Ignition system special for gas turbine

Abstract

The invention relates to the technical field of ignition systems of gas turbines, and discloses a special ignition system for a gas turbine, which comprises a monitoring control layer, an energy distribution layer and a combustion optimization layer. The monitoring control layer monitors parameters such as combustion chamber pressure fluctuation in real time through multiple types of sensors and executes phase synchronization and injection control. The energy distribution layer is provided with a multi-stage energy release strategy and a self-adaptive time sequence matching mechanism, a stable ignition energy field is constructed, and physical and digital space linkage regulation and control are achieved through a multi-modal energy interface; and the combustion optimization layer establishes a multi-dimensional mapping combustion state control model based on a dynamic combustion algorithm and real-time data, and adopts a collaborative optimization algorithm to calibrate an ignition time sequence, regulate and control a fuel mixing ratio and maintain flame stability. According to the system, through multi-dimensional energy fusion, model parameter dynamic correction and intelligent algorithm application, the accuracy, stability and multi-working-condition adaptability of the gas turbine ignition process are remarkably improved, and the system is suitable for gas turbine efficient combustion control in the fields of aerospace, electric power and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of combustion engine ignition systems, in particular to a special ignition system for combustion engines. Background Art

[0002] In the energy and power sector, gas turbines, as core power equipment, are widely used in aerospace, power generation, industrial drives, and other applications. The performance of their ignition systems directly determines the turbine's startup reliability, combustion efficiency, and pollutant emissions. Traditional gas turbine ignition systems generally suffer from insufficient energy matching accuracy, weak dynamic combustion state control capabilities, and poor adaptability to multiple operating conditions. These systems fail to meet the stringent requirements of modern gas turbines for efficient combustion, low emissions, and long life.

[0003] From an energy control perspective, traditional ignition systems often employ a fixed energy output mode, making it impossible to precisely match and gradient-release ignition energy based on dynamic parameters such as engine speed and combustion chamber pressure. For example, at low speeds, a fixed high energy output can cause the fuel to ignite before it is fully mixed, leading to unstable combustion and concentrated heat loads. Meanwhile, at high loads, insufficient energy supply can lead to risks such as delayed ignition, flame destabilization, and even flameout. Furthermore, traditional systems lack a multimodal energy interface design and a coordinated control mechanism between the physical combustion space and the digital control space, making it difficult to achieve real-time monitoring and dynamic correction of the energy release process.

[0004] When it comes to combustion state monitoring and optimization, traditional technologies primarily rely on a single type of sensor to acquire combustion parameters. This fails to fully capture the energy distribution characteristics of the combustion chamber's three-dimensional space (such as pressure fluctuations, heat flux density, and fuel mixture uniformity), resulting in insufficient accuracy in combustion model construction. Furthermore, combustion optimization strategies are often based on empirical rules or fixed control algorithms, lacking the ability to deeply mine historical operating data and dynamically learn from real-time combustion data. This makes it difficult to establish an accurate combustion state control model, and it is impossible to achieve coordinated optimization control of ignition timing, fuel mixture ratio, and flame stability.

[0005] With the penetration of intelligent and digital technologies in the power sector, the development of gas turbine ignition systems towards precision and adaptability has become an inevitable trend. In existing technologies, although some systems have introduced sensor arrays and digital control modules, there are still significant defects in key technical links such as multi-source data fusion, energy release timing matching, and iterative correction of combustion models under complex working conditions. For example, the lack of an effective multi-level energy release strategy and an adaptive timing matching mechanism makes it difficult to achieve a stable energy field construction within a wide speed range; the single nature of the combustion optimization algorithm makes it impossible to take into account both combustion efficiency and emission control requirements under different load conditions. In particular, during the transition process of changing operating conditions, combustion state fluctuations and pollutant emission peaks are prone to occur.

[0006] Furthermore, traditional ignition systems suffer from inefficient data exchange between various functional modules (such as monitoring, control, and energy distribution) and a lack of standardized linkage mechanisms. This results in a lag in overall system response and an inability to meet the real-time control requirements of rapid turbine startup and shutdown and sudden load changes. Developing an ignition system that integrates high-precision monitoring, dynamic energy distribution, and intelligent combustion optimization, while achieving a deep integration of the physical combustion process and digital control models, has become a critical challenge in the field of gas turbine technology. Summary of the Invention

[0007] The object of the present invention is to provide a dedicated ignition system for a combustion engine to solve the problems raised in the above background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: a dedicated ignition system for a combustion engine, the system comprising:

[0009] Monitoring and control layer, energy distribution layer and combustion optimization layer;

[0010] The monitoring and control layer includes an ignition energy monitoring device, a phase regulator, a combustion state sensor, and a fuel injection controller, which are used to monitor the pressure fluctuations in the combustion chamber of the combustion engine in real time, analyze the fuel mixing parameters, and perform phase synchronization and injection control operations according to the operating conditions;

[0011] The energy distribution layer includes a high-voltage energy storage module, a pulse generator, an energy coupler, and a dynamic distributor. It also sets a multi-level energy release strategy and an adaptive timing matching mechanism to build a stable ignition energy field, achieve multi-level energy precise matching and directional release, and use the timing matching mechanism to perform gradient release and waveform shaping of energy according to the changes in the engine speed and combustion chamber pressure. By configuring a multi-modal energy interface, the physical combustion space and the digital control space are linked and controlled.

[0012] The combustion optimization layer is used to iteratively correct the ignition energy distribution model using a dynamic combustion algorithm combined with a historical operating condition feature library and a real-time combustion data set, establish a multi-dimensional mapping combustion state control model, and use a collaborative optimization algorithm based on this control model to calibrate the ignition timing, control the fuel mixture ratio, and maintain the flame stability state.

[0013] Preferably, the ignition energy monitoring device includes at least an ionization sensor, an optical flame detector, a pressure fluctuation recorder and a heat flux density meter, which are used to obtain the three-dimensional spatial energy distribution characteristics of the combustion chamber; the phase regulator includes at least a digital timing controller, a waveform shaping module, an energy synchronization unit and a pulse distributor, which are used to realize the phase characteristic reconstruction of the ignition trigger pulse; the fuel injection controller includes at least a proportional control valve, an atomization parameter detector, and a flow dynamic balancing device, which are used to execute the injection control strategy issued by the combustion optimization layer.

[0014] Preferably, in the energy distribution layer, the data collected by the monitoring and control layer is transmitted from the pulse generator to the energy coupler via a synchronous link. The energy coupler then performs energy feature matching and transmits the data together with the dynamic parameters recorded by the combustion state sensor to the dynamic distributor for storage and processing. The energy processing adopts multi-dimensional energy fusion technology to input the collected combustion signals into the parameter space established by different control models for collaborative analysis.

[0015] The method of realizing the linkage regulation of the physical combustion space and the digital control space by configuring a multimodal energy interface includes: configuring the multimodal energy interface to realize the linkage regulation of the physical combustion space and the digital control space, exchanging the real-time status data of the combustion chamber, and jointly analyzing the acquired energy characteristics, while realizing dynamic mapping of control parameters, completing instruction transmission, status feedback and strategy execution, and sending phase adjustment parameters to the monitoring and control layer.

[0016] Preferably, the establishment of a multi-dimensional mapping combustion state control model includes:

[0017] Establish parameter mapping channels and real-time update mechanisms between the physical combustion environment and the digital control space;

[0018] The actual combustion process is analyzed through feature extraction and state matching. Based on the pressure waveform, heat flux distribution, and fuel mixture parameters obtained by the ignition energy monitoring device, a baseline feature library of the combustion environment and an abnormal operating condition template library are constructed. Model parameter correction and operating state simulation are performed based on dynamic monitoring data to transform the actual combustion environment into a high-precision digital control space.

[0019] Calibrate the parameters of the combustion state control model, input the real-time monitored combustion environment data into the established control model, and use the state matching algorithm to dynamically calibrate the analytical results of the model to obtain the optimized combustion state control model;

[0020] The combustion state control model includes a physical combustion space, a digital control space, a feature database, and a linkage mechanism between modules;

[0021] The physical combustion space is the data source of the control model, which contains the original state characteristics of the combustion environment; the digital control space forms a mapping relationship with the physical combustion space, and mathematically represents the combustion state characteristics through multi-dimensional parametric modeling; the feature database integrates historical state data and real-time monitoring information, and provides a benchmark data set including an energy distribution library, a phase feature library, and an injection parameter library; the linkage mechanism realizes data interaction between modules, and the physical combustion space and the feature database realize real-time collection and model update of state parameters through a standardized protocol, the physical combustion space and the digital control space transfer state parameters through a data interface, and the digital control space and the feature database realize information interaction through a data bus.

[0022] Preferably, the ignition timing calibration is performed using a collaborative optimization algorithm based on the control model, including:

[0023] Based on the combustion state control model, historical combustion baseline data, equipment operating parameter records, and abnormal operating condition characteristic data are obtained to construct a state sample set;

[0024] After normalizing the features of the state sample set, it is divided into a training set and a validation set;

[0025] Establish an SVM-GA-BP hybrid model architecture, set the model's initial parameters, input the training set into the hybrid model for collaborative training, perform feature selection using a support vector machine, optimize parameters using a genetic algorithm, perform nonlinear mapping using a neural network, and utilize an adaptive adjustment mechanism to balance the computational errors of different algorithms under specific working conditions until the model convergence speed reaches the set threshold or the scheduled training rounds are completed;

[0026] Input the validation set into the trained hybrid model, calculate the comprehensive performance indicators of the model, and select the optimal time series calibration model;

[0027] The phase characteristics of the ignition trigger pulse are output based on the optimal timing calibration model, and the optimal ignition time coordinates are determined in combination with the combustion chamber pressure fluctuation model.

[0028] Preferably, the fuel mixing ratio control using a collaborative optimization algorithm based on the control model includes:

[0029] Based on the combustion state control model, the dynamic characteristics of fuel flow, atomized particle size distribution, and mixing uniformity parameters are extracted to construct a fuel feature vector set;

[0030] The kernel function mapping method is used to perform nonlinear transformation on the fuel characteristic vector to obtain the key control characteristic components;

[0031] A mixed state classification model based on fuzzy clustering was established, and the optimal number of classifications was determined using the silhouette coefficient method.

[0032] The classified mixed features are input into the preset control strategy library to match the optimal control scheme and generate a targeted set of injection control parameters.

[0033] Preferably, the flame stability state maintenance is performed using a collaborative optimization algorithm based on the control model, including:

[0034] Based on the combustion state control model, the combustion oscillation frequency characteristics, flame propagation speed, and temperature gradient distribution are collected to build a flame stability feature library;

[0035] Perform time series segmentation on the data in the flame stability feature library to generate combustion cycle sample segments;

[0036] Establish an LSTM-CNN hybrid network model, set the number of memory units and convolution kernel size, and obtain the spatiotemporal characteristics of the combustion process through backpropagation calculation;

[0037] The fused features are input into the regression layer for state prediction, and the corresponding relationship between the combustion stability index and the adjustment parameters is output.

[0038] Preferably, the establishment of a multi-dimensional mapping combustion state control model further includes:

[0039] The sliding combustion cycle mechanism is used to segment the continuous monitoring data, and the data segment within each combustion cycle is independently evaluated;

[0040] Establish a correlation matrix between data segment characteristics and equipment load, and record the corresponding combustion state patterns under different load conditions;

[0041] The control model is adapted to the working conditions through the transfer learning algorithm, and the model structure reconstruction mechanism is triggered when an unrecorded operating mode is detected.

[0042] Preferably, the method for generating the injection control strategy includes:

[0043] Establish a mapping relationship table between fuel parameters and control methods, including the control methods of stratified injection corresponding to lean combustion and homogeneous mixing corresponding to rich combustion;

[0044] A simulated annealing algorithm was used to search for the optimal combination of control parameters, including spray pulse width, atomization pressure, and mixing time;

[0045] The control error is corrected in real time through the feedforward compensation mechanism, and the parameter recalculation process is triggered when the actual combustion parameters deviate from the target values.

[0046] Preferably, the training process of the hybrid model architecture includes:

[0047] The Boosting method is used to generate training sequences of multiple base models. The prediction accuracy of each base model is evaluated by the holdout method. The gradient descent algorithm is used to calculate the combined optimization coefficient of each base model. The weighted fusion mechanism is used to make a comprehensive decision on the output results of the base models.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] At the monitoring and control level, the integration of multiple monitoring devices, including ionization sensors, optical flame detectors, and pressure fluctuation recorders, enables real-time acquisition of the three-dimensional energy distribution characteristics of the combustion chamber (including pressure waveforms, heat flux density, and fuel mixture parameters), overcoming the limitations of traditional single-sensor monitoring. The phase regulator, through components such as a digital timing controller and a waveform shaping module, reconstructs the phase characteristics of the ignition trigger pulse. Combined with the dynamic adjustment capabilities of the fuel injection controller, this system precisely executes phase synchronization and injection control strategies based on real-time operating conditions, significantly improving the accuracy of combustion parameter analysis and control response speed.

[0050] One of the core innovations is the multi-level energy release strategy and adaptive timing matching mechanism of the energy distribution layer. The coordinated operation of the high-voltage energy storage module, pulse generator, and dynamic distributor can achieve gradient energy release and waveform shaping according to the changes in the engine speed and combustion chamber pressure through the timing matching mechanism, thereby constructing a stable ignition energy field. The configuration of the multimodal energy interface opens up the linkage channel between the physical combustion space and the digital control space. Through real-time data exchange and joint analysis of energy characteristics, it realizes the dynamic mapping of control parameters and the precise transmission of instructions, solving the problems of energy matching lag and single control dimension in traditional systems. Especially under variable operating conditions, it can significantly reduce energy loss and improve ignition reliability.

[0051] The combustion optimization layer establishes a multidimensional combustion state control model by deeply integrating a dynamic combustion algorithm with a historical operating condition feature library and real-time combustion datasets. This model transforms the actual combustion process into a high-precision digital model through a parameter mapping channel and real-time update mechanism between the physical combustion environment and the digital control space. Dynamic calibration of model parameters and operating condition adaptation are achieved through a state matching algorithm and transfer learning techniques. Collaborative optimization algorithms based on this model (such as the SVM-GA-BP hybrid model, fuzzy clustering algorithm, and LSTM-CNN hybrid network model) enable intelligent control of ignition timing calibration, fuel mixture ratio control, and flame stability maintenance, respectively. For example, in ignition timing calibration, the hybrid model learns from historical data to accurately determine the optimal ignition time coordinates, reducing the risk of combustion oscillation. In fuel mixture control, a fuzzy clustering-based state classification model matches the optimal injection control scheme, achieving precise control in lean / rich combustion modes. For flame stability maintenance, the LSTM-CNN model's ability to extract spatiotemporal features effectively predicts the combustion stability index and outputs adjustment parameters, improving flame stability under variable load conditions.

[0052] Furthermore, the system's innovative design features, including multi-dimensional energy fusion technology, a sliding combustion cycle mechanism, and a feedforward compensation mechanism, further enhance the timeliness of data processing and the robustness of the control strategy. For example, the sliding combustion cycle mechanism enables segmented evaluation of continuous monitoring data, combining it with the equipment load correlation matrix to accurately identify combustion state patterns under different loads. The feedforward compensation mechanism corrects control errors in real time, ensuring the dynamic accuracy of injection parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a working principle diagram of the dedicated ignition system for a combustion engine according to the present invention;

[0054] Figure 2 Workflow diagram of the collaborative optimization algorithm for ignition timing calibration;

[0055] Figure 3 Flowchart of the collaborative optimization algorithm for fuel mixture ratio control;

[0056] Figure 4 Flowchart of the collaborative optimization algorithm for flame stability maintenance. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] See also Figure 1-Figure 4 The present invention relates to a dedicated ignition system for gas turbines. The system comprises a monitoring and control layer, an energy distribution layer, and a combustion optimization layer. Each layer cooperates with each other to achieve precise control of the gas turbine ignition and combustion process. The specific implementation steps are as follows:

[0059] The monitoring and control layer includes an ignition energy monitoring device, a phase regulator, a combustion state sensor, and a fuel injection controller. The ignition energy monitoring device collects data such as combustion chamber pressure fluctuations and fuel mixing parameters in real time. The phase regulator synchronously adjusts the phase of the ignition trigger pulse according to the operating conditions. The combustion state sensor continuously monitors the combustion state and feeds back to the system. The fuel injection controller performs fuel injection control operations according to the instructions of the combustion optimization layer. The energy distribution layer is equipped with a high-voltage energy storage module, a pulse generator, an energy coupler, and a dynamic distributor. Through a multi-level energy release strategy and an adaptive timing matching mechanism, a stable ignition energy field is constructed to achieve precise matching and directional release of multi-level energy. According to the changes in the engine speed and combustion chamber pressure, a timing matching mechanism is used to perform gradient release and waveform shaping of energy. At the same time, a multi-modal energy interface is used to realize the linkage control of the physical combustion space and the digital control space. The combustion optimization layer uses a dynamic combustion algorithm, combined with a historical operating condition feature library and a real-time combustion data set, to iteratively correct the ignition energy distribution model, establish a multi-dimensional mapping combustion state control model, and adopt a collaborative optimization algorithm based on this model to achieve ignition timing calibration, fuel mixture ratio control, and flame stability maintenance.

[0060] The technical solution of the present invention is further described in detail below with reference to specific embodiments.

[0061] Example 1:

[0062] As one of the core control modules of a dedicated gas turbine ignition system, the monitoring and control layer's components collaborate closely to achieve real-time monitoring and precise control of the combustion process. This layer includes an ignition energy monitor, phase regulator, combustion state sensor, and fuel injection controller. These components are both independent and closely linked in hardware structure and functional logic, working together to monitor combustion chamber pressure fluctuations in real time, analyze fuel mixing parameters, and implement phase synchronization and injection control based on operating conditions.

[0063] The ignition energy monitoring device is the fundamental data acquisition unit of the monitoring and control layer. It comprises at least an ionization sensor, an optical flame detector, a pressure fluctuation recorder, and a heat flux density meter. The ionization sensor utilizes a hardware structure based on the principle of gas ionization. By placing a pair of electrodes within the combustion chamber, it uses the high-temperature ionized gas generated during combustion as a conductive medium to detect the intensity and frequency of the ion current in real time. The sensor's signal output is connected to the system's data processing module and continuously outputs an electrical signal indicating the presence, stability, and energy intensity of the flame. The optical flame detector, based on spectral analysis technology, features a built-in multi-band photoelectric sensor array covering the ultraviolet, visible, and infrared spectral ranges. By capturing specific wavelength spectral signals emitted by the flame, it analyzes the flame's temperature distribution, combustion product composition, and energy radiation characteristics. The pressure fluctuation recorder utilizes a highly sensitive piezoresistive sensor. Its measuring probe is embedded directly into the combustion chamber wall. It records the dynamic temporal changes in the combustion chamber's internal pressure at a microsecond sampling rate, effectively capturing pressure fluctuations caused by turbulence, detonation, and other phenomena during combustion. The heat flux density meter utilizes an array of heat flux sensors to measure the combined heat flux intensity of thermal radiation and heat conduction within the combustion chamber, acquiring heat flux distribution data at different spatial locations and providing key parameters for systematic analysis of combustion energy distribution. These four sensors are spatially distributed, forming a three-dimensional monitoring network that captures the three-dimensional energy distribution characteristics of the combustion chamber. The output signals of each sensor undergo anti-interference filtering and analog-to-digital conversion before being transmitted via a parallel data bus to the central processing unit of the monitoring and control layer.

[0064] The phase adjuster is the core functional module for reconstructing the phase characteristics of the ignition trigger pulse. It comprises at least a digital timing controller, a waveform shaping module, an energy synchronization unit, and a pulse distributor. Designed based on a field-programmable gate array (FPGA) architecture, the digital timing controller incorporates a high-precision clock oscillator and logic control unit. It generates a sequence of time-control signals with nanosecond precision based on the system's preset ignition timing logic and real-time collected operating data (such as engine speed and crankshaft phase). The controller interacts with the system's main controller via a serial communication interface, receiving phase adjustment commands from the combustion optimization layer and dynamically adjusting the ignition timing parameters accordingly. The waveform shaping module utilizes a hardware architecture combining a programmable logic device and a power amplifier circuit. Its input is connected to the output of the digital timing controller and modulates the waveform of the original ignition trigger pulse according to a preset waveform template (such as a square wave, spike pulse, or sine wave). By adjusting parameters such as the rising and falling edge slopes, pulse width, and amplitude, it achieves precise control of the ignition energy release pattern. The energy synchronization unit includes a phase-locked loop (PLL) circuit and a synchronous trigger interface. By tracking the rotational phase signal of the engine crankshaft in real time, it ensures that the ignition trigger pulse maintains strict phase synchronization with the engine's operating cycle, avoiding ignition failure or unstable combustion due to phase deviation. The pulse distributor uses a multi-channel power drive circuit to evenly distribute the phase-adjusted and waveform-shaped ignition trigger pulse to the igniters of each cylinder of the engine. It has integrated overload protection and fault diagnosis functions, and can monitor the current and voltage status of each output channel in real time. When an abnormal signal is detected, it automatically disconnects the faulty channel and issues an alarm signal.

[0065] The fuel injection controller (FIC) executes the injection control strategy issued by the combustion optimization layer. It comprises at least a proportional control valve, an atomization parameter detector, and a flow dynamic balancing device. The proportional control valve utilizes electro-pneumatic proportional control technology, and its valve body features a high-precision throttling mechanism. By receiving analog voltage or digital signals (such as 4-20mA current or 0-5V voltage) from the control module, it linearly adjusts the cross-sectional area of the fuel channel, achieving precise control of the fuel injection quantity. The control valve incorporates a built-in position feedback sensor that detects the actual valve opening in real time. A closed-loop control algorithm corrects for control signal errors, ensuring that the injection quantity closely matches the commanded value. The atomization parameter detector utilizes the principles of a laser particle size analyzer. By placing a laser transmitter and receiver downstream of the fuel injection nozzle, it uses Mie scattering theory to measure the particle size distribution and velocity field of the atomized fuel particles in real time. After signal processing, the measured data is converted into characteristic parameters such as the mean atomized particle size and the particle size distribution variance, which are fed back to the control module for use in adjusting injection parameters. The flow dynamic balancing device consists of multiple parallel flow compensation branches and pressure sensors. When the system detects flow deviations between the fuel injection channels, it automatically adjusts the opening of the compensation branches to balance the fluid resistance of each channel, ensuring that the fuel injection amount of different cylinders is uniform and consistent, avoiding incomplete combustion or imbalance between cylinders due to uneven fuel distribution.

[0066] During system operation, the ignition energy monitoring device continuously collects three-dimensional energy distribution data from the combustion chamber, including pressure waveforms, heat flux density, and spectral characteristics, and transmits this data to the central processing unit (CPU) in the monitoring and control layer. After preprocessing the data (such as noise reduction, feature extraction, and normalization), the CPU provides real-time combustion status information via the combustion state sensor module and transmits the data to the energy distribution layer and combustion optimization layer. Based on the operating parameters and phase adjustment instructions output by the CPU, the phase adjuster generates precise ignition timing signals through a digital timing controller. After processing by the waveform shaping module and energy synchronization unit, the signals are output by the pulse distributor to each igniter, reconstructing the phase characteristics of the ignition trigger pulse. The fuel injection controller, based on the injection control strategy issued by the combustion optimization layer, adjusts the fuel injection amount via a proportional control valve, monitors the atomization status in real time using an atomization parameter detector, and ensures stable flow in each channel using a dynamic flow balancing device. The entire monitoring and control layer realizes dynamic monitoring and precise control of the ignition and combustion process of the gas turbine through multi-sensor fusion, high-precision timing control and closed-loop feedback adjustment mechanism, providing basic data support and execution guarantee for the functional realization of the energy distribution layer and combustion optimization layer.

[0067] The specific hardware selection and parameter configuration of each of the above components can be adaptively adjusted according to the type of gas turbine (such as gas turbine, gasoline engine, diesel engine, etc.), power level, and combustion mode (such as premixed combustion, diffusion combustion, etc.). For example, for high-pressure gas turbines, the sensor of the ignition energy monitoring device must have the characteristics of high temperature resistance, corrosion resistance, and high pressure resistance, the phase adjuster must adopt an anti-electromagnetic interference design, and the fuel injection controller must meet the sealing and response speed requirements of high-pressure fuel injection. In addition, the electrical connections between the components use shielded cables and anti-interference connectors, and the data communication protocol uses industrial-grade real-time communication standards (such as EtherCAT, PROFINET, etc.) to ensure the reliability and real-time performance of the system in complex electromagnetic environments. Through the above structural design and functional configuration, the monitoring and control layer can efficiently and stably complete the monitoring and control tasks of the gas turbine combustion process, laying the foundation for the performance optimization of the entire ignition system.

[0068] Example 2:

[0069] The energy distribution layer, the energy management core of the dedicated gas turbine ignition system, achieves stable storage, precise distribution, and intelligent control of ignition energy through the coordinated operation of high-voltage energy storage modules, pulse generators, energy couplers, and dynamic distributors. This layer implements a multi-level energy release strategy and adaptive timing matching mechanism to dynamically adjust energy output modes based on the gas turbine's real-time operating conditions. It also achieves cross-domain linkage between the physical combustion space and the digital control space through a multimodal energy interface, providing efficient and stable ignition energy support for the gas turbine.

[0070] The high-voltage energy storage module is an energy storage unit in the energy distribution layer. Its core is a high-voltage capacitor bank or inductive energy storage device, which can store electrical energy at a specific energy level according to the ignition requirements of the gas turbine. The module adopts a modular design and contains multiple independent energy storage units. Each unit is equipped with a voltage sensor and overvoltage protection circuit to ensure the safety and stability of the energy storage process. The charging process of the energy storage module is controlled by the system main controller through a pulse width modulation (PWM) signal. The charging voltage and charging rate can be dynamically adjusted according to the real-time operating conditions. For example, a fast charging mode is used during the startup phase of the gas turbine to shorten the preparation time, and a constant voltage charging mode is used during the normal operation phase to maintain the energy storage level. The output end of the energy storage module is connected to the pulse generator through a high-voltage insulated cable, which can release the stored energy in a very short time to form a high-intensity ignition pulse.

[0071] The pulse generator is a key component that converts the electrical energy of the energy storage module into an ignition pulse of a specific waveform. It is built based on power electronic switching devices (such as insulated gate bipolar transistors IGBTs, metal-oxide semiconductor field-effect transistors MOSFETs, etc.) and has the ability to quickly turn on and off. The control signal of the pulse generator comes from the phase regulator of the monitoring and control layer, and receives the ignition trigger signal generated by the digital timing controller through a synchronous link. This component can generate ignition pulses of different energy levels and different waveform characteristics (such as single pulses, multi-pulse trains, high-frequency pulse sequences, etc.) by adjusting the conduction time and frequency of the switching device according to the timing and waveform requirements of the input signal. The pulse generator has built-in current sensors and temperature sensors to monitor the working status in real time and feedback to the dynamic distributor. When abnormal temperature rise or overcurrent is detected, the protection mechanism is automatically triggered to cut off the power supply to avoid device damage.

[0072] The energy coupler is the core hub for matching monitoring data with energy characteristics, and its hardware structure includes a signal conditioning circuit and an energy conversion module. The data such as combustion chamber pressure fluctuations and fuel mixing parameters collected by the monitoring and control layer are transmitted by the pulse generator through a synchronous link to the signal conditioning circuit of the energy coupler. The circuit performs pre-processing such as amplification, filtering, and analog-to-digital conversion on the original data to extract key parameters related to energy characteristics (such as pressure peak, frequency component, fuel air-fuel ratio, etc.). The energy conversion module performs feature matching on the energy released by the high-voltage energy storage module based on the pre-processed parameters. For example, by adjusting the pulse width, amplitude or frequency, the waveform characteristics of the output energy are consistent with the optimal ignition energy model under the current combustion conditions. The energy coupler also has a multi-channel input and output interface, which can simultaneously handle the energy matching needs of multiple combustion chambers or multiple ignition points, and realize parallel energy regulation.

[0073] The dynamic allocator is the end-effector of the energy distribution layer, assuming the multiple functions of data storage, energy processing, and instruction execution. It integrates a high-speed data memory, a multi-dimensional energy fusion processing module, and an intelligent allocation algorithm. The energy signature matching data output by the energy coupler and the dynamic parameters recorded by the combustion state sensor (such as flame propagation velocity and temperature gradient) are transmitted to the dynamic allocator's high-speed data memory, forming a real-time combustion state dataset. Multi-dimensional energy fusion technology integrates the collected combustion signals into parameter spaces established by different control models (such as the pressure-energy space and the temperature-time space), performing collaborative analysis. For example, in the pressure-energy space, the system analyzes the coupling relationship between combustion chamber pressure fluctuations and ignition energy, while in the temperature-time space, it establishes a mapping model between combustion temperature distribution and energy release timing. Through this cross-model collaborative analysis, the system can fully understand the real-time energy demand of the combustion state, providing a basis for precise energy allocation.

[0074] The configuration of a multimodal energy interface is a key technical path for achieving coordinated control between the physical combustion space and the digital control space. This interface comprises a hardware communication module and a software protocol stack. At the hardware level, an industrial fieldbus (such as CANopen, ModbusTCP, etc.) or a dedicated communication interface is used to enable bidirectional transmission of real-time combustion chamber status data (such as pressure, temperature, and fuel flow) between the physical and digital spaces. At the software level, a standardized data exchange protocol is established, defining parameters such as data format, transmission frequency, and verification mechanism to ensure the accuracy and real-time nature of data transmission. The coordinated control process specifically involves uploading real-time status data from the physical combustion space to the digital control space via the multimodal energy interface. The digital control space analyzes the data using a combustion state control model, extracting energy characteristics (such as areas of uneven energy distribution and peak energy demand times) and generating control instructions (such as phase adjustment parameters and energy release patterns). These control instructions are then transmitted back to the physical combustion space via the interface, where they are executed by the energy distribution layer and the monitoring and control layer, completing a closed-loop control process encompassing command transmission, status feedback, and strategy execution. For example, when digital control space analysis finds that there is combustion instability in a certain area of the combustion chamber, an energy-directed release instruction is sent to the dynamic distributor through the multimodal energy interface. The dynamic distributor adjusts the output parameters of the energy coupler according to the instruction and accurately delivers additional energy to the area to enhance the ignition effect and stabilize combustion.

[0075] In terms of the implementation of the timing matching mechanism, the energy distribution layer dynamically adjusts the timing and gradient of energy release by monitoring the changes in the engine speed and combustion chamber pressure in real time. Specifically, when the engine speed increases, the system determines that the combustion process is accelerating and needs to shorten the energy release interval and increase the energy release frequency. The pulse generator generates a high-frequency pulse sequence, which cooperates with the waveform shaping function of the energy coupler to achieve rapid gradient release of energy. When the combustion chamber pressure rises abnormally, the system identifies the possible risk of detonation and immediately triggers the energy release suppression strategy through the dynamic distributor to reduce the single release of energy and extend the release interval. At the same time, the phase advance adjustment parameters are sent to the monitoring and control layer through the multimodal energy interface to alleviate pressure fluctuations. This timing matching mechanism achieves microsecond dynamic response through the synergy of the hardware real-time computing unit and the software algorithm, ensuring the precise matching of energy release and engine operating conditions.

[0076] Components at the energy distribution layer are connected to other system levels via standardized interfaces, forming a highly integrated control network. High-voltage energy storage modules, pulse generators, energy couplers, and dynamic distributors are compactly arranged and installed in independent shielded cabinets to reduce the impact of electromagnetic interference on precision electronic components. The cabinets are equipped with a highly efficient heat dissipation system, which uses temperature sensors and intelligent fan speed adjustment algorithms to ensure that component operating temperatures remain within a safe range. Data transmission paths utilize a redundant design, with critical signals (such as ignition trigger signals and energy status signals) transmitted via dual buses, enhancing system reliability and fault tolerance.

[0077] Example 3:

[0078] The combustion optimization layer, the intelligent decision-making core of the gas turbine-specific ignition system, achieves in-depth control of ignition energy distribution and the combustion process by establishing a multi-dimensional combustion state control model and collaborative optimization algorithm. Based on a parameter mapping mechanism between the physical combustion environment and the digital control space, this layer constructs an intelligent decision-making system consisting of a baseline feature library, an abnormal operating condition template library, and a dynamic correction algorithm. It also achieves adaptive optimization for complex operating conditions through sliding combustion cycles and transfer learning techniques.

[0079] The parameter mapping channel between the physical combustion environment and the digital control space is built based on standardized data interfaces and real-time communication protocols. Combustion state parameters in the physical space (such as pressure waveforms, heat flux distribution, and fuel mixture ratio) are collected in real time by a sensor network at the monitoring and control layer. After analog-to-digital conversion and signal conditioning, they are transmitted to the digital control space via industrial-grade Ethernet (such as PROFINET, EtherCAT) or fieldbus (such as CANopen). The core of the digital control space is a high-performance computing platform, utilizing a combination of multi-core processors and dedicated algorithm acceleration chips (such as GPUs and FPGAs). This architecture enables the processing and analysis of massive amounts of combustion data within milliseconds. The parameter mapping process utilizes multi-dimensional feature extraction algorithms to convert the original state parameters in the physical space into feature vectors in the digital space. For example, fast Fourier transforms (FFTs) are used to convert the time-domain pressure waveform into frequency-domain features, wavelet analysis is used to extract local singularity features in the heat flux distribution, and principal component analysis (PCA) is used to reduce the dimensionality of high-dimensional fuel mixture parameters.

[0080] The construction of the baseline feature library and the abnormal operating condition template library is based on the historical data accumulation and real-time data update mechanism. The baseline feature library stores a set of feature vectors under normal combustion conditions, and is classified and indexed according to operating parameters such as engine load, speed, and ambient temperature. Each category contains feature templates for multiple typical operating conditions. The abnormal operating condition template library collects feature patterns of various abnormal combustion states (such as knock, flameout, incomplete combustion, etc.), and automatically labels the abnormal type and severity through pattern recognition algorithms. Both libraries adopt an incremental update strategy. When the system detects new operating condition characteristics or abnormal patterns, the library content is expanded and optimized through machine learning algorithms. For example, when the system encounters a slight knock phenomenon under a specific load for the first time, the feature vector of the operating condition is associated with the knock feature and stored, and the discrimination threshold of the abnormal condition is updated.

[0081] The parameter calibration process of the control model adopts a multi-stage optimization strategy. First, the historical operating condition data is divided into a training set and a validation set. The training set is used to initialize the model parameters, and the validation set is used to evaluate the model performance. The model training adopts an ensemble learning method, combining the advantages of multiple algorithms such as support vector machine (SVM), random forest (RF) and long short-term memory network (LSTM) to construct a hybrid prediction model. For example, SVM is used to handle small sample high-dimensional data classification problems, RF is used to extract nonlinear relationships between features, and LSTM is used to capture the temporal dependencies of the combustion process. During the training process, an adaptive learning rate adjustment strategy and regularization techniques (such as L1 / L2 regularization) are used to prevent overfitting, and the optimal model parameter combination is selected through cross-validation.

[0082] The real-time correction mechanism is implemented based on Kalman filtering and particle filtering algorithms. When the system acquires new combustion state data, it first uses the Kalman filter to predict the current state estimate. It then compares the actual measured value with the predicted value and calculates the residual. If the residual exceeds the preset threshold, the particle filter algorithm is triggered to correct the model parameters. The particle filter generates a set of random samples (particles) to represent the probability distribution of the state space, weights each particle according to the measured value, retains high-weighted particles, and generates a new set of particles, thereby dynamically updating the model parameters. This mechanism can effectively handle nonlinear and non-Gaussian noise interference in the combustion process, improving the robustness of the model and the prediction accuracy.

[0083] The sliding combustion cycle mechanism employs a time window sliding sampling strategy, dividing the continuous combustion process into multiple overlapping time segments. Each time segment serves as an independent combustion cycle, encompassing the complete intake, compression, combustion, and exhaust phases. The length of the time window is dynamically adjusted based on the engine speed to ensure sufficient characteristic information within each cycle. Within each cycle, the system independently performs state assessment and control decisions while preserving temporal correlations between adjacent cycles. This approach captures the dynamics of the combustion process while reducing computational complexity and improving the system's real-time responsiveness.

[0084] The process for establishing an association matrix between data segment characteristics and equipment load is as follows: First, the equipment load is divided into several discrete levels (such as low load, medium load, and high load), each corresponding to a load interval. Next, feature extraction and cluster analysis are performed on the combustion state data segments within each load interval to identify the typical combustion mode under that load. Finally, an association matrix is constructed, with the rows representing different load levels and the columns representing various combustion modes. The value of the matrix element represents the probability of each combustion mode occurring at that load level. The association matrix uses an online update mechanism, continuously adjusting the element values as new operating condition data accumulates to reflect the dynamic changes in combustion characteristics with load.

[0085] The application process of the transfer learning algorithm in the control model working condition adaptation is as follows: when the system detects a new working condition that has not been recorded, it first retrieves the most similar known working condition from the benchmark feature library as the source domain, and takes the new working condition as the target domain. Then, through feature mapping and distribution alignment technology, the knowledge of the source domain (such as model parameters, feature representation, etc.) is transferred to the target domain. Specifically, the maximum mean difference (MMD) algorithm is used to measure the distribution difference between the source domain and the target domain, and the alignment of the feature space is achieved by minimizing the difference. At the same time, using the idea of adversarial training, a discriminator is designed to distinguish between source domain and target domain samples, forcing the generator to learn feature representations that can eliminate domain differences. In the process of model parameter migration, a parameter fine-tuning strategy is adopted to freeze some underlying network parameters and only update the high-level decision layer parameters to retain the common features learned in the source domain and accelerate the convergence of the target domain model.

[0086] When transfer learning cannot effectively handle new working conditions, the system triggers the model structure reconstruction mechanism. This mechanism first performs anomaly detection on the new working condition data to confirm whether it belongs to an undefined working condition type. If so, the model structure search process is initiated, and the neural architecture search (NAS) technology is used to automatically design a model architecture suitable for the new working condition. The NAS algorithm performs heuristic search in a predefined search space, and gradually finds the optimal model structure by evaluating the performance of different architectures on the validation set. A weight sharing strategy is adopted in the search process to significantly reduce computational overhead. After the new model structure is constructed, historical data and new working condition data are mixed for training to generate a unified model that can adapt to both new and old working conditions.

[0087] The multidimensional mapping combustion state control model consists of four core components: a physical combustion space, a digital control space, a feature database, and a linkage mechanism. The physical combustion space serves as the data source, collecting real-time raw state characteristics of the combustion environment through a distributed sensor network, including multidimensional physical quantities such as pressure, temperature, heat flow, and fuel concentration. The digital control space forms a mapping relationship with the physical combustion space. Using multidimensional parametric modeling techniques, the digital control space transforms continuous state changes in the physical space into mathematical representations in the digital space. The feature database integrates historical state data and real-time monitoring information, and includes multiple sub-libraries such as an energy distribution library, a phase feature library, and an injection parameter library, providing a rich benchmark dataset and knowledge support for the model. The linkage mechanism enables data exchange and collaborative operation between the modules: The physical combustion space and the feature database utilize a standardized data acquisition protocol for real-time state parameter acquisition and model updates. A high-speed data interface is used between the physical combustion space and the digital control space for state parameter transmission and control command issuance. A data bus connects the digital control space and the feature database for information query and knowledge retrieval.

[0088] During actual operation, the combustion optimization layer continuously monitors combustion state parameters, dynamically updates the control model, and generates the optimal ignition energy distribution and fuel injection strategy in real time. When operating conditions change, the system first quickly matches similar operating modes using the correlation matrix and invokes the corresponding control strategy. If the match fails, transfer learning or model reconstruction mechanisms are initiated to ensure efficient and stable combustion control across the entire operating range. This layer collaborates closely with the monitoring and control layer and the energy distribution layer through standardized interfaces to form a closed-loop control system, jointly improving the combustion efficiency and operational stability of the gas turbine.

[0089] Example 4:

[0090] The collaborative optimization algorithm in the combustion optimization layer integrates a multi-agent model with timing analysis techniques to achieve precise calibration and dynamic optimization of the ignition timing. This algorithm constructs a multidimensional feature space based on the combustion state control model and integrates the advantages of multiple machine learning algorithms through a hybrid modeling architecture, forming a complete optimization chain from feature extraction to parameter output.

[0091] The historical operating condition data obtained based on the combustion state control model contains multi-dimensional information. The system first pre-processes and features these data. The historical operating condition combustion baseline data records the normal combustion characteristics of the gas turbine under different loads and environmental conditions, including pressure fluctuation curves, flame propagation speed, temperature distribution gradient, etc.; the equipment operating parameter records include real-time operating variables such as speed, intake flow, and fuel supply; the abnormal operating condition characteristic data stores the characteristic patterns of abnormal combustion states such as detonation and flameout. These raw data are aligned by time stamps and constructed into a three-dimensional data set containing time domain, frequency domain, and spatial domain features. The system uses wavelet transform to perform multi-resolution analysis on the pressure fluctuation signal and extract the energy distribution characteristics of different frequency bands; uses principal component analysis (PCA) to reduce the dimensionality of high-dimensional operating parameters to reduce feature redundancy; and uses spatial interpolation algorithms to reconstruct discrete temperature measurement point data into a continuous temperature field distribution.

[0092] The preprocessed dataset was divided into a training set and a validation set, using a stratified sampling strategy to ensure that each operating condition type had a similar distribution ratio in both sets. To eliminate the influence of different feature dimensions, all features were normalized and mapped to the interval [0, 1]. On this basis, a hybrid SVM-GA-BP model architecture was constructed, combining the small sample classification advantages of a support vector machine (SVM), the global optimization capabilities of a genetic algorithm (GA), and the nonlinear mapping properties of a neural network (BP).

[0093] The model training process consists of three stages: feature selection, parameter optimization, and nonlinear mapping. During the feature selection stage, the SVM constructs a hyperplane to divide the high-dimensional feature space into distinct classification regions. The SVM calculates the contribution of each feature to the classification result, selecting a subset of features that significantly influence the ignition timing. This process utilizes a recursive feature elimination (RFE) algorithm to gradually eliminate features with low contributions until the number of retained features reaches a preset threshold. Experiments have shown that feature selection can effectively reduce model input dimensionality, improving training efficiency and generalization capabilities.

[0094] The parameter optimization phase is performed by the Generative Adversarial Network (GA) algorithm, which encodes the BP neural network's weights, thresholds, and other parameters into chromosomes and searches for the optimal parameter combination through genetic operations such as selection, crossover, and mutation. The fitness function is designed to be the inverse of the prediction error on the validation set, and a regularization term is introduced to prevent overfitting. To improve search efficiency, an elite retention strategy is employed to ensure that the best individuals in each evolutionary generation advance directly to the next. During the evolutionary process, the crossover and mutation rates are dynamically adjusted. A higher crossover rate is initially used to promote global search, while a lower crossover rate and a higher mutation rate are later used to enhance local search capabilities.

[0095] During the nonlinear mapping phase, an optimized BP neural network is used to establish the complex relationship between feature parameters and ignition timing. The BP network utilizes a multi-layer perceptron architecture. The number of nodes in the input layer is equal to the dimension of the feature subset. Two hidden layers are set to enhance the model's expressiveness. The output layer contains two parameters: the ignition phase angle and the ignition advance angle. The Sigmoid function is used as the activation function to address nonlinearities. The training algorithm utilizes gradient descent with momentum to accelerate convergence and avoid local minima. During training, the system monitors the validation set error in real time. If the error stops decreasing after multiple training cycles, an early stopping mechanism is triggered to prevent overfitting.

[0096] To balance the computational errors of different algorithms under specific operating conditions, the system introduces an adaptive adjustment mechanism. This mechanism, based on a weighted fusion strategy, dynamically assigns weights to each sub-model based on its performance on the validation set. Specifically, after each training cycle, the prediction errors of the SVM, GA, and BP sub-models on the validation set are calculated. The sub-model with the smaller error receives a larger weight. Weight updates utilize an exponential moving average method to ensure a rapid response to changes in model performance. Furthermore, when a sub-model's error is detected to be continuously increasing, the system automatically triggers a model fine-tuning process to locally optimize the sub-model's parameters.

[0097] The trained hybrid model was evaluated using a validation set, calculating metrics such as mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²). After selecting the optimal timing calibration model, it was deployed in the real-time control system. In practice, the model receives real-time combustion state data from the monitoring and control layer, converts it into a model input format through a feature extraction module, and then outputs the phase characteristic parameters of the ignition trigger pulse through model inference.

[0098] The process for determining the optimal ignition timing coordinates by combining a combustion chamber pressure fluctuation model is as follows: First, a mathematical model of combustion chamber pressure fluctuations is established based on fluid mechanics principles. This model accounts for the influence of multiple factors on pressure variations, such as fuel injection, turbulent mixing, and flame propagation. The model is discretized using the finite element method, dividing the combustion chamber space into multiple tiny cells. The pressure variations in each cell are described by the conservation of mass, momentum, and energy equations. The phase characteristic parameters output by the timing calibration model are then substituted into the pressure fluctuation model as boundary conditions, and the pressure distribution evolution at different ignition times is calculated using a numerical solution method.

[0099] The system defines metrics such as pressure rise rate, peak pressure, and pressure oscillation amplitude to evaluate combustion performance, constructing a multi-objective optimization function. The non-dominated sorting genetic algorithm (NSGA-II) is used to solve the optimization problem, generating a set of Pareto-optimal solutions, each corresponding to a candidate ignition timing. The optimal ignition timing coordinates are determined by analyzing the performance of these candidate solutions under different operating conditions and incorporating engineering experience. These coordinates not only maximize combustion efficiency but also balance combustion stability and emissions performance.

[0100] During actual operation, the system continuously monitors the combustion state and updates the model parameters. When a change in operating conditions is detected, it first determines whether it falls within an existing operating condition category. If so, the corresponding calibration parameters are directly called. If not, the online learning mechanism is activated, using transfer learning technology to quickly adapt to the new operating conditions. At the same time, the system records the actual combustion results after each ignition adjustment, forming a closed-loop feedback loop and providing data support for subsequent model optimization. In this way, the collaborative optimization algorithm can ensure the accuracy and robustness of ignition timing calibration while maintaining computational efficiency, meeting the requirements for efficient and stable operation of the gas turbine under different operating conditions.

[0101] Example 5:

[0102] The combustion optimization layer's fuel mixing and flame stability control mechanism achieves refined control of the combustion process through the collaborative work of multi-dimensional feature extraction, state classification, and prediction models. This mechanism constructs a fuel feature vector space based on the combustion state control model, employs a fuzzy clustering algorithm for mixing state classification, and combines an LSTM-CNN hybrid network for flame stability prediction. Furthermore, it improves control accuracy through feedforward compensation and base model fusion strategies.

[0103] During the fuel feature vector construction process, the system first extracts raw data such as the dynamic characteristics of the fuel flow, atomized particle size distribution, and mixing uniformity parameters from the combustion state control model. These data exhibit complex nonlinear relationships in the time and space dimensions and require effective characterization through feature engineering. The system uses short-time Fourier transform (STFT) to analyze the time-frequency characteristics of the fuel flow, converting the time domain signal into a two-dimensional time-frequency spectrum, and extracting characteristic parameters such as peak frequency and frequency band energy distribution. For the atomized particle size distribution data, statistics such as the average particle size, particle size distribution variance, and Sauter mean diameter (SMD) are calculated to quantify the atomization quality. The mixing uniformity parameter is characterized by calculating the coefficient of variation of the fuel concentration at different spatial locations. A smaller coefficient of variation indicates a more uniform mixing.

[0104] To handle the nonlinear relationship between these characteristic parameters, the system uses the kernel function mapping method to map the low-dimensional feature vector to the high-dimensional Hilbert space. The Gaussian radial basis function (RBF) is selected as the kernel function, and its expression is:

[0105]

[0106] Among them, x i and x j is the input feature vector, and σ is the kernel function width parameter, which controls the complexity of the mapped feature space. Through this mapping, the nonlinear separable problem in the original feature space is transformed into a linear separable problem in a high-dimensional space, facilitating subsequent classification and regression analysis.

[0107] Based on the eigenvectors after kernel mapping, the system establishes a fuzzy clustering model to classify the fuel mixture state. The fuzzy C-means (FCM) algorithm is used, which allows samples to belong to multiple cluster centers with different memberships and is more suitable for handling uncertainties in the combustion process. The algorithm iteratively updates the cluster centers and the membership matrix by minimizing the objective function. The objective function is defined as the sum of the squares of the weighted distances from the sample to each cluster center. To determine the optimal number of classifications, the system calculates the silhouette coefficient (Silhouette Coefficient) under different numbers of clusters. This coefficient comprehensively considers the cohesion and separation of the samples. The larger the value, the better the clustering effect. By traversing the preset range of cluster numbers, the number of clusters when the silhouette coefficient is the largest is selected as the optimal number of classifications.

[0108] The classified mixed features are input into the preset control strategy library for matching. The control strategy library adopts a rule-case hybrid representation method to summarize expert experience into deterministic rules, and stores historical successful cases as a reference. Each strategy entry contains a mixed state characteristic pattern, a corresponding control parameter combination (such as injection pulse width, atomization pressure, mixing time, etc.) and an applicable operating range. The system retrieves the most matching control scheme from the strategy library through similarity calculation. If the similarity is lower than the threshold, the case reasoning mechanism is triggered, and a new control scheme is generated by combining the nearest neighbor algorithm and the genetic algorithm.

[0109] The flame stability feature library is constructed based on multi-sensor fusion data. The system collects parameters such as combustion oscillation frequency characteristics, flame propagation velocity, and temperature gradient distribution. Using time series segmentation techniques, the continuous monitoring data is divided into fixed-length combustion cycle sample segments. Each sample segment contains information about the complete combustion cycle, and a sliding window method is used to generate overlapping sample segments to preserve temporal continuity. To reduce data redundancy, principal component analysis is performed on each sample segment, extracting the principal component that represents the majority of the variability in the original data as the feature vector.

[0110] The LSTM-CNN hybrid network model combines the spatial feature extraction capabilities of convolutional neural networks (CNNs) with the temporal modeling advantages of long-short-term memory networks (LSTMs). The network architecture consists of three modules: a feature extraction module, a temporal analysis module, and a regression prediction module. The feature extraction module employs a multi-layer convolutional structure, using convolution kernels of varying sizes to capture the local spatial patterns of flame stability characteristics. Each convolutional layer is followed by batch normalization and a ReLU activation function to enhance the model's nonlinear expression capabilities. The temporal analysis module, consisting of LSTM layers, processes the temporal feature sequences output by the feature extraction module and captures the long- and short-term dependencies within the combustion process. The regression prediction module uses a fully connected layer to map the LSTM outputs to a combustion stability index space, outputting the corresponding relationship between the combustion stability index and the adjustment parameters.

[0111] During model training, the system uses a backpropagation algorithm to calculate gradients and update network parameters. To prevent overfitting, the Dropout technique randomly ignores some neurons, and an L2 regularization term is added to the loss function. An adaptive learning rate adjustment strategy dynamically adjusts the learning rate based on the validation set error. If the validation set error stops decreasing after multiple training cycles, the learning rate is reduced by an order of magnitude. To evaluate model performance, metrics such as mean squared error (MSE) and mean absolute error (MAE) are used, and cross-validation is used to ensure the model's generalization ability.

[0112] The injection control strategy is generated based on a mapping table of fuel parameters and control methods. This table, constructed through analysis of extensive experimental data and expert experience, contains the correspondence between different fuel characteristics (such as calorific value, viscosity, volatility, etc.) and the optimal control method. For example, for highly volatile fuels, a homogeneous mixed injection mode is preferred, while for high-viscosity fuels, a stratified injection mode is more suitable. The system selects an initial control method from the mapping table based on the current fuel feature vector. It then uses a simulated annealing algorithm to optimize the specific control parameters (such as injection pulse width, atomization pressure, mixing time, etc.).

[0113] The simulated annealing algorithm simulates the thermal equilibrium phenomenon of the solid annealing process, conducting a probabilistic search within the solution space. Starting from a random solution, the algorithm generates a neighborhood solution at each step, deciding whether to accept the solution based on the Metropolis criterion. The acceptance probability is related to the temperature parameter and the change in the objective function. The temperature gradually decreases with the number of iterations, transitioning the algorithm from a global to a local search. During the parameter optimization process, the objective function is defined as the deviation between the combustion stability index and the desired stability threshold. Constrained optimization is performed in conjunction with constraints (such as maximum injection pressure and minimum mixing time).

[0114] The feedforward compensation mechanism, based on the principle of a disturbance observer, is used to correct control errors in real time. The system treats uncertainties in the combustion process (such as ambient temperature changes and fluctuations in fuel quality) as system disturbances and estimates their impact on the combustion state by establishing a disturbance model. When actual combustion parameters deviate from the target values, the disturbance observer calculates the required compensation and adds it to the original control signal to form a corrected control command. To improve compensation accuracy, an adaptive disturbance observer is used to adjust the observer parameters in real time based on the system's dynamic characteristics.

[0115] The hybrid model architecture is trained using a Boosting ensemble learning approach, which iteratively generates multiple base models and then weights them together. In each iteration, the system adjusts the weights of training samples based on the prediction errors of the previous model, giving more attention to samples with incorrect predictions in subsequent training. A holdout approach is used to partition the dataset into training, validation, and test sets. The training set is used for base model training, the validation set is used for model selection and parameter tuning, and the test set is used for final performance evaluation. A gradient descent algorithm is used to calculate the combined optimization coefficients for each base model, optimizing the performance of the ensemble model on the validation set.

[0116] In practical applications, the system first determines the current fuel mixture state using a fuzzy clustering model and retrieves an initial control solution from the control strategy library. Simultaneously, a LSTM-CNN hybrid network predicts the stability index of the current combustion state and generates adjustment recommendations after comparing it with a desired threshold. The injection control strategy generation module combines these two pieces of information, optimizes control parameters using a simulated annealing algorithm, and corrects errors in real time using a feedforward compensation mechanism. This hybrid model architecture performs a weighted fusion of the outputs of multiple base models to improve prediction accuracy and robustness. The entire control process forms a closed-loop feedback loop, continuously adjusting the control strategy based on real-time monitoring data to ensure stable and efficient combustion under all turbine operating conditions.

[0117] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0118] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A special ignition system for a combustion engine, characterized in that: It includes: Monitoring and control layer, energy distribution layer and combustion optimization layer; The monitoring and control layer includes an ignition energy monitoring device, a phase regulator, a combustion state sensor, and a fuel injection controller, which are used to monitor the pressure fluctuations in the combustion chamber of the combustion engine in real time, analyze the fuel mixing parameters, and perform phase synchronization and injection control operations according to the operating conditions; The energy distribution layer includes a high-voltage energy storage module, a pulse generator, an energy coupler, and a dynamic distributor. It also sets a multi-level energy release strategy and an adaptive timing matching mechanism to build a stable ignition energy field, achieve multi-level energy precise matching and directional release, and use the timing matching mechanism to perform gradient release and waveform shaping of energy according to the changes in the engine speed and combustion chamber pressure. By configuring a multi-modal energy interface, the physical combustion space and the digital control space are linked and controlled. The combustion optimization layer is used to iteratively correct the ignition energy distribution model using a dynamic combustion algorithm combined with a historical operating condition feature library and a real-time combustion data set, establish a multi-dimensional mapping combustion state control model, and use a collaborative optimization algorithm based on this control model to calibrate the ignition timing, control the fuel mixture ratio, and maintain the flame stability state.

2. A combustion engine-specific ignition system according to claim 1, characterized in that: The ignition energy monitoring device includes at least an ionization sensor, an optical flame detector, a pressure fluctuation recorder and a heat flux meter, which are used to obtain the three-dimensional spatial energy distribution characteristics of the combustion chamber; the phase regulator includes at least a digital timing controller, a waveform shaping module, an energy synchronization unit and a pulse distributor, which are used to realize the phase characteristic reconstruction of the ignition trigger pulse; the fuel injection controller includes at least a proportional control valve, an atomization parameter detector, and a flow dynamic balancing device, which are used to execute the injection control strategy issued by the combustion optimization layer.

3. A combustion engine-specific ignition system according to claim 1, characterized in that: In the energy distribution layer, the data collected by the monitoring and control layer is transmitted from the pulse generator to the energy coupler via a synchronous link. The energy coupler then performs energy feature matching and transmits the data together with the dynamic parameters recorded by the combustion state sensor to the dynamic distributor for storage and processing. The energy processing adopts multi-dimensional energy fusion technology to input the collected combustion signals into the parameter space established by different control models for collaborative analysis. The method of realizing the linkage regulation of the physical combustion space and the digital control space by configuring a multimodal energy interface includes: configuring the multimodal energy interface to realize the linkage regulation of the physical combustion space and the digital control space, exchanging the real-time status data of the combustion chamber, and jointly analyzing the acquired energy characteristics, while realizing dynamic mapping of control parameters, completing instruction transmission, status feedback and strategy execution, and sending phase adjustment parameters to the monitoring and control layer.

4. A combustion engine-specific ignition system according to claim 1, characterized in that: The multi-dimensional mapping combustion state control model is established, including: Establish parameter mapping channels and real-time update mechanisms between the physical combustion environment and the digital control space; The actual combustion process is analyzed through feature extraction and state matching. Based on the pressure waveform, heat flux distribution, and fuel mixture parameters obtained by the ignition energy monitoring device, a baseline feature library of the combustion environment and an abnormal operating condition template library are constructed. Model parameter correction and operating state simulation are performed based on dynamic monitoring data to transform the actual combustion environment into a high-precision digital control space. Calibrate the parameters of the combustion state control model, input the real-time monitored combustion environment data into the established control model, and use the state matching algorithm to dynamically calibrate the analytical results of the model to obtain the optimized combustion state control model; The combustion state control model includes a physical combustion space, a digital control space, a feature database, and a linkage mechanism between modules; The physical combustion space is the data source of the control model, which contains the original state characteristics of the combustion environment; the digital control space forms a mapping relationship with the physical combustion space, and mathematically represents the combustion state characteristics through multi-dimensional parametric modeling; the feature database integrates historical state data and real-time monitoring information, and provides a benchmark data set including an energy distribution library, a phase feature library, and an injection parameter library; the linkage mechanism realizes data interaction between modules, and the physical combustion space and the feature database realize real-time collection and model update of state parameters through a standardized protocol, the physical combustion space and the digital control space transfer state parameters through a data interface, and the digital control space and the feature database realize information interaction through a data bus.

5. The combustion engine-specific ignition system according to claim 1, characterized in that: The ignition timing calibration is performed using a collaborative optimization algorithm based on the control model, including: Based on the combustion state control model, historical combustion baseline data, equipment operating parameter records, and abnormal operating condition characteristic data are obtained to construct a state sample set; After normalizing the features of the state sample set, it is divided into a training set and a validation set; Establish an SVM-GA-BP hybrid model architecture, set the model's initial parameters, input the training set into the hybrid model for collaborative training, perform feature selection using a support vector machine, optimize parameters using a genetic algorithm, perform nonlinear mapping using a neural network, and utilize an adaptive adjustment mechanism to balance the computational errors of different algorithms under specific working conditions until the model convergence speed reaches the set threshold or the scheduled training rounds are completed; Input the validation set into the trained hybrid model, calculate the comprehensive performance indicators of the model, and select the optimal time series calibration model; The phase characteristics of the ignition trigger pulse are output based on the optimal timing calibration model, and the optimal ignition time coordinates are determined in combination with the combustion chamber pressure fluctuation model.

6. A combustion engine-specific ignition system according to claim 1, characterized in that: The fuel mixing ratio is controlled by using a collaborative optimization algorithm based on the control model, including: Based on the combustion state control model, the dynamic characteristics of fuel flow, atomized particle size distribution, and mixing uniformity parameters are extracted to construct a fuel feature vector set; The kernel function mapping method is used to perform nonlinear transformation on the fuel characteristic vector to obtain the key control characteristic components; A mixed state classification model based on fuzzy clustering was established, and the optimal number of classifications was determined using the silhouette coefficient method. The classified mixed features are input into the preset control strategy library to match the optimal control scheme and generate a targeted set of injection control parameters.

7. The combustion engine-specific ignition system according to claim 1, characterized in that: The method of maintaining the flame stability state by using a collaborative optimization algorithm based on the control model includes: Based on the combustion state control model, the combustion oscillation frequency characteristics, flame propagation speed, and temperature gradient distribution are collected to build a flame stability feature library; Perform time series segmentation on the data in the flame stability feature library to generate combustion cycle sample segments; Establish an LSTM-CNN hybrid network model, set the number of memory units and convolution kernel size, and obtain the spatiotemporal characteristics of the combustion process through backpropagation calculation; The fused features are input into the regression layer for state prediction, and the corresponding relationship between the combustion stability index and the adjustment parameters is output.

8. The combustion engine-specific ignition system according to claim 1, characterized in that: The said establishing a multi-dimensional mapping combustion state control model further includes: The sliding combustion cycle mechanism is used to segment the continuous monitoring data, and the data segment within each combustion cycle is independently evaluated; Establish a correlation matrix between data segment characteristics and equipment load, and record the corresponding combustion state patterns under different load conditions; The control model is adapted to the working conditions through the transfer learning algorithm, and the model structure reconstruction mechanism is triggered when an unrecorded operating mode is detected.

9. The combustion engine-specific ignition system according to claim 1, characterized in that: The method for generating the injection control strategy includes: Establish a mapping relationship table between fuel parameters and control methods, including the control methods of stratified injection corresponding to lean combustion and homogeneous mixing corresponding to rich combustion; A simulated annealing algorithm was used to search for the optimal combination of control parameters, including spray pulse width, atomization pressure, and mixing time; The control error is corrected in real time through the feedforward compensation mechanism, and the parameter recalculation process is triggered when the actual combustion parameters deviate from the target values.

10. The combustion engine-specific ignition system according to claim 1, characterized in that: The training process of the hybrid model architecture includes: The Boosting method is used to generate training sequences of multiple base models. The prediction accuracy of each base model is evaluated by the holdout method. The gradient descent algorithm is used to calculate the combined optimization coefficient of each base model. The weighted fusion mechanism is used to make a comprehensive decision on the output results of the base models.

Images