PPU-based simulation optimization method and system for gas turbine combined cycle DCS

Abstract

The application provides a PPU-based gas turbine combined cycle DCS simulation optimization method and system, which comprises the following steps: constructing a parallel simulation system of a gas turbine and a combined cycle DCS based on a parallel controller (PPU), establishing a simulation mirror system running in parallel with a physical DCS system, realizing high-speed synchronization of I / O measurement point data between the physical DCS and the PPU through an FPGA acceleration unit, and supporting the PPU to write back the optimized control parameters to the physical DCS without disturbance; a fractional calculus model is used to model the gas turbine thermal system inside the PPU; the parallel computing capability of the PPU is used to run multiple simulation instances in advance of the physical time in the background; a multi-objective genetic algorithm is used to optimize the target functions of maximizing the unit thermal efficiency and minimizing the pollutant emission. A heterogeneous data synchronization interface based on shared memory and FPGA acceleration is designed to solve the data interaction bottleneck between the physical DCS and the simulation system.

Description

Technical Field

[0001] This specification relates to the field of gas turbine simulation technology, specifically to a PPU-based DCS simulation optimization method and system for combined cycle gas turbines. Background Technology

[0002] Currently, thermal power and gas turbine simulation systems are widely used in scenarios such as unit operation simulation, control logic verification, operating condition simulation, and operation and maintenance training, providing important technical support for the safe and stable operation of gas-steam combined cycle units. However, in practical engineering applications, existing simulation and control systems still have many technical defects, such as poor model adaptability, lagging control strategies, insufficient ultra-real-time simulation capabilities, and lack of system coupling optimization, making it difficult to meet the operation and control requirements of combined cycle units under varying operating conditions, dynamic operation, high precision, and high energy efficiency.

[0003] First, the core models of existing thermal power and gas turbine simulation systems are mostly built once based on the unit's factory design data, with fixed internal parameters that cannot be dynamically updated. This type of static modeling can only adapt to the unit's rated design conditions and cannot match the actual changes in operating characteristics caused by factors such as equipment wear, component aging, environmental parameter fluctuations, and maintenance modifications during long-term operation. It lacks the ability to self-evolve and dynamically calibrate based on real-time on-site operating conditions, resulting in a continuous accumulation of deviations between the simulation model and the actual unit operating state. This significantly reduces simulation accuracy and operating condition reproducibility, failing to provide reliable model support for dynamic control and operating condition optimization of the unit.

[0004] Secondly, existing DCS control systems for generating units generally adopt a fixed "black box" logic control architecture. The control logic and adjustment parameters are all preset fixed values, allowing only passive feedback adjustment based on current operating conditions, lacking the ability to predict future operating conditions. However, gas-steam combined cycle units are typical strongly coupled, nonlinear dynamic systems. During variable operating conditions, the parameters of various devices and systems are interconnected, exhibiting significant dynamic fluctuations. Traditional passive control strategies cannot predict changes in operating conditions in advance, making it difficult to achieve precise pre-emptive parameter adjustment. This directly leads to poor adaptability of the control strategy and significant adjustment lag during variable operating conditions, preventing the unit's overall operating efficiency and control stability from reaching their optimal state.

[0005] Furthermore, current mainstream gas turbine simulation systems are all built and run on general-purpose workstations, without dedicated parallel computing hardware architectures, resulting in limited computing power. When conducting simulations of complex combined cycle operating conditions, multiple scenarios, and the entire process, the serial computing mode of general-purpose workstations cannot meet the parallel processing requirements of large-scale, high-dimensional simulation data, making it difficult to achieve true ultra-real-time simulation computing. This significantly limits the efficiency of rapid simulation of complex operating conditions and iterative optimization of control strategies, and cannot support the real-time requirements of dynamic closed-loop optimization control of the unit.

[0006] To address the aforementioned computing power bottleneck, the applicant has previously developed a parallel processing unit based on the intelligent islands algorithm (patent number: CN111427690B) and a real-time computing method for distributed processing units (patent number: CN111427318B). These methods effectively solve the industry pain point of insufficient basic computing power and inability to achieve real-time computing in traditional simulation systems, providing a hardware and algorithmic foundation for high-precision and rapid simulation of gas turbine combined cycle systems.

[0007] However, under the current technological framework, the industry still lacks a dedicated architecture for parallel control and closed-loop optimization that can be adapted to the DCS level, given the strongly coupled and nonlinear operating characteristics of gas turbine systems. Existing technologies cannot achieve deep integration of mechanistic simulation models and data-driven algorithms. Traditional pure mechanistic models have limited adaptability, while pure data-driven models suffer from poor interpretability and insufficient stability. The two are disconnected, failing to take into account both the physical constraints of unit operation and the dynamic characteristics of actual field data. This makes it difficult to achieve accurate parallel control, dynamic closed-loop calibration, and energy efficiency optimization under all operating conditions of the gas turbine, becoming the core problem for the implementation of intelligent simulation and efficient, precise control technologies for combined cycle gas turbines. Summary of the Invention

[0008] In view of this, the embodiments of this specification provide a DCS simulation optimization method and system for combined cycle gas turbines based on PPU.

[0009] This specification provides the following technical solution in its embodiments: a PPU-based DCS simulation optimization method for combined cycle gas turbines, comprising: S100: Construct a parallel simulation system for gas turbine and combined cycle DCS based on a parallel controller PPU. The PPU adopts a heterogeneous computing architecture including a CPU unit and an FPGA acceleration unit to establish a simulation mirror system that runs in parallel with the physical DCS system. The FPGA acceleration unit enables high-speed synchronization of I / O measurement data between the physical DCS and the PPU, and supports the PPU to write back the optimized control parameters to the physical DCS without disturbance. S200: The gas turbine thermodynamic system is modeled using a fractional-order calculus model inside the PPU. The residual between the simulation model output value and the physical entity feedback value is calculated in real time. When the residual exceeds a preset threshold, the model parameter correction mechanism is automatically triggered to update the key thermodynamic parameters of the model online, thereby realizing the adaptive evolution of the simulation model. S300: Utilizing the parallel computing capabilities of the PPU, multiple simulation instances ahead of physical time are run in parallel in the background to test different control strategies and quickly evaluate the impact of each strategy on unit efficiency and safety indicators. S400: Based on the deduction results of step S300, a multi-objective genetic algorithm is used to optimize the unit's thermal efficiency and minimize pollutant emissions as objective functions to generate the optimal control setpoint or control command. Before sending the control command to the physical DCS, safety boundary verification and smooth switching processing are performed.

[0010] Preferably, the specific method by which the FPGA acceleration unit in step S100 achieves high-speed synchronization of I / O measurement point data is as follows: A custom state machine and logic circuit are built into the FPGA to identify the industrial communication protocol of the DCS system. When a data frame enters the FPGA, frame header detection, CRC check and data splitting steps are triggered in parallel within the same clock cycle. According to the pre-written mapping table between DCS measurement points and the shared memory address of the simulation system, the parsed data is directly written to the specified address of the shared memory through the DMA controller, bypassing the CPU's software scheduling and operation. The PPU converts the optimized control parameters into a signal format recognizable by the DCS via the FPGA and writes them into the input register of the physical DCS without interference. The FPGA uses fixed physical circuits burned onto the chip to achieve signal transmission, which makes the time from data reception to writing to shared memory deterministic and does not cause delay jitter with changes in the DCS system load.

[0011] Preferably, the fractional-order calculus model mentioned in step S200 is specifically as follows: The evolution of the gas turbine temperature field T(x,t) is described using a fractional-order heat conduction equation: in, Let represent the fractional derivative of order α (0 < α < 1), where α is the thermal index, k is the thermal diffusivity, and Q(t) is the heat source term; The key thermodynamic parameters include thermal inertia coefficient and transfer efficiency; The model parameter correction mechanism uses fractional-order UKF or improved EKF estimation algorithms to identify parameters and uses historical operating data to update the key thermodynamic parameters online.

[0012] Preferably, in steps S300 and S400: The parallel simulation example sets the prediction step ahead of physical time to 5-10 minutes, and tests different guide vane opening and fuel quantity combination strategies respectively. The multi-objective genetic algorithm adopts a solution-based multi-objective genetic algorithm, which simultaneously takes maximizing thermal efficiency and minimizing NOx emissions as optimization objectives, uses fuel mass flow rate and compressor guide vane IGV opening as decision variables, and finds an optimal solution set through Pareto optimization. The system automatically searches for the combination of operating parameters with the highest thermal efficiency under the current operating conditions, or finds the combustion strategy with the lowest emissions when meeting specific load requirements, while ensuring that NOx emissions do not exceed the standard.

[0013] Preferably, the security boundary verification and smooth handover process in step S400 includes: Real-time monitoring of the discrepancy between the simulation model output and the feedback from the physical entity; Safety boundary verification: Check whether the optimization instructions exceed the safe operating envelope of the unit, which includes the upper limit constraint of exhaust temperature and surge boundary constraint; Smooth switching: When switching between PPU control mode and manual control mode or traditional DCS control mode, incremental PID or ramp function is used to smooth the control signal.

[0014] Preferably, when constructing the simulation mirror system in step S100: A simulation model consistent with the DCS logic of the physical gas turbine is built using configuration software, and a verification and prediction configuration module is introduced into the configuration software. The non-parallel system is transformed into a parallel system through the verification and prediction configuration module. The I / O measurement data includes speed, exhaust temperature, and guide vane opening feedback signals, which are read from the physical DCS at 10ms intervals and synchronized to the PPU's shared memory area via DMA technology for use by the simulation model.

[0015] Preferably, the parallel simulation system is logically divided into a physical entity layer, a parallel simulation layer, and an optimization decision layer: The physical entity layer includes the gas turbine body, waste heat boiler, steam turbine generator and DCS system, and the DCS system is connected to sensors and actuators through hardwiring or fieldbus; The PPU of the parallel simulation layer is connected to the physical DCS system via a high-speed industrial Ethernet. The CPU unit is used to run the mechanism model, and the FPGA acceleration unit is used to process high-speed IO signal synchronization and logic verification. The parallel simulation layer also includes a storage module for storing historical running data and model parameters. The optimization decision layer is deployed in the high-performance computing unit of the PPU or in the edge server connected to it, and includes an ultra-real-time parallel hypothesis analysis engine and a multi-objective genetic algorithm optimization module.

[0016] Preferably, the preset threshold in step S200 includes an exhaust temperature deviation threshold, and the parameter correction mechanism is triggered when the exhaust temperature deviation exceeds 5°C. The method is applied to gas turbines and combined cycle generator sets. The simulation mirror system covers the full operating conditions of the combined cycle of gas turbines, waste heat boilers, and steam turbine generators, including start-up, load variation, and shutdown conditions.

[0017] A PPU-based DCS parallel simulation optimization system, the system being used to execute the PPU-based gas turbine combined cycle DCS simulation optimization method described above, comprising: The physical entity layer includes the gas turbine body, waste heat boiler, steam turbine generator and DCS system, wherein the DCS system is connected to sensors and actuators via hardwired or fieldbus; The parallel simulation layer uses a parallel controller PPU as the computing core. The PPU contains a multi-core CPU unit and an FPGA acceleration unit, which is connected to the physical DCS system via Ethernet. The FPGA acceleration unit is used to achieve high-speed synchronization of I / O measurement point data between the physical DCS and the PPU. The CPU unit is used to run the simulation model based on fractional calculus. The FPGA acceleration unit has a customized state machine and logic circuit, and the parsed data is directly written to shared memory through a DMA controller. The optimized decision layer, deployed in the computing unit of the PPU or the edge server connected to it, includes an ultra-real-time parallel hypothesis analysis engine and a multi-objective genetic algorithm optimization module.

[0018] Compared with the prior art, the beneficial effects that this application can achieve include at least the following: By designing a heterogeneous data synchronization interface based on shared memory and FPGA acceleration, the DCS protocol parsing logic is hardened into a parallel circuit inside the FPGA, and the write channel of shared memory is directly opened through DMA. The data interaction latency is compressed from hundreds of milliseconds at the software level to hundreds of microseconds at the hardware level, achieving a millisecond-level data synchronization cycle. This ensures strict alignment of physical entities and parallel simulation entities in terms of timestamps, overcomes the control oscillation problem caused by data latency in traditional simulation systems, and supports uninterrupted write-back of optimization parameters. It opens up a high-speed path from perception to execution, laying the hardware foundation for real-time closed-loop optimization.

[0019] A model parameter identification algorithm based on fractional calculus is introduced. Fractional differential operators have infinite memory characteristics, which can accurately describe the historical dependence of heat and nonlocal heat transfer characteristics in the gas turbine thermodynamic system. This overcomes the problems of missing physical meaning and low fitting accuracy of integer-order models. It can calculate residuals in real time and automatically trigger parameter correction for strongly nonlinear operating conditions such as gas turbine start-up and variable load. It realizes online updates of key parameters such as thermal inertia coefficient and transfer efficiency, enabling the digital twin to self-evolve with the aging and operating condition changes of the physical unit, ensuring high accuracy of the simulation model throughout its entire life cycle.

[0020] Leveraging the parallel computing advantages of the PPU, the control strategy is upgraded from post-event feedback to pre-event prediction. Through an ultra-real-time parallel hypothesis analysis engine, multiple control strategies (such as different guide vane openings and fuel quantity combinations) are tested in parallel on a timeline ahead of physical time. Combined with a solution-oriented multi-objective genetic algorithm, the contradiction between maximizing thermal efficiency and minimizing NOx emissions is transformed into a Pareto optimization problem, automatically finding the optimal solution set that satisfies safety constraints. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the implementation of a PPU-based DCS simulation optimization method for combined cycle gas turbines. Figure 2 yes Figure 1 An enlarged view of the upper part; Figure 3 yes Figure 1 A magnified view of the middle section; Figure 4 yes Figure 1 Enlarged view of the lower part; Figure 5 This is an architecture diagram of a gas turbine combined cycle DCS simulation optimization system based on PPU. Detailed Implementation

[0023] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0024] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this application, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0026] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0027] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0028] The technical solutions provided by the various embodiments of this application are described below with reference to the accompanying drawings.

[0029] like Figures 1-4 As shown, a PPU-based DCS simulation optimization method for combined cycle gas turbines constructs a parallel operation architecture between the physical gas turbine and the combined cycle unit and its digital twin. Utilizing the heterogeneous computing power and fractional-order calculus model of the parallel controller (PPU), it achieves automatic control optimization and equipment performance optimization of the unit, including: Step S100: Construct a parallel simulation system for the gas turbine and combined cycle DCS based on a parallel controller PPU. The PPU adopts a heterogeneous computing architecture including a multi-core CPU unit and an FPGA acceleration unit. The CPU unit is responsible for running complex gas turbine thermodynamic mechanism models and optimization decision algorithms, while the FPGA acceleration unit is responsible for high-speed synchronization and logic verification of I / O measurement data between the physical DCS and the PPU. The system establishes a simulation mirror system running in parallel with the physical DCS system. High-speed synchronization of I / O measurement data between the physical DCS and the PPU is achieved through the FPGA acceleration unit, and the system supports the PPU in writing back the optimized control parameters to the physical DCS without disturbance.

[0030] The PPU decouples data synchronization and model computation tasks through a heterogeneous computing architecture. The FPGA handles data transmission and reception and protocol parsing in a pure hardware pipeline, while the CPU focuses on high-precision model computation. The two work in parallel without blocking each other, thus ensuring real-time data synchronization without affecting the computational performance of the simulation model. This breaks through the latency bottleneck of traditional simulation systems that rely on software-layer data interaction, laying the foundation for a millisecond-level high-speed data path for closed-loop control between physical entities and virtual models.

[0031] Step S200: The gas turbine thermodynamic system is modeled using a fractional-order calculus model within the PPU. The residual between the simulation model output value and the physical entity feedback value is calculated in real time. When the residual exceeds a preset threshold, the model parameter correction mechanism is automatically triggered to update the key thermodynamic parameters of the model online, thereby realizing the adaptive evolution of the simulation model.

[0032] Gas turbine metal components (such as turbine blades and combustion chamber walls) possess significant heat capacity. Their temperature changes depend not only on the current gas temperature but also on the heat accumulation process over a past period, exhibiting historical dependence and nonlocal heat transfer characteristics. Fractional-order calculus models, by introducing fractional-order parameters, can mathematically and accurately describe this thermal memory effect, overcoming the limitation of traditional integer-order models that only reflect the rate of change at the current moment. When the deviation between the simulation model and the physical entity exceeds a threshold, the system uses fractional-order UKF or improved EKF estimation algorithms to identify and update key parameters such as thermal inertia coefficient and transfer efficiency online based on historical operating data. This allows the digital twin to self-calibrate as the physical unit ages and its operating conditions change. This solves the industry problem of traditional simulation models having fixed parameters and becoming distorted with equipment aging, ensuring that the simulation model maintains high fidelity throughout its entire lifecycle.

[0033] Step S300: Utilize the parallel computing capability of the PPU to run multiple simulation instances ahead of physical time in the background, test different control strategies, and quickly evaluate the impact of each strategy on unit efficiency and safety indicators.

[0034] The PPU's parallel computing architecture supports the simultaneous running of multiple independent simulation instances. Each instance performs simulations at ultra-real-time speeds on a timeline ahead of physical time, taking into account different control strategy parameters (such as different guide vane opening and fuel quantity combinations) to simulate the impact of these strategies on the unit's thermal efficiency, emissions levels, and safety margins over the next 5 to 10 minutes. Because the simulation speed is much faster than the passage of physical time, the system can pre-compute multi-scenario evaluations before the physical unit actually executes a certain strategy, providing predictive data support for optimization decisions. This elevates the control strategy from a traditional "post-event feedback" mode to a "pre-event prediction" mode, enabling the system to anticipate changes in operating conditions and proactively adjust control parameters.

[0035] Step S400: Based on the deduction results of step S300, a multi-objective genetic algorithm is used to optimize the unit's thermal efficiency and minimize pollutant emissions as objective functions to generate the optimal control setpoint or control command. Before sending the control command to the physical DCS, safety boundary verification and smooth switching processing are performed.

[0036] Improving thermal efficiency often requires increasing combustion temperature, but high-temperature combustion easily leads to excessive nitrogen oxide (NOx) emissions. The multi-objective genetic algorithm overcomes the limitations of traditional single-objective weighted summation, simultaneously optimizing for both maximizing thermal efficiency and minimizing NOx emissions. Using fuel mass flow rate and compressor guide vane (IGV) opening as decision variables, it seeks a set of optimal solutions through Pareto optimization rather than a single compromise. After generating optimization commands, the system performs dual safety checks: first, it checks whether the commands exceed the unit's safe operating envelope (including exhaust temperature upper limit constraints and surge boundary constraints); then, it uses incremental PID or ramp functions for signal smoothing when switching between PPU control mode and manual / traditional DCS control mode. Under the premise of ensuring absolute unit safety, it automatically searches for the optimal combination of operating parameters under the current operating conditions, ensuring the unit always operates within the optimal economic range, while eliminating the risk of load surges or tripping due to mode switching.

[0037] Through the coordinated operation of steps S100 to S400 described above, the embodiments of this specification achieve at least the following technical effects: First, through the heterogeneous data synchronization interface based on the FPGA acceleration unit, millisecond-level data synchronization between the physical DCS and the simulation system is achieved, ensuring that the virtual and real data are strictly aligned in timestamps. This effectively overcomes the control oscillation problem caused by data delay in traditional simulation systems, and opens up a high-speed path from "perception" to "execution" for disturbance-free closed-loop control.

[0038] Second, by introducing a fractional-order calculus model and an online parameter correction mechanism, the simulation model can automatically calibrate key thermodynamic parameters based on real-time operating data, realizing the self-evolution of the digital twin throughout its entire life cycle and improving simulation accuracy and operational condition reproducibility.

[0039] Third, by combining the ultra-real-time parallel hypothesis analysis engine with the multi-objective genetic algorithm, the system has the ability to predict and extrapolate future operating conditions. It can pre-complete multi-scheme evaluation and global optimization before the physical unit executes the control strategy, so that the unit always operates at the optimal balance between energy efficiency and environmental protection under complex and variable operating conditions.

[0040] Fourth, through triple protection of virtual-real deviation monitoring, safety boundary verification, and smooth switching mechanism, the system ensures the absolute safety of industrial control while introducing advanced optimization algorithms, solving the problem of the difficulty of implementing advanced algorithms in industrial fields.

[0041] The above four aspects support each other and form a closed loop, jointly constructing a gas turbine combined cycle DCS parallel simulation and optimization system that is "symbiotic between virtual and real, self-evolving, predictive and optimized, and safe and controllable".

[0042] In some implementations, the specific method by which the FPGA acceleration unit in step S100 achieves high-speed synchronization of I / O measurement point data is as follows: a state machine and logic circuit are customized inside the FPGA to identify the industrial communication protocols commonly used in the DCS system. When a data frame enters the FPGA from the physical DCS via a high-speed industrial Ethernet, the hardware state machine inside the FPGA triggers frame header detection, CRC check, and data splitting in parallel within the same clock cycle, quickly completing protocol parsing and obtaining valid data. Subsequently, the hardware circuit, based on a pre-written mapping table of DCS measurement points and shared memory addresses of the simulation system, directly writes the parsed data to the designated address of the shared memory through the DMA (Direct Memory Access) controller, completely bypassing the software scheduling and CPU operation of the DCS. In the data uplink direction, the optimization parameters calculated by the PPU are converted into a signal format recognizable by the DCS via the FPGA and written into the input register of the physical DCS without disturbance, realizing the downlink write-back of the optimization instructions.

[0043] The FPGA uses fixed physical circuits programmed onto the chip to implement signal transmission paths, and it has tens of thousands of logic units that can work simultaneously. For example, the FPGA can simultaneously handle data reception at port A, data transmission at port B, and data writing to shared memory, without interference between the three processes, eliminating the "queueing" and "context switching" overhead of traditional CPUs. Because the FPGA logic is physical circuitry embedded on the chip, and the signal transmission path is fixed, the time from data reception to writing to shared memory is absolutely certain regardless of the DCS system load. This avoids the latency and jitter common in software systems and prevents state divergence in simulation models caused by data instability.

[0044] Traditional simulation systems employ a software-level "read-determine-process" data interaction process. This process involves multiple steps, including operating system scheduling, CPU computation, and memory copying, resulting in latency typically in the hundreds of milliseconds and exhibiting jitter. This implementation hardens the DCS protocol parsing logic into parallel digital circuits within the FPGA and directly establishes a write channel to shared memory via DMA, compressing the data interaction latency from hundreds of milliseconds at the software level to hundreds of microseconds at the hardware level. This establishes a high-speed, lossless data bridge between the physical DCS and the simulation system, ensuring the absolute stability of the data flow in high-precision hardware-in-the-loop simulation and providing a reliable real-time data foundation for subsequent model adaptive evolution and ultra-real-time simulation.

[0045] In some implementations, the fractional calculus model described in step S200 is specifically: The evolution of the gas turbine temperature field T(x,t) is described using a fractional-order heat conduction equation: in, Let represent the fractional derivative of order α (0 < α < 1), where α is the thermal index, k is the thermal diffusivity, and Q(t) is the heat source term; Integer-order differential operators only represent the rate of change at the current moment and are local; however, the metal components of a gas turbine (such as turbine blades and combustion chamber walls) have significant heat capacity, and their temperature changes depend not only on the current gas temperature but also on the heat accumulation process over a past period. Only fractional-order models can reflect this historical dependence. In the fractional-order heat conduction equation, the fractional order α (0 < α < 1) is the thermal index, used to characterize the strength of the memory effect of the heat transfer process: when α is close to 1, the system exhibits an instantaneous response characteristic of approximately integer order; when α is close to 0, the system exhibits a strong memory effect and diffusion damping characteristics; in actual gas turbine engineering, the common value range of α is 0.6 to 0.8, representing that the gas turbine thermodynamic object has a significant thermal memory effect, and its dynamic response process is smoother than that of the integer-order model, which is consistent with the actual operation of gas turbine engineering.

[0046] The key thermodynamic parameters include thermal inertia coefficient and heat transfer efficiency. The model parameter correction mechanism employs fractional-order UKF (Unscented Kalman Filter) or improved EKF (Extended Kalman Filter) estimation algorithms for parameter identification. While the traditional EKF method solves nonlinear problems through first-order Taylor series expansion, it has inherent limitations when dealing with objects like gas turbines, which exhibit significant memory effects and complex nonlinearities. This implementation combines fractional-order modeling with fractional-order UKF or improved EKF estimation algorithms. By introducing a fractional order α, the model can mathematically accurately describe the historical characteristics of the heat transfer process, possessing infinite memory characteristics, thus more closely resembling the actual object in physical essence. The system uses historical operating data to update the key thermodynamic parameters online, enabling the simulation model to adaptively correct itself as the physical unit ages and its operating conditions change.

[0047] It overcomes the problems of missing physical meaning and low fitting accuracy of integer-order models. Under highly nonlinear operating conditions such as gas turbine start-up and rapid load increase and decrease, the deviation between the dynamic response (such as overshoot and settling time) predicted by the model and the actual physical object is significantly reduced, providing a high-fidelity model basis for subsequent optimization decisions.

[0048] In some implementations, the ultra-real-time parallel simulation and multi-objective optimization in steps S300 and S400 are specifically performed as follows: the parallel simulation instance sets a prediction step size 5 to 10 minutes ahead of physical time, and tests different combinations of control strategies in parallel in virtual space. For example, the following multiple schemes are simulated simultaneously: Scheme A: increase fuel quantity but decrease guide vane opening; Scheme B: maintain fuel quantity but open guide vanes earlier; Scheme C: simultaneously adjust fuel quantity and guide vane opening to intermediate values, etc. Each simulation instance rapidly extrapolates on a timeline ahead of physical time, evaluating the impact of different guide vane opening and fuel quantity combinations on the unit's future thermal efficiency, NOx emission levels, and safety margin.

[0049] The multi-objective genetic algorithm employs a solution-based multi-objective genetic algorithm, simultaneously optimizing thermal efficiency and minimizing NOx emissions, using fuel mass flow rate and compressor guide vane (IGV) opening as decision variables. This algorithm transforms the complex nonlinear coupling between energy conservation and emission reduction into a Pareto optimization problem, moving away from seeking a single compromise solution and instead aiming to find a set of optimal solutions (Pareto front). Through iterative evolution of the genetic algorithm, including selection, crossover, and mutation operations, it automatically searches for the optimal set of control setpoints that satisfy all constraints (including exhaust temperature limits, surge boundaries, and other safety constraints) from the massive amounts of extrapolation data generated by the ultra-real-time parallel hypothesis analysis engine. The system can automatically search for the combination of operating parameters with the highest thermal efficiency under current operating conditions, or find the combustion strategy with the lowest emissions when meeting specific load requirements, while ensuring that NOx emissions do not exceed limits.

[0050] Traditional single-objective optimization methods typically use weighted summation to transform multiple objectives into a single objective. However, the selection of weights is subjective and fails to reflect the true game-theoretic relationships between the objectives. This implementation employs a solution-based multi-objective genetic algorithm to directly find the optimal solution set in a Pareto sense, preserving the true trade-offs between objectives. An intelligent control model is constructed at the DCS system level that accurately reflects the complex physical characteristics of the gas turbine and achieves multi-objective synergistic optimization, ensuring the unit always operates at the optimal balance between energy efficiency and environmental protection under complex and variable operating conditions.

[0051] In some implementations, the security boundary verification and smooth handover process in step S400 includes the following three levels: The first level is virtual-real deviation monitoring: The system monitors the deviation between the output of the simulation model and the feedback of the physical entity in real time. When the deviation continues to increase or exceeds the safe range, the system determines that the credibility of the current simulation model has decreased, automatically reduces the weight of optimization instructions or suspends the issuance of optimization instructions, and switches to manual or traditional DCS control mode to prevent erroneous control instructions caused by model inaccuracy from affecting the safety of the unit.

[0052] The second level is safety boundary verification: Before sending the optimization command generated by the PPU to the physical DCS, the system checks whether the optimization command exceeds the safe operating envelope of the unit. The safe operating envelope includes the upper limit constraint of exhaust temperature and surge boundary constraint, etc. If the optimization command causes the exhaust temperature to exceed the material's allowable upper limit or the compressor to enter the surge range, the system will automatically cut off or correct the command to ensure that the unit always operates within the safe operating range.

[0053] The third level is smooth switching: When switching between PPU control mode and manual control mode or traditional DCS control mode, incremental PID or ramp function is used to smooth the control signal. Specifically, when switching from PPU optimized control to manual control, the system gradually transfers control to the operator using a ramp function, allowing the control signal to transition smoothly within a set time window; when switching from manual control to PPU optimized control, the incremental PID uses the current manual control output as the initial value and gradually adds the optimization correction, avoiding abrupt changes in the control signal.

[0054] Industrial gas turbine control has stringent safety requirements; any discontinuity in control signals can trigger sudden load changes or even tripping of the unit. This implementation method achieves real-time assessment of model reliability through virtual-real deviation monitoring, implements hard constraint protection at the command level through safety boundary verification, and ensures signal continuity during control mode transitions through smooth switching. These three elements constitute a triple safety defense line, from software to hardware, and from monitoring to execution. This completely eliminates the risk of sudden load changes or tripping of the unit due to model errors or mode switching, enabling advanced optimization algorithms to be reliably implemented while ensuring absolute safety in industrial settings.

[0055] In some implementations, the specific method for constructing the simulation mirror system in step S100 is as follows: A simulation model consistent with the logic of the physical gas turbine DCS is built using configuration software conforming to the IEC-61131-3 standard. IEC-61131-3 is a widely adopted programming standard in the industrial control field, supporting multiple programming languages ​​such as ladder diagrams (LD), function block diagrams (FBD), and structured text (ST), ensuring that the control logic of the simulation model is highly consistent with the physical DCS and guaranteeing the reliability of the simulation results. A "verification and prediction" configuration module is introduced into the configuration software to transform the non-parallel system into a parallel system. Specifically, the verification module is responsible for comparing the received physical DCS data with the simulation model calculation results in real time, outputting a deviation signal to trigger model correction; the prediction module, based on the current operating conditions and optimization algorithm output, generates predictive control commands that lead physical time, realizing the transformation from passive simulation to active parallel control.

[0056] The I / O measurement data includes key parameters such as engine speed, exhaust temperature, and guide vane opening feedback signals. This data is read from the physical DCS at 10ms intervals and synchronized to the PPU's shared memory area via DMA technology for use by the simulation model. The 10ms synchronization period is an optimal value considering both control real-time performance and data throughput: a period that is too short will increase communication load and FPGA resource consumption; a period that is too long will fail to capture the dynamic characteristics of the gas turbine during rapid changes in operating conditions. By directly writing data to shared memory using DMA technology, the simulation model can read the latest data in a zero-copy manner, further reducing data access latency.

[0057] By constructing a PPU using configuration software and introducing verification and prediction modules, the system can be easily integrated into the existing DCS architecture without large-scale modifications to the physical DCS, achieving low-cost, low-intrusion deployment of a parallel simulation system. Simultaneously, a 10ms-level data synchronization cycle and a zero-copy data access mechanism ensure that the simulation model is always calculated based on the latest physical data, providing a reliable data foundation for high-precision parallel control.

[0058] In some implementations, the preset threshold in step S200 includes an exhaust temperature deviation threshold, which triggers the parameter correction mechanism when the exhaust temperature deviation exceeds 5°C. Exhaust temperature is a core parameter reflecting the operating state of the gas turbine, and the deviation between its simulated and measured values ​​directly reflects the model's accuracy. Setting the threshold to 5°C takes into account the following factors: when the deviation is less than 5°C, the model accuracy is within an acceptable range, and frequent correction triggers increase the computational burden with limited benefits; when the deviation reaches 5°C, it indicates that the model has experienced significant drift, and failure to correct it in time will affect the reliability of subsequent optimization decisions. Of course, the threshold can be adaptively adjusted according to different unit types and operating conditions. For example, the threshold can be appropriately relaxed during the unit startup phase to adapt to a wide range of operating condition changes, while the threshold can be tightened during steady-state operation to pursue higher model accuracy.

[0059] The method is applied to gas turbines and combined cycle generator sets. The simulation mirror system covers the entire operating range of the combined cycle of gas turbines, waste heat boilers, and steam turbine generators, including start-up, variable load, and shutdown conditions. During start-up, the gas turbine undergoes the entire process from turning gear to grid connection, with drastic changes in parameters and significant nonlinear characteristics. The fractional-order calculus model can effectively characterize the thermal memory effect and heat transfer hysteresis characteristics during this stage. Under variable load conditions, the unit needs to frequently respond to grid peak-shaving and frequency regulation commands. The ultra-real-time parallel hypothesis analysis engine can predict the unit's response under different load change rates, providing predictive support for optimal load tracking strategies. Under shutdown conditions, the system can optimize the shutdown curve based on historical data, reducing thermal stress shocks and extending equipment life.

[0060] By setting reasonable thresholds, a balance between the timeliness and economy of model correction is achieved. At the same time, full operating condition coverage ensures that the parallel simulation system can provide effective optimization support at all stages of unit operation, giving full play to the value of the virtual-real symbiotic architecture in the whole life cycle management.

[0061] like Figure 5 As shown, based on the same inventive concept, this specification provides a PPU-based DCS parallel simulation optimization system. The system is used to execute the PPU-based gas turbine combined cycle DCS simulation optimization method described above. In terms of logical architecture, it includes three levels: physical entity layer, parallel simulation layer, and optimization decision layer.

[0062] The physical entity layer includes the gas turbine itself, waste heat boiler, steam turbine generator, and DCS system. The DCS system is connected to sensors and actuators via hardwiring or fieldbus. Sensors include temperature transmitters, pressure transmitters, speed sensors, etc., used to collect operating parameters of key components of the unit in real time. Actuators include fuel control valves, inlet guide vane IGV servo motors, etc., used to execute control commands issued by the DCS or PPU. The physical entity layer is the data source and the final execution end of the control commands for the parallel simulation system.

[0063] The parallel simulation layer employs a parallel controller (PPU) as its computational core, which internally includes a multi-core CPU unit and an FPGA acceleration unit. The PPU is connected to the physical DCS system via a high-speed industrial Ethernet. The FPGA acceleration unit enables high-speed synchronization of I / O measurement data between the physical DCS and the PPU. The CPU unit runs the simulation model based on fractional calculus. The FPGA acceleration unit incorporates a custom state machine and logic circuits, and uses a DMA controller to directly write the parsed data into shared memory, bypassing the CPU's software scheduling and computation stages, achieving data synchronization latency at the microsecond level. The parallel simulation layer also includes a storage module for storing historical running data and model parameters, providing data support for model adaptive evolution and offline analysis.

[0064] The optimization decision layer is deployed in the computing unit of the PPU or an edge server connected to it, and includes a real-time parallel hypothesis analysis engine and a multi-objective genetic algorithm optimization module. The real-time parallel hypothesis analysis engine utilizes the parallel computing capabilities of the PPU to run multiple simulation instances simultaneously in the background, ahead of physical time, to test different combinations of control strategies and evaluate their impact on unit efficiency and safety indicators. The multi-objective genetic algorithm optimization module uses maximizing unit thermal efficiency and minimizing pollutant emissions as objective functions, and automatically searches for the optimal control setpoint or generates control commands from the simulation results through Pareto optimization. The optimization decision layer is the intelligent central hub of the system, responsible for strategy generation, global optimization, and safety decision-making.

[0065] Through the collaborative work of the three-layer architecture, the virtual-real mapping between the physical gas turbine and the digital twin, the adaptive evolution of the model, the ultra-real-time prediction and inference, and the multi-objective safety optimization are realized, providing a system-level solution for the efficient, safe and environmentally friendly operation of gas turbine combined cycle units.

[0066] In implementation, the heterogeneous computing architecture of the PPU can be flexibly configured according to specific application scenarios. For small gas turbine units, a single SoC chip integrating a CPU and FPGA can be used to implement the PPU function, reducing hardware costs and system complexity. For large combined cycle units, a combination architecture of multi-core CPU servers and independent FPGA acceleration cards can be used to provide stronger computing power and higher data throughput. The storage module can use an NVMe solid-state drive array, supporting high-speed read and write and redundant backup to ensure the security and traceability of historical operating data.

[0067] Regarding communication connectivity, the connection method between the PPU and the physical DCS can be selected based on site conditions: when the PPU and DCS are deployed in the same rack or server room, a direct Ethernet connection or PCIe bus can be used to achieve the highest bandwidth; when the PPU and DCS are deployed in different areas, a redundant ring network composed of industrial switches can be used to achieve high availability. The synchronization period of the I / O measurement point data can be dynamically adjusted according to the unit's operating status: a shorter synchronization period (e.g., 5ms) is used during periods of drastic changes in operating conditions such as startup and shutdown to capture rapid dynamics; a longer synchronization period (e.g., 20ms) is used during steady-state operation to reduce communication load.

[0068] At the optimization decision-making level, the parameters of the multi-objective genetic algorithm can be tuned on-site: parameters such as population size, crossover probability, mutation probability, and maximum number of iterations can be optimized according to unit characteristics and real-time requirements. The safe operating envelope can be customized based on design data provided by the unit manufacturer and on-site operating experience to ensure the accuracy and practicality of safety constraints.

[0069] It should be noted that the technical features in the above embodiments can be combined with each other to form new embodiments without contradiction. For example, the redundancy verification mechanism of the FPGA acceleration unit can be combined with the threshold adaptive adjustment mechanism of the fractional-order model to achieve dynamic optimization of model accuracy while ensuring data integrity; the adaptive mutation rate mechanism of the multi-objective genetic algorithm can be combined with the multi-level protection mechanism of security boundary verification to ensure the absolute security of control commands while ensuring optimization efficiency.

[0070] In summary, it has at least the following effects: 1. Achieved high-fidelity, low-latency "virtual-real symbiosis" and precise synchronization. By designing a heterogeneous data synchronization interface based on shared memory and FPGA acceleration, the data interaction bottleneck between the physical DCS and the simulation system was solved.

[0071] Eliminating control oscillations: Achieving millisecond-level data synchronization cycles ensures strict alignment of physical entities and parallel simulation objects in terms of timestamps, effectively overcoming the control oscillation problem caused by data delays in traditional simulation systems.

[0072] Disruptive closed-loop control: It supports the uninterrupted writing back of optimized parameters, opening up a high-speed path from "sensing" to "execution" and laying the hardware foundation for real-time closed-loop optimization.

[0073] 2. Possesses the ability to "self-evolve" throughout the entire model lifecycle, significantly improving simulation accuracy. It breaks through the limitations of traditional simulation models, which have fixed parameters and cannot adapt to equipment aging.

[0074] Adaptive correction: By introducing a model parameter identification algorithm based on fractional calculus, it is possible to calculate residuals in real time and automatically trigger parameter correction for highly nonlinear operating conditions such as gas turbine start-up and variable load.

[0075] Long-term high fidelity: It enables online updates of key parameters such as thermal inertia and transmission efficiency, allowing the digital twin to "self-evolve" as the physical unit ages and its operating conditions change. This ensures the high accuracy of the simulation model throughout its entire lifecycle and solves the industry problem of model distortion over time.

[0076] 3. Based on real-time simulation-based "predictive" optimization, excavator units achieve ultimate energy efficiency. By leveraging the parallel computing capabilities of the PPU, the control strategy can be upgraded from "post-event feedback" to "pre-event prediction".

[0077] Multi-strategy parallel optimization: Through the ultra-real-time parallel hypothesis analysis engine, multiple control strategies (such as different guide vane openings and fuel quantity combinations) can be tested in parallel ahead of physical time.

[0078] Multi-objective global optimization: Combining multi-objective genetic algorithms, the system automatically finds the optimal setpoint that maximizes thermal efficiency and minimizes emissions while ensuring safety. This ensures that the unit always operates in the best economic range under complex and variable operating conditions, significantly improving the overall energy efficiency and environmental protection level of the combined cycle unit.

[0079] 4. Industrial Site Safety and System Compatibility It fully considers the stringent safety requirements of industrial control and solves the problem of the difficulty in implementing advanced algorithms.

[0080] Dual safety safeguards: A virtual-to-real deviation monitoring and safety boundary verification mechanism is designed, and incremental PID or ramp function is used for smoothing during mode switching, completely eliminating the risk of sudden load changes or tripping of the unit due to model errors or mode switching.

[0081] Low-cost retrofit: By using configuration software to build the PPU and introducing a "verification and prediction" module, the system can be easily integrated into the existing DCS architecture. The same or similar parts between the various embodiments in this specification can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the method embodiments described later are relatively simple in description since they correspond to the system, and relevant parts can be referred to the descriptions in the system embodiments.

[0082] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A PPU-based gas turbine combined cycle DCS simulation optimization method, characterized in that, include: S100: Construct a parallel simulation system for gas turbine and combined cycle DCS based on a parallel controller PPU. The PPU adopts a heterogeneous computing architecture including a CPU unit and an FPGA acceleration unit to establish a simulation mirror system that runs in parallel with the physical DCS system. The FPGA acceleration unit enables high-speed synchronization of I / O measurement data between the physical DCS and the PPU, and supports the PPU to write back the optimized control parameters to the physical DCS without disturbance. S200: The gas turbine thermodynamic system is modeled using a fractional-order calculus model inside the PPU. The residual between the simulation model output value and the physical entity feedback value is calculated in real time. When the residual exceeds a preset threshold, the model parameter correction mechanism is automatically triggered to update the key thermodynamic parameters of the model online, thereby realizing the adaptive evolution of the simulation model. S300: Utilizing the parallel computing capabilities of the PPU, multiple simulation instances ahead of physical time are run in parallel in the background to test different control strategies and quickly evaluate the impact of each strategy on unit efficiency and safety indicators. S400: Based on the deduction results of step S300, a multi-objective genetic algorithm is used to optimize the unit's thermal efficiency and minimize pollutant emissions as objective functions to generate the optimal control setpoint or control command. Before sending the control command to the physical DCS, safety boundary verification and smooth switching processing are performed.

2. The method of claim 1, wherein, The specific method by which the FPGA acceleration unit in step S100 achieves high-speed synchronization of I / O measurement point data is as follows: A custom state machine and logic circuit are built into the FPGA to identify the industrial communication protocol of the DCS system. When a data frame enters the FPGA, frame header detection, CRC check and data splitting steps are triggered in parallel within the same clock cycle. According to the pre-written mapping table between DCS measurement points and the shared memory address of the simulation system, the parsed data is directly written to the specified address of the shared memory through the DMA controller, bypassing the CPU's software scheduling and operation. The PPU converts the optimized control parameters into a signal format recognizable by the DCS via the FPGA and writes them into the input register of the physical DCS without interference. The FPGA uses fixed physical circuits burned onto the chip to achieve signal transmission, which makes the time from data reception to writing to shared memory deterministic and does not cause delay jitter with changes in the DCS system load.

3. The method of claim 1, wherein, The fractional calculus model mentioned in step S200 is specifically as follows: The evolution of the gas turbine temperature field T(x,t) is described using a fractional-order heat conduction equation: wherein, denotes a fractional derivative of order a (0 < a < 1), a is the thermal index, k is the thermal diffusivity, and Q(t) is the heat source term; The key thermodynamic parameters include thermal inertia coefficient and transfer efficiency; The model parameter correction mechanism uses fractional-order UKF or improved EKF estimation algorithms to identify parameters and uses historical operating data to update the key thermodynamic parameters online.

4. The method of claim 1, wherein, In steps S300 and S400: the parallel simulation instance is set to a prediction step size 5-10 minutes ahead of physical time, and different guide vane opening and fuel quantity combination strategies are tested respectively. The multi-objective genetic algorithm adopts a solution-based multi-objective genetic algorithm, which simultaneously takes maximizing thermal efficiency and minimizing NOx emissions as optimization objectives, uses fuel mass flow rate and compressor guide vane IGV opening as decision variables, and finds an optimal solution set through Pareto optimization. The system automatically searches for the combination of operating parameters with the highest thermal efficiency under the current operating conditions, or finds the combustion strategy with the lowest emissions when meeting specific load requirements, while ensuring that NOx emissions do not exceed the standard.

5. The method according to claim 4, characterized in that, The security boundary verification and smooth handover process described in step S400 includes: Real-time monitoring of the discrepancy between the simulation model output and the feedback from the physical entity; Safety boundary verification: Check whether the optimization instructions exceed the safe operating envelope of the unit, which includes the upper limit constraint of exhaust temperature and surge boundary constraint; Smooth switching: When switching between PPU control mode and manual control mode or traditional DCS control mode, incremental PID or ramp function is used to smooth the control signal.

6. The method according to claim 1, characterized in that, When constructing the simulation mirror system in step S100: A simulation model consistent with the DCS logic of the physical gas turbine is built using configuration software, and a verification and prediction configuration module is introduced into the configuration software. The non-parallel system is transformed into a parallel system through the verification and prediction configuration module. The I / O measurement data includes speed, exhaust temperature, and guide vane opening feedback signals, which are read from the physical DCS at 10ms intervals and synchronized to the PPU's shared memory area via DMA technology for use by the simulation model.

7. The method according to any one of claims 1-6, characterized in that, The parallel simulation system is logically divided into a physical entity layer, a parallel simulation layer, and an optimization decision layer: The physical entity layer includes the gas turbine body, waste heat boiler, steam turbine generator and DCS system, and the DCS system is connected to sensors and actuators through hardwiring or fieldbus; The PPU of the parallel simulation layer is connected to the physical DCS system via a high-speed industrial Ethernet. The CPU unit is used to run the mechanism model, and the FPGA acceleration unit is used to process high-speed IO signal synchronization and logic verification. The parallel simulation layer also includes a storage module for storing historical running data and model parameters. The optimization decision layer is deployed in the high-performance computing unit of the PPU or in the edge server connected to it, and includes an ultra-real-time parallel hypothesis analysis engine and a multi-objective genetic algorithm optimization module.

8. The method according to claim 1, characterized in that, The preset threshold mentioned in step S200 includes an exhaust temperature deviation threshold. When the exhaust temperature deviation exceeds 5°C, the parameter correction mechanism is triggered. The method is applied to gas turbines and combined cycle generator sets. The simulation mirror system covers the full operating range of the combined cycle of gas turbines, waste heat boilers, and steam turbine generators, including start-up, variable load, and shutdown conditions.

9. A DCS parallel simulation optimization system based on PPU, characterized in that, The system is used to execute the PPU-based DCS simulation optimization method for combined cycle gas turbines as described in any one of claims 1 to 8, including: The physical entity layer includes the gas turbine body, waste heat boiler, steam turbine generator and DCS system, wherein the DCS system is connected to sensors and actuators via hardwired or fieldbus; The parallel simulation layer uses a parallel controller PPU as the computing core. The PPU contains a multi-core CPU unit and an FPGA acceleration unit, which is connected to the physical DCS system via Ethernet. The FPGA acceleration unit is used to achieve high-speed synchronization of I / O measurement point data between the physical DCS and the PPU. The CPU unit is used to run the simulation model based on fractional calculus. The FPGA acceleration unit has a customized state machine and logic circuit, and the parsed data is directly written to shared memory through a DMA controller. The optimized decision layer, deployed in the computing unit of the PPU or the edge server connected to it, includes an ultra-real-time parallel hypothesis analysis engine and a multi-objective genetic algorithm optimization module.

Images