performance evaluation unit, filters candidate cases using a case quality filtering algorithm to eliminate redundant and low-quality cases, standardizes and encapsulates high-quality cases, and utilizes a generative adversarial network framework. The generator takes the coal quality fluctuation index and equipment health index as input conditions to generate marginal cases located on the boundary of the real working condition distribution. The discriminator performs a true / false judgment task based on real case data in the historical case library. Through adversarial training, the generator continuously optimizes the generation of highly realistic virtual cases. The virtual cases generated by the adversarial training are then integrated with high-quality cases extracted from actual operation and integrated into the historical case library through a case library management mechanism to establish semantic relationships and multi-dimensional retrieval indexes between cases.

[0032] Furthermore, the adaptive learning unit specifically performs the following operations: The system continuously monitors the differences between the current operating condition and the feature distribution of all operating conditions in the historical case library. When the current operating condition is detected to belong to a sparse data region or a completely new operating condition, the transfer learning algorithm is automatically activated. It selects useful knowledge parameters from source operating condition models with similar features and integrates them into the evaluation model of the current target operating condition through dynamically calculated transferability weights. The formula for calculating the transferability weights is as follows: ; in, To start from the source condition To the target working condition The transferability weights, The attenuation coefficient controls the sensitivity to the influence of distribution differences. To measure the maximum mean difference, the distribution of source operating condition data is... Distribution of target operating condition data The differences between them For source model Historical performance scores under its corresponding operating conditions These are the parameters for all candidate source models.

[0033] The technical effects and advantages of this invention are as follows: In this invention, the real-time operating condition perception module achieves real-time perception of boundary conditions such as coal quality, equipment performance, and load requirements through dynamic feature extraction; the adaptive strategy generation module generates dynamic strategy packages through intelligent decision-making, solving the problems of static logic and fixed parameters in sequential control programs; the sequential control dynamic execution module completes real-time intervention in the sequential control program through the execution engine, solving the problem of the program's inability to dynamically adjust; the execution performance evaluation module completes closed-loop verification of strategy execution effects through multi-dimensional performance evaluation; and the case generation unit completes the automated accumulation of operational experience through case generation and knowledge management. The system as a whole realizes the leap from static execution to dynamic adaptation of sequential control programs, significantly improving the operational safety, stability, and economy of Shell units under complex operating conditions, and making a breakthrough contribution to the intelligent upgrading of sequential control technology in the process industry. Attached Figure Description

[0034] 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 system block diagram of the present invention. Detailed Implementation

[0035] 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.

[0036] Reference Figure 1 As shown, the present invention provides a technical solution: a Shell device multi-scenario sequential control integrated system based on the PKS system, comprising: The real-time operating condition sensing module receives operating condition sensor data in real time, including data from coal composition analyzers, equipment vibration sensor readings, and pressure and temperature transmitter signals. It cleans and removes outliers, such as jump signals exceeding the measurement range. The module then uses time-series alignment to process the multi-source data, extracting the coal quality fluctuation index and equipment health index. The coal quality fluctuation index reflects the quantitative changes in coal composition, while the equipment health index is a comprehensive parameter characterizing the degree of equipment performance degradation. Finally, it outputs an operating condition feature vector to the adaptive strategy generation module.

[0037] The adaptive strategy generation module receives operating condition feature vectors, identifies operating modes through a machine learning classifier (e.g., high-load stable combustion mode, coal quality drastic change mode), retrieves similar historical case operation strategies from the historical case library, and uses reinforcement learning algorithms to optimize the current strategy parameters, generating a dynamic strategy package containing program branch selection and delay parameters (e.g., selecting an economic optimization branch or a safety protection branch, and delay parameters such as equipment start-stop interval time). This package is then output to the sequential control dynamic execution module.

[0038] The sequential control dynamic execution module acquires dynamic strategy packages, parses execution instructions, and modifies the transfer conditions and timer settings of the running sequential control program through the PKS control interface. Transfer conditions include step jump logic judgment conditions, and timer settings include step execution waiting time. It outputs equipment control instructions such as motor start / stop commands and valve opening signals, and sends the equipment control instructions to the field actuators.

[0039] The performance evaluation module collects process parameters and equipment status data after execution. Process parameters include steady-state values ​​such as gasifier temperature and pressure, while equipment status data includes equipment operating current and vibration values. The module calculates performance scores through a multi-objective evaluation model, generates case data, and outputs it to the adaptive strategy generation module to complete closed-loop optimization.

[0040] In this embodiment, the real-time operating condition sensing module includes: The data cleaning unit receives real-time data from operating condition sensors, including coal quality data, equipment operating parameters, and load commands. It applies limit checks and rate change checks to identify and remove abnormal data points, outputs verified valid data, and sends it to the data synchronization unit.

[0041] The data synchronization unit receives the verified multi-source data, uses a timestamp alignment algorithm to unify data with different sampling frequencies to the same time base, outputs a time-synchronized data sequence, and outputs it to the feature calculation unit.

[0042] The feature calculation unit receives the synchronized data sequence, calculates dynamic feature parameters in real time using the coal quality impact factor calculation model and equipment performance evaluation algorithm, including but not limited to the coal quality fluctuation index and equipment health index, and outputs a set of dynamic feature parameters to the vector assembly unit.

[0043] The vector assembly unit receives a set of feature parameters, performs normalization and vectorization encapsulation, and outputs a high-dimensional working condition feature vector, which is then fed to the adaptive strategy generation module.

[0044] Furthermore, the feature calculation unit specifically performs the following operations: The system receives time-synchronized data sequences from the data synchronization unit, including real-time coal composition data, equipment operating current, and vibration spectrum data. Using a coal quality influencing factor calculation model, it calculates the comprehensive influencing factor of ash and volatile matter, the coal quality reactivity index, and the calorific value fluctuation rate in real time, and integrates them to generate a coal quality fluctuation index. Simultaneously, using an equipment performance evaluation algorithm based on a real-time pump efficiency calculation model, a fan vibration trend analysis algorithm, and a heat exchanger scaling monitoring model, it calculates the pump efficiency attenuation coefficient, fan vibration health, and heat exchanger scaling index, and comprehensively generates an equipment health index. The coal quality fluctuation index, equipment health index, and various sub-feature parameters are integrated into a dynamic feature parameter set and output to the vector assembly unit for further processing.

[0045] In this embodiment, the adaptive strategy generation module includes: The working condition classification unit receives the working condition feature vector, performs real-time pattern recognition through a pre-trained neural network model, outputs the current working condition type identifier, and sends it to the case retrieval unit.

[0046] The case retrieval unit receives the working condition type identifier and feature vector, performs similarity matching in the historical case database, outputs relevant historical strategies, and sends them to the strategy optimization unit.

[0047] The strategy optimization unit receives historical strategies, operating condition feature vectors, and operating condition type identifiers. It performs strategy effect deduction through a simplified process model, adjusts strategy parameters by comprehensively considering economic and safety indicators, and outputs the optimized strategy parameters to the strategy verification unit.

[0048] The strategy verification unit receives the optimized strategy, performs logical consistency and security checks, and outputs the final strategy instruction package that has passed verification, which includes program branch selection and delay parameters, and outputs it to the sequential control dynamic execution module.

[0049] Furthermore, the working condition classification unit specifically performs the following operations: The system receives feature vectors from the operating condition perception module, performs standardized preprocessing and feature dimensionality reduction on the input data, calls a pre-trained deep neural network classification model to perform operating condition pattern recognition, classifies the current operating state into a preset typical operating condition category, calculates the classification confidence and outputs the category probability distribution, and packages the operating condition category identifier and related feature vectors and outputs them to the case retrieval unit and strategy optimization unit.

[0050] Furthermore, the strategy optimization unit specifically performs the following operations: The feature vectors of the previous operating conditions and the retrieved historical strategies together constitute a reinforcement learning state space. By simplifying the process model, such as a process model based on mass and energy conservation, the strategy effect is deduced with the goal of maximizing long-term cumulative reward. A constrained Markov decision algorithm is used for strategy search, where the reward function comprehensively considers economic and safety indicators. Economic indicators include effective gas production and oxygen consumption, while safety indicators include temperature fluctuation range and equipment load rate. Constraints are embedded in the algorithm to ensure that all explored strategies meet the safe operating window of key process parameters, such as the safe operating window of temperature and the maximum allowable load of equipment. The output is the strategy parameters that converge after hundreds of simulation iterations and have the highest expected return. The strategy search formula is: , ; in, Strategies to be optimized The expected value of the following mathematical expression This is a mapping from state to parameter adjustment. For time step The state at any given time is composed of operating condition characteristics and historical strategy characteristics. For time step The actions taken at that time, namely, the fine-tuning of the sequential control program parameters, The reward function is calculated by weighting economic indicators, safety indicators, and operational penalty. As a discount factor, it weighs the importance of current and future rewards. For example, the difference between the predicted temperature and the safe temperature.

[0051] In this embodiment, the sequential dynamic execution module includes: The instruction parsing unit receives dynamic policy packets, parses the target program identifier and parameter modification requirements, outputs structured execution instructions, and outputs them to the interface adaptation unit.

[0052] The interface adaptation unit receives structured execution instructions, converts them into control commands that the PKS system can recognize, modifies the transfer conditions and timer settings of the running sequential control program through the PKS control interface, and outputs standardized control commands to the conflict detection unit.

[0053] The conflict detection unit receives control commands, checks multiple sequential control programs for control requests to the same device, outputs conflict detection results, and sends them to the execution scheduling unit.

[0054] The execution scheduling unit receives control commands and conflict detection results, sorts the commands according to preset priorities, outputs ordered equipment control commands, and outputs them to the field actuators.

[0055] Furthermore, the instruction parsing unit specifically performs the following steps: The strategy package is format-decoded and integrity-verified. The target program identifier in the strategy package is parsed to determine the sequential control program instance that needs to be modified. At the same time, the parameter modification requirements are extracted, including program branch selection instructions and delay parameter adjustment values. The parsed instruction elements are converted into structured execution instructions containing operation objects, operation types and parameter values ​​according to preset mapping rules. The structured execution instructions are sorted and encapsulated according to execution priority and output to the interface adaptation unit for further processing.

[0056] Furthermore, the interface adaptation unit specifically performs the following steps: The operation objects, operation types, and parameter values ​​in the structured execution instructions are parsed and verified, and then converted into control commands that the PKS system can recognize. This includes mapping program branch selection instructions to sequential control program step jump commands, converting delay parameter adjustment values ​​into timer setting commands, modifying the transfer condition register and timer setting value register of the running sequential control program through the API interface provided by the PKS system, encapsulating the modification results into standardized control instructions, including execution status feedback and timing identifiers, and outputting them to the conflict detection unit for further processing.

[0057] In this embodiment, the performance evaluation module includes: The data acquisition unit receives post-execution process parameters and equipment status data from the operating condition data, performs data quality checks, outputs standardized evaluation data, and sends it to the performance evaluation unit.

[0058] The performance evaluation unit receives standardized evaluation data, calculates and outputs a comprehensive performance score through a multi-objective evaluation model, including economic indicators and stability indicators, and outputs it to the case generation unit.

[0059] The case generation unit receives performance scores and operating condition data, automatically extracts effective features, outputs standardized case data, and sends it to the knowledge base management unit.

[0060] The knowledge base management unit receives case data, performs deduplication and quality screening of similar cases, updates the historical case library, and outputs the results to the adaptive strategy generation module.

[0061] The adaptive learning unit receives long-term operating data, including historical operating condition data, historical strategy data, historical performance data, and long-term performance degradation trends of the equipment. It adjusts the evaluation model parameters and weight coefficients, outputs the optimized model parameters, and sends them to the performance evaluation unit.

[0062] Furthermore, the performance evaluation unit specifically performs the following operations: receiving post-execution process parameters and equipment status data collected by the real-time operating condition perception module, including key performance indicators such as gasifier effective gas production, carbon conversion rate, and oxygen consumption; calling benchmark data from the historical case library for comparative analysis; calculating the economic index score using a multi-objective evaluation model; and combining the stability index of equipment safety operation data and environmental emission data, weighting and summing the two to obtain a comprehensive performance score; generating a detailed evaluation report containing the comprehensive performance score, performance analysis of each indicator, and improvement suggestions; outputting the report to the case generation unit as a case quality criterion; and feeding it back to the adaptive strategy generation module.

[0063] Furthermore, the case generation unit specifically performs the following operations: The system receives the execution performance score output by the performance evaluation unit. Candidate cases are comprehensively evaluated and screened based on case novelty, strategy effectiveness, and operational condition coverage using a case quality filtering algorithm. Redundant and low-quality cases are eliminated, and high-quality cases are standardized and encapsulated. A generative adversarial network (GAN) framework is used, where the generator takes coal quality fluctuation index and equipment health index as input conditions to produce edge cases located on the boundary of real operational condition distribution, such as operational strategies under extremely high ash coal conditions. The discriminator performs a true / false judgment task based on real case data in the historical case library. Through adversarial training, the generator continuously optimizes to produce highly realistic virtual cases. The virtual cases generated by adversarial training are then integrated with high-quality cases extracted from actual operation and integrated into the historical case library through a case library management mechanism, establishing semantic relationships and multi-dimensional retrieval indexes between cases.

[0064] Furthermore, the adaptive learning unit specifically performs the following operations: The system continuously monitors the differences between the current operating condition and the feature distribution of all operating conditions in the historical case library. When the current operating condition is detected to belong to a sparse data region or a completely new operating condition, the transfer learning algorithm is automatically activated. It selects useful knowledge parameters from source operating condition models with similar features and integrates them into the evaluation model of the current target operating condition through dynamically calculated transferability weights. The formula for calculating the transferability weights is as follows: ; in, To start from the source condition To the target working condition The transferability weights, The attenuation coefficient controls the sensitivity to the influence of distribution differences. To measure the maximum mean difference, the distribution of source operating condition data is... Distribution of target operating condition data The differences between them For source model Historical performance scores under its corresponding operating conditions These are the parameters for all candidate source models.

[0065] 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 Shell plant multi-scenario sequence control integration system based on a PKS system, comprising: a real-time working condition perception module that receives working condition sensor data in real time, removes abnormal values through data cleaning, aligns multiple source data using a time sequence, extracts coal quality fluctuation indexes and equipment health degree indexes, and outputs working condition feature vectors to an adaptive strategy generation module; the adaptive strategy generation module receives working condition feature vectors, identifies operating modes through a machine learning classifier, retrieves historical similar case operation strategies in a historical case library, optimizes current strategy parameters based on the historical similar case operation strategies, generates a dynamic strategy package containing program branch selection and delay parameters, and outputs the dynamic strategy package to a sequence control dynamic execution module; the sequence control dynamic execution module acquires the dynamic strategy package, parses execution instructions, modifies transfer conditions and timer setting values of a running sequence control program through a PKS control interface, and outputs device control instructions to a field execution mechanism; an execution performance evaluation module collects process parameters and equipment state data after execution, calculates performance scores through a multi-objective evaluation model, generates case data, and outputs the case data to the adaptive strategy generation module for closed-loop optimization.

2. The PKS system based Shell unit multi-scenario sequence control integrated system according to claim 1, characterized in that: The real-time working condition perception module comprises: a data cleaning unit that receives working condition sensor data in real time, applies limit value checking and rate change checking rules, identifies and removes abnormal data points, and outputs valid data that has passed verification; a data synchronization unit that receives multiple source data that has passed verification, aligns data of different sampling frequencies to the same time reference using a timestamp alignment algorithm, and outputs time-synchronized data sequences; a feature calculation unit that receives synchronized data sequences, calculates dynamic feature parameters in real time through a coal quality influence factor calculation model and an equipment performance evaluation algorithm, and outputs a dynamic feature parameter set; a vector assembly unit that receives the dynamic feature parameter set, performs normalization processing and vectorization packaging, and outputs a high-dimensional working condition feature vector.

3. The PKS system based SCOP integrated system for Shell unit according to claim 1, characterized in that: The adaptive strategy generation module comprises: a working condition classification unit that receives working condition feature vectors, performs real-time mode recognition through a pre-trained neural network model, and outputs a current working condition type identifier; a case retrieval unit that receives the working condition type identifier and the feature vectors, performs similarity matching in a historical case library, and outputs related historical strategies; a strategy optimization unit that receives historical strategies, working condition feature vectors, and working condition type identifiers, performs strategy effect deduction through a simplified process model, adjusts strategy parameters based on economic indicators and safety indicators, and outputs optimized strategy parameters; a strategy verification unit that receives optimized strategies, performs logic consistency and safety checks, and outputs a final strategy instruction package that has passed verification, including program branch selection and delay parameters.

4. The PKS system based Shell unit multi-scenario sequence control integrated system according to claim 3, characterized in that: The working condition classification unit specifically performs the following operations: receives feature vectors from the working condition perception module, performs standardization preprocessing and feature dimension reduction on input data, calls a pre-trained deep neural network classification model to perform working condition mode recognition, classifies the current operating state into a preset typical working condition category, calculates a classification confidence and outputs a category probability distribution, and packs the working condition category identifier and related feature vectors and outputs them to the case retrieval unit and the strategy optimization unit.

5. The PKS system based SCOP integrated system for Shell unit according to claim 3, characterized in that: The strategy optimization unit specifically performs the following operations: The current working condition feature vector and the retrieved historical strategy jointly constitute a reinforcement learning state space, a strategy effect deduction is performed by simplifying a process model to maximize long-term cumulative reward, a constrained Markov decision algorithm is used for strategy search, a reward function comprehensively considers economic indicators and safety indicators, and a constraint condition is embedded in the algorithm to ensure that all exploration strategies meet the safe operation window of key process parameters, and the strategy parameters of the strategy with the highest expected return after hundreds of simulation iterations are output, and the strategy search formula is: ， ； in, Strategies to be optimized The expected value of the following mathematical expression For time step The state at that time, For time step The actions taken at that time For the reward function, As a discount factor, For constraint functions.

6. The PKS system based SCOP integrated system for Shell unit according to claim 1, characterized in that: The sequence control dynamic execution module comprises: An instruction analysis unit receives a dynamic strategy package, analyzes a target program identifier and a parameter modification requirement, and outputs a structured execution instruction; An interface adaptation unit receives the structured execution instruction, converts it into a control command recognizable by the PKS system, modifies the transfer condition and timer set value of the sequence control program in operation through the PKS control interface, and outputs a standardized control instruction; A conflict detection unit receives the control instruction, checks the control requests of multiple sequence control programs to the same device, and outputs a conflict detection result; An execution scheduling unit receives the control instruction and the conflict detection result, sorts the instructions according to a preset priority, outputs an ordered device control instruction, and outputs it to a field execution mechanism.

7. The PKS system based SCOP integrated system for Shell unit according to claim 1, characterized in that: The execution performance evaluation module comprises: A data acquisition unit receives post-execution process parameters and device state data in working condition data, performs data quality inspection, and outputs standardized evaluation data; An efficiency evaluation unit receives the standardized evaluation data, calculates and outputs a comprehensive efficiency score including economic indicators and stability indicators through a multi-objective evaluation model; A case generation unit receives the efficiency score and the working condition data, automatically extracts effective features, and outputs standardized case data; A knowledge base management unit receives the case data, performs similar case deduplication and quality screening, and updates a historical case library; An adaptive learning unit receives long-term operation data, adjusts evaluation model parameters and weight coefficients, and outputs model optimization parameters.

8. The PKS system based Shell unit multi-scenario sequence control integrated system according to claim 7, characterized in that: The working condition classification unit specifically performs the following operations: Receives a feature vector from a working condition perception module, performs standardization preprocessing and feature dimension reduction on input data, calls a pre-trained deep neural network classification model to perform working condition mode recognition, classifies the current running state into a preset typical working condition category, calculates a classification confidence and outputs a category probability distribution, and packs the working condition category identifier and related feature vectors and outputs them to a case retrieval unit and a strategy optimization unit.

9. The PKS system based SCOP integrated system for Shell unit according to claim 7, characterized in that: The case generation unit specifically performs the following operations: The execution effect score output by the receiving performance evaluation unit is used to filter the candidate cases through a case quality filtering algorithm, redundant and low-quality cases are removed, high-quality cases are standardized and packaged, and a generative adversarial network framework is used, wherein the generator takes the coal quality fluctuation index and the equipment health index as the condition input to generate edge cases located on the boundary of the real working condition distribution, and the discriminator performs a true or false discrimination task based on the real case data in the historical case library, the generator is continuously optimized through the adversarial training to generate high-fidelity virtual cases, the virtual cases generated through the adversarial training are fused with the high-quality cases extracted from the actual operation, and the fused cases are integrated into the historical case library through a case library management mechanism to establish a semantic association relationship between the cases and a multi-dimensional retrieval index.

10. The PKS system based SCOP integrated system for Shell unit according to claim 1, characterized in that: The adaptive learning unit specifically performs the following operations: The adaptive learning unit specifically performs the following operations: The adaptive learning unit specifically performs the following operations: ； wherein, is the migratability weight for migrating from a source working condition to a target working condition , is the decay coefficient, controlling the sensitivity of the distribution difference impact, is the maximum mean difference, measuring the difference between the source working condition data distribution and the target working condition data distribution , is the historical performance score of the source model on its corresponding working condition, is the parameter of all candidate source models.

Images